diff --git a/benchmarks/Benchmark_comms.cc b/benchmarks/Benchmark_comms.cc
index 5a85269e..216172f9 100644
--- a/benchmarks/Benchmark_comms.cc
+++ b/benchmarks/Benchmark_comms.cc
@@ -11,15 +11,15 @@ int main (int argc, char ** argv)
   std::vector<int> simd_layout = GridDefaultSimd(Nd,vComplexD::Nsimd());
   std::vector<int> mpi_layout  = GridDefaultMpi();
   int threads = GridThread::GetThreads();
-  std::cout << "Grid is setup to use "<<threads<<" threads"<<std::endl;
+  std::cout<<GridLogMessage << "Grid is setup to use "<<threads<<" threads"<<std::endl;
 
   int Nloop=10;
   int nmu=0;
   for(int mu=0;mu<4;mu++) if (mpi_layout[mu]>1) nmu++;
-  std::cout << "===================================================================================================="<<std::endl;
-  std::cout << "= Benchmarking concurrent halo exchange in "<<nmu<<" dimensions"<<std::endl;
-  std::cout << "===================================================================================================="<<std::endl;
-  std::cout << "  L  "<<"\t\t"<<" Ls  "<<"\t\t"<<"bytes"<<"\t\t"<<"MB/s uni"<<"\t\t"<<"MB/s bidi"<<std::endl;
+  std::cout<<GridLogMessage << "===================================================================================================="<<std::endl;
+  std::cout<<GridLogMessage << "= Benchmarking concurrent halo exchange in "<<nmu<<" dimensions"<<std::endl;
+  std::cout<<GridLogMessage << "===================================================================================================="<<std::endl;
+  std::cout<<GridLogMessage << "  L  "<<"\t\t"<<" Ls  "<<"\t\t"<<"bytes"<<"\t\t"<<"MB/s uni"<<"\t\t"<<"MB/s bidi"<<std::endl;
 
 
 
@@ -87,15 +87,15 @@ int main (int argc, char ** argv)
 
       double time = stop-start; // microseconds
 
-      std::cout << lat<<"\t\t"<<Ls<<"\t\t"<<bytes<<"\t\t"<<xbytes/time<<"\t\t"<<bidibytes/time<<std::endl;
+      std::cout<<GridLogMessage << lat<<"\t\t"<<Ls<<"\t\t"<<bytes<<"\t\t"<<xbytes/time<<"\t\t"<<bidibytes/time<<std::endl;
     }
   }    
 
 
-  std::cout << "===================================================================================================="<<std::endl;
-  std::cout << "= Benchmarking sequential halo exchange in "<<nmu<<" dimensions"<<std::endl;
-  std::cout << "===================================================================================================="<<std::endl;
-  std::cout << "  L  "<<"\t\t"<<" Ls  "<<"\t\t"<<"bytes"<<"\t\t"<<"MB/s uni"<<"\t\t"<<"MB/s bidi"<<std::endl;
+  std::cout<<GridLogMessage << "===================================================================================================="<<std::endl;
+  std::cout<<GridLogMessage << "= Benchmarking sequential halo exchange in "<<nmu<<" dimensions"<<std::endl;
+  std::cout<<GridLogMessage << "===================================================================================================="<<std::endl;
+  std::cout<<GridLogMessage << "  L  "<<"\t\t"<<" Ls  "<<"\t\t"<<"bytes"<<"\t\t"<<"MB/s uni"<<"\t\t"<<"MB/s bidi"<<std::endl;
 
 
   for(int lat=4;lat<=32;lat+=2){
@@ -163,7 +163,7 @@ int main (int argc, char ** argv)
 
       double time = stop-start;
 
-      std::cout << lat<<"\t\t"<<Ls<<"\t\t"<<bytes<<"\t\t"<<xbytes/time<<"\t\t"<<bidibytes/time<<std::endl;
+      std::cout<<GridLogMessage << lat<<"\t\t"<<Ls<<"\t\t"<<bytes<<"\t\t"<<xbytes/time<<"\t\t"<<bidibytes/time<<std::endl;
     }
   }  
 
diff --git a/benchmarks/Benchmark_dwf.cc b/benchmarks/Benchmark_dwf.cc
index 9ce8f0db..9d9303ae 100644
--- a/benchmarks/Benchmark_dwf.cc
+++ b/benchmarks/Benchmark_dwf.cc
@@ -21,7 +21,7 @@ int main (int argc, char ** argv)
   Grid_init(&argc,&argv);
 
   int threads = GridThread::GetThreads();
-  std::cout << "Grid is setup to use "<<threads<<" threads"<<std::endl;
+  std::cout<<GridLogMessage << "Grid is setup to use "<<threads<<" threads"<<std::endl;
 
   std::vector<int> latt4 = GridDefaultLatt();
   const int Ls=8;
@@ -81,7 +81,7 @@ int main (int argc, char ** argv)
   RealD M5  =1.8;
   DomainWallFermion Dw(Umu,*FGrid,*FrbGrid,*UGrid,*UrbGrid,mass,M5);
   
-  std::cout << "Calling Dw"<<std::endl;
+  std::cout<<GridLogMessage << "Calling Dw"<<std::endl;
   int ncall=10;
   double t0=usecond();
   for(int i=0;i<ncall;i++){
@@ -92,12 +92,12 @@ int main (int argc, char ** argv)
   double volume=Ls;  for(int mu=0;mu<Nd;mu++) volume=volume*latt4[mu];
   double flops=1344*volume*ncall;
   
-  std::cout << "Called Dw"<<std::endl;
-  std::cout << "norm result "<< norm2(result)<<std::endl;
-  std::cout << "norm ref    "<< norm2(ref)<<std::endl;
-  std::cout << "mflop/s =   "<< flops/(t1-t0)<<std::endl;
+  std::cout<<GridLogMessage << "Called Dw"<<std::endl;
+  std::cout<<GridLogMessage << "norm result "<< norm2(result)<<std::endl;
+  std::cout<<GridLogMessage << "norm ref    "<< norm2(ref)<<std::endl;
+  std::cout<<GridLogMessage << "mflop/s =   "<< flops/(t1-t0)<<std::endl;
   err = ref-result; 
-  std::cout << "norm diff   "<< norm2(err)<<std::endl;
+  std::cout<<GridLogMessage << "norm diff   "<< norm2(err)<<std::endl;
 
 
   if (1)
@@ -120,11 +120,11 @@ int main (int argc, char ** argv)
     ref = -0.5*ref;
   }
   Dw.Dhop(src,result,1);
-  std::cout << "Called DwDag"<<std::endl;
-  std::cout << "norm result "<< norm2(result)<<std::endl;
-  std::cout << "norm ref    "<< norm2(ref)<<std::endl;
+  std::cout<<GridLogMessage << "Called DwDag"<<std::endl;
+  std::cout<<GridLogMessage << "norm result "<< norm2(result)<<std::endl;
+  std::cout<<GridLogMessage << "norm ref    "<< norm2(ref)<<std::endl;
   err = ref-result; 
-  std::cout << "norm diff   "<< norm2(err)<<std::endl;
+  std::cout<<GridLogMessage << "norm diff   "<< norm2(err)<<std::endl;
 
   LatticeFermion src_e (FrbGrid);
   LatticeFermion src_o (FrbGrid);
@@ -133,24 +133,32 @@ int main (int argc, char ** argv)
   LatticeFermion r_eo  (FGrid);
 
 
-  std::cout << "Calling Deo and Doe"<<std::endl;
+  std::cout<<GridLogMessage << "Calling Deo and Doe"<<std::endl;
   pickCheckerboard(Even,src_e,src);
   pickCheckerboard(Odd,src_o,src);
 
+  std::cout<<GridLogMessage << "src_e"<<norm2(src_e)<<std::endl;
+  std::cout<<GridLogMessage << "src_o"<<norm2(src_o)<<std::endl;
+
+
   Dw.DhopEO(src_o,r_e,DaggerNo);
   Dw.DhopOE(src_e,r_o,DaggerNo);
-  Dw.Dhop(src,result,DaggerNo);
+  Dw.Dhop  (src  ,result,DaggerNo);
+
+  std::cout<<GridLogMessage << "r_e"<<norm2(r_e)<<std::endl;
+  std::cout<<GridLogMessage << "r_o"<<norm2(r_o)<<std::endl;
+  std::cout<<GridLogMessage << "res"<<norm2(result)<<std::endl;
 
   setCheckerboard(r_eo,r_o);
   setCheckerboard(r_eo,r_e);
 
   err = r_eo-result; 
-  std::cout << "norm diff   "<< norm2(err)<<std::endl;
+  std::cout<<GridLogMessage << "norm diff   "<< norm2(err)<<std::endl;
 
   pickCheckerboard(Even,src_e,err);
   pickCheckerboard(Odd,src_o,err);
-  std::cout << "norm diff even  "<< norm2(src_e)<<std::endl;
-  std::cout << "norm diff odd   "<< norm2(src_o)<<std::endl;
+  std::cout<<GridLogMessage << "norm diff even  "<< norm2(src_e)<<std::endl;
+  std::cout<<GridLogMessage << "norm diff odd   "<< norm2(src_o)<<std::endl;
 
   Grid_finalize();
 }
diff --git a/benchmarks/Benchmark_memory_bandwidth.cc b/benchmarks/Benchmark_memory_bandwidth.cc
index 5b436001..16dad525 100644
--- a/benchmarks/Benchmark_memory_bandwidth.cc
+++ b/benchmarks/Benchmark_memory_bandwidth.cc
@@ -17,13 +17,13 @@ int main (int argc, char ** argv)
   std::vector<int> mpi_layout  = GridDefaultMpi();
 
   int threads = GridThread::GetThreads();
-  std::cout << "Grid is setup to use "<<threads<<" threads"<<std::endl;
+  std::cout<<GridLogMessage << "Grid is setup to use "<<threads<<" threads"<<std::endl;
   
-  std::cout << "===================================================================================================="<<std::endl;
-  std::cout << "= Benchmarking fused AXPY bandwidth ; sizeof(Real) "<<sizeof(Real)<<std::endl;
-  std::cout << "===================================================================================================="<<std::endl;
-  std::cout << "  L  "<<"\t\t"<<"bytes"<<"\t\t\t"<<"GB/s"<<"\t\t"<<"Gflop/s"<<std::endl;
-  std::cout << "----------------------------------------------------------"<<std::endl;
+  std::cout<<GridLogMessage << "===================================================================================================="<<std::endl;
+  std::cout<<GridLogMessage << "= Benchmarking fused AXPY bandwidth ; sizeof(Real) "<<sizeof(Real)<<std::endl;
+  std::cout<<GridLogMessage << "===================================================================================================="<<std::endl;
+  std::cout<<GridLogMessage << "  L  "<<"\t\t"<<"bytes"<<"\t\t\t"<<"GB/s"<<"\t\t"<<"Gflop/s"<<std::endl;
+  std::cout<<GridLogMessage << "----------------------------------------------------------"<<std::endl;
 
   for(int lat=4;lat<=32;lat+=4){
 
@@ -49,15 +49,15 @@ int main (int argc, char ** argv)
       
       double flops=vol*Nvec*2;// mul,add
       double bytes=3*vol*Nvec*sizeof(Real);
-      std::cout<<std::setprecision(3) << lat<<"\t\t"<<bytes<<"   \t\t"<<bytes/time<<"\t\t"<<flops/time<<std::endl;
+      std::cout<<GridLogMessage<<std::setprecision(3) << lat<<"\t\t"<<bytes<<"   \t\t"<<bytes/time<<"\t\t"<<flops/time<<std::endl;
 
     }
 
-  std::cout << "===================================================================================================="<<std::endl;
-  std::cout << "= Benchmarking a*x + y bandwidth"<<std::endl;
-  std::cout << "===================================================================================================="<<std::endl;
-  std::cout << "  L  "<<"\t\t"<<"bytes"<<"\t\t\t"<<"GB/s"<<"\t\t"<<"Gflop/s"<<std::endl;
-  std::cout << "----------------------------------------------------------"<<std::endl;
+  std::cout<<GridLogMessage << "===================================================================================================="<<std::endl;
+  std::cout<<GridLogMessage << "= Benchmarking a*x + y bandwidth"<<std::endl;
+  std::cout<<GridLogMessage << "===================================================================================================="<<std::endl;
+  std::cout<<GridLogMessage << "  L  "<<"\t\t"<<"bytes"<<"\t\t\t"<<"GB/s"<<"\t\t"<<"Gflop/s"<<std::endl;
+  std::cout<<GridLogMessage << "----------------------------------------------------------"<<std::endl;
   
   for(int lat=4;lat<=32;lat+=4){
 
@@ -81,14 +81,14 @@ int main (int argc, char ** argv)
      
       double flops=vol*Nvec*2;// mul,add
       double bytes=3*vol*Nvec*sizeof(Real);
-      std::cout<<std::setprecision(3) << lat<<"\t\t"<<bytes<<"   \t\t"<<bytes/time<<"\t\t"<<flops/time<<std::endl;
+      std::cout<<GridLogMessage<<std::setprecision(3) << lat<<"\t\t"<<bytes<<"   \t\t"<<bytes/time<<"\t\t"<<flops/time<<std::endl;
 
     }
 
-  std::cout << "===================================================================================================="<<std::endl;
-  std::cout << "= Benchmarking SCALE bandwidth"<<std::endl;
-  std::cout << "===================================================================================================="<<std::endl;
-  std::cout << "  L  "<<"\t\t"<<"bytes"<<"\t\t\t"<<"GB/s"<<"\t\t"<<"Gflop/s"<<std::endl;
+  std::cout<<GridLogMessage << "===================================================================================================="<<std::endl;
+  std::cout<<GridLogMessage << "= Benchmarking SCALE bandwidth"<<std::endl;
+  std::cout<<GridLogMessage << "===================================================================================================="<<std::endl;
+  std::cout<<GridLogMessage << "  L  "<<"\t\t"<<"bytes"<<"\t\t\t"<<"GB/s"<<"\t\t"<<"Gflop/s"<<std::endl;
 
   for(int lat=4;lat<=32;lat+=4){
 
@@ -114,15 +114,15 @@ int main (int argc, char ** argv)
       
       double bytes=2*vol*Nvec*sizeof(Real);
       double flops=vol*Nvec*1;// mul
-      std::cout <<std::setprecision(3) << lat<<"\t\t"<<bytes<<"   \t\t"<<bytes/time<<"\t\t"<<flops/time<<std::endl;
+      std::cout<<GridLogMessage <<std::setprecision(3) << lat<<"\t\t"<<bytes<<"   \t\t"<<bytes/time<<"\t\t"<<flops/time<<std::endl;
 
   }
 
-  std::cout << "===================================================================================================="<<std::endl;
-  std::cout << "= Benchmarking READ bandwidth"<<std::endl;
-  std::cout << "===================================================================================================="<<std::endl;
-  std::cout << "  L  "<<"\t\t"<<"bytes"<<"\t\t\t"<<"GB/s"<<"\t\t"<<"Gflop/s"<<std::endl;
-  std::cout << "----------------------------------------------------------"<<std::endl;
+  std::cout<<GridLogMessage << "===================================================================================================="<<std::endl;
+  std::cout<<GridLogMessage << "= Benchmarking READ bandwidth"<<std::endl;
+  std::cout<<GridLogMessage << "===================================================================================================="<<std::endl;
+  std::cout<<GridLogMessage << "  L  "<<"\t\t"<<"bytes"<<"\t\t\t"<<"GB/s"<<"\t\t"<<"Gflop/s"<<std::endl;
+  std::cout<<GridLogMessage << "----------------------------------------------------------"<<std::endl;
 
   for(int lat=4;lat<=32;lat+=4){
 
@@ -147,7 +147,7 @@ int main (int argc, char ** argv)
       
       double bytes=vol*Nvec*sizeof(Real);
       double flops=vol*Nvec*2;// mul,add
-      std::cout<<std::setprecision(3) << lat<<"\t\t"<<bytes<<"  \t\t"<<bytes/time<<"\t\t"<<flops/time<<std::endl;
+      std::cout<<GridLogMessage<<std::setprecision(3) << lat<<"\t\t"<<bytes<<"  \t\t"<<bytes/time<<"\t\t"<<flops/time<<std::endl;
 
   }    
 
diff --git a/benchmarks/Benchmark_su3.cc b/benchmarks/Benchmark_su3.cc
index 903ea348..3312e3c8 100644
--- a/benchmarks/Benchmark_su3.cc
+++ b/benchmarks/Benchmark_su3.cc
@@ -14,13 +14,13 @@ int main (int argc, char ** argv)
   std::vector<int> mpi_layout  = GridDefaultMpi();
 
   int threads = GridThread::GetThreads();
-  std::cout << "Grid is setup to use "<<threads<<" threads"<<std::endl;
+  std::cout<<GridLogMessage << "Grid is setup to use "<<threads<<" threads"<<std::endl;
 
-  std::cout << "===================================================================================================="<<std::endl;
-  std::cout << "= Benchmarking SU3xSU3  x= x*y"<<std::endl;
-  std::cout << "===================================================================================================="<<std::endl;
-  std::cout << "  L  "<<"\t\t"<<"bytes"<<"\t\t\t"<<"GB/s\t\t GFlop/s"<<std::endl;
-  std::cout << "----------------------------------------------------------"<<std::endl;
+  std::cout<<GridLogMessage << "===================================================================================================="<<std::endl;
+  std::cout<<GridLogMessage << "= Benchmarking SU3xSU3  x= x*y"<<std::endl;
+  std::cout<<GridLogMessage << "===================================================================================================="<<std::endl;
+  std::cout<<GridLogMessage << "  L  "<<"\t\t"<<"bytes"<<"\t\t\t"<<"GB/s\t\t GFlop/s"<<std::endl;
+  std::cout<<GridLogMessage << "----------------------------------------------------------"<<std::endl;
 
   for(int lat=2;lat<=24;lat+=2){
 
@@ -43,16 +43,16 @@ int main (int argc, char ** argv)
       double bytes=3.0*vol*Nc*Nc*sizeof(Complex);
       double footprint=2.0*vol*Nc*Nc*sizeof(Complex);
       double flops=Nc*Nc*(6.0+8.0+8.0)*vol;
-      std::cout<<std::setprecision(3) << lat<<"\t\t"<<footprint<<"    \t\t"<<bytes/time<<"\t\t" << flops/time<<std::endl;
+      std::cout<<GridLogMessage<<std::setprecision(3) << lat<<"\t\t"<<footprint<<"    \t\t"<<bytes/time<<"\t\t" << flops/time<<std::endl;
 
     }
 
 
-  std::cout << "===================================================================================================="<<std::endl;
-  std::cout << "= Benchmarking SU3xSU3  z= x*y"<<std::endl;
-  std::cout << "===================================================================================================="<<std::endl;
-  std::cout << "  L  "<<"\t\t"<<"bytes"<<"\t\t\t"<<"GB/s\t\t GFlop/s"<<std::endl;
-  std::cout << "----------------------------------------------------------"<<std::endl;
+  std::cout<<GridLogMessage << "===================================================================================================="<<std::endl;
+  std::cout<<GridLogMessage << "= Benchmarking SU3xSU3  z= x*y"<<std::endl;
+  std::cout<<GridLogMessage << "===================================================================================================="<<std::endl;
+  std::cout<<GridLogMessage << "  L  "<<"\t\t"<<"bytes"<<"\t\t\t"<<"GB/s\t\t GFlop/s"<<std::endl;
+  std::cout<<GridLogMessage << "----------------------------------------------------------"<<std::endl;
 
   for(int lat=2;lat<=24;lat+=2){
 
@@ -75,15 +75,15 @@ int main (int argc, char ** argv)
       
       double bytes=3*vol*Nc*Nc*sizeof(Complex);
       double flops=Nc*Nc*(6+8+8)*vol;
-      std::cout<<std::setprecision(3) << lat<<"\t\t"<<bytes<<"    \t\t"<<bytes/time<<"\t\t" << flops/time<<std::endl;
+      std::cout<<GridLogMessage<<std::setprecision(3) << lat<<"\t\t"<<bytes<<"    \t\t"<<bytes/time<<"\t\t" << flops/time<<std::endl;
 
     }
 
-  std::cout << "===================================================================================================="<<std::endl;
-  std::cout << "= Benchmarking SU3xSU3  mult(z,x,y)"<<std::endl;
-  std::cout << "===================================================================================================="<<std::endl;
-  std::cout << "  L  "<<"\t\t"<<"bytes"<<"\t\t\t"<<"GB/s\t\t GFlop/s"<<std::endl;
-  std::cout << "----------------------------------------------------------"<<std::endl;
+  std::cout<<GridLogMessage << "===================================================================================================="<<std::endl;
+  std::cout<<GridLogMessage << "= Benchmarking SU3xSU3  mult(z,x,y)"<<std::endl;
+  std::cout<<GridLogMessage << "===================================================================================================="<<std::endl;
+  std::cout<<GridLogMessage << "  L  "<<"\t\t"<<"bytes"<<"\t\t\t"<<"GB/s\t\t GFlop/s"<<std::endl;
+  std::cout<<GridLogMessage << "----------------------------------------------------------"<<std::endl;
 
   for(int lat=2;lat<=24;lat+=2){
 
@@ -106,15 +106,15 @@ int main (int argc, char ** argv)
       
       double bytes=3*vol*Nc*Nc*sizeof(Complex);
       double flops=Nc*Nc*(6+8+8)*vol;
-      std::cout<<std::setprecision(3) << lat<<"\t\t"<<bytes<<"    \t\t"<<bytes/time<<"\t\t" << flops/time<<std::endl;
+      std::cout<<GridLogMessage<<std::setprecision(3) << lat<<"\t\t"<<bytes<<"    \t\t"<<bytes/time<<"\t\t" << flops/time<<std::endl;
 
     }
 
-  std::cout << "===================================================================================================="<<std::endl;
-  std::cout << "= Benchmarking SU3xSU3  mac(z,x,y)"<<std::endl;
-  std::cout << "===================================================================================================="<<std::endl;
-  std::cout << "  L  "<<"\t\t"<<"bytes"<<"\t\t\t"<<"GB/s\t\t GFlop/s"<<std::endl;
-  std::cout << "----------------------------------------------------------"<<std::endl;
+  std::cout<<GridLogMessage << "===================================================================================================="<<std::endl;
+  std::cout<<GridLogMessage << "= Benchmarking SU3xSU3  mac(z,x,y)"<<std::endl;
+  std::cout<<GridLogMessage << "===================================================================================================="<<std::endl;
+  std::cout<<GridLogMessage << "  L  "<<"\t\t"<<"bytes"<<"\t\t\t"<<"GB/s\t\t GFlop/s"<<std::endl;
+  std::cout<<GridLogMessage << "----------------------------------------------------------"<<std::endl;
 
   for(int lat=2;lat<=24;lat+=2){
 
@@ -137,7 +137,7 @@ int main (int argc, char ** argv)
       
       double bytes=3*vol*Nc*Nc*sizeof(Complex);
       double flops=Nc*Nc*(8+8+8)*vol;
-      std::cout<<std::setprecision(3) << lat<<"\t\t"<<bytes<<"   \t\t"<<bytes/time<<"\t\t" << flops/time<<std::endl;
+      std::cout<<GridLogMessage<<std::setprecision(3) << lat<<"\t\t"<<bytes<<"   \t\t"<<bytes/time<<"\t\t" << flops/time<<std::endl;
 
     }
 
diff --git a/benchmarks/Benchmark_wilson.cc b/benchmarks/Benchmark_wilson.cc
index aa34ccf4..7d23dec3 100644
--- a/benchmarks/Benchmark_wilson.cc
+++ b/benchmarks/Benchmark_wilson.cc
@@ -28,10 +28,10 @@ int main (int argc, char ** argv)
   GridRedBlackCartesian     RBGrid(latt_size,simd_layout,mpi_layout);
 
   int threads = GridThread::GetThreads();
-  std::cout << "Grid is setup to use "<<threads<<" threads"<<std::endl;
-  std::cout << "Grid floating point word size is REALF"<< sizeof(RealF)<<std::endl;
-  std::cout << "Grid floating point word size is REALD"<< sizeof(RealD)<<std::endl;
-  std::cout << "Grid floating point word size is REAL"<< sizeof(Real)<<std::endl;
+  std::cout<<GridLogMessage << "Grid is setup to use "<<threads<<" threads"<<std::endl;
+  std::cout<<GridLogMessage << "Grid floating point word size is REALF"<< sizeof(RealF)<<std::endl;
+  std::cout<<GridLogMessage << "Grid floating point word size is REALD"<< sizeof(RealD)<<std::endl;
+  std::cout<<GridLogMessage << "Grid floating point word size is REAL"<< sizeof(Real)<<std::endl;
 
   std::vector<int> seeds({1,2,3,4});
   GridParallelRNG          pRNG(&Grid);
@@ -58,8 +58,8 @@ int main (int argc, char ** argv)
   for(int nn=0;nn<Nd;nn++){
     random(pRNG,U[nn]);
     if(0) {
-      if (nn==-1) { U[nn]=zero; std::cout << "zeroing gauge field in dir "<<nn<<std::endl; }
-      else       { U[nn] = cone;std::cout << "unit gauge field in dir "<<nn<<std::endl; }
+      if (nn==-1) { U[nn]=zero; std::cout<<GridLogMessage << "zeroing gauge field in dir "<<nn<<std::endl; }
+      else       { U[nn] = cone;std::cout<<GridLogMessage << "unit gauge field in dir "<<nn<<std::endl; }
     }
     pokeIndex<LorentzIndex>(Umu,U[nn],nn);
   }
@@ -89,7 +89,7 @@ int main (int argc, char ** argv)
   RealD mass=0.1;
   WilsonFermion Dw(Umu,Grid,RBGrid,mass);
   
-  std::cout << "Calling Dw"<<std::endl;
+  std::cout<<GridLogMessage << "Calling Dw"<<std::endl;
   int ncall=10000;
   double t0=usecond();
   for(int i=0;i<ncall;i++){
@@ -98,12 +98,12 @@ int main (int argc, char ** argv)
   double t1=usecond();
   double flops=1344*volume*ncall;
   
-  std::cout << "Called Dw"<<std::endl;
-  std::cout << "norm result "<< norm2(result)<<std::endl;
-  std::cout << "norm ref    "<< norm2(ref)<<std::endl;
-  std::cout << "mflop/s =   "<< flops/(t1-t0)<<std::endl;
+  std::cout<<GridLogMessage << "Called Dw"<<std::endl;
+  std::cout<<GridLogMessage << "norm result "<< norm2(result)<<std::endl;
+  std::cout<<GridLogMessage << "norm ref    "<< norm2(ref)<<std::endl;
+  std::cout<<GridLogMessage << "mflop/s =   "<< flops/(t1-t0)<<std::endl;
   err = ref-result; 
-  std::cout << "norm diff   "<< norm2(err)<<std::endl;
+  std::cout<<GridLogMessage << "norm diff   "<< norm2(err)<<std::endl;
 
 
   //  for(int ss=0;ss<10;ss++ ){
@@ -112,7 +112,7 @@ int main (int argc, char ** argv)
       for(int j=0;j<Nc;j++){
 	ComplexF * ref_p = (ComplexF *)&ref._odata[ss]()(i)(j);
 	ComplexF * res_p = (ComplexF *)&result._odata[ss]()(i)(j);
-	std::cout << ss<< " "<<i<<" "<<j<<" "<< (*ref_p)<<" " <<(*res_p)<<std::endl;
+	std::cout<<GridLogMessage << ss<< " "<<i<<" "<<j<<" "<< (*ref_p)<<" " <<(*res_p)<<std::endl;
       }
     }
   }
@@ -136,11 +136,11 @@ int main (int argc, char ** argv)
   }
   ref = -0.5*ref;
   Dw.Dhop(src,result,1);
-  std::cout << "Called DwDag"<<std::endl;
-  std::cout << "norm result "<< norm2(result)<<std::endl;
-  std::cout << "norm ref    "<< norm2(ref)<<std::endl;
+  std::cout<<GridLogMessage << "Called DwDag"<<std::endl;
+  std::cout<<GridLogMessage << "norm result "<< norm2(result)<<std::endl;
+  std::cout<<GridLogMessage << "norm ref    "<< norm2(ref)<<std::endl;
   err = ref-result; 
-  std::cout << "norm diff   "<< norm2(err)<<std::endl;
+  std::cout<<GridLogMessage << "norm diff   "<< norm2(err)<<std::endl;
 
   Grid_finalize();
 }
diff --git a/configure b/configure
index 64770a5c..e02b9e7e 100755
--- a/configure
+++ b/configure
@@ -626,13 +626,6 @@ ac_subst_vars='am__EXEEXT_FALSE
 am__EXEEXT_TRUE
 LTLIBOBJS
 LIBOBJS
-enable_latex_docs
-enable_html_docs
-enable_dot
-DOXYGEN_DOC_FALSE
-DOXYGEN_DOC_TRUE
-DOT
-DOXYGEN
 BUILD_COMMS_NONE_FALSE
 BUILD_COMMS_NONE_TRUE
 BUILD_COMMS_MPI_FALSE
@@ -647,7 +640,6 @@ CCDEPMODE
 ac_ct_CC
 CFLAGS
 CC
-HAVE_CXX11
 RANLIB
 OPENMP_CXXFLAGS
 am__fastdepCXX_FALSE
@@ -753,10 +745,6 @@ enable_openmp
 enable_simd
 enable_precision
 enable_comms
-enable_doxygen
-enable_dot
-enable_html_docs
-enable_latex_docs
 '
       ac_precious_vars='build_alias
 host_alias
@@ -1402,11 +1390,6 @@ Optional Features:
   --enable-precision=single|double
                           Select default word size of Real
   --enable-comms=none|mpi Select communications
-  --enable-doxygen        enable documentation generation with doxygen (auto)
-  --enable-dot            use 'dot' to generate graphs in doxygen (auto)
-  --enable-html-docs      enable HTML generation with doxygen (yes)
-  --enable-latex-docs     enable LaTeX documentation generation with doxygen
-                          (no)
 
 Some influential environment variables:
   CXX         C++ compiler command
@@ -3157,7 +3140,7 @@ END
 fi
 
 
-ac_config_headers="$ac_config_headers lib/GridConfig.h"
+ac_config_headers="$ac_config_headers lib/Config.h"
 
 # Check whether --enable-silent-rules was given.
 if test "${enable_silent_rules+set}" = set; then :
@@ -4076,190 +4059,7 @@ else
   RANLIB="$ac_cv_prog_RANLIB"
 fi
 
-    ax_cxx_compile_cxx11_required=true
-  ac_ext=cpp
-ac_cpp='$CXXCPP $CPPFLAGS'
-ac_compile='$CXX -c $CXXFLAGS $CPPFLAGS conftest.$ac_ext >&5'
-ac_link='$CXX -o conftest$ac_exeext $CXXFLAGS $CPPFLAGS $LDFLAGS conftest.$ac_ext $LIBS >&5'
-ac_compiler_gnu=$ac_cv_cxx_compiler_gnu
-  ac_success=no
-  { $as_echo "$as_me:${as_lineno-$LINENO}: checking whether $CXX supports C++11 features by default" >&5
-$as_echo_n "checking whether $CXX supports C++11 features by default... " >&6; }
-if ${ax_cv_cxx_compile_cxx11+:} false; then :
-  $as_echo_n "(cached) " >&6
-else
-  cat confdefs.h - <<_ACEOF >conftest.$ac_ext
-/* end confdefs.h.  */
-
-  template <typename T>
-    struct check
-    {
-      static_assert(sizeof(int) <= sizeof(T), "not big enough");
-    };
-
-    struct Base {
-    virtual void f() {}
-    };
-    struct Child : public Base {
-    virtual void f() override {}
-    };
-
-    typedef check<check<bool>> right_angle_brackets;
-
-    int a;
-    decltype(a) b;
-
-    typedef check<int> check_type;
-    check_type c;
-    check_type&& cr = static_cast<check_type&&>(c);
-
-    auto d = a;
-    auto l = [](){};
-    // Prevent Clang error: unused variable 'l' [-Werror,-Wunused-variable]
-    struct use_l { use_l() { l(); } };
-
-    // http://stackoverflow.com/questions/13728184/template-aliases-and-sfinae
-    // Clang 3.1 fails with headers of libstd++ 4.8.3 when using std::function because of this
-    namespace test_template_alias_sfinae {
-        struct foo {};
-
-        template<typename T>
-        using member = typename T::member_type;
-
-        template<typename T>
-        void func(...) {}
-
-        template<typename T>
-        void func(member<T>*) {}
-
-        void test();
-
-        void test() {
-            func<foo>(0);
-        }
-    }
-
-_ACEOF
-if ac_fn_cxx_try_compile "$LINENO"; then :
-  ax_cv_cxx_compile_cxx11=yes
-else
-  ax_cv_cxx_compile_cxx11=no
-fi
-rm -f core conftest.err conftest.$ac_objext conftest.$ac_ext
-fi
-{ $as_echo "$as_me:${as_lineno-$LINENO}: result: $ax_cv_cxx_compile_cxx11" >&5
-$as_echo "$ax_cv_cxx_compile_cxx11" >&6; }
-  if test x$ax_cv_cxx_compile_cxx11 = xyes; then
-    ac_success=yes
-  fi
-
-
-
-    if test x$ac_success = xno; then
-            for switch in -std=c++11 -std=c++0x +std=c++11; do
-      cachevar=`$as_echo "ax_cv_cxx_compile_cxx11_$switch" | $as_tr_sh`
-      { $as_echo "$as_me:${as_lineno-$LINENO}: checking whether $CXX supports C++11 features with $switch" >&5
-$as_echo_n "checking whether $CXX supports C++11 features with $switch... " >&6; }
-if eval \${$cachevar+:} false; then :
-  $as_echo_n "(cached) " >&6
-else
-  ac_save_CXXFLAGS="$CXXFLAGS"
-         CXXFLAGS="$CXXFLAGS $switch"
-         cat confdefs.h - <<_ACEOF >conftest.$ac_ext
-/* end confdefs.h.  */
-
-  template <typename T>
-    struct check
-    {
-      static_assert(sizeof(int) <= sizeof(T), "not big enough");
-    };
-
-    struct Base {
-    virtual void f() {}
-    };
-    struct Child : public Base {
-    virtual void f() override {}
-    };
-
-    typedef check<check<bool>> right_angle_brackets;
-
-    int a;
-    decltype(a) b;
-
-    typedef check<int> check_type;
-    check_type c;
-    check_type&& cr = static_cast<check_type&&>(c);
-
-    auto d = a;
-    auto l = [](){};
-    // Prevent Clang error: unused variable 'l' [-Werror,-Wunused-variable]
-    struct use_l { use_l() { l(); } };
-
-    // http://stackoverflow.com/questions/13728184/template-aliases-and-sfinae
-    // Clang 3.1 fails with headers of libstd++ 4.8.3 when using std::function because of this
-    namespace test_template_alias_sfinae {
-        struct foo {};
-
-        template<typename T>
-        using member = typename T::member_type;
-
-        template<typename T>
-        void func(...) {}
-
-        template<typename T>
-        void func(member<T>*) {}
-
-        void test();
-
-        void test() {
-            func<foo>(0);
-        }
-    }
-
-_ACEOF
-if ac_fn_cxx_try_compile "$LINENO"; then :
-  eval $cachevar=yes
-else
-  eval $cachevar=no
-fi
-rm -f core conftest.err conftest.$ac_objext conftest.$ac_ext
-         CXXFLAGS="$ac_save_CXXFLAGS"
-fi
-eval ac_res=\$$cachevar
-	       { $as_echo "$as_me:${as_lineno-$LINENO}: result: $ac_res" >&5
-$as_echo "$ac_res" >&6; }
-      if eval test x\$$cachevar = xyes; then
-        CXXFLAGS="$CXXFLAGS $switch"
-        ac_success=yes
-        break
-      fi
-    done
-  fi
-  ac_ext=cpp
-ac_cpp='$CXXCPP $CPPFLAGS'
-ac_compile='$CXX -c $CXXFLAGS $CPPFLAGS conftest.$ac_ext >&5'
-ac_link='$CXX -o conftest$ac_exeext $CXXFLAGS $CPPFLAGS $LDFLAGS conftest.$ac_ext $LIBS >&5'
-ac_compiler_gnu=$ac_cv_cxx_compiler_gnu
-
-  if test x$ax_cxx_compile_cxx11_required = xtrue; then
-    if test x$ac_success = xno; then
-      as_fn_error $? "*** A compiler with support for C++11 language features is required." "$LINENO" 5
-    fi
-  else
-    if test x$ac_success = xno; then
-      HAVE_CXX11=0
-      { $as_echo "$as_me:${as_lineno-$LINENO}: No compiler with C++11 support was found" >&5
-$as_echo "$as_me: No compiler with C++11 support was found" >&6;}
-    else
-      HAVE_CXX11=1
-
-$as_echo "#define HAVE_CXX11 1" >>confdefs.h
-
-    fi
-
-
-  fi
-
+#AX_CXX_COMPILE_STDCXX_11(noext, mandatory)
 ac_ext=c
 ac_cpp='$CPP $CPPFLAGS'
 ac_compile='$CC -c $CFLAGS $CPPFLAGS conftest.$ac_ext >&5'
@@ -6005,7 +5805,6 @@ $as_echo "$as_me: WARNING: Your processor supports fma instructions but not your
 
 
 
-
 # Checks for libraries.
 #AX_GCC_VAR_ATTRIBUTE(aligned)
 
@@ -6659,7 +6458,6 @@ Info at: http://www.mpfr.org/" "$LINENO" 5
 fi
 
 
-
 # Check whether --enable-simd was given.
 if test "${enable_simd+set}" = set; then :
   enableval=$enable_simd; \
@@ -6799,160 +6597,15 @@ fi
 ###################################################################
 # Checks for doxygen support
 # if present enables the "make doxyfile" command
-echo
-echo Checking doxygen support
-echo :::::::::::::::::::::::::::::::::::::::::::
+#echo
+#echo Checking doxygen support
+#echo :::::::::::::::::::::::::::::::::::::::::::
+#AC_PROG_DOXYGEN
 
-# Check whether --enable-doxygen was given.
-if test "${enable_doxygen+set}" = set; then :
-  enableval=$enable_doxygen;
-fi
-
-
-# Check whether --enable-dot was given.
-if test "${enable_dot+set}" = set; then :
-  enableval=$enable_dot;
-fi
-
-# Check whether --enable-html-docs was given.
-if test "${enable_html_docs+set}" = set; then :
-  enableval=$enable_html_docs;
-else
-   enable_html_docs=yes
-fi
-
-# Check whether --enable-latex-docs was given.
-if test "${enable_latex_docs+set}" = set; then :
-  enableval=$enable_latex_docs;
-else
-  enable_latex_docs=no
-fi
-
-
-if test "x$enable_doxygen" = xno; then
- enable_doc=no
-else
- # Extract the first word of "doxygen", so it can be a program name with args.
-set dummy doxygen; ac_word=$2
-{ $as_echo "$as_me:${as_lineno-$LINENO}: checking for $ac_word" >&5
-$as_echo_n "checking for $ac_word... " >&6; }
-if ${ac_cv_prog_DOXYGEN+:} false; then :
-  $as_echo_n "(cached) " >&6
-else
-  if test -n "$DOXYGEN"; then
-  ac_cv_prog_DOXYGEN="$DOXYGEN" # Let the user override the test.
-else
-as_save_IFS=$IFS; IFS=$PATH_SEPARATOR
-for as_dir in $PATH
-do
-  IFS=$as_save_IFS
-  test -z "$as_dir" && as_dir=.
-    for ac_exec_ext in '' $ac_executable_extensions; do
-  if as_fn_executable_p "$as_dir/$ac_word$ac_exec_ext"; then
-    ac_cv_prog_DOXYGEN="doxygen"
-    $as_echo "$as_me:${as_lineno-$LINENO}: found $as_dir/$ac_word$ac_exec_ext" >&5
-    break 2
-  fi
-done
-  done
-IFS=$as_save_IFS
-
-fi
-fi
-DOXYGEN=$ac_cv_prog_DOXYGEN
-if test -n "$DOXYGEN"; then
-  { $as_echo "$as_me:${as_lineno-$LINENO}: result: $DOXYGEN" >&5
-$as_echo "$DOXYGEN" >&6; }
-else
-  { $as_echo "$as_me:${as_lineno-$LINENO}: result: no" >&5
-$as_echo "no" >&6; }
-fi
-
-
- if test x$DOXYGEN = x; then
-  if test "x$enable_doxygen" = xyes; then
-   as_fn_error $? "could not find doxygen" "$LINENO" 5
-  fi
-  enable_doc=no
- else
-  doxy_ver=`doxygen --version`
-  doxy_major=`expr "$doxy_ver" : '\([0-9]\)\..*'`
-  doxy_minor=`expr "$doxy_ver" : '[0-9]\.\([0-9]\).*'`
-  if test $doxy_major -eq "1" -a $doxy_minor -ge "3" ; then
-   enable_doc=yes
-   # Extract the first word of "dot", so it can be a program name with args.
-set dummy dot; ac_word=$2
-{ $as_echo "$as_me:${as_lineno-$LINENO}: checking for $ac_word" >&5
-$as_echo_n "checking for $ac_word... " >&6; }
-if ${ac_cv_prog_DOT+:} false; then :
-  $as_echo_n "(cached) " >&6
-else
-  if test -n "$DOT"; then
-  ac_cv_prog_DOT="$DOT" # Let the user override the test.
-else
-as_save_IFS=$IFS; IFS=$PATH_SEPARATOR
-for as_dir in $PATH
-do
-  IFS=$as_save_IFS
-  test -z "$as_dir" && as_dir=.
-    for ac_exec_ext in '' $ac_executable_extensions; do
-  if as_fn_executable_p "$as_dir/$ac_word$ac_exec_ext"; then
-    ac_cv_prog_DOT="dot"
-    $as_echo "$as_me:${as_lineno-$LINENO}: found $as_dir/$ac_word$ac_exec_ext" >&5
-    break 2
-  fi
-done
-  done
-IFS=$as_save_IFS
-
-fi
-fi
-DOT=$ac_cv_prog_DOT
-if test -n "$DOT"; then
-  { $as_echo "$as_me:${as_lineno-$LINENO}: result: $DOT" >&5
-$as_echo "$DOT" >&6; }
-else
-  { $as_echo "$as_me:${as_lineno-$LINENO}: result: no" >&5
-$as_echo "no" >&6; }
-fi
-
-
-  else
-   { $as_echo "$as_me:${as_lineno-$LINENO}: WARNING: doxygen version $doxy_ver too old, doxygen will not be used." >&5
-$as_echo "$as_me: WARNING: doxygen version $doxy_ver too old, doxygen will not be used." >&2;}
-   enable_doc=no
-  fi
- fi
-fi
-
- if test x$enable_doc = xyes; then
-  DOXYGEN_DOC_TRUE=
-  DOXYGEN_DOC_FALSE='#'
-else
-  DOXYGEN_DOC_TRUE='#'
-  DOXYGEN_DOC_FALSE=
-fi
-
-
-if test x$DOT = x; then
- if test "x$enable_dot" = xyes; then
-  as_fn_error $? "could not find dot" "$LINENO" 5
- fi
- enable_dot=no
-else
- enable_dot=yes
-fi
-
-
-
-
-
-
-if test -n "$DOXYGEN"
-then
-ac_config_files="$ac_config_files docs/doxy.cfg"
-
-fi
+#if test -n "$DOXYGEN"
+#then
+#AC_CONFIG_FILES([docs/doxy.cfg])
+#fi
 
 echo
 echo Creating configuration files
@@ -7110,10 +6763,6 @@ if test -z "${BUILD_COMMS_NONE_TRUE}" && test -z "${BUILD_COMMS_NONE_FALSE}"; th
   as_fn_error $? "conditional \"BUILD_COMMS_NONE\" was never defined.
 Usually this means the macro was only invoked conditionally." "$LINENO" 5
 fi
-if test -z "${DOXYGEN_DOC_TRUE}" && test -z "${DOXYGEN_DOC_FALSE}"; then
-  as_fn_error $? "conditional \"DOXYGEN_DOC\" was never defined.
-Usually this means the macro was only invoked conditionally." "$LINENO" 5
-fi
 
 : "${CONFIG_STATUS=./config.status}"
 ac_write_fail=0
@@ -7706,9 +7355,8 @@ cat >>$CONFIG_STATUS <<\_ACEOF || ac_write_fail=1
 for ac_config_target in $ac_config_targets
 do
   case $ac_config_target in
-    "lib/GridConfig.h") CONFIG_HEADERS="$CONFIG_HEADERS lib/GridConfig.h" ;;
+    "lib/Config.h") CONFIG_HEADERS="$CONFIG_HEADERS lib/Config.h" ;;
     "depfiles") CONFIG_COMMANDS="$CONFIG_COMMANDS depfiles" ;;
-    "docs/doxy.cfg") CONFIG_FILES="$CONFIG_FILES docs/doxy.cfg" ;;
     "Makefile") CONFIG_FILES="$CONFIG_FILES Makefile" ;;
     "lib/Makefile") CONFIG_FILES="$CONFIG_FILES lib/Makefile" ;;
     "tests/Makefile") CONFIG_FILES="$CONFIG_FILES tests/Makefile" ;;
diff --git a/configure.ac b/configure.ac
index be01228c..88f2d1f4 100644
--- a/configure.ac
+++ b/configure.ac
@@ -11,7 +11,7 @@ AC_CANONICAL_SYSTEM
 AM_INIT_AUTOMAKE(subdir-objects)
 AC_CONFIG_MACRO_DIR([m4])
 AC_CONFIG_SRCDIR([lib/Grid.h])
-AC_CONFIG_HEADERS([lib/GridConfig.h])
+AC_CONFIG_HEADERS([lib/Config.h])
 m4_ifdef([AM_SILENT_RULES], [AM_SILENT_RULES([yes])])
 
 AC_MSG_NOTICE([
@@ -26,10 +26,9 @@ AC_LANG(C++)
 AC_PROG_CXX
 AC_OPENMP
 AC_PROG_RANLIB
-AX_CXX_COMPILE_STDCXX_11(noext, mandatory)
+#AX_CXX_COMPILE_STDCXX_11(noext, mandatory)
 AX_EXT
 
-
 # Checks for libraries.
 #AX_GCC_VAR_ATTRIBUTE(aligned)
 
@@ -66,7 +65,6 @@ AC_CHECK_LIB([mpfr],[mpfr_init],,
 Please install or provide the correct path to your installation
 Info at: http://www.mpfr.org/)])
 
-
 AC_ARG_ENABLE([simd],[AC_HELP_STRING([--enable-simd=SSE4|AVX|AVX2|AVX512|MIC],\
 	[Select instructions to be SSE4.0, AVX 1.0, AVX 2.0+FMA, AVX 512, MIC])],\
 	[ac_SIMD=${enable_simd}],[ac_SIMD=AVX2])
@@ -155,15 +153,15 @@ AM_CONDITIONAL(BUILD_COMMS_NONE,[ test "X${ac_COMMS}X" == "XnoneX" ])
 ###################################################################
 # Checks for doxygen support
 # if present enables the "make doxyfile" command
-echo
-echo Checking doxygen support 
-echo :::::::::::::::::::::::::::::::::::::::::::
-AC_PROG_DOXYGEN
+#echo
+#echo Checking doxygen support 
+#echo :::::::::::::::::::::::::::::::::::::::::::
+#AC_PROG_DOXYGEN
 
-if test -n "$DOXYGEN"
-then
-AC_CONFIG_FILES([docs/doxy.cfg])
-fi
+#if test -n "$DOXYGEN"
+#then
+#AC_CONFIG_FILES([docs/doxy.cfg])
+#fi
 
 echo
 echo Creating configuration files
diff --git a/gcc-bug-report/broken.cc b/gcc-bug-report/broken.cc
index d23572d3..edf878c5 100644
--- a/gcc-bug-report/broken.cc
+++ b/gcc-bug-report/broken.cc
@@ -29,12 +29,12 @@ public:
 
 template<int N,class obj,typename std::enable_if<N==obj::NestLevel >::type * = nullptr > auto function(const obj &arg)-> obj
 {
-  std::cout<<"Leaf "<<obj::NestLevel<<std::endl;
+  std::cout<<GridLogMessage<<"Leaf "<<obj::NestLevel<<std::endl;
   return arg;
 }
 template<int N,class obj,typename std::enable_if<N!=obj::NestLevel >::type * = nullptr > auto function(const obj &arg)-> obj
 {
-  std::cout<<"Node "<<obj::NestLevel<<std::endl;
+  std::cout<<GridLogMessage<<"Node "<<obj::NestLevel<<std::endl;
   obj ret;
   ret.internal=function<N>(arg.internal);
   return ret;
diff --git a/lib/AlignedAllocator.h b/lib/AlignedAllocator.h
index 3e90371a..ea25f6a8 100644
--- a/lib/AlignedAllocator.h
+++ b/lib/AlignedAllocator.h
@@ -1,6 +1,13 @@
 #ifndef GRID_ALIGNED_ALLOCATOR_H
 #define GRID_ALIGNED_ALLOCATOR_H
 
+#ifdef HAVE_MALLOC_MALLOC_H
+#include <malloc/malloc.h>
+#endif
+#ifdef HAVE_MALLOC_H
+#include <malloc.h>
+#endif
+
 #include <immintrin.h>
 #ifdef HAVE_MM_MALLOC_H
 #include <mm_malloc.h>
diff --git a/lib/GridConfig.h b/lib/Config.h
similarity index 95%
rename from lib/GridConfig.h
rename to lib/Config.h
index 6976175c..3507e6b7 100644
--- a/lib/GridConfig.h
+++ b/lib/Config.h
@@ -1,5 +1,5 @@
-/* lib/GridConfig.h.  Generated from GridConfig.h.in by configure.  */
-/* lib/GridConfig.h.in.  Generated from configure.ac by autoheader.  */
+/* lib/Config.h.  Generated from Config.h.in by configure.  */
+/* lib/Config.h.in.  Generated from configure.ac by autoheader.  */
 
 /* AVX Intrinsics */
 /* #undef AVX1 */
@@ -34,9 +34,6 @@
 /* Support AVX2 (Advanced Vector Extensions 2) instructions */
 /* #undef HAVE_AVX2 */
 
-/* define if the compiler supports basic C++11 syntax */
-/* #undef HAVE_CXX11 */
-
 /* Define to 1 if you have the declaration of `be64toh', and to 0 if you
    don't. */
 #define HAVE_DECL_BE64TOH 1
diff --git a/lib/GridConfig.h.in b/lib/Config.h.in
similarity index 96%
rename from lib/GridConfig.h.in
rename to lib/Config.h.in
index 136d7bfb..3528a9f7 100644
--- a/lib/GridConfig.h.in
+++ b/lib/Config.h.in
@@ -1,4 +1,4 @@
-/* lib/GridConfig.h.in.  Generated from configure.ac by autoheader.  */
+/* lib/Config.h.in.  Generated from configure.ac by autoheader.  */
 
 /* AVX Intrinsics */
 #undef AVX1
@@ -33,9 +33,6 @@
 /* Support AVX2 (Advanced Vector Extensions 2) instructions */
 #undef HAVE_AVX2
 
-/* define if the compiler supports basic C++11 syntax */
-#undef HAVE_CXX11
-
 /* Define to 1 if you have the declaration of `be64toh', and to 0 if you
    don't. */
 #undef HAVE_DECL_BE64TOH
diff --git a/lib/Grid.h b/lib/Grid.h
index aec0e9ac..624bc5a7 100644
--- a/lib/Grid.h
+++ b/lib/Grid.h
@@ -6,92 +6,48 @@
 //  Copyright (c) 2014 University of Edinburgh. All rights reserved.
 //
 
-
 #ifndef GRID_H
 #define GRID_H
 
+///////////////////
+// Std C++ dependencies
+///////////////////
 #include <cassert>
-
 #include <complex>
 #include <vector>
-
 #include <iostream>
 #include <iomanip>
 #include <random>
 #include <functional>
-
 #include <stdio.h>
 #include <stdlib.h>
-#include <sys/time.h>
 #include <stdio.h>
 #include <signal.h>
+#include <ctime>
+#include <sys/time.h>
+#include <chrono>
 
-#ifndef MAX
-#define MAX(x,y) ((x)>(y)?(x):(y))
-#define MIN(x,y) ((x)>(y)?(y):(x))
-#endif
-
-#define strong_inline __attribute__((always_inline)) inline
-
-#include <GridConfig.h>
-
-////////////////////////////////////////////////////////////
-// Tunable header includes
-////////////////////////////////////////////////////////////
-
-#ifdef HAVE_MALLOC_MALLOC_H
-#include <malloc/malloc.h>
-#endif
-#ifdef HAVE_MALLOC_H
-#include <malloc.h>
-#endif
-
+///////////////////
+// Grid headers
+///////////////////
+#include <MacroMagic.h>
+#include <Config.h>
+#include <Timer.h>
+#include <Log.h>
 #include <AlignedAllocator.h>
-
 #include <Simd.h>
 #include <Threads.h>
-
-#include <Communicator.h> // subdir aggregate
-#include <Cartesian.h> // subdir aggregate
-#include <Tensors.h>   // subdir aggregate
-#include <Lattice.h>   // subdir aggregate
-#include <Cshift.h>    // subdir aggregate
-#include <Stencil.h>   // subdir aggregate
-#include <Algorithms.h>// subdir aggregate
-
+#include <Communicator.h> 
+#include <Cartesian.h>    
+#include <Tensors.h>      
+#include <Lattice.h>      
+#include <Cshift.h>       
+#include <Stencil.h>      
+#include <Algorithms.h>   
 #include <qcd/QCD.h>
 #include <parallelIO/NerscIO.h>
 
-namespace Grid {
+#include <Init.h>
 
-  void Grid_init(int *argc,char ***argv);
-  void Grid_finalize(void);
-  // internal, controled with --handle
-  void Grid_sa_signal_handler(int sig,siginfo_t *si,void * ptr);
-  void Grid_debug_handler_init(void);
-  void Grid_quiesce_nodes(void);
-  void Grid_unquiesce_nodes(void);
-
-  // C++11 time facilities better?
-  double usecond(void);
-
-  const std::vector<int> GridDefaultSimd(int dims,int nsimd);
-  const std::vector<int> &GridDefaultLatt(void);
-  const std::vector<int> &GridDefaultMpi(void);
-  const int              &GridThreads(void)  ;
-  void                 GridSetThreads(int t) ;
-
-  // Common parsing chores
-  std::string GridCmdOptionPayload(char ** begin, char ** end, const std::string & option);
-  bool        GridCmdOptionExists(char** begin, char** end, const std::string& option);
-  std::string GridCmdVectorIntToString(const std::vector<int> & vec);
-
-  void GridParseLayout(char **argv,int argc,
-		       std::vector<int> &latt,
-		       std::vector<int> &simd,
-		       std::vector<int> &mpi);
-
-
-};
 
 #endif
diff --git a/lib/GridInit.cc b/lib/Init.cc
similarity index 58%
rename from lib/GridInit.cc
rename to lib/Init.cc
index 20371a4b..194fd589 100644
--- a/lib/GridInit.cc
+++ b/lib/Init.cc
@@ -25,17 +25,19 @@
 
 namespace Grid {
 
-  //////////////////////////////////////////////////////
-  // Convenience functions to access stadard command line arg
-  // driven parallelism controls
-  //////////////////////////////////////////////////////
-  static std::vector<int> Grid_default_latt;
-  static std::vector<int> Grid_default_mpi;
+//////////////////////////////////////////////////////
+// Convenience functions to access stadard command line arg
+// driven parallelism controls
+//////////////////////////////////////////////////////
+static std::vector<int> Grid_default_latt;
+static std::vector<int> Grid_default_mpi;
+int GridThread::_threads;
 
-  int GridThread::_threads;
 
-  const std::vector<int> GridDefaultSimd(int dims,int nsimd)
-  {
+const std::vector<int> &GridDefaultLatt(void)     {return Grid_default_latt;};
+const std::vector<int> &GridDefaultMpi(void)      {return Grid_default_mpi;};
+const std::vector<int> GridDefaultSimd(int dims,int nsimd)
+{
     std::vector<int> layout(dims);
     int nn=nsimd;
     for(int d=dims-1;d>=0;d--){
@@ -48,15 +50,11 @@ namespace Grid {
     }
     assert(nn==1);
     return layout;
-  }
-
+}
   
-  const std::vector<int> &GridDefaultLatt(void)     {return Grid_default_latt;};
-  const std::vector<int> &GridDefaultMpi(void)      {return Grid_default_mpi;};
-
-  ////////////////////////////////////////////////////////////
-  // Command line parsing assist for stock controls
-  ////////////////////////////////////////////////////////////
+////////////////////////////////////////////////////////////
+// Command line parsing assist for stock controls
+////////////////////////////////////////////////////////////
 std::string GridCmdOptionPayload(char ** begin, char ** end, const std::string & option)
 {
   char ** itr = std::find(begin, end, option);
@@ -70,6 +68,23 @@ bool GridCmdOptionExists(char** begin, char** end, const std::string& option)
 {
   return std::find(begin, end, option) != end;
 }
+  // Comma separated list
+void GridCmdOptionCSL(std::string str,std::vector<std::string> & vec)
+{
+  size_t pos = 0;
+  std::string token;
+  std::string delimiter(",");
+
+  vec.resize(0);
+  while ((pos = str.find(delimiter)) != std::string::npos) {
+    token = str.substr(0, pos);
+    vec.push_back(token);
+    str.erase(0, pos + delimiter.length());
+  }
+  token = str;
+  vec.push_back(token);
+  return;
+}
 
 void GridCmdOptionIntVector(std::string &str,std::vector<int> & vec)
 {
@@ -84,6 +99,7 @@ void GridCmdOptionIntVector(std::string &str,std::vector<int> & vec)
   return;
 }
 
+
 void GridParseLayout(char **argv,int argc,
 		     std::vector<int> &latt,
 		     std::vector<int> &mpi)
@@ -117,8 +133,9 @@ std::string GridCmdVectorIntToString(const std::vector<int> & vec){
   std::copy(vec.begin(), vec.end(),std::ostream_iterator<int>(oss, " "));
   return oss.str();
 }
-  /////////////////////////////////////////////////////////
-  /////////////////////////////////////////////////////////
+/////////////////////////////////////////////////////////
+//
+/////////////////////////////////////////////////////////
 void Grid_init(int *argc,char ***argv)
 {
 #ifdef GRID_COMMS_MPI
@@ -126,15 +143,33 @@ void Grid_init(int *argc,char ***argv)
 #endif
   // Parse command line args.
 
+  GridLogger::StopWatch.Start();
+
+  std::string arg;
+  std::vector<std::string> logstreams;
+  std::string defaultLog("Error,Warning,Message,Performance");
+
+  GridCmdOptionCSL(defaultLog,logstreams);
+  GridLogConfigure(logstreams);
+
   if( GridCmdOptionExists(*argv,*argv+*argc,"--help") ){
-    std::cout<<"--help : this message"<<std::endl;
-    std::cout<<"--debug-signals : catch sigsegv and print a blame report"<<std::endl;
-    std::cout<<"--debug-stdout  : print stdout from EVERY node"<<std::endl;    
-    std::cout<<"--decomposition : report on default omp,mpi and simd decomposition"<<std::endl;    
-    std::cout<<"--mpi n.n.n.n   : default MPI decomposition"<<std::endl;    
-    std::cout<<"--omp n         : default number of OMP threads"<<std::endl;    
-    std::cout<<"--grid n.n.n.n  : default Grid size"<<std::endl;    
+    std::cout<<GridLogMessage<<"--help : this message"<<std::endl;
+    std::cout<<GridLogMessage<<"--debug-signals : catch sigsegv and print a blame report"<<std::endl;
+    std::cout<<GridLogMessage<<"--debug-stdout  : print stdout from EVERY node"<<std::endl;    
+    std::cout<<GridLogMessage<<"--decomposition : report on default omp,mpi and simd decomposition"<<std::endl;    
+    std::cout<<GridLogMessage<<"--mpi n.n.n.n   : default MPI decomposition"<<std::endl;    
+    std::cout<<GridLogMessage<<"--omp n         : default number of OMP threads"<<std::endl;    
+    std::cout<<GridLogMessage<<"--grid n.n.n.n  : default Grid size"<<std::endl;    
+    std::cout<<GridLogMessage<<"--log list      : comma separted list of streams from Error,Warning,Message,Performance,Iterative,Debug"<<std::endl;    
   }
+
+  if( GridCmdOptionExists(*argv,*argv+*argc,"--log") ){
+    arg = GridCmdOptionPayload(*argv,*argv+*argc,"--log");
+    GridCmdOptionCSL(arg,logstreams);
+    GridLogConfigure(logstreams);
+  }
+
+
   if( GridCmdOptionExists(*argv,*argv+*argc,"--debug-signals") ){
     Grid_debug_handler_init();
   }
@@ -152,38 +187,18 @@ void Grid_init(int *argc,char ***argv)
 		  Grid_default_latt,
 		  Grid_default_mpi);
   if( GridCmdOptionExists(*argv,*argv+*argc,"--decomposition") ){
-    std::cout<<"Grid Decomposition\n";
-    std::cout<<"\tOpenMP threads : "<<GridThread::GetThreads()<<std::endl;
-    std::cout<<"\tMPI tasks      : "<<GridCmdVectorIntToString(GridDefaultMpi())<<std::endl;
-    std::cout<<"\tvRealF         : "<<sizeof(vRealF)*8    <<"bits ; " <<GridCmdVectorIntToString(GridDefaultSimd(4,vRealF::Nsimd()))<<std::endl;
-    std::cout<<"\tvRealD         : "<<sizeof(vRealD)*8    <<"bits ; " <<GridCmdVectorIntToString(GridDefaultSimd(4,vRealD::Nsimd()))<<std::endl;
-    std::cout<<"\tvComplexF      : "<<sizeof(vComplexF)*8 <<"bits ; " <<GridCmdVectorIntToString(GridDefaultSimd(4,vComplexF::Nsimd()))<<std::endl;
-    std::cout<<"\tvComplexD      : "<<sizeof(vComplexD)*8 <<"bits ; " <<GridCmdVectorIntToString(GridDefaultSimd(4,vComplexD::Nsimd()))<<std::endl;
+    std::cout<<GridLogMessage<<"Grid Decomposition\n";
+    std::cout<<GridLogMessage<<"\tOpenMP threads : "<<GridThread::GetThreads()<<std::endl;
+    std::cout<<GridLogMessage<<"\tMPI tasks      : "<<GridCmdVectorIntToString(GridDefaultMpi())<<std::endl;
+    std::cout<<GridLogMessage<<"\tvRealF         : "<<sizeof(vRealF)*8    <<"bits ; " <<GridCmdVectorIntToString(GridDefaultSimd(4,vRealF::Nsimd()))<<std::endl;
+    std::cout<<GridLogMessage<<"\tvRealD         : "<<sizeof(vRealD)*8    <<"bits ; " <<GridCmdVectorIntToString(GridDefaultSimd(4,vRealD::Nsimd()))<<std::endl;
+    std::cout<<GridLogMessage<<"\tvComplexF      : "<<sizeof(vComplexF)*8 <<"bits ; " <<GridCmdVectorIntToString(GridDefaultSimd(4,vComplexF::Nsimd()))<<std::endl;
+    std::cout<<GridLogMessage<<"\tvComplexD      : "<<sizeof(vComplexD)*8 <<"bits ; " <<GridCmdVectorIntToString(GridDefaultSimd(4,vComplexD::Nsimd()))<<std::endl;
   }
 
+
 }
 
-
-  ////////////////////////////////////////////////////////////
-  // Verbose limiter on MPI tasks
-  ////////////////////////////////////////////////////////////
-  void Grid_quiesce_nodes(void)
-  {
-#ifdef GRID_COMMS_MPI
-    int me;
-    MPI_Comm_rank(MPI_COMM_WORLD,&me);
-    if ( me ) { 
-      std::cout.setstate(std::ios::badbit);
-    }
-#endif
-  }
-  void Grid_unquiesce_nodes(void)
-  {
-#ifdef GRID_COMMS_MPI
-    std::cout.clear();
-#endif
-  }
-
   
 void Grid_finalize(void)
 {
diff --git a/lib/Init.h b/lib/Init.h
new file mode 100644
index 00000000..aa0b74cc
--- /dev/null
+++ b/lib/Init.h
@@ -0,0 +1,32 @@
+#ifndef GRID_INIT_H
+#define GRID_INIT_H
+
+namespace Grid {
+
+  void Grid_init(int *argc,char ***argv);
+  void Grid_finalize(void);
+  // internal, controled with --handle
+  void Grid_sa_signal_handler(int sig,siginfo_t *si,void * ptr);
+  void Grid_debug_handler_init(void);
+  void Grid_quiesce_nodes(void);
+  void Grid_unquiesce_nodes(void);
+
+  const std::vector<int> GridDefaultSimd(int dims,int nsimd);
+  const std::vector<int> &GridDefaultLatt(void);
+  const std::vector<int> &GridDefaultMpi(void);
+  const int              &GridThreads(void)  ;
+  void                    GridSetThreads(int t) ;
+
+  // Common parsing chores
+  std::string GridCmdOptionPayload(char ** begin, char ** end, const std::string & option);
+  bool        GridCmdOptionExists(char** begin, char** end, const std::string& option);
+  std::string GridCmdVectorIntToString(const std::vector<int> & vec);
+
+  void GridParseLayout(char **argv,int argc,
+		       std::vector<int> &latt,
+		       std::vector<int> &simd,
+		       std::vector<int> &mpi);
+
+
+};
+#endif
diff --git a/lib/Log.cc b/lib/Log.cc
new file mode 100644
index 00000000..468d9116
--- /dev/null
+++ b/lib/Log.cc
@@ -0,0 +1,62 @@
+#include <Grid.h>
+
+namespace Grid {
+
+GridStopWatch GridLogger::StopWatch;
+std::ostream  GridLogger::devnull(0);
+
+GridLogger GridLogError      (1,"Error");
+GridLogger GridLogWarning    (1,"Warning");
+GridLogger GridLogMessage    (1,"Message");
+GridLogger GridLogDebug      (1,"Debug");
+GridLogger GridLogPerformance(1,"Performance");
+GridLogger GridLogIterative  (1,"Iterative");
+
+void GridLogConfigure(std::vector<std::string> &logstreams)
+{
+  GridLogError.Active(0);
+  GridLogWarning.Active(0);
+  GridLogMessage.Active(0);
+  GridLogIterative.Active(0);
+  GridLogDebug.Active(0);
+  GridLogPerformance.Active(0);
+
+  for(int i=0;i<logstreams.size();i++){
+    if ( logstreams[i]== std::string("Error")       ) GridLogError.Active(1);
+    if ( logstreams[i]== std::string("Warning")     ) GridLogWarning.Active(1);
+    if ( logstreams[i]== std::string("Message")     ) GridLogMessage.Active(1);
+    if ( logstreams[i]== std::string("Iterative")   ) GridLogIterative.Active(1);
+    if ( logstreams[i]== std::string("Debug")       ) GridLogDebug.Active(1);
+    if ( logstreams[i]== std::string("Performance") ) GridLogPerformance.Active(1);
+  }
+}
+
+////////////////////////////////////////////////////////////
+// Verbose limiter on MPI tasks
+////////////////////////////////////////////////////////////
+void Grid_quiesce_nodes(void)
+{
+#ifdef GRID_COMMS_MPI
+  int me;
+  MPI_Comm_rank(MPI_COMM_WORLD,&me);
+  if ( me ) { 
+    std::cout.setstate(std::ios::badbit);
+  }
+#endif
+}
+
+void Grid_unquiesce_nodes(void)
+{
+#ifdef GRID_COMMS_MPI
+    std::cout.clear();
+#endif
+}
+
+std::ostream& operator<< (std::ostream& stream, const GridTime& time)
+{
+  stream << time.count()<<" ms";
+  return stream;
+}
+
+}
+
diff --git a/lib/Log.h b/lib/Log.h
new file mode 100644
index 00000000..4cf094bc
--- /dev/null
+++ b/lib/Log.h
@@ -0,0 +1,46 @@
+#ifndef GRID_LOG_H
+#define GRID_LOG_H
+namespace Grid {
+
+// Dress the output; use std::chrono for time stamping via the StopWatch class
+
+std::ostream& operator<< (std::ostream& stream, const GridTime& time);
+
+class GridLogger { 
+  int active;
+  std::string name;
+public:
+
+  static GridStopWatch StopWatch;
+  static std::ostream devnull;
+  
+  GridLogger(int on, std::string nm): active(on), name(nm) { 
+  };
+  
+  void Active(int on) {active = on;};
+
+  friend std::ostream& operator<< (std::ostream& stream, const GridLogger& log){
+    if ( log.active ) {
+      StopWatch.Stop();
+      GridTime now = StopWatch.Elapsed();
+      StopWatch.Start();
+      stream << "Grid : "<<log.name << " : " << now << " : ";
+      return stream;
+    } else { 
+      return devnull;
+    }
+  }
+
+};
+
+void GridLogConfigure(std::vector<std::string> &logstreams);
+
+extern GridLogger GridLogError;
+extern GridLogger GridLogWarning;
+extern GridLogger GridLogMessage;
+extern GridLogger GridLogDebug  ;
+extern GridLogger GridLogPerformance;
+extern GridLogger GridLogIterative  ;
+
+}
+#endif
diff --git a/lib/MacroMagic.h b/lib/MacroMagic.h
new file mode 100644
index 00000000..00f60089
--- /dev/null
+++ b/lib/MacroMagic.h
@@ -0,0 +1,78 @@
+#ifndef GRID_MACRO_MAGIC_H
+#define GRID_MACRO_MAGIC_H
+
+#define strong_inline __attribute__((always_inline)) inline
+
+#ifndef MAX
+#define MAX(x,y) ((x)>(y)?(x):(y))
+#define MIN(x,y) ((x)>(y)?(y):(x))
+#endif
+
+#define GRID_MACRO_FIRST(a, ...) a
+#define GRID_MACRO_SECOND(a, b, ...) b
+
+#define GRID_MACRO_EMPTY()
+
+#define GRID_MACRO_EVAL(...)     GRID_MACRO_EVAL1024(__VA_ARGS__)
+#define GRID_MACRO_EVAL1024(...) GRID_MACRO_EVAL512(GRID_MACRO_EVAL512(__VA_ARGS__))
+#define GRID_MACRO_EVAL512(...)  GRID_MACRO_EVAL256(GRID_MACRO_EVAL256(__VA_ARGS__))
+#define GRID_MACRO_EVAL256(...)  GRID_MACRO_EVAL128(GRID_MACRO_EVAL128(__VA_ARGS__))
+#define GRID_MACRO_EVAL128(...)  GRID_MACRO_EVAL64(GRID_MACRO_EVAL64(__VA_ARGS__))
+#define GRID_MACRO_EVAL64(...)   GRID_MACRO_EVAL32(GRID_MACRO_EVAL32(__VA_ARGS__))
+#define GRID_MACRO_EVAL32(...)   GRID_MACRO_EVAL16(GRID_MACRO_EVAL16(__VA_ARGS__))
+#define GRID_MACRO_EVAL16(...)   GRID_MACRO_EVAL8(GRID_MACRO_EVAL8(__VA_ARGS__))
+#define GRID_MACRO_EVAL8(...)    GRID_MACRO_EVAL4(GRID_MACRO_EVAL4(__VA_ARGS__))
+#define GRID_MACRO_EVAL4(...)    GRID_MACRO_EVAL2(GRID_MACRO_EVAL2(__VA_ARGS__))
+#define GRID_MACRO_EVAL2(...)    GRID_MACRO_EVAL1(GRID_MACRO_EVAL1(__VA_ARGS__))
+#define GRID_MACRO_EVAL1(...) __VA_ARGS__
+
+#define GRID_MACRO_DEFER1(m) m GRID_MACRO_EMPTY()
+#define GRID_MACRO_DEFER2(m) m GRID_MACRO_EMPTY GRID_MACRO_EMPTY()()
+#define GRID_MACRO_DEFER3(m) m GRID_MACRO_EMPTY GRID_MACRO_EMPTY GRID_MACRO_EMPTY()()()
+#define GRID_MACRO_DEFER4(m) m GRID_MACRO_EMPTY GRID_MACRO_EMPTY GRID_MACRO_EMPTY GRID_MACRO_EMPTY()()()()
+
+#define GRID_MACRO_IS_PROBE(...) GRID_MACRO_SECOND(__VA_ARGS__, 0)
+#define GRID_MACRO_PROBE() ~, 1
+
+#define GRID_MACRO_CAT(a,b) a ## b
+
+#define GRID_MACRO_NOT(x) GRID_MACRO_IS_PROBE(GRID_MACRO_CAT(_GRID_MACRO_NOT_, x))
+#define _GRID_MACRO_NOT_0 GRID_MACRO_PROBE()
+
+#define GRID_MACRO_BOOL(x) GRID_MACRO_NOT(GRID_MACRO_NOT(x))
+
+#define GRID_MACRO_IF_ELSE(condition) _GRID_MACRO_IF_ELSE(GRID_MACRO_BOOL(condition))
+#define _GRID_MACRO_IF_ELSE(condition) GRID_MACRO_CAT(_GRID_MACRO_IF_, condition)
+
+#define _GRID_MACRO_IF_1(...) __VA_ARGS__ _GRID_MACRO_IF_1_ELSE
+#define _GRID_MACRO_IF_0(...)             _GRID_MACRO_IF_0_ELSE
+
+#define _GRID_MACRO_IF_1_ELSE(...)
+#define _GRID_MACRO_IF_0_ELSE(...) __VA_ARGS__
+
+#define GRID_MACRO_HAS_ARGS(...) GRID_MACRO_BOOL(GRID_MACRO_FIRST(_GRID_MACRO_END_OF_ARGUMENTS_ __VA_ARGS__)())
+#define _GRID_MACRO_END_OF_ARGUMENTS_() 0
+
+#define GRID_MACRO_MAP(m, first, second, ...)   \
+  m(first,second)                           \
+  GRID_MACRO_IF_ELSE(GRID_MACRO_HAS_ARGS(__VA_ARGS__))(				       \
+				 GRID_MACRO_DEFER4(_GRID_MACRO_MAP)()(m, __VA_ARGS__)   \
+				     )(                                 \
+				       /* Do nothing, just terminate */ \
+									)
+#define _GRID_MACRO_MAP() GRID_MACRO_MAP
+
+#define GRID_MACRO_MEMBER(A,B)        A B;
+
+#define GRID_MACRO_OS_WRITE_MEMBER(A,B) os<< #A <<" "#B <<" = "<< obj. B <<" ; " <<std::endl;
+
+#define GRID_DECL_CLASS_MEMBERS(cname,...)		\
+  GRID_MACRO_EVAL(GRID_MACRO_MAP(GRID_MACRO_MEMBER,__VA_ARGS__))		\
+  friend std::ostream & operator << (std::ostream &os, const cname &obj ) {	\
+    os<<"class "<<#cname<<" {"<<std::endl;\
+    GRID_MACRO_EVAL(GRID_MACRO_MAP(GRID_MACRO_OS_WRITE_MEMBER,__VA_ARGS__))	\
+      os<<"}";								\
+    return os;\
+  };  
+
+#endif
diff --git a/lib/Make.inc b/lib/Make.inc
index dfa8401b..a54ce3c7 100644
--- a/lib/Make.inc
+++ b/lib/Make.inc
@@ -1,4 +1,4 @@
 
-HFILES=./Cshift.h ./simd/Grid_avx.h ./simd/Grid_vector_types.h ./simd/Grid_sse4.h ./simd/Grid_avx512.h ./simd/Grid_empty.h ./simd/Grid_vector_unops.h ./simd/Grid_neon.h ./simd/Grid_qpx.h ./Tensors.h ./Algorithms.h ./communicator/Communicator_base.h ./lattice/Lattice_rng.h ./lattice/Lattice_reduction.h ./lattice/Lattice_transfer.h ./lattice/Lattice_unary.h ./lattice/Lattice_peekpoke.h ./lattice/Lattice_coordinate.h ./lattice/Lattice_comparison.h ./lattice/Lattice_overload.h ./lattice/Lattice_reality.h ./lattice/Lattice_local.h ./lattice/Lattice_conformable.h ./lattice/Lattice_where.h ./lattice/Lattice_comparison_utils.h ./lattice/Lattice_arith.h ./lattice/Lattice_base.h ./lattice/Lattice_ET.h ./lattice/Lattice_transpose.h ./lattice/Lattice_trace.h ./Stencil.h ./tensors/Tensor_arith_sub.h ./tensors/Tensor_exp.h ./tensors/Tensor_arith_mul.h ./tensors/Tensor_class.h ./tensors/Tensor_logical.h ./tensors/Tensor_transpose.h ./tensors/Tensor_arith_mac.h ./tensors/Tensor_arith_scalar.h ./tensors/Tensor_reality.h ./tensors/Tensor_trace.h ./tensors/Tensor_index.h ./tensors/Tensor_arith_add.h ./tensors/Tensor_outer.h ./tensors/Tensor_inner.h ./tensors/Tensor_traits.h ./tensors/Tensor_Ta.h ./tensors/Tensor_unary.h ./tensors/Tensor_determinant.h ./tensors/Tensor_arith.h ./tensors/Tensor_extract_merge.h ./Communicator.h ./Cartesian.h ./parallelIO/NerscIO.h ./qcd/QCD.h ./qcd/hmc/integrators/Integrator.h ./qcd/hmc/integrators/Integrator_algorithm.h ./qcd/hmc/HMC.h ./qcd/utils/SpaceTimeGrid.h ./qcd/utils/SUn.h ./qcd/utils/LinalgUtils.h ./qcd/utils/CovariantCshift.h ./qcd/utils/WilsonLoops.h ./qcd/action/ActionBase.h ./qcd/action/gauge/WilsonGaugeAction.h ./qcd/action/Actions.h ./qcd/action/fermion/CayleyFermion5D.h ./qcd/action/fermion/ScaledShamirFermion.h ./qcd/action/fermion/MobiusFermion.h ./qcd/action/fermion/OverlapWilsonContfracTanhFermion.h ./qcd/action/fermion/PartialFractionFermion5D.h ./qcd/action/fermion/ShamirZolotarevFermion.h ./qcd/action/fermion/FermionOperator.h ./qcd/action/fermion/WilsonFermion5D.h ./qcd/action/fermion/WilsonCompressor.h ./qcd/action/fermion/OverlapWilsonPartialFractionZolotarevFermion.h ./qcd/action/fermion/WilsonKernels.h ./qcd/action/fermion/DomainWallFermion.h ./qcd/action/fermion/OverlapWilsonPartialFractionTanhFermion.h ./qcd/action/fermion/OverlapWilsonContfracZolotarevFermion.h ./qcd/action/fermion/MobiusZolotarevFermion.h ./qcd/action/fermion/g5HermitianLinop.h ./qcd/action/fermion/OverlapWilsonCayleyTanhFermion.h ./qcd/action/fermion/WilsonFermion.h ./qcd/action/fermion/ContinuedFractionFermion5D.h ./qcd/action/fermion/OverlapWilsonCayleyZolotarevFermion.h ./qcd/spin/TwoSpinor.h ./qcd/spin/Dirac.h ./cshift/Cshift_common.h ./cshift/Cshift_none.h ./cshift/Cshift_mpi.h ./Simd.h ./GridConfig.h ./cartesian/Cartesian_base.h ./cartesian/Cartesian_red_black.h ./cartesian/Cartesian_full.h ./AlignedAllocator.h ./Lattice.h ./Old/Tensor_poke.h ./Old/Tensor_peek.h ./Threads.h ./Grid.h ./algorithms/Preconditioner.h ./algorithms/iterative/ConjugateResidual.h ./algorithms/iterative/PrecGeneralisedConjugateResidual.h ./algorithms/iterative/ConjugateGradientMultiShift.h ./algorithms/iterative/SchurRedBlack.h ./algorithms/iterative/NormalEquations.h ./algorithms/iterative/ConjugateGradient.h ./algorithms/iterative/AdefGeneric.h ./algorithms/approx/Chebyshev.h ./algorithms/approx/Zolotarev.h ./algorithms/approx/MultiShiftFunction.h ./algorithms/approx/bigfloat.h ./algorithms/approx/bigfloat_double.h ./algorithms/approx/Remez.h ./algorithms/LinearOperator.h ./algorithms/SparseMatrix.h ./algorithms/CoarsenedMatrix.h ./stencil/Lebesgue.h
+HFILES=./Cshift.h ./simd/Grid_avx.h ./simd/Grid_vector_types.h ./simd/Grid_sse4.h ./simd/Grid_avx512.h ./simd/Grid_empty.h ./simd/Grid_vector_unops.h ./simd/Grid_neon.h ./simd/Grid_qpx.h ./Init.h ./Tensors.h ./Algorithms.h ./communicator/Communicator_base.h ./lattice/Lattice_rng.h ./lattice/Lattice_reduction.h ./lattice/Lattice_transfer.h ./lattice/Lattice_unary.h ./lattice/Lattice_peekpoke.h ./lattice/Lattice_coordinate.h ./lattice/Lattice_comparison.h ./lattice/Lattice_overload.h ./lattice/Lattice_reality.h ./lattice/Lattice_local.h ./lattice/Lattice_conformable.h ./lattice/Lattice_where.h ./lattice/Lattice_comparison_utils.h ./lattice/Lattice_arith.h ./lattice/Lattice_base.h ./lattice/Lattice_ET.h ./lattice/Lattice_transpose.h ./lattice/Lattice_trace.h ./Stencil.h ./tensors/Tensor_arith_sub.h ./tensors/Tensor_exp.h ./tensors/Tensor_arith_mul.h ./tensors/Tensor_class.h ./tensors/Tensor_logical.h ./tensors/Tensor_transpose.h ./tensors/Tensor_arith_mac.h ./tensors/Tensor_arith_scalar.h ./tensors/Tensor_reality.h ./tensors/Tensor_trace.h ./tensors/Tensor_index.h ./tensors/Tensor_arith_add.h ./tensors/Tensor_outer.h ./tensors/Tensor_inner.h ./tensors/Tensor_traits.h ./tensors/Tensor_Ta.h ./tensors/Tensor_unary.h ./tensors/Tensor_determinant.h ./tensors/Tensor_arith.h ./tensors/Tensor_extract_merge.h ./Communicator.h ./Cartesian.h ./parallelIO/NerscIO.h ./Timer.h ./qcd/QCD.h ./qcd/hmc/integrators/Integrator_base.h ./qcd/hmc/integrators/Integrator.h ./qcd/hmc/integrators/Integrator_algorithm.h ./qcd/hmc/HMC.h ./qcd/utils/SpaceTimeGrid.h ./qcd/utils/SUn.h ./qcd/utils/LinalgUtils.h ./qcd/utils/CovariantCshift.h ./qcd/utils/WilsonLoops.h ./qcd/action/ActionBase.h ./qcd/action/gauge/WilsonGaugeAction.h ./qcd/action/Actions.h ./qcd/action/pseudofermion/TwoFlavour.h ./qcd/action/fermion/CayleyFermion5D.h ./qcd/action/fermion/ScaledShamirFermion.h ./qcd/action/fermion/MobiusFermion.h ./qcd/action/fermion/OverlapWilsonContfracTanhFermion.h ./qcd/action/fermion/PartialFractionFermion5D.h ./qcd/action/fermion/ShamirZolotarevFermion.h ./qcd/action/fermion/FermionOperator.h ./qcd/action/fermion/WilsonFermion5D.h ./qcd/action/fermion/WilsonCompressor.h ./qcd/action/fermion/OverlapWilsonPartialFractionZolotarevFermion.h ./qcd/action/fermion/WilsonKernels.h ./qcd/action/fermion/DomainWallFermion.h ./qcd/action/fermion/OverlapWilsonPartialFractionTanhFermion.h ./qcd/action/fermion/OverlapWilsonContfracZolotarevFermion.h ./qcd/action/fermion/MobiusZolotarevFermion.h ./qcd/action/fermion/g5HermitianLinop.h ./qcd/action/fermion/OverlapWilsonCayleyTanhFermion.h ./qcd/action/fermion/WilsonFermion.h ./qcd/action/fermion/ContinuedFractionFermion5D.h ./qcd/action/fermion/OverlapWilsonCayleyZolotarevFermion.h ./qcd/spin/TwoSpinor.h ./qcd/spin/Dirac.h ./Log.h ./cshift/Cshift_common.h ./cshift/Cshift_none.h ./cshift/Cshift_mpi.h ./Simd.h ./MacroMagic.h ./Config.h ./cartesian/Cartesian_base.h ./cartesian/Cartesian_red_black.h ./cartesian/Cartesian_full.h ./AlignedAllocator.h ./Lattice.h ./Old/Tensor_poke.h ./Old/Tensor_peek.h ./Threads.h ./Grid.h ./algorithms/Preconditioner.h ./algorithms/iterative/ConjugateResidual.h ./algorithms/iterative/PrecGeneralisedConjugateResidual.h ./algorithms/iterative/ConjugateGradientMultiShift.h ./algorithms/iterative/SchurRedBlack.h ./algorithms/iterative/NormalEquations.h ./algorithms/iterative/ConjugateGradient.h ./algorithms/iterative/AdefGeneric.h ./algorithms/approx/Chebyshev.h ./algorithms/approx/Zolotarev.h ./algorithms/approx/MultiShiftFunction.h ./algorithms/approx/bigfloat.h ./algorithms/approx/bigfloat_double.h ./algorithms/approx/Remez.h ./algorithms/LinearOperator.h ./algorithms/SparseMatrix.h ./algorithms/CoarsenedMatrix.h ./stencil/Lebesgue.h
 
-CCFILES=./qcd/hmc/integrators/Integrator.cc ./qcd/hmc/HMC.cc ./qcd/utils/SpaceTimeGrid.cc ./qcd/action/fermion/WilsonKernels.cc ./qcd/action/fermion/PartialFractionFermion5D.cc ./qcd/action/fermion/CayleyFermion5D.cc ./qcd/action/fermion/WilsonKernelsHand.cc ./qcd/action/fermion/WilsonFermion.cc ./qcd/action/fermion/ContinuedFractionFermion5D.cc ./qcd/action/fermion/WilsonFermion5D.cc ./qcd/spin/Dirac.cc ./GridInit.cc ./algorithms/approx/MultiShiftFunction.cc ./algorithms/approx/Remez.cc ./algorithms/approx/Zolotarev.cc ./stencil/Lebesgue.cc ./stencil/Stencil_common.cc
+CCFILES=./qcd/hmc/integrators/Integrator.cc ./qcd/hmc/HMC.cc ./qcd/utils/SpaceTimeGrid.cc ./qcd/action/fermion/WilsonKernels.cc ./qcd/action/fermion/PartialFractionFermion5D.cc ./qcd/action/fermion/CayleyFermion5D.cc ./qcd/action/fermion/WilsonKernelsHand.cc ./qcd/action/fermion/WilsonFermion.cc ./qcd/action/fermion/ContinuedFractionFermion5D.cc ./qcd/action/fermion/WilsonFermion5D.cc ./qcd/spin/Dirac.cc ./Init.cc ./Log.cc ./algorithms/approx/MultiShiftFunction.cc ./algorithms/approx/Remez.cc ./algorithms/approx/Zolotarev.cc ./stencil/Lebesgue.cc ./stencil/Stencil_common.cc
diff --git a/lib/Timer.h b/lib/Timer.h
new file mode 100644
index 00000000..43ace890
--- /dev/null
+++ b/lib/Timer.h
@@ -0,0 +1,52 @@
+#ifndef GRID_TIME_H
+#define GRID_TIME_H
+
+#include <sys/time.h>
+#include <ctime>
+#include <chrono>
+
+namespace Grid {
+
+
+  // Dress the output; use std::chrono
+
+// C++11 time facilities better?
+double usecond(void);
+
+typedef  std::chrono::system_clock          GridClock;
+typedef  std::chrono::time_point<GridClock> GridTimePoint;
+typedef  std::chrono::milliseconds          GridTime;
+
+ 
+class GridStopWatch {
+private:
+  bool running;
+  GridTimePoint start;
+  GridTime accumulator;
+public:
+  GridStopWatch () { 
+    Reset();
+  }
+  void     Start(void) { 
+    assert(running == false);
+    start = GridClock::now(); 
+    running = true;
+  }
+  void     Stop(void)  { 
+    assert(running == true);
+    accumulator+= std::chrono::duration_cast<GridTime>(GridClock::now()-start); 
+    running = false; 
+  };
+  void     Reset(void){
+    running = false;
+    start = GridClock::now();
+    accumulator = std::chrono::duration_cast<GridTime>(start-start); 
+  }
+  GridTime Elapsed(void) {
+    assert(running == false);
+    return accumulator;
+  }
+};
+
+}
+#endif
diff --git a/lib/algorithms/CoarsenedMatrix.h b/lib/algorithms/CoarsenedMatrix.h
index 998292b2..c4b039f1 100644
--- a/lib/algorithms/CoarsenedMatrix.h
+++ b/lib/algorithms/CoarsenedMatrix.h
@@ -32,12 +32,12 @@ namespace Grid {
       displacements[2*_d]=0;
       
       //// report back
-      std::cout<<"directions    :";
+      std::cout<<GridLogMessage<<"directions    :";
       for(int d=0;d<npoint;d++) std::cout<< directions[d]<< " ";
       std::cout <<std::endl;
-      std::cout<<"displacements :";
+      std::cout<<GridLogMessage<<"displacements :";
       for(int d=0;d<npoint;d++) std::cout<< displacements[d]<< " ";
-      std::cout <<std::endl;
+      std::cout<<std::endl;
     }
   
     /*
@@ -100,9 +100,9 @@ namespace Grid {
 	  eProj._odata[ss](i)=CComplex(1.0);
 	}
 	eProj=eProj - iProj;
-	std::cout<<"Orthog check error "<<i<<" " << norm2(eProj)<<std::endl;
+	std::cout<<GridLogMessage<<"Orthog check error "<<i<<" " << norm2(eProj)<<std::endl;
       }
-      std::cout <<"CheckOrthog done"<<std::endl;
+      std::cout<<GridLogMessage <<"CheckOrthog done"<<std::endl;
     }
     void ProjectToSubspace(CoarseVector &CoarseVec,const FineField &FineVec){
       blockProject(CoarseVec,FineVec,subspace);
@@ -113,7 +113,7 @@ namespace Grid {
     void CreateSubspaceRandom(GridParallelRNG &RNG){
       for(int i=0;i<nbasis;i++){
 	random(RNG,subspace[i]);
-	std::cout<<" norm subspace["<<i<<"] "<<norm2(subspace[i])<<std::endl;
+	std::cout<<GridLogMessage<<" norm subspace["<<i<<"] "<<norm2(subspace[i])<<std::endl;
       }
       Orthogonalise();
     }
@@ -121,7 +121,7 @@ namespace Grid {
 
       RealD scale;
 
-      ConjugateGradient<FineField> CG(2.0e-3,10000);
+      ConjugateGradient<FineField> CG(1.0e-2,10000);
       FineField noise(FineGrid);
       FineField Mn(FineGrid);
 
@@ -131,7 +131,7 @@ namespace Grid {
 	scale = std::pow(norm2(noise),-0.5); 
 	noise=noise*scale;
 
-	hermop.Op(noise,Mn); std::cout << "noise   ["<<b<<"] <n|MdagM|n> "<<norm2(Mn)<<std::endl;
+	hermop.Op(noise,Mn); std::cout<<GridLogMessage << "noise   ["<<b<<"] <n|MdagM|n> "<<norm2(Mn)<<std::endl;
 
 	for(int i=0;i<1;i++){
 
@@ -143,7 +143,7 @@ namespace Grid {
 
 	}
 
-	hermop.Op(noise,Mn); std::cout << "filtered["<<b<<"] <f|MdagM|f> "<<norm2(Mn)<<std::endl;
+	hermop.Op(noise,Mn); std::cout<<GridLogMessage << "filtered["<<b<<"] <f|MdagM|f> "<<norm2(Mn)<<std::endl;
 	subspace[b]   = noise;
 
       }
@@ -189,7 +189,7 @@ namespace Grid {
       SimpleCompressor<siteVector> compressor;
       Stencil.HaloExchange(in,comm_buf,compressor);
 
-      //PARALLEL_FOR_LOOP
+PARALLEL_FOR_LOOP
       for(int ss=0;ss<Grid()->oSites();ss++){
         siteVector res = zero;
 	siteVector nbr;
@@ -252,10 +252,6 @@ namespace Grid {
 
       // Orthogonalise the subblocks over the basis
       blockOrthogonalise(InnerProd,Subspace.subspace);
-      //Subspace.Orthogonalise();
-      //      Subspace.CheckOrthogonal();
-      //Subspace.Orthogonalise();
-      //      Subspace.CheckOrthogonal();
 
       // Compute the matrix elements of linop between this orthonormal
       // set of vectors.
@@ -306,6 +302,7 @@ namespace Grid {
 	  Subspace.ProjectToSubspace(oProj,oblock);
 	  //	  blockProject(iProj,iblock,Subspace.subspace);
 	  //	  blockProject(oProj,oblock,Subspace.subspace);
+PARALLEL_FOR_LOOP
 	  for(int ss=0;ss<Grid()->oSites();ss++){
 	    for(int j=0;j<nbasis;j++){
 	      if( disp!= 0 ) {
@@ -321,12 +318,12 @@ namespace Grid {
       ///////////////////////////
       // test code worth preserving in if block
       ///////////////////////////
-      std::cout<< " Computed matrix elements "<< self_stencil <<std::endl;
+      std::cout<<GridLogMessage<< " Computed matrix elements "<< self_stencil <<std::endl;
       for(int p=0;p<geom.npoint;p++){
-	std::cout<< "A["<<p<<"]" << std::endl;
-	std::cout<< A[p] << std::endl;
+	std::cout<<GridLogMessage<< "A["<<p<<"]" << std::endl;
+	std::cout<<GridLogMessage<< A[p] << std::endl;
       }
-      std::cout<< " picking by block0 "<< self_stencil <<std::endl;
+      std::cout<<GridLogMessage<< " picking by block0 "<< self_stencil <<std::endl;
 
       phi=Subspace.subspace[0];
       std::vector<int> bc(FineGrid->_ndimension,0);
@@ -334,9 +331,9 @@ namespace Grid {
       blockPick(Grid(),phi,tmp,bc);      // Pick out a block
       linop.Op(tmp,Mphi);                // Apply big dop
       blockProject(iProj,Mphi,Subspace.subspace); // project it and print it
-      std::cout<< " Computed matrix elements from block zero only "<<std::endl;
-      std::cout<< iProj <<std::endl;
-      std::cout<<"Computed Coarse Operator"<<std::endl;
+      std::cout<<GridLogMessage<< " Computed matrix elements from block zero only "<<std::endl;
+      std::cout<<GridLogMessage<< iProj <<std::endl;
+      std::cout<<GridLogMessage<<"Computed Coarse Operator"<<std::endl;
 #endif
       //      ForceHermitian();
       AssertHermitian();
@@ -345,9 +342,9 @@ namespace Grid {
     void ForceDiagonal(void) {
 
 
-      std::cout<<"**************************************************"<<std::endl;
-      std::cout<<"****   Forcing coarse operator to be diagonal ****"<<std::endl;
-      std::cout<<"**************************************************"<<std::endl;
+      std::cout<<GridLogMessage<<"**************************************************"<<std::endl;
+      std::cout<<GridLogMessage<<"****   Forcing coarse operator to be diagonal ****"<<std::endl;
+      std::cout<<GridLogMessage<<"**************************************************"<<std::endl;
       for(int p=0;p<8;p++){
 	A[p]=zero;
       }
@@ -387,13 +384,13 @@ namespace Grid {
 	
 	Diff = AA - adj(AAc);
 
-	std::cout<<"Norm diff dim "<<d<<" "<< norm2(Diff)<<std::endl;
-	std::cout<<"Norm dim "<<d<<" "<< norm2(AA)<<std::endl;
+	std::cout<<GridLogMessage<<"Norm diff dim "<<d<<" "<< norm2(Diff)<<std::endl;
+	std::cout<<GridLogMessage<<"Norm dim "<<d<<" "<< norm2(AA)<<std::endl;
 	  
       }
       Diff = A[8] - adj(A[8]);
-      std::cout<<"Norm diff local "<< norm2(Diff)<<std::endl;
-      std::cout<<"Norm local "<< norm2(A[8])<<std::endl;
+      std::cout<<GridLogMessage<<"Norm diff local "<< norm2(Diff)<<std::endl;
+      std::cout<<GridLogMessage<<"Norm local "<< norm2(A[8])<<std::endl;
     }
     
   };
diff --git a/lib/algorithms/approx/Chebyshev.h b/lib/algorithms/approx/Chebyshev.h
index f629ab08..e4fa0b61 100644
--- a/lib/algorithms/approx/Chebyshev.h
+++ b/lib/algorithms/approx/Chebyshev.h
@@ -120,7 +120,7 @@ namespace Grid {
       Field *Tn  = &T1;
       Field *Tnp = &T2;
 
-      std::cout << "Chebyshev ["<<lo<<","<<hi<<"]"<< " order "<<order <<std::endl;
+      std::cout<<GridLogMessage << "Chebyshev ["<<lo<<","<<hi<<"]"<< " order "<<order <<std::endl;
       // Tn=T1 = (xscale M + mscale)in
       double xscale = 2.0/(hi-lo);
       double mscale = -(hi+lo)/(hi-lo);
diff --git a/lib/algorithms/approx/MultiShiftFunction.h b/lib/algorithms/approx/MultiShiftFunction.h
index 4136736e..fceed2eb 100644
--- a/lib/algorithms/approx/MultiShiftFunction.h
+++ b/lib/algorithms/approx/MultiShiftFunction.h
@@ -1,6 +1,8 @@
 #ifndef MULTI_SHIFT_FUNCTION
 #define MULTI_SHIFT_FUNCTION
+
 namespace Grid {
+
 class MultiShiftFunction {
 public:
   int order;
diff --git a/lib/algorithms/approx/Remez.cc b/lib/algorithms/approx/Remez.cc
index 4ec2270d..38d60088 100644
--- a/lib/algorithms/approx/Remez.cc
+++ b/lib/algorithms/approx/Remez.cc
@@ -757,3 +757,4 @@ void AlgRemez::csv(std::ostream & os)
   }
   return;
 }
+
diff --git a/lib/algorithms/approx/Remez.h b/lib/algorithms/approx/Remez.h
index 336382d2..f88c5675 100644
--- a/lib/algorithms/approx/Remez.h
+++ b/lib/algorithms/approx/Remez.h
@@ -15,6 +15,8 @@
 #ifndef INCLUDED_ALG_REMEZ_H
 #define INCLUDED_ALG_REMEZ_H
 
+#include <stddef.h>
+
 #include <algorithms/approx/bigfloat.h>
 
 #define JMAX 10000 //Maximum number of iterations of Newton's approximation
@@ -28,6 +30,7 @@
   remez.getIPFE(res,pole,&norm);
   remez.csv(ostream &os);
  */
+
 class AlgRemez
 {
  private:
diff --git a/lib/algorithms/iterative/AdefGeneric.h b/lib/algorithms/iterative/AdefGeneric.h
index 0d1c3d91..6b839d7a 100644
--- a/lib/algorithms/iterative/AdefGeneric.h
+++ b/lib/algorithms/iterative/AdefGeneric.h
@@ -149,7 +149,7 @@ class TwoLevelFlexiblePcg : public LinearFunction<Field>
       }
 
       RealD rrn=sqrt(rn/ssq);
-      std::cout<<"TwoLevelfPcg: k= "<<k<<" residual = "<<rrn<<std::endl;
+      std::cout<<GridLogMessage<<"TwoLevelfPcg: k= "<<k<<" residual = "<<rrn<<std::endl;
 
       // Stopping condition
       if ( rn <= rsq ) { 
@@ -161,8 +161,8 @@ class TwoLevelFlexiblePcg : public LinearFunction<Field>
 	RealD srcnorm = sqrt(norm2(src));
 	RealD tmpnorm = sqrt(norm2(tmp));
 	RealD true_residual = tmpnorm/srcnorm;
-	std::cout<<"TwoLevelfPcg:   true residual is "<<true_residual<<std::endl;
-	std::cout<<"TwoLevelfPcg: target residual was"<<Tolerance<<std::endl;
+	std::cout<<GridLogMessage<<"TwoLevelfPcg:   true residual is "<<true_residual<<std::endl;
+	std::cout<<GridLogMessage<<"TwoLevelfPcg: target residual was"<<Tolerance<<std::endl;
 	return k;
       }
     }
diff --git a/lib/algorithms/iterative/ConjugateGradient.h b/lib/algorithms/iterative/ConjugateGradient.h
index e281d765..b919dd91 100644
--- a/lib/algorithms/iterative/ConjugateGradient.h
+++ b/lib/algorithms/iterative/ConjugateGradient.h
@@ -13,9 +13,7 @@ namespace Grid {
 public:                                                
     RealD   Tolerance;
     Integer MaxIterations;
-    int verbose;
     ConjugateGradient(RealD tol,Integer maxit) : Tolerance(tol), MaxIterations(maxit) { 
-      verbose=0;
     };
 
 
@@ -42,14 +40,12 @@ public:
       cp =a;
       ssq=norm2(src);
 
-      if ( verbose ) {
-	std::cout <<std::setprecision(4)<< "ConjugateGradient: guess "<<guess<<std::endl;
-	std::cout <<std::setprecision(4)<< "ConjugateGradient:   src "<<ssq  <<std::endl;
-	std::cout <<std::setprecision(4)<< "ConjugateGradient:    mp "<<d    <<std::endl;
-	std::cout <<std::setprecision(4)<< "ConjugateGradient:   mmp "<<b    <<std::endl;
-	std::cout <<std::setprecision(4)<< "ConjugateGradient:  cp,r "<<cp   <<std::endl;
-	std::cout <<std::setprecision(4)<< "ConjugateGradient:     p "<<a    <<std::endl;
-      }
+      std::cout<<GridLogIterative <<std::setprecision(4)<< "ConjugateGradient: guess "<<guess<<std::endl;
+      std::cout<<GridLogIterative <<std::setprecision(4)<< "ConjugateGradient:   src "<<ssq  <<std::endl;
+      std::cout<<GridLogIterative <<std::setprecision(4)<< "ConjugateGradient:    mp "<<d    <<std::endl;
+      std::cout<<GridLogIterative <<std::setprecision(4)<< "ConjugateGradient:   mmp "<<b    <<std::endl;
+      std::cout<<GridLogIterative <<std::setprecision(4)<< "ConjugateGradient:  cp,r "<<cp   <<std::endl;
+      std::cout<<GridLogIterative <<std::setprecision(4)<< "ConjugateGradient:     p "<<a    <<std::endl;
 
       RealD rsq =  Tolerance* Tolerance*ssq;
       
@@ -58,7 +54,7 @@ public:
 	return;
       }
       
-      if(verbose) std::cout << std::setprecision(4)<< "ConjugateGradient: k=0 residual "<<cp<<" rsq"<<rsq<<std::endl;
+      std::cout<<GridLogIterative << std::setprecision(4)<< "ConjugateGradient: k=0 residual "<<cp<<" rsq"<<rsq<<std::endl;
       
       int k;
       for (k=1;k<=MaxIterations;k++){
@@ -80,7 +76,7 @@ public:
 	psi= a*p+psi;
 	p  = p*b+r;
 	  
-	if (verbose) std::cout<<"ConjugateGradient: Iteration " <<k<<" residual "<<cp<< " target"<< rsq<<std::endl;
+	std::cout<<GridLogIterative<<"ConjugateGradient: Iteration " <<k<<" residual "<<cp<< " target"<< rsq<<std::endl;
 	
 	// Stopping condition
 	if ( cp <= rsq ) { 
@@ -94,14 +90,14 @@ public:
 	  RealD resnorm = sqrt(norm2(p));
 	  RealD true_residual = resnorm/srcnorm;
 
-	  std::cout<<"ConjugateGradient: Converged on iteration " <<k
+	  std::cout<<GridLogMessage<<"ConjugateGradient: Converged on iteration " <<k
 		   <<" computed residual "<<sqrt(cp/ssq)
 		   <<" true residual     "<<true_residual
 		   <<" target "<<Tolerance<<std::endl;
 	  return;
 	}
       }
-      std::cout<<"ConjugateGradient did NOT converge"<<std::endl;
+      std::cout<<GridLogMessage<<"ConjugateGradient did NOT converge"<<std::endl;
       assert(0);
     }
   };
diff --git a/lib/algorithms/iterative/ConjugateGradientMultiShift.h b/lib/algorithms/iterative/ConjugateGradientMultiShift.h
index 580ab838..c6e7650d 100644
--- a/lib/algorithms/iterative/ConjugateGradientMultiShift.h
+++ b/lib/algorithms/iterative/ConjugateGradientMultiShift.h
@@ -91,7 +91,7 @@ void operator() (LinearOperatorBase<Field> &Linop, const Field &src, std::vector
   cp = norm2(src);
   for(int s=0;s<nshift;s++){
     rsq[s] = cp * mresidual[s] * mresidual[s];
-    std::cout<<"ConjugateGradientMultiShift: shift "<<s
+    std::cout<<GridLogMessage<<"ConjugateGradientMultiShift: shift "<<s
 	     <<" target resid "<<rsq[s]<<std::endl;
     ps[s] = src;
   }
@@ -109,7 +109,7 @@ void operator() (LinearOperatorBase<Field> &Linop, const Field &src, std::vector
   // p and mmp is equal to d after this since
   // the d computation is tricky
   //  qq = real(innerProduct(p,mmp));
-  //  std::cout << "debug equal ?  qq "<<qq<<" d "<< d<<std::endl;
+  //  std::cout<<GridLogMessage << "debug equal ?  qq "<<qq<<" d "<< d<<std::endl;
   
   b = -cp /d;
   
@@ -214,7 +214,7 @@ void operator() (LinearOperatorBase<Field> &Linop, const Field &src, std::vector
 	
 	if(css<rsq[s]){
 	  if ( ! converged[s] )
-	    std::cout<<"ConjugateGradientMultiShift k="<<k<<" Shift "<<s<<" has converged"<<std::endl;
+	    std::cout<<GridLogMessage<<"ConjugateGradientMultiShift k="<<k<<" Shift "<<s<<" has converged"<<std::endl;
 	      converged[s]=1;
 	} else {
 	  all_converged=0;
@@ -225,8 +225,8 @@ void operator() (LinearOperatorBase<Field> &Linop, const Field &src, std::vector
     
     if ( all_converged ){
 
-      std::cout<< "CGMultiShift: All shifts have converged iteration "<<k<<std::endl;
-      std::cout<< "CGMultiShift: Checking solutions"<<std::endl;
+      std::cout<<GridLogMessage<< "CGMultiShift: All shifts have converged iteration "<<k<<std::endl;
+      std::cout<<GridLogMessage<< "CGMultiShift: Checking solutions"<<std::endl;
       
       // Check answers 
       for(int s=0; s < nshift; s++) { 
@@ -235,13 +235,13 @@ void operator() (LinearOperatorBase<Field> &Linop, const Field &src, std::vector
 	axpy(r,-alpha[s],src,tmp);
 	RealD rn = norm2(r);
 	RealD cn = norm2(src);
-	std::cout<<"CGMultiShift: shift["<<s<<"] true residual "<<std::sqrt(rn/cn)<<std::endl;
+	std::cout<<GridLogMessage<<"CGMultiShift: shift["<<s<<"] true residual "<<std::sqrt(rn/cn)<<std::endl;
       }
       return;
     }
   }
   // ugly hack
-  std::cout<<"CG multi shift did not converge"<<std::endl;
+  std::cout<<GridLogMessage<<"CG multi shift did not converge"<<std::endl;
   assert(0);
 }
 
diff --git a/lib/algorithms/iterative/ConjugateResidual.h b/lib/algorithms/iterative/ConjugateResidual.h
index 197a0b15..1c3c6a91 100644
--- a/lib/algorithms/iterative/ConjugateResidual.h
+++ b/lib/algorithms/iterative/ConjugateResidual.h
@@ -41,7 +41,7 @@ namespace Grid {
       ssq=norm2(src);
       rsq=Tolerance*Tolerance*ssq;
 
-      if (verbose) std::cout<<"ConjugateResidual: iteration " <<0<<" residual "<<cp<< " target"<< rsq<<std::endl;
+      if (verbose) std::cout<<GridLogMessage<<"ConjugateResidual: iteration " <<0<<" residual "<<cp<< " target"<< rsq<<std::endl;
 
       for(int k=1;k<MaxIterations;k++){
 
@@ -60,13 +60,13 @@ namespace Grid {
 	axpy(p,b,p,r);
 	pAAp=axpy_norm(Ap,b,Ap,Ar);
 	
-	if(verbose) std::cout<<"ConjugateResidual: iteration " <<k<<" residual "<<cp<< " target"<< rsq<<std::endl;
+	if(verbose) std::cout<<GridLogMessage<<"ConjugateResidual: iteration " <<k<<" residual "<<cp<< " target"<< rsq<<std::endl;
 
 	if(cp<rsq) {
 	  Linop.HermOp(psi,Ap);
 	  axpy(r,-1.0,src,Ap);
 	  RealD true_resid = norm2(r)/ssq;
-	  std::cout<<"ConjugateResidual: Converged on iteration " <<k
+	  std::cout<<GridLogMessage<<"ConjugateResidual: Converged on iteration " <<k
 		   << " computed residual "<<sqrt(cp/ssq)
 	           << " true residual "<<sqrt(true_resid)
 	           << " target "       <<Tolerance <<std::endl;
@@ -75,7 +75,7 @@ namespace Grid {
 
       }
 
-      std::cout<<"ConjugateResidual did NOT converge"<<std::endl;
+      std::cout<<GridLogMessage<<"ConjugateResidual did NOT converge"<<std::endl;
       assert(0);
     }
   };
diff --git a/lib/algorithms/iterative/PrecGeneralisedConjugateResidual.h b/lib/algorithms/iterative/PrecGeneralisedConjugateResidual.h
index fee41fa8..3a9bb731 100644
--- a/lib/algorithms/iterative/PrecGeneralisedConjugateResidual.h
+++ b/lib/algorithms/iterative/PrecGeneralisedConjugateResidual.h
@@ -45,13 +45,13 @@ namespace Grid {
 
 	cp=GCRnStep(Linop,src,psi,rsq);
 
-	if ( verbose ) std::cout<<"VPGCR("<<mmax<<","<<nstep<<") "<< steps <<" steps cp = "<<cp<<std::endl;
+	if ( verbose ) std::cout<<GridLogMessage<<"VPGCR("<<mmax<<","<<nstep<<") "<< steps <<" steps cp = "<<cp<<std::endl;
 
 	if(cp<rsq) {
 	  Linop.HermOp(psi,r);
 	  axpy(r,-1.0,src,r);
 	  RealD tr = norm2(r);
-	  std::cout<<"PrecGeneralisedConjugateResidual: Converged on iteration " <<steps
+	  std::cout<<GridLogMessage<<"PrecGeneralisedConjugateResidual: Converged on iteration " <<steps
 		   << " computed residual "<<sqrt(cp/ssq)
 	           << " true residual "    <<sqrt(tr/ssq)
 	           << " target "           <<Tolerance <<std::endl;
@@ -59,7 +59,7 @@ namespace Grid {
 	}
 
       }
-      std::cout<<"Variable Preconditioned GCR did not converge"<<std::endl;
+      std::cout<<GridLogMessage<<"Variable Preconditioned GCR did not converge"<<std::endl;
       assert(0);
     }
     RealD GCRnStep(LinearOperatorBase<Field> &Linop,const Field &src, Field &psi,RealD rsq){
@@ -96,21 +96,21 @@ namespace Grid {
       /////////////////////
       Preconditioner(r,z);
 
-      std::cout<< " Preconditioner in " << norm2(r)<<std::endl; 
-      std::cout<< " Preconditioner out " << norm2(z)<<std::endl; 
+      std::cout<<GridLogMessage<< " Preconditioner in " << norm2(r)<<std::endl; 
+      std::cout<<GridLogMessage<< " Preconditioner out " << norm2(z)<<std::endl; 
       
       Linop.HermOp(z,tmp); 
 
-      std::cout<< " Preconditioner Aout " << norm2(tmp)<<std::endl; 
+      std::cout<<GridLogMessage<< " Preconditioner Aout " << norm2(tmp)<<std::endl; 
       ttmp=tmp;
       tmp=tmp-r;
 
-      std::cout<< " Preconditioner resid " << std::sqrt(norm2(tmp)/norm2(r))<<std::endl; 
+      std::cout<<GridLogMessage<< " Preconditioner resid " << std::sqrt(norm2(tmp)/norm2(r))<<std::endl; 
       /*
-      std::cout<<r<<std::endl;
-      std::cout<<z<<std::endl;
-      std::cout<<ttmp<<std::endl;
-      std::cout<<tmp<<std::endl;
+      std::cout<<GridLogMessage<<r<<std::endl;
+      std::cout<<GridLogMessage<<z<<std::endl;
+      std::cout<<GridLogMessage<<ttmp<<std::endl;
+      std::cout<<GridLogMessage<<tmp<<std::endl;
       */
 
       Linop.HermOpAndNorm(z,Az,zAz,zAAz); 
@@ -137,7 +137,7 @@ namespace Grid {
 
 	cp = axpy_norm(r,-a,q[peri_k],r);  
 
-	std::cout<< " VPGCR_step resid" <<sqrt(cp/rsq)<<std::endl; 
+	std::cout<<GridLogMessage<< " VPGCR_step resid" <<sqrt(cp/rsq)<<std::endl; 
 	if((k==nstep-1)||(cp<rsq)){
 	  return cp;
 	}
@@ -148,7 +148,7 @@ namespace Grid {
 
 	Linop.HermOp(z,tmp);
         tmp=tmp-r;
-	std::cout<< " Preconditioner resid" <<sqrt(norm2(tmp)/norm2(r))<<std::endl; 
+	std::cout<<GridLogMessage<< " Preconditioner resid" <<sqrt(norm2(tmp)/norm2(r))<<std::endl; 
 
 	q[peri_kp]=Az;
 	p[peri_kp]=z;
diff --git a/lib/algorithms/iterative/SchurRedBlack.h b/lib/algorithms/iterative/SchurRedBlack.h
index 3ebc5f1d..b023fb7b 100644
--- a/lib/algorithms/iterative/SchurRedBlack.h
+++ b/lib/algorithms/iterative/SchurRedBlack.h
@@ -89,7 +89,7 @@ namespace Grid {
       //////////////////////////////////////////////////////////////
       // Call the red-black solver
       //////////////////////////////////////////////////////////////
-      std::cout << "SchurRedBlack solver calling the MpcDagMp solver" <<std::endl;
+      std::cout<<GridLogMessage << "SchurRedBlack solver calling the MpcDagMp solver" <<std::endl;
       _HermitianRBSolver(_HermOpEO,src_o,sol_o);  assert(sol_o.checkerboard==Odd);
 
       ///////////////////////////////////////////////////
@@ -108,7 +108,7 @@ namespace Grid {
       RealD ns = norm2(in);
       RealD nr = norm2(resid);
 
-      std::cout << "SchurRedBlackDiagMooee solver true unprec resid "<< std::sqrt(nr/ns) <<" nr "<< nr <<" ns "<<ns << std::endl;
+      std::cout<<GridLogMessage << "SchurRedBlackDiagMooee solver true unprec resid "<< std::sqrt(nr/ns) <<" nr "<< nr <<" ns "<<ns << std::endl;
     }     
   };
 
diff --git a/lib/lattice/Lattice_ET.h b/lib/lattice/Lattice_ET.h
index cbcca2ce..a7910747 100644
--- a/lib/lattice/Lattice_ET.h
+++ b/lib/lattice/Lattice_ET.h
@@ -132,18 +132,18 @@ inline void CBFromExpression(int &cb,const T1& lat)   // Lattice leaf
     assert(cb==lat.checkerboard);
   } 
   cb=lat.checkerboard;
-  //  std::cout<<"Lattice leaf cb "<<cb<<std::endl;
+  //  std::cout<<GridLogMessage<<"Lattice leaf cb "<<cb<<std::endl;
 }
 template<class T1,typename std::enable_if<!is_lattice<T1>::value, T1>::type * = nullptr >
 inline void CBFromExpression(int &cb,const T1& notlat)   // non-lattice leaf
 {
-  //  std::cout<<"Non lattice leaf cb"<<cb<<std::endl;
+  //  std::cout<<GridLogMessage<<"Non lattice leaf cb"<<cb<<std::endl;
 }
 template <typename Op, typename T1>
 inline void CBFromExpression(int &cb,const LatticeUnaryExpression<Op,T1 > &expr)
 {
   CBFromExpression(cb,std::get<0>(expr.second));// recurse 
-  //  std::cout<<"Unary node cb "<<cb<<std::endl;
+  //  std::cout<<GridLogMessage<<"Unary node cb "<<cb<<std::endl;
 }
 
 template <typename Op, typename T1, typename T2>
@@ -151,7 +151,7 @@ inline void CBFromExpression(int &cb,const LatticeBinaryExpression<Op,T1,T2> &ex
 {
   CBFromExpression(cb,std::get<0>(expr.second));// recurse
   CBFromExpression(cb,std::get<1>(expr.second));
-  //  std::cout<<"Binary node cb "<<cb<<std::endl;
+  //  std::cout<<GridLogMessage<<"Binary node cb "<<cb<<std::endl;
 }
 template <typename Op, typename T1, typename T2, typename T3>
 inline void CBFromExpression( int &cb,const LatticeTrinaryExpression<Op,T1,T2,T3 > &expr) 
@@ -159,7 +159,7 @@ inline void CBFromExpression( int &cb,const LatticeTrinaryExpression<Op,T1,T2,T3
   CBFromExpression(cb,std::get<0>(expr.second));// recurse
   CBFromExpression(cb,std::get<1>(expr.second));
   CBFromExpression(cb,std::get<2>(expr.second));
-  //  std::cout<<"Trinary node cb "<<cb<<std::endl;
+  //  std::cout<<GridLogMessage<<"Trinary node cb "<<cb<<std::endl;
 }
 
 ////////////////////////////////////////////
@@ -370,7 +370,7 @@ using namespace Grid;
    tmp.func(eval(0,v1),eval(0,v2));
 
    auto var = v1+v2;
-   std::cout<<typeid(var).name()<<std::endl;
+   std::cout<<GridLogMessage<<typeid(var).name()<<std::endl;
 
    v3=v1+v2;
    v3=v1+v2+v1*v2;
diff --git a/lib/lattice/Lattice_base.h b/lib/lattice/Lattice_base.h
index 0503732c..9c9ef044 100644
--- a/lib/lattice/Lattice_base.h
+++ b/lib/lattice/Lattice_base.h
@@ -60,6 +60,11 @@ public:
     GridBase *_grid;
     int checkerboard;
     std::vector<vobj,alignedAllocator<vobj> > _odata;
+    
+    // to pthread need a computable loop where loop induction is not required
+    int begin(void) { return 0;};
+    int end(void)   { return _odata.size(); }
+    vobj & operator[](int i) { return _odata[i]; };
 
 public:
     typedef typename vobj::scalar_type scalar_type;
@@ -221,7 +226,7 @@ PARALLEL_FOR_LOOP
     template<class robj> strong_inline Lattice<vobj> & operator = (const Lattice<robj> & r){
       this->checkerboard = r.checkerboard;
       conformable(*this,r);
-      std::cout<<"Lattice operator ="<<std::endl;
+      std::cout<<GridLogMessage<<"Lattice operator ="<<std::endl;
 PARALLEL_FOR_LOOP
         for(int ss=0;ss<_grid->oSites();ss++){
             this->_odata[ss]=r._odata[ss];
diff --git a/lib/lattice/Lattice_reduction.h b/lib/lattice/Lattice_reduction.h
index 922821ed..e6287b41 100644
--- a/lib/lattice/Lattice_reduction.h
+++ b/lib/lattice/Lattice_reduction.h
@@ -125,7 +125,7 @@ template<class vobj> inline void sliceSum(const Lattice<vobj> &Data,std::vector<
   assert(grid!=NULL);
 
   // FIXME
-  std::cout<<"WARNING ! SliceSum is unthreaded "<<grid->SumArraySize()<<" threads "<<std::endl;
+  std::cout<<GridLogMessage<<"WARNING ! SliceSum is unthreaded "<<grid->SumArraySize()<<" threads "<<std::endl;
 
   const int    Nd = grid->_ndimension;
   const int Nsimd = grid->Nsimd();
diff --git a/lib/lattice/Lattice_rng.h b/lib/lattice/Lattice_rng.h
index f3cc0ddc..55246016 100644
--- a/lib/lattice/Lattice_rng.h
+++ b/lib/lattice/Lattice_rng.h
@@ -233,7 +233,8 @@ namespace Grid {
       int words=sizeof(scalar_object)/sizeof(scalar_type);
 
       std::vector<scalar_object> buf(Nsimd);
-      
+
+PARALLEL_FOR_LOOP
       for(int ss=0;ss<osites;ss++){
 	for(int si=0;si<Nsimd;si++){
 
diff --git a/lib/lattice/Lattice_transfer.h b/lib/lattice/Lattice_transfer.h
index ac628faa..b5c43e43 100644
--- a/lib/lattice/Lattice_transfer.h
+++ b/lib/lattice/Lattice_transfer.h
@@ -23,7 +23,7 @@ inline void subdivides(GridBase *coarse,GridBase *fine)
   template<class vobj> inline void pickCheckerboard(int cb,Lattice<vobj> &half,const Lattice<vobj> &full){
     half.checkerboard = cb;
     int ssh=0;
-PARALLEL_FOR_LOOP
+    //PARALLEL_FOR_LOOP
     for(int ss=0;ss<full._grid->oSites();ss++){
       std::vector<int> coor;
       int cbos;
@@ -40,7 +40,7 @@ PARALLEL_FOR_LOOP
   template<class vobj> inline void setCheckerboard(Lattice<vobj> &full,const Lattice<vobj> &half){
     int cb = half.checkerboard;
     int ssh=0;
-PARALLEL_FOR_LOOP
+    //PARALLEL_FOR_LOOP
     for(int ss=0;ss<full._grid->oSites();ss++){
       std::vector<int> coor;
       int cbos;
@@ -158,6 +158,7 @@ template<class vobj,class CComplex>
 
   fine_inner = localInnerProduct(fineX,fineY);
   blockSum(coarse_inner,fine_inner);
+PARALLEL_FOR_LOOP
   for(int ss=0;ss<coarse->oSites();ss++){
     CoarseInner._odata[ss] = coarse_inner._odata[ss];
   }
diff --git a/lib/pugixml/README.md b/lib/pugixml/README.md
new file mode 100644
index 00000000..9d8a935f
--- /dev/null
+++ b/lib/pugixml/README.md
@@ -0,0 +1,44 @@
+pugixml [![Build Status](https://travis-ci.org/zeux/pugixml.svg?branch=master)](https://travis-ci.org/zeux/pugixml) [![Build status](https://ci.appveyor.com/api/projects/status/9hdks1doqvq8pwe7/branch/master?svg=true)](https://ci.appveyor.com/project/zeux/pugixml)
+=======
+
+pugixml is a C++ XML processing library, which consists of a DOM-like interface with rich traversal/modification
+capabilities, an extremely fast XML parser which constructs the DOM tree from an XML file/buffer, and an XPath 1.0
+implementation for complex data-driven tree queries. Full Unicode support is also available, with Unicode interface
+variants and conversions between different Unicode encodings (which happen automatically during parsing/saving).
+
+pugixml is used by a lot of projects, both open-source and proprietary, for performance and easy-to-use interface.
+
+## Documentation
+
+Documentation for the current release of pugixml is available on-line as two separate documents:
+
+* [Quick-start guide](http://pugixml.org/docs/quickstart.html), that aims to provide enough information to start using the library;
+* [Complete reference manual](http://pugixml.org/docs/manual.html), that describes all features of the library in detail.
+
+You’re advised to start with the quick-start guide; however, many important library features are either not described in it at all or only mentioned briefly; if you require more information you should read the complete manual.
+
+## License
+This library is available to anybody free of charge, under the terms of MIT License:
+
+Copyright (c) 2006-2015 Arseny Kapoulkine
+
+Permission is hereby granted, free of charge, to any person
+obtaining a copy of this software and associated documentation
+files (the "Software"), to deal in the Software without
+restriction, including without limitation the rights to use,
+copy, modify, merge, publish, distribute, sublicense, and/or sell
+copies of the Software, and to permit persons to whom the
+Software is furnished to do so, subject to the following
+conditions:
+
+The above copyright notice and this permission notice shall be
+included in all copies or substantial portions of the Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES
+OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT
+HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY,
+WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
+FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+OTHER DEALINGS IN THE SOFTWARE.
diff --git a/lib/pugixml/pugiconfig.hpp b/lib/pugixml/pugiconfig.hpp
new file mode 100644
index 00000000..5ee5131f
--- /dev/null
+++ b/lib/pugixml/pugiconfig.hpp
@@ -0,0 +1,71 @@
+/**
+ * pugixml parser - version 1.6
+ * --------------------------------------------------------
+ * Copyright (C) 2006-2015, by Arseny Kapoulkine (arseny.kapoulkine@gmail.com)
+ * Report bugs and download new versions at http://pugixml.org/
+ *
+ * This library is distributed under the MIT License. See notice at the end
+ * of this file.
+ *
+ * This work is based on the pugxml parser, which is:
+ * Copyright (C) 2003, by Kristen Wegner (kristen@tima.net)
+ */
+
+#ifndef HEADER_PUGICONFIG_HPP
+#define HEADER_PUGICONFIG_HPP
+
+// Uncomment this to enable wchar_t mode
+// #define PUGIXML_WCHAR_MODE
+
+// Uncomment this to disable XPath
+// #define PUGIXML_NO_XPATH
+
+// Uncomment this to disable STL
+// #define PUGIXML_NO_STL
+
+// Uncomment this to disable exceptions
+// #define PUGIXML_NO_EXCEPTIONS
+
+// Set this to control attributes for public classes/functions, i.e.:
+// #define PUGIXML_API __declspec(dllexport) // to export all public symbols from DLL
+// #define PUGIXML_CLASS __declspec(dllimport) // to import all classes from DLL
+// #define PUGIXML_FUNCTION __fastcall // to set calling conventions to all public functions to fastcall
+// In absence of PUGIXML_CLASS/PUGIXML_FUNCTION definitions PUGIXML_API is used instead
+
+// Tune these constants to adjust memory-related behavior
+// #define PUGIXML_MEMORY_PAGE_SIZE 32768
+// #define PUGIXML_MEMORY_OUTPUT_STACK 10240
+// #define PUGIXML_MEMORY_XPATH_PAGE_SIZE 4096
+
+// Uncomment this to switch to header-only version
+// #define PUGIXML_HEADER_ONLY
+
+// Uncomment this to enable long long support
+// #define PUGIXML_HAS_LONG_LONG
+
+#endif
+
+/**
+ * Copyright (c) 2006-2015 Arseny Kapoulkine
+ *
+ * Permission is hereby granted, free of charge, to any person
+ * obtaining a copy of this software and associated documentation
+ * files (the "Software"), to deal in the Software without
+ * restriction, including without limitation the rights to use,
+ * copy, modify, merge, publish, distribute, sublicense, and/or sell
+ * copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following
+ * conditions:
+ *
+ * The above copyright notice and this permission notice shall be
+ * included in all copies or substantial portions of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES
+ * OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT
+ * HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY,
+ * WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
+ * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ */
diff --git a/lib/pugixml/pugixml.cpp b/lib/pugixml/pugixml.cpp
new file mode 100644
index 00000000..07f3a336
--- /dev/null
+++ b/lib/pugixml/pugixml.cpp
@@ -0,0 +1,12485 @@
+/**
+ * pugixml parser - version 1.6
+ * --------------------------------------------------------
+ * Copyright (C) 2006-2015, by Arseny Kapoulkine (arseny.kapoulkine@gmail.com)
+ * Report bugs and download new versions at http://pugixml.org/
+ *
+ * This library is distributed under the MIT License. See notice at the end
+ * of this file.
+ *
+ * This work is based on the pugxml parser, which is:
+ * Copyright (C) 2003, by Kristen Wegner (kristen@tima.net)
+ */
+
+#ifndef SOURCE_PUGIXML_CPP
+#define SOURCE_PUGIXML_CPP
+
+#include "pugixml.hpp"
+
+#include <stdlib.h>
+#include <stdio.h>
+#include <string.h>
+#include <assert.h>
+
+#ifdef PUGIXML_WCHAR_MODE
+#	include <wchar.h>
+#endif
+
+#ifndef PUGIXML_NO_XPATH
+#	include <math.h>
+#	include <float.h>
+#	ifdef PUGIXML_NO_EXCEPTIONS
+#		include <setjmp.h>
+#	endif
+#endif
+
+#ifndef PUGIXML_NO_STL
+#	include <istream>
+#	include <ostream>
+#	include <string>
+#endif
+
+// For placement new
+#include <new>
+
+#ifdef _MSC_VER
+#	pragma warning(push)
+#	pragma warning(disable: 4127) // conditional expression is constant
+#	pragma warning(disable: 4324) // structure was padded due to __declspec(align())
+#	pragma warning(disable: 4611) // interaction between '_setjmp' and C++ object destruction is non-portable
+#	pragma warning(disable: 4702) // unreachable code
+#	pragma warning(disable: 4996) // this function or variable may be unsafe
+#	pragma warning(disable: 4793) // function compiled as native: presence of '_setjmp' makes a function unmanaged
+#endif
+
+#ifdef __INTEL_COMPILER
+#	pragma warning(disable: 177) // function was declared but never referenced 
+#	pragma warning(disable: 279) // controlling expression is constant
+#	pragma warning(disable: 1478 1786) // function was declared "deprecated"
+#	pragma warning(disable: 1684) // conversion from pointer to same-sized integral type
+#endif
+
+#if defined(__BORLANDC__) && defined(PUGIXML_HEADER_ONLY)
+#	pragma warn -8080 // symbol is declared but never used; disabling this inside push/pop bracket does not make the warning go away
+#endif
+
+#ifdef __BORLANDC__
+#	pragma option push
+#	pragma warn -8008 // condition is always false
+#	pragma warn -8066 // unreachable code
+#endif
+
+#ifdef __SNC__
+// Using diag_push/diag_pop does not disable the warnings inside templates due to a compiler bug
+#	pragma diag_suppress=178 // function was declared but never referenced
+#	pragma diag_suppress=237 // controlling expression is constant
+#endif
+
+// Inlining controls
+#if defined(_MSC_VER) && _MSC_VER >= 1300
+#	define PUGI__NO_INLINE __declspec(noinline)
+#elif defined(__GNUC__)
+#	define PUGI__NO_INLINE __attribute__((noinline))
+#else
+#	define PUGI__NO_INLINE 
+#endif
+
+// Branch weight controls
+#if defined(__GNUC__)
+#	define PUGI__UNLIKELY(cond) __builtin_expect(cond, 0)
+#else
+#	define PUGI__UNLIKELY(cond) (cond)
+#endif
+
+// Simple static assertion
+#define PUGI__STATIC_ASSERT(cond) { static const char condition_failed[(cond) ? 1 : -1] = {0}; (void)condition_failed[0]; }
+
+// Digital Mars C++ bug workaround for passing char loaded from memory via stack
+#ifdef __DMC__
+#	define PUGI__DMC_VOLATILE volatile
+#else
+#	define PUGI__DMC_VOLATILE
+#endif
+
+// Borland C++ bug workaround for not defining ::memcpy depending on header include order (can't always use std::memcpy because some compilers don't have it at all)
+#if defined(__BORLANDC__) && !defined(__MEM_H_USING_LIST)
+using std::memcpy;
+using std::memmove;
+#endif
+
+// In some environments MSVC is a compiler but the CRT lacks certain MSVC-specific features
+#if defined(_MSC_VER) && !defined(__S3E__)
+#	define PUGI__MSVC_CRT_VERSION _MSC_VER
+#endif
+
+#ifdef PUGIXML_HEADER_ONLY
+#	define PUGI__NS_BEGIN namespace pugi { namespace impl {
+#	define PUGI__NS_END } }
+#	define PUGI__FN inline
+#	define PUGI__FN_NO_INLINE inline
+#else
+#	if defined(_MSC_VER) && _MSC_VER < 1300 // MSVC6 seems to have an amusing bug with anonymous namespaces inside namespaces
+#		define PUGI__NS_BEGIN namespace pugi { namespace impl {
+#		define PUGI__NS_END } }
+#	else
+#		define PUGI__NS_BEGIN namespace pugi { namespace impl { namespace {
+#		define PUGI__NS_END } } }
+#	endif
+#	define PUGI__FN
+#	define PUGI__FN_NO_INLINE PUGI__NO_INLINE
+#endif
+
+// uintptr_t
+#if !defined(_MSC_VER) || _MSC_VER >= 1600
+#	include <stdint.h>
+#else
+namespace pugi
+{
+#	ifndef _UINTPTR_T_DEFINED
+	typedef size_t uintptr_t;
+#	endif
+
+	typedef unsigned __int8 uint8_t;
+	typedef unsigned __int16 uint16_t;
+	typedef unsigned __int32 uint32_t;
+}
+#endif
+
+// Memory allocation
+PUGI__NS_BEGIN
+	PUGI__FN void* default_allocate(size_t size)
+	{
+		return malloc(size);
+	}
+
+	PUGI__FN void default_deallocate(void* ptr)
+	{
+		free(ptr);
+	}
+
+	template <typename T>
+	struct xml_memory_management_function_storage
+	{
+		static allocation_function allocate;
+		static deallocation_function deallocate;
+	};
+
+	// Global allocation functions are stored in class statics so that in header mode linker deduplicates them
+	// Without a template<> we'll get multiple definitions of the same static
+	template <typename T> allocation_function xml_memory_management_function_storage<T>::allocate = default_allocate;
+	template <typename T> deallocation_function xml_memory_management_function_storage<T>::deallocate = default_deallocate;
+
+	typedef xml_memory_management_function_storage<int> xml_memory;
+PUGI__NS_END
+
+// String utilities
+PUGI__NS_BEGIN
+	// Get string length
+	PUGI__FN size_t strlength(const char_t* s)
+	{
+		assert(s);
+
+	#ifdef PUGIXML_WCHAR_MODE
+		return wcslen(s);
+	#else
+		return strlen(s);
+	#endif
+	}
+
+	// Compare two strings
+	PUGI__FN bool strequal(const char_t* src, const char_t* dst)
+	{
+		assert(src && dst);
+
+	#ifdef PUGIXML_WCHAR_MODE
+		return wcscmp(src, dst) == 0;
+	#else
+		return strcmp(src, dst) == 0;
+	#endif
+	}
+
+	// Compare lhs with [rhs_begin, rhs_end)
+	PUGI__FN bool strequalrange(const char_t* lhs, const char_t* rhs, size_t count)
+	{
+		for (size_t i = 0; i < count; ++i)
+			if (lhs[i] != rhs[i])
+				return false;
+	
+		return lhs[count] == 0;
+	}
+
+	// Get length of wide string, even if CRT lacks wide character support
+	PUGI__FN size_t strlength_wide(const wchar_t* s)
+	{
+		assert(s);
+
+	#ifdef PUGIXML_WCHAR_MODE
+		return wcslen(s);
+	#else
+		const wchar_t* end = s;
+		while (*end) end++;
+		return static_cast<size_t>(end - s);
+	#endif
+	}
+
+#ifdef PUGIXML_WCHAR_MODE
+	// Convert string to wide string, assuming all symbols are ASCII
+	PUGI__FN void widen_ascii(wchar_t* dest, const char* source)
+	{
+		for (const char* i = source; *i; ++i) *dest++ = *i;
+		*dest = 0;
+	}
+#endif
+PUGI__NS_END
+
+// auto_ptr-like object for exception recovery
+PUGI__NS_BEGIN
+	template <typename T, typename D = void(*)(T*)> struct auto_deleter
+	{
+		T* data;
+		D deleter;
+
+		auto_deleter(T* data_, D deleter_): data(data_), deleter(deleter_)
+		{
+		}
+
+		~auto_deleter()
+		{
+			if (data) deleter(data);
+		}
+
+		T* release()
+		{
+			T* result = data;
+			data = 0;
+			return result;
+		}
+	};
+PUGI__NS_END
+
+#ifdef PUGIXML_COMPACT
+PUGI__NS_BEGIN
+	class compact_hash_table
+	{
+	public:
+		compact_hash_table(): _items(0), _capacity(0), _count(0)
+		{
+		}
+
+		void clear()
+		{
+			if (_items)
+			{
+				xml_memory::deallocate(_items);
+				_items = 0;
+				_capacity = 0;
+				_count = 0;
+			}
+		}
+
+		void** find(const void* key)
+		{
+			assert(key);
+
+			if (_capacity == 0) return 0;
+
+			size_t hashmod = _capacity - 1;
+			size_t bucket = hash(key) & hashmod;
+
+			for (size_t probe = 0; probe <= hashmod; ++probe)
+			{
+				item_t& probe_item = _items[bucket];
+
+				if (probe_item.key == key)
+					return &probe_item.value;
+
+				if (probe_item.key == 0)
+					return 0;
+
+				// hash collision, quadratic probing
+				bucket = (bucket + probe + 1) & hashmod;
+			}
+
+			assert(!"Hash table is full");
+			return 0;
+		}
+
+		void** insert(const void* key)
+		{
+			assert(key);
+			assert(_count < _capacity * 3 / 4);
+
+			size_t hashmod = _capacity - 1;
+			size_t bucket = hash(key) & hashmod;
+
+			for (size_t probe = 0; probe <= hashmod; ++probe)
+			{
+				item_t& probe_item = _items[bucket];
+
+				if (probe_item.key == 0)
+				{
+					probe_item.key = key;
+					_count++;
+					return &probe_item.value;
+				}
+
+				if (probe_item.key == key)
+					return &probe_item.value;
+
+				// hash collision, quadratic probing
+				bucket = (bucket + probe + 1) & hashmod;
+			}
+
+			assert(!"Hash table is full");
+			return 0;
+		}
+
+		bool reserve()
+		{
+			if (_count + 16 >= _capacity - _capacity / 4)
+				return rehash();
+
+			return true;
+		}
+
+	private:
+		struct item_t
+		{
+			const void* key;
+			void* value;
+		};
+
+		item_t* _items;
+		size_t _capacity;
+
+		size_t _count;
+
+		bool rehash();
+
+		static unsigned int hash(const void* key)
+		{
+			unsigned int h = static_cast<unsigned int>(reinterpret_cast<uintptr_t>(key));
+
+			// MurmurHash3 32-bit finalizer
+			h ^= h >> 16;
+			h *= 0x85ebca6bu;
+			h ^= h >> 13;
+			h *= 0xc2b2ae35u;
+			h ^= h >> 16;
+
+			return h;
+		}
+	};
+
+	PUGI__FN_NO_INLINE bool compact_hash_table::rehash()
+	{
+		compact_hash_table rt;
+		rt._capacity = (_capacity == 0) ? 32 : _capacity * 2;
+		rt._items = static_cast<item_t*>(xml_memory::allocate(sizeof(item_t) * rt._capacity));
+
+		if (!rt._items)
+			return false;
+
+		memset(rt._items, 0, sizeof(item_t) * rt._capacity);
+
+		for (size_t i = 0; i < _capacity; ++i)
+			if (_items[i].key)
+				*rt.insert(_items[i].key) = _items[i].value;
+
+		if (_items)
+			xml_memory::deallocate(_items);
+
+		_capacity = rt._capacity;
+		_items = rt._items;
+
+		return true;
+	}
+
+PUGI__NS_END
+#endif
+
+PUGI__NS_BEGIN
+	static const size_t xml_memory_page_size =
+	#ifdef PUGIXML_MEMORY_PAGE_SIZE
+		PUGIXML_MEMORY_PAGE_SIZE
+	#else
+		32768
+	#endif
+		;
+
+#ifdef PUGIXML_COMPACT
+	static const uintptr_t xml_memory_block_alignment = 4;
+
+	static const uintptr_t xml_memory_page_alignment = sizeof(void*);
+#else
+	static const uintptr_t xml_memory_block_alignment = sizeof(void*);
+
+	static const uintptr_t xml_memory_page_alignment = 64;
+	static const uintptr_t xml_memory_page_pointer_mask = ~(xml_memory_page_alignment - 1);
+#endif
+
+	// extra metadata bits
+	static const uintptr_t xml_memory_page_contents_shared_mask = 32;
+	static const uintptr_t xml_memory_page_name_allocated_mask = 16;
+	static const uintptr_t xml_memory_page_value_allocated_mask = 8;
+	static const uintptr_t xml_memory_page_type_mask = 7;
+
+	// combined masks for string uniqueness
+	static const uintptr_t xml_memory_page_name_allocated_or_shared_mask = xml_memory_page_name_allocated_mask | xml_memory_page_contents_shared_mask;
+	static const uintptr_t xml_memory_page_value_allocated_or_shared_mask = xml_memory_page_value_allocated_mask | xml_memory_page_contents_shared_mask;
+
+#ifdef PUGIXML_COMPACT
+	#define PUGI__GETPAGE_IMPL(header) (header).get_page()
+#else
+	#define PUGI__GETPAGE_IMPL(header) reinterpret_cast<impl::xml_memory_page*>((header) & impl::xml_memory_page_pointer_mask)
+#endif
+
+	#define PUGI__GETPAGE(n) PUGI__GETPAGE_IMPL((n)->header)
+	#define PUGI__NODETYPE(n) static_cast<xml_node_type>(((n)->header & impl::xml_memory_page_type_mask) + 1)
+
+	struct xml_allocator;
+
+	struct xml_memory_page
+	{
+		static xml_memory_page* construct(void* memory)
+		{
+			xml_memory_page* result = static_cast<xml_memory_page*>(memory);
+
+			result->allocator = 0;
+			result->prev = 0;
+			result->next = 0;
+			result->busy_size = 0;
+			result->freed_size = 0;
+
+		#ifdef PUGIXML_COMPACT
+			result->compact_string_base = 0;
+			result->compact_shared_parent = 0;
+			result->compact_page_marker = 0;
+		#endif
+
+			return result;
+		}
+
+		xml_allocator* allocator;
+
+		xml_memory_page* prev;
+		xml_memory_page* next;
+
+		size_t busy_size;
+		size_t freed_size;
+
+	#ifdef PUGIXML_COMPACT
+		char_t* compact_string_base;
+		void* compact_shared_parent;
+		uint32_t* compact_page_marker;
+	#endif
+	};
+
+	struct xml_memory_string_header
+	{
+		uint16_t page_offset; // offset from page->data
+		uint16_t full_size; // 0 if string occupies whole page
+	};
+
+	struct xml_allocator
+	{
+		xml_allocator(xml_memory_page* root): _root(root), _busy_size(root->busy_size)
+		{
+		#ifdef PUGIXML_COMPACT
+			_hash = 0;
+		#endif
+		}
+
+		xml_memory_page* allocate_page(size_t data_size)
+		{
+			size_t size = sizeof(xml_memory_page) + data_size;
+
+			// allocate block with some alignment, leaving memory for worst-case padding
+			void* memory = xml_memory::allocate(size + xml_memory_page_alignment);
+			if (!memory) return 0;
+
+			// align to next page boundary (note: this guarantees at least 1 usable byte before the page)
+			char* page_memory = reinterpret_cast<char*>((reinterpret_cast<uintptr_t>(memory) + xml_memory_page_alignment) & ~(xml_memory_page_alignment - 1));
+
+			// prepare page structure
+			xml_memory_page* page = xml_memory_page::construct(page_memory);
+			assert(page);
+
+			page->allocator = _root->allocator;
+
+			// record the offset for freeing the memory block
+			assert(page_memory > memory && page_memory - static_cast<char*>(memory) <= 127);
+			page_memory[-1] = static_cast<char>(page_memory - static_cast<char*>(memory));
+
+			return page;
+		}
+
+		static void deallocate_page(xml_memory_page* page)
+		{
+			char* page_memory = reinterpret_cast<char*>(page);
+
+			xml_memory::deallocate(page_memory - page_memory[-1]);
+		}
+
+		void* allocate_memory_oob(size_t size, xml_memory_page*& out_page);
+
+		void* allocate_memory(size_t size, xml_memory_page*& out_page)
+		{
+			if (PUGI__UNLIKELY(_busy_size + size > xml_memory_page_size))
+				return allocate_memory_oob(size, out_page);
+
+			void* buf = reinterpret_cast<char*>(_root) + sizeof(xml_memory_page) + _busy_size;
+
+			_busy_size += size;
+
+			out_page = _root;
+
+			return buf;
+		}
+
+	#ifdef PUGIXML_COMPACT
+		void* allocate_object(size_t size, xml_memory_page*& out_page)
+		{
+			void* result = allocate_memory(size + sizeof(uint32_t), out_page);
+			if (!result) return 0;
+
+			// adjust for marker
+			ptrdiff_t offset = static_cast<char*>(result) - reinterpret_cast<char*>(out_page->compact_page_marker);
+
+			if (PUGI__UNLIKELY(static_cast<uintptr_t>(offset) >= 256 * xml_memory_block_alignment))
+			{
+				// insert new marker
+				uint32_t* marker = static_cast<uint32_t*>(result);
+
+				*marker = static_cast<uint32_t>(reinterpret_cast<char*>(marker) - reinterpret_cast<char*>(out_page));
+				out_page->compact_page_marker = marker;
+
+				// since we don't reuse the page space until we reallocate it, we can just pretend that we freed the marker block
+				// this will make sure deallocate_memory correctly tracks the size
+				out_page->freed_size += sizeof(uint32_t);
+
+				return marker + 1;
+			}
+			else
+			{
+				// roll back uint32_t part
+				_busy_size -= sizeof(uint32_t);
+
+				return result;
+			}
+		}
+	#else
+		void* allocate_object(size_t size, xml_memory_page*& out_page)
+		{
+			return allocate_memory(size, out_page);
+		}
+	#endif
+
+		void deallocate_memory(void* ptr, size_t size, xml_memory_page* page)
+		{
+			if (page == _root) page->busy_size = _busy_size;
+
+			assert(ptr >= reinterpret_cast<char*>(page) + sizeof(xml_memory_page) && ptr < reinterpret_cast<char*>(page) + sizeof(xml_memory_page) + page->busy_size);
+			(void)!ptr;
+
+			page->freed_size += size;
+			assert(page->freed_size <= page->busy_size);
+
+			if (page->freed_size == page->busy_size)
+			{
+				if (page->next == 0)
+				{
+					assert(_root == page);
+
+					// top page freed, just reset sizes
+					page->busy_size = 0;
+					page->freed_size = 0;
+
+				#ifdef PUGIXML_COMPACT
+					// reset compact state to maximize efficiency
+					page->compact_string_base = 0;
+					page->compact_shared_parent = 0;
+					page->compact_page_marker = 0;
+				#endif
+
+					_busy_size = 0;
+				}
+				else
+				{
+					assert(_root != page);
+					assert(page->prev);
+
+					// remove from the list
+					page->prev->next = page->next;
+					page->next->prev = page->prev;
+
+					// deallocate
+					deallocate_page(page);
+				}
+			}
+		}
+
+		char_t* allocate_string(size_t length)
+		{
+			static const size_t max_encoded_offset = (1 << 16) * xml_memory_block_alignment;
+
+			PUGI__STATIC_ASSERT(xml_memory_page_size <= max_encoded_offset);
+
+			// allocate memory for string and header block
+			size_t size = sizeof(xml_memory_string_header) + length * sizeof(char_t);
+			
+			// round size up to block alignment boundary
+			size_t full_size = (size + (xml_memory_block_alignment - 1)) & ~(xml_memory_block_alignment - 1);
+
+			xml_memory_page* page;
+			xml_memory_string_header* header = static_cast<xml_memory_string_header*>(allocate_memory(full_size, page));
+
+			if (!header) return 0;
+
+			// setup header
+			ptrdiff_t page_offset = reinterpret_cast<char*>(header) - reinterpret_cast<char*>(page) - sizeof(xml_memory_page);
+
+			assert(page_offset % xml_memory_block_alignment == 0);
+			assert(page_offset >= 0 && static_cast<size_t>(page_offset) < max_encoded_offset);
+			header->page_offset = static_cast<uint16_t>(static_cast<size_t>(page_offset) / xml_memory_block_alignment);
+
+			// full_size == 0 for large strings that occupy the whole page
+			assert(full_size % xml_memory_block_alignment == 0);
+			assert(full_size < max_encoded_offset || (page->busy_size == full_size && page_offset == 0));
+			header->full_size = static_cast<uint16_t>(full_size < max_encoded_offset ? full_size / xml_memory_block_alignment : 0);
+
+			// round-trip through void* to avoid 'cast increases required alignment of target type' warning
+			// header is guaranteed a pointer-sized alignment, which should be enough for char_t
+			return static_cast<char_t*>(static_cast<void*>(header + 1));
+		}
+
+		void deallocate_string(char_t* string)
+		{
+			// this function casts pointers through void* to avoid 'cast increases required alignment of target type' warnings
+			// we're guaranteed the proper (pointer-sized) alignment on the input string if it was allocated via allocate_string
+
+			// get header
+			xml_memory_string_header* header = static_cast<xml_memory_string_header*>(static_cast<void*>(string)) - 1;
+			assert(header);
+
+			// deallocate
+			size_t page_offset = sizeof(xml_memory_page) + header->page_offset * xml_memory_block_alignment;
+			xml_memory_page* page = reinterpret_cast<xml_memory_page*>(static_cast<void*>(reinterpret_cast<char*>(header) - page_offset));
+
+			// if full_size == 0 then this string occupies the whole page
+			size_t full_size = header->full_size == 0 ? page->busy_size : header->full_size * xml_memory_block_alignment;
+
+			deallocate_memory(header, full_size, page);
+		}
+
+		bool reserve()
+		{
+		#ifdef PUGIXML_COMPACT
+			return _hash->reserve();
+		#else
+			return true;
+		#endif
+		}
+
+		xml_memory_page* _root;
+		size_t _busy_size;
+
+	#ifdef PUGIXML_COMPACT
+		compact_hash_table* _hash;
+	#endif
+	};
+
+	PUGI__FN_NO_INLINE void* xml_allocator::allocate_memory_oob(size_t size, xml_memory_page*& out_page)
+	{
+		const size_t large_allocation_threshold = xml_memory_page_size / 4;
+
+		xml_memory_page* page = allocate_page(size <= large_allocation_threshold ? xml_memory_page_size : size);
+		out_page = page;
+
+		if (!page) return 0;
+
+		if (size <= large_allocation_threshold)
+		{
+			_root->busy_size = _busy_size;
+
+			// insert page at the end of linked list
+			page->prev = _root;
+			_root->next = page;
+			_root = page;
+
+			_busy_size = size;
+		}
+		else
+		{
+			// insert page before the end of linked list, so that it is deleted as soon as possible
+			// the last page is not deleted even if it's empty (see deallocate_memory)
+			assert(_root->prev);
+
+			page->prev = _root->prev;
+			page->next = _root;
+
+			_root->prev->next = page;
+			_root->prev = page;
+
+			page->busy_size = size;
+		}
+
+		return reinterpret_cast<char*>(page) + sizeof(xml_memory_page);
+	}
+PUGI__NS_END
+
+#ifdef PUGIXML_COMPACT
+PUGI__NS_BEGIN
+	static const uintptr_t compact_alignment_log2 = 2;
+	static const uintptr_t compact_alignment = 1 << compact_alignment_log2;
+
+	class compact_header
+	{
+	public:
+		compact_header(xml_memory_page* page, unsigned int flags)
+		{
+			PUGI__STATIC_ASSERT(xml_memory_block_alignment == compact_alignment);
+
+			ptrdiff_t offset = (reinterpret_cast<char*>(this) - reinterpret_cast<char*>(page->compact_page_marker));
+			assert(offset % compact_alignment == 0 && static_cast<uintptr_t>(offset) < 256 * compact_alignment);
+
+			_page = static_cast<unsigned char>(offset >> compact_alignment_log2);
+			_flags = static_cast<unsigned char>(flags);
+		}
+
+		void operator&=(uintptr_t mod)
+		{
+			_flags &= mod;
+		}
+
+		void operator|=(uintptr_t mod)
+		{
+			_flags |= mod;
+		}
+
+		uintptr_t operator&(uintptr_t mod) const
+		{
+			return _flags & mod;
+		}
+
+		xml_memory_page* get_page() const
+		{
+			const char* page_marker = reinterpret_cast<const char*>(this) - (_page << compact_alignment_log2);
+			const char* page = page_marker - *reinterpret_cast<const uint32_t*>(page_marker);
+
+			return const_cast<xml_memory_page*>(reinterpret_cast<const xml_memory_page*>(page));
+		}
+
+	private:
+		unsigned char _page;
+		unsigned char _flags;
+	};
+
+	PUGI__FN xml_memory_page* compact_get_page(const void* object, int header_offset)
+	{
+		const compact_header* header = reinterpret_cast<const compact_header*>(static_cast<const char*>(object) - header_offset);
+
+		return header->get_page();
+	}
+
+	template <int header_offset, typename T> PUGI__FN_NO_INLINE T* compact_get_value(const void* object)
+	{
+		return static_cast<T*>(*compact_get_page(object, header_offset)->allocator->_hash->find(object));
+	}
+
+	template <int header_offset, typename T> PUGI__FN_NO_INLINE void compact_set_value(const void* object, T* value)
+	{
+		*compact_get_page(object, header_offset)->allocator->_hash->insert(object) = value;
+	}
+
+	template <typename T, int header_offset, int start = -126> class compact_pointer
+	{
+	public:
+		compact_pointer(): _data(0)
+		{
+		}
+
+		void operator=(const compact_pointer& rhs)
+		{
+			*this = rhs + 0;
+		}
+
+		void operator=(T* value)
+		{
+			if (value)
+			{
+				// value is guaranteed to be compact-aligned; 'this' is not
+				// our decoding is based on 'this' aligned to compact alignment downwards (see operator T*)
+				// so for negative offsets (e.g. -3) we need to adjust the diff by compact_alignment - 1 to
+				// compensate for arithmetic shift rounding for negative values
+				ptrdiff_t diff = reinterpret_cast<char*>(value) - reinterpret_cast<char*>(this);
+				ptrdiff_t offset = ((diff + int(compact_alignment - 1)) >> compact_alignment_log2) - start;
+
+				if (static_cast<uintptr_t>(offset) <= 253)
+					_data = static_cast<unsigned char>(offset + 1);
+				else
+				{
+					compact_set_value<header_offset>(this, value);
+
+					_data = 255;
+				}
+			}
+			else
+				_data = 0;
+		}
+
+		operator T*() const
+		{
+			if (_data)
+			{
+				if (_data < 255)
+				{
+					uintptr_t base = reinterpret_cast<uintptr_t>(this) & ~(compact_alignment - 1);
+
+					return reinterpret_cast<T*>(base + ((_data - 1 + start) << compact_alignment_log2));
+				}
+				else
+					return compact_get_value<header_offset, T>(this);
+			}
+			else
+				return 0;
+		}
+
+		T* operator->() const
+		{
+			return operator T*();
+		}
+
+	private:
+		unsigned char _data;
+	};
+
+	template <typename T, int header_offset> class compact_pointer_parent
+	{
+	public:
+		compact_pointer_parent(): _data(0)
+		{
+		}
+
+		void operator=(const compact_pointer_parent& rhs)
+		{
+			*this = rhs + 0;
+		}
+
+		void operator=(T* value)
+		{
+			if (value)
+			{
+				// value is guaranteed to be compact-aligned; 'this' is not
+				// our decoding is based on 'this' aligned to compact alignment downwards (see operator T*)
+				// so for negative offsets (e.g. -3) we need to adjust the diff by compact_alignment - 1 to
+				// compensate for arithmetic shift behavior for negative values
+				ptrdiff_t diff = reinterpret_cast<char*>(value) - reinterpret_cast<char*>(this);
+				ptrdiff_t offset = ((diff + int(compact_alignment - 1)) >> compact_alignment_log2) + 65533;
+
+				if (static_cast<uintptr_t>(offset) <= 65533)
+				{
+					_data = static_cast<unsigned short>(offset + 1);
+				}
+				else
+				{
+					xml_memory_page* page = compact_get_page(this, header_offset);
+
+					if (PUGI__UNLIKELY(page->compact_shared_parent == 0))
+						page->compact_shared_parent = value;
+
+					if (page->compact_shared_parent == value)
+					{
+						_data = 65534;
+					}
+					else
+					{
+						compact_set_value<header_offset>(this, value);
+
+						_data = 65535;
+					}
+				}
+			}
+			else
+			{
+				_data = 0;
+			}
+		}
+
+		operator T*() const
+		{
+			if (_data)
+			{
+				if (_data < 65534)
+				{
+					uintptr_t base = reinterpret_cast<uintptr_t>(this) & ~(compact_alignment - 1);
+
+					return reinterpret_cast<T*>(base + ((_data - 1 - 65533) << compact_alignment_log2));
+				}
+				else if (_data == 65534)
+					return static_cast<T*>(compact_get_page(this, header_offset)->compact_shared_parent);
+				else
+					return compact_get_value<header_offset, T>(this);
+			}
+			else
+				return 0;
+		}
+
+		T* operator->() const
+		{
+			return operator T*();
+		}
+
+	private:
+		uint16_t _data;
+	};
+
+	template <int header_offset, int base_offset> class compact_string
+	{
+	public:
+		compact_string(): _data(0)
+		{
+		}
+
+		void operator=(const compact_string& rhs)
+		{
+			*this = rhs + 0;
+		}
+
+		void operator=(char_t* value)
+		{
+			if (value)
+			{
+				xml_memory_page* page = compact_get_page(this, header_offset);
+
+				if (PUGI__UNLIKELY(page->compact_string_base == 0))
+					page->compact_string_base = value;
+
+				ptrdiff_t offset = value - page->compact_string_base;
+
+				if (static_cast<uintptr_t>(offset) < (65535 << 7))
+				{
+					uint16_t* base = reinterpret_cast<uint16_t*>(reinterpret_cast<char*>(this) - base_offset);
+
+					if (*base == 0)
+					{
+						*base = static_cast<uint16_t>((offset >> 7) + 1);
+						_data = static_cast<unsigned char>((offset & 127) + 1);
+					}
+					else
+					{
+						ptrdiff_t remainder = offset - ((*base - 1) << 7);
+
+						if (static_cast<uintptr_t>(remainder) <= 253)
+						{
+							_data = static_cast<unsigned char>(remainder + 1);
+						}
+						else
+						{
+							compact_set_value<header_offset>(this, value);
+
+							_data = 255;
+						}
+					}
+				}
+				else
+				{
+					compact_set_value<header_offset>(this, value);
+
+					_data = 255;
+				}
+			}
+			else
+			{
+				_data = 0;
+			}
+		}
+
+		operator char_t*() const
+		{
+			if (_data)
+			{
+				if (_data < 255)
+				{
+					xml_memory_page* page = compact_get_page(this, header_offset);
+
+					const uint16_t* base = reinterpret_cast<const uint16_t*>(reinterpret_cast<const char*>(this) - base_offset);
+					assert(*base);
+
+					ptrdiff_t offset = ((*base - 1) << 7) + (_data - 1);
+
+					return page->compact_string_base + offset;
+				}
+				else
+				{
+					return compact_get_value<header_offset, char_t>(this);
+				}
+			}
+			else
+				return 0;
+		}
+
+	private:
+		unsigned char _data;
+	};
+PUGI__NS_END
+#endif
+
+#ifdef PUGIXML_COMPACT
+namespace pugi
+{
+	struct xml_attribute_struct
+	{
+		xml_attribute_struct(impl::xml_memory_page* page): header(page, 0), namevalue_base(0)
+		{
+			PUGI__STATIC_ASSERT(sizeof(xml_attribute_struct) == 8);
+		}
+
+		impl::compact_header header;
+
+		uint16_t namevalue_base;
+
+		impl::compact_string<4, 2> name;
+		impl::compact_string<5, 3> value;
+
+		impl::compact_pointer<xml_attribute_struct, 6> prev_attribute_c;
+		impl::compact_pointer<xml_attribute_struct, 7, 0> next_attribute;
+	};
+
+	struct xml_node_struct
+	{
+		xml_node_struct(impl::xml_memory_page* page, xml_node_type type): header(page, type - 1), namevalue_base(0)
+		{
+			PUGI__STATIC_ASSERT(sizeof(xml_node_struct) == 12);
+		}
+
+		impl::compact_header header;
+
+		uint16_t namevalue_base;
+
+		impl::compact_string<4, 2> name;
+		impl::compact_string<5, 3> value;
+
+		impl::compact_pointer_parent<xml_node_struct, 6> parent;
+
+		impl::compact_pointer<xml_node_struct, 8, 0> first_child;
+
+		impl::compact_pointer<xml_node_struct,  9>    prev_sibling_c;
+		impl::compact_pointer<xml_node_struct, 10, 0> next_sibling;
+
+		impl::compact_pointer<xml_attribute_struct, 11, 0> first_attribute;
+	};
+}
+#else
+namespace pugi
+{
+	struct xml_attribute_struct
+	{
+		xml_attribute_struct(impl::xml_memory_page* page): header(reinterpret_cast<uintptr_t>(page)), name(0), value(0), prev_attribute_c(0), next_attribute(0)
+		{
+		}
+
+		uintptr_t header;
+
+		char_t*	name;
+		char_t*	value;
+
+		xml_attribute_struct* prev_attribute_c;
+		xml_attribute_struct* next_attribute;
+	};
+
+	struct xml_node_struct
+	{
+		xml_node_struct(impl::xml_memory_page* page, xml_node_type type): header(reinterpret_cast<uintptr_t>(page) | (type - 1)), name(0), value(0), parent(0), first_child(0), prev_sibling_c(0), next_sibling(0), first_attribute(0)
+		{
+		}
+
+		uintptr_t header;
+
+		char_t* name;
+		char_t* value;
+
+		xml_node_struct* parent;
+
+		xml_node_struct* first_child;
+
+		xml_node_struct* prev_sibling_c;
+		xml_node_struct* next_sibling;
+
+		xml_attribute_struct* first_attribute;
+	};
+}
+#endif
+
+PUGI__NS_BEGIN
+	struct xml_extra_buffer
+	{
+		char_t* buffer;
+		xml_extra_buffer* next;
+	};
+
+	struct xml_document_struct: public xml_node_struct, public xml_allocator
+	{
+		xml_document_struct(xml_memory_page* page): xml_node_struct(page, node_document), xml_allocator(page), buffer(0), extra_buffers(0)
+		{
+		#ifdef PUGIXML_COMPACT
+			_hash = &hash;
+		#endif
+		}
+
+		const char_t* buffer;
+
+		xml_extra_buffer* extra_buffers;
+
+	#ifdef PUGIXML_COMPACT
+		compact_hash_table hash;
+	#endif
+	};
+
+	template <typename Object> inline xml_allocator& get_allocator(const Object* object)
+	{
+		assert(object);
+
+		return *PUGI__GETPAGE(object)->allocator;
+	}
+
+	template <typename Object> inline xml_document_struct& get_document(const Object* object)
+	{
+		assert(object);
+
+		return *static_cast<xml_document_struct*>(PUGI__GETPAGE(object)->allocator);
+	}
+PUGI__NS_END
+
+// Low-level DOM operations
+PUGI__NS_BEGIN
+	inline xml_attribute_struct* allocate_attribute(xml_allocator& alloc)
+	{
+		xml_memory_page* page;
+		void* memory = alloc.allocate_object(sizeof(xml_attribute_struct), page);
+		if (!memory) return 0;
+
+		return new (memory) xml_attribute_struct(page);
+	}
+
+	inline xml_node_struct* allocate_node(xml_allocator& alloc, xml_node_type type)
+	{
+		xml_memory_page* page;
+		void* memory = alloc.allocate_object(sizeof(xml_node_struct), page);
+		if (!memory) return 0;
+
+		return new (memory) xml_node_struct(page, type);
+	}
+
+	inline void destroy_attribute(xml_attribute_struct* a, xml_allocator& alloc)
+	{
+		if (a->header & impl::xml_memory_page_name_allocated_mask)
+			alloc.deallocate_string(a->name);
+
+		if (a->header & impl::xml_memory_page_value_allocated_mask)
+			alloc.deallocate_string(a->value);
+
+		alloc.deallocate_memory(a, sizeof(xml_attribute_struct), PUGI__GETPAGE(a));
+	}
+
+	inline void destroy_node(xml_node_struct* n, xml_allocator& alloc)
+	{
+		if (n->header & impl::xml_memory_page_name_allocated_mask)
+			alloc.deallocate_string(n->name);
+
+		if (n->header & impl::xml_memory_page_value_allocated_mask)
+			alloc.deallocate_string(n->value);
+
+		for (xml_attribute_struct* attr = n->first_attribute; attr; )
+		{
+			xml_attribute_struct* next = attr->next_attribute;
+
+			destroy_attribute(attr, alloc);
+
+			attr = next;
+		}
+
+		for (xml_node_struct* child = n->first_child; child; )
+		{
+			xml_node_struct* next = child->next_sibling;
+
+			destroy_node(child, alloc);
+
+			child = next;
+		}
+
+		alloc.deallocate_memory(n, sizeof(xml_node_struct), PUGI__GETPAGE(n));
+	}
+
+	inline void append_node(xml_node_struct* child, xml_node_struct* node)
+	{
+		child->parent = node;
+
+		xml_node_struct* head = node->first_child;
+
+		if (head)
+		{
+			xml_node_struct* tail = head->prev_sibling_c;
+
+			tail->next_sibling = child;
+			child->prev_sibling_c = tail;
+			head->prev_sibling_c = child;
+		}
+		else
+		{
+			node->first_child = child;
+			child->prev_sibling_c = child;
+		}
+	}
+
+	inline void prepend_node(xml_node_struct* child, xml_node_struct* node)
+	{
+		child->parent = node;
+
+		xml_node_struct* head = node->first_child;
+
+		if (head)
+		{
+			child->prev_sibling_c = head->prev_sibling_c;
+			head->prev_sibling_c = child;
+		}
+		else
+			child->prev_sibling_c = child;
+
+		child->next_sibling = head;
+		node->first_child = child;
+	}
+
+	inline void insert_node_after(xml_node_struct* child, xml_node_struct* node)
+	{
+		xml_node_struct* parent = node->parent;
+
+		child->parent = parent;
+
+		if (node->next_sibling)
+			node->next_sibling->prev_sibling_c = child;
+		else
+			parent->first_child->prev_sibling_c = child;
+
+		child->next_sibling = node->next_sibling;
+		child->prev_sibling_c = node;
+
+		node->next_sibling = child;
+	}
+
+	inline void insert_node_before(xml_node_struct* child, xml_node_struct* node)
+	{
+		xml_node_struct* parent = node->parent;
+
+		child->parent = parent;
+
+		if (node->prev_sibling_c->next_sibling)
+			node->prev_sibling_c->next_sibling = child;
+		else
+			parent->first_child = child;
+
+		child->prev_sibling_c = node->prev_sibling_c;
+		child->next_sibling = node;
+
+		node->prev_sibling_c = child;
+	}
+
+	inline void remove_node(xml_node_struct* node)
+	{
+		xml_node_struct* parent = node->parent;
+
+		if (node->next_sibling)
+			node->next_sibling->prev_sibling_c = node->prev_sibling_c;
+		else
+			parent->first_child->prev_sibling_c = node->prev_sibling_c;
+
+		if (node->prev_sibling_c->next_sibling)
+			node->prev_sibling_c->next_sibling = node->next_sibling;
+		else
+			parent->first_child = node->next_sibling;
+
+		node->parent = 0;
+		node->prev_sibling_c = 0;
+		node->next_sibling = 0;
+	}
+
+	inline void append_attribute(xml_attribute_struct* attr, xml_node_struct* node)
+	{
+		xml_attribute_struct* head = node->first_attribute;
+
+		if (head)
+		{
+			xml_attribute_struct* tail = head->prev_attribute_c;
+
+			tail->next_attribute = attr;
+			attr->prev_attribute_c = tail;
+			head->prev_attribute_c = attr;
+		}
+		else
+		{
+			node->first_attribute = attr;
+			attr->prev_attribute_c = attr;
+		}
+	}
+
+	inline void prepend_attribute(xml_attribute_struct* attr, xml_node_struct* node)
+	{
+		xml_attribute_struct* head = node->first_attribute;
+
+		if (head)
+		{
+			attr->prev_attribute_c = head->prev_attribute_c;
+			head->prev_attribute_c = attr;
+		}
+		else
+			attr->prev_attribute_c = attr;
+
+		attr->next_attribute = head;
+		node->first_attribute = attr;
+	}
+
+	inline void insert_attribute_after(xml_attribute_struct* attr, xml_attribute_struct* place, xml_node_struct* node)
+	{
+		if (place->next_attribute)
+			place->next_attribute->prev_attribute_c = attr;
+		else
+			node->first_attribute->prev_attribute_c = attr;
+
+		attr->next_attribute = place->next_attribute;
+		attr->prev_attribute_c = place;
+		place->next_attribute = attr;
+	}
+
+	inline void insert_attribute_before(xml_attribute_struct* attr, xml_attribute_struct* place, xml_node_struct* node)
+	{
+		if (place->prev_attribute_c->next_attribute)
+			place->prev_attribute_c->next_attribute = attr;
+		else
+			node->first_attribute = attr;
+
+		attr->prev_attribute_c = place->prev_attribute_c;
+		attr->next_attribute = place;
+		place->prev_attribute_c = attr;
+	}
+
+	inline void remove_attribute(xml_attribute_struct* attr, xml_node_struct* node)
+	{
+		if (attr->next_attribute)
+			attr->next_attribute->prev_attribute_c = attr->prev_attribute_c;
+		else
+			node->first_attribute->prev_attribute_c = attr->prev_attribute_c;
+
+		if (attr->prev_attribute_c->next_attribute)
+			attr->prev_attribute_c->next_attribute = attr->next_attribute;
+		else
+			node->first_attribute = attr->next_attribute;
+
+		attr->prev_attribute_c = 0;
+		attr->next_attribute = 0;
+	}
+
+	PUGI__FN_NO_INLINE xml_node_struct* append_new_node(xml_node_struct* node, xml_allocator& alloc, xml_node_type type = node_element)
+	{
+		if (!alloc.reserve()) return 0;
+
+		xml_node_struct* child = allocate_node(alloc, type);
+		if (!child) return 0;
+
+		append_node(child, node);
+
+		return child;
+	}
+
+	PUGI__FN_NO_INLINE xml_attribute_struct* append_new_attribute(xml_node_struct* node, xml_allocator& alloc)
+	{
+		if (!alloc.reserve()) return 0;
+
+		xml_attribute_struct* attr = allocate_attribute(alloc);
+		if (!attr) return 0;
+
+		append_attribute(attr, node);
+
+		return attr;
+	}
+PUGI__NS_END
+
+// Helper classes for code generation
+PUGI__NS_BEGIN
+	struct opt_false
+	{
+		enum { value = 0 };
+	};
+
+	struct opt_true
+	{
+		enum { value = 1 };
+	};
+PUGI__NS_END
+
+// Unicode utilities
+PUGI__NS_BEGIN
+	inline uint16_t endian_swap(uint16_t value)
+	{
+		return static_cast<uint16_t>(((value & 0xff) << 8) | (value >> 8));
+	}
+
+	inline uint32_t endian_swap(uint32_t value)
+	{
+		return ((value & 0xff) << 24) | ((value & 0xff00) << 8) | ((value & 0xff0000) >> 8) | (value >> 24);
+	}
+
+	struct utf8_counter
+	{
+		typedef size_t value_type;
+
+		static value_type low(value_type result, uint32_t ch)
+		{
+			// U+0000..U+007F
+			if (ch < 0x80) return result + 1;
+			// U+0080..U+07FF
+			else if (ch < 0x800) return result + 2;
+			// U+0800..U+FFFF
+			else return result + 3;
+		}
+
+		static value_type high(value_type result, uint32_t)
+		{
+			// U+10000..U+10FFFF
+			return result + 4;
+		}
+	};
+
+	struct utf8_writer
+	{
+		typedef uint8_t* value_type;
+
+		static value_type low(value_type result, uint32_t ch)
+		{
+			// U+0000..U+007F
+			if (ch < 0x80)
+			{
+				*result = static_cast<uint8_t>(ch);
+				return result + 1;
+			}
+			// U+0080..U+07FF
+			else if (ch < 0x800)
+			{
+				result[0] = static_cast<uint8_t>(0xC0 | (ch >> 6));
+				result[1] = static_cast<uint8_t>(0x80 | (ch & 0x3F));
+				return result + 2;
+			}
+			// U+0800..U+FFFF
+			else
+			{
+				result[0] = static_cast<uint8_t>(0xE0 | (ch >> 12));
+				result[1] = static_cast<uint8_t>(0x80 | ((ch >> 6) & 0x3F));
+				result[2] = static_cast<uint8_t>(0x80 | (ch & 0x3F));
+				return result + 3;
+			}
+		}
+
+		static value_type high(value_type result, uint32_t ch)
+		{
+			// U+10000..U+10FFFF
+			result[0] = static_cast<uint8_t>(0xF0 | (ch >> 18));
+			result[1] = static_cast<uint8_t>(0x80 | ((ch >> 12) & 0x3F));
+			result[2] = static_cast<uint8_t>(0x80 | ((ch >> 6) & 0x3F));
+			result[3] = static_cast<uint8_t>(0x80 | (ch & 0x3F));
+			return result + 4;
+		}
+
+		static value_type any(value_type result, uint32_t ch)
+		{
+			return (ch < 0x10000) ? low(result, ch) : high(result, ch);
+		}
+	};
+
+	struct utf16_counter
+	{
+		typedef size_t value_type;
+
+		static value_type low(value_type result, uint32_t)
+		{
+			return result + 1;
+		}
+
+		static value_type high(value_type result, uint32_t)
+		{
+			return result + 2;
+		}
+	};
+
+	struct utf16_writer
+	{
+		typedef uint16_t* value_type;
+
+		static value_type low(value_type result, uint32_t ch)
+		{
+			*result = static_cast<uint16_t>(ch);
+
+			return result + 1;
+		}
+
+		static value_type high(value_type result, uint32_t ch)
+		{
+			uint32_t msh = static_cast<uint32_t>(ch - 0x10000) >> 10;
+			uint32_t lsh = static_cast<uint32_t>(ch - 0x10000) & 0x3ff;
+
+			result[0] = static_cast<uint16_t>(0xD800 + msh);
+			result[1] = static_cast<uint16_t>(0xDC00 + lsh);
+
+			return result + 2;
+		}
+
+		static value_type any(value_type result, uint32_t ch)
+		{
+			return (ch < 0x10000) ? low(result, ch) : high(result, ch);
+		}
+	};
+
+	struct utf32_counter
+	{
+		typedef size_t value_type;
+
+		static value_type low(value_type result, uint32_t)
+		{
+			return result + 1;
+		}
+
+		static value_type high(value_type result, uint32_t)
+		{
+			return result + 1;
+		}
+	};
+
+	struct utf32_writer
+	{
+		typedef uint32_t* value_type;
+
+		static value_type low(value_type result, uint32_t ch)
+		{
+			*result = ch;
+
+			return result + 1;
+		}
+
+		static value_type high(value_type result, uint32_t ch)
+		{
+			*result = ch;
+
+			return result + 1;
+		}
+
+		static value_type any(value_type result, uint32_t ch)
+		{
+			*result = ch;
+
+			return result + 1;
+		}
+	};
+
+	struct latin1_writer
+	{
+		typedef uint8_t* value_type;
+
+		static value_type low(value_type result, uint32_t ch)
+		{
+			*result = static_cast<uint8_t>(ch > 255 ? '?' : ch);
+
+			return result + 1;
+		}
+
+		static value_type high(value_type result, uint32_t ch)
+		{
+			(void)ch;
+
+			*result = '?';
+
+			return result + 1;
+		}
+	};
+
+	template <size_t size> struct wchar_selector;
+
+	template <> struct wchar_selector<2>
+	{
+		typedef uint16_t type;
+		typedef utf16_counter counter;
+		typedef utf16_writer writer;
+	};
+
+	template <> struct wchar_selector<4>
+	{
+		typedef uint32_t type;
+		typedef utf32_counter counter;
+		typedef utf32_writer writer;
+	};
+
+	typedef wchar_selector<sizeof(wchar_t)>::counter wchar_counter;
+	typedef wchar_selector<sizeof(wchar_t)>::writer wchar_writer;
+
+	template <typename Traits, typename opt_swap = opt_false> struct utf_decoder
+	{
+		static inline typename Traits::value_type decode_utf8_block(const uint8_t* data, size_t size, typename Traits::value_type result)
+		{
+			const uint8_t utf8_byte_mask = 0x3f;
+
+			while (size)
+			{
+				uint8_t lead = *data;
+
+				// 0xxxxxxx -> U+0000..U+007F
+				if (lead < 0x80)
+				{
+					result = Traits::low(result, lead);
+					data += 1;
+					size -= 1;
+
+					// process aligned single-byte (ascii) blocks
+					if ((reinterpret_cast<uintptr_t>(data) & 3) == 0)
+					{
+						// round-trip through void* to silence 'cast increases required alignment of target type' warnings
+						while (size >= 4 && (*static_cast<const uint32_t*>(static_cast<const void*>(data)) & 0x80808080) == 0)
+						{
+							result = Traits::low(result, data[0]);
+							result = Traits::low(result, data[1]);
+							result = Traits::low(result, data[2]);
+							result = Traits::low(result, data[3]);
+							data += 4;
+							size -= 4;
+						}
+					}
+				}
+				// 110xxxxx -> U+0080..U+07FF
+				else if (static_cast<unsigned int>(lead - 0xC0) < 0x20 && size >= 2 && (data[1] & 0xc0) == 0x80)
+				{
+					result = Traits::low(result, ((lead & ~0xC0) << 6) | (data[1] & utf8_byte_mask));
+					data += 2;
+					size -= 2;
+				}
+				// 1110xxxx -> U+0800-U+FFFF
+				else if (static_cast<unsigned int>(lead - 0xE0) < 0x10 && size >= 3 && (data[1] & 0xc0) == 0x80 && (data[2] & 0xc0) == 0x80)
+				{
+					result = Traits::low(result, ((lead & ~0xE0) << 12) | ((data[1] & utf8_byte_mask) << 6) | (data[2] & utf8_byte_mask));
+					data += 3;
+					size -= 3;
+				}
+				// 11110xxx -> U+10000..U+10FFFF
+				else if (static_cast<unsigned int>(lead - 0xF0) < 0x08 && size >= 4 && (data[1] & 0xc0) == 0x80 && (data[2] & 0xc0) == 0x80 && (data[3] & 0xc0) == 0x80)
+				{
+					result = Traits::high(result, ((lead & ~0xF0) << 18) | ((data[1] & utf8_byte_mask) << 12) | ((data[2] & utf8_byte_mask) << 6) | (data[3] & utf8_byte_mask));
+					data += 4;
+					size -= 4;
+				}
+				// 10xxxxxx or 11111xxx -> invalid
+				else
+				{
+					data += 1;
+					size -= 1;
+				}
+			}
+
+			return result;
+		}
+
+		static inline typename Traits::value_type decode_utf16_block(const uint16_t* data, size_t size, typename Traits::value_type result)
+		{
+			const uint16_t* end = data + size;
+
+			while (data < end)
+			{
+				unsigned int lead = opt_swap::value ? endian_swap(*data) : *data;
+
+				// U+0000..U+D7FF
+				if (lead < 0xD800)
+				{
+					result = Traits::low(result, lead);
+					data += 1;
+				}
+				// U+E000..U+FFFF
+				else if (static_cast<unsigned int>(lead - 0xE000) < 0x2000)
+				{
+					result = Traits::low(result, lead);
+					data += 1;
+				}
+				// surrogate pair lead
+				else if (static_cast<unsigned int>(lead - 0xD800) < 0x400 && data + 1 < end)
+				{
+					uint16_t next = opt_swap::value ? endian_swap(data[1]) : data[1];
+
+					if (static_cast<unsigned int>(next - 0xDC00) < 0x400)
+					{
+						result = Traits::high(result, 0x10000 + ((lead & 0x3ff) << 10) + (next & 0x3ff));
+						data += 2;
+					}
+					else
+					{
+						data += 1;
+					}
+				}
+				else
+				{
+					data += 1;
+				}
+			}
+
+			return result;
+		}
+
+		static inline typename Traits::value_type decode_utf32_block(const uint32_t* data, size_t size, typename Traits::value_type result)
+		{
+			const uint32_t* end = data + size;
+
+			while (data < end)
+			{
+				uint32_t lead = opt_swap::value ? endian_swap(*data) : *data;
+
+				// U+0000..U+FFFF
+				if (lead < 0x10000)
+				{
+					result = Traits::low(result, lead);
+					data += 1;
+				}
+				// U+10000..U+10FFFF
+				else
+				{
+					result = Traits::high(result, lead);
+					data += 1;
+				}
+			}
+
+			return result;
+		}
+
+		static inline typename Traits::value_type decode_latin1_block(const uint8_t* data, size_t size, typename Traits::value_type result)
+		{
+			for (size_t i = 0; i < size; ++i)
+			{
+				result = Traits::low(result, data[i]);
+			}
+
+			return result;
+		}
+
+		static inline typename Traits::value_type decode_wchar_block_impl(const uint16_t* data, size_t size, typename Traits::value_type result)
+		{
+			return decode_utf16_block(data, size, result);
+		}
+
+		static inline typename Traits::value_type decode_wchar_block_impl(const uint32_t* data, size_t size, typename Traits::value_type result)
+		{
+			return decode_utf32_block(data, size, result);
+		}
+
+		static inline typename Traits::value_type decode_wchar_block(const wchar_t* data, size_t size, typename Traits::value_type result)
+		{
+			return decode_wchar_block_impl(reinterpret_cast<const wchar_selector<sizeof(wchar_t)>::type*>(data), size, result);
+		}
+	};
+
+	template <typename T> PUGI__FN void convert_utf_endian_swap(T* result, const T* data, size_t length)
+	{
+		for (size_t i = 0; i < length; ++i) result[i] = endian_swap(data[i]);
+	}
+
+#ifdef PUGIXML_WCHAR_MODE
+	PUGI__FN void convert_wchar_endian_swap(wchar_t* result, const wchar_t* data, size_t length)
+	{
+		for (size_t i = 0; i < length; ++i) result[i] = static_cast<wchar_t>(endian_swap(static_cast<wchar_selector<sizeof(wchar_t)>::type>(data[i])));
+	}
+#endif
+PUGI__NS_END
+
+PUGI__NS_BEGIN
+	enum chartype_t
+	{
+		ct_parse_pcdata = 1,	// \0, &, \r, <
+		ct_parse_attr = 2,		// \0, &, \r, ', "
+		ct_parse_attr_ws = 4,	// \0, &, \r, ', ", \n, tab
+		ct_space = 8,			// \r, \n, space, tab
+		ct_parse_cdata = 16,	// \0, ], >, \r
+		ct_parse_comment = 32,	// \0, -, >, \r
+		ct_symbol = 64,			// Any symbol > 127, a-z, A-Z, 0-9, _, :, -, .
+		ct_start_symbol = 128	// Any symbol > 127, a-z, A-Z, _, :
+	};
+
+	static const unsigned char chartype_table[256] =
+	{
+		55,  0,   0,   0,   0,   0,   0,   0,      0,   12,  12,  0,   0,   63,  0,   0,   // 0-15
+		0,   0,   0,   0,   0,   0,   0,   0,      0,   0,   0,   0,   0,   0,   0,   0,   // 16-31
+		8,   0,   6,   0,   0,   0,   7,   6,      0,   0,   0,   0,   0,   96,  64,  0,   // 32-47
+		64,  64,  64,  64,  64,  64,  64,  64,     64,  64,  192, 0,   1,   0,   48,  0,   // 48-63
+		0,   192, 192, 192, 192, 192, 192, 192,    192, 192, 192, 192, 192, 192, 192, 192, // 64-79
+		192, 192, 192, 192, 192, 192, 192, 192,    192, 192, 192, 0,   0,   16,  0,   192, // 80-95
+		0,   192, 192, 192, 192, 192, 192, 192,    192, 192, 192, 192, 192, 192, 192, 192, // 96-111
+		192, 192, 192, 192, 192, 192, 192, 192,    192, 192, 192, 0, 0, 0, 0, 0,           // 112-127
+
+		192, 192, 192, 192, 192, 192, 192, 192,    192, 192, 192, 192, 192, 192, 192, 192, // 128+
+		192, 192, 192, 192, 192, 192, 192, 192,    192, 192, 192, 192, 192, 192, 192, 192,
+		192, 192, 192, 192, 192, 192, 192, 192,    192, 192, 192, 192, 192, 192, 192, 192,
+		192, 192, 192, 192, 192, 192, 192, 192,    192, 192, 192, 192, 192, 192, 192, 192,
+		192, 192, 192, 192, 192, 192, 192, 192,    192, 192, 192, 192, 192, 192, 192, 192,
+		192, 192, 192, 192, 192, 192, 192, 192,    192, 192, 192, 192, 192, 192, 192, 192,
+		192, 192, 192, 192, 192, 192, 192, 192,    192, 192, 192, 192, 192, 192, 192, 192,
+		192, 192, 192, 192, 192, 192, 192, 192,    192, 192, 192, 192, 192, 192, 192, 192
+	};
+
+	enum chartypex_t
+	{
+		ctx_special_pcdata = 1,   // Any symbol >= 0 and < 32 (except \t, \r, \n), &, <, >
+		ctx_special_attr = 2,     // Any symbol >= 0 and < 32 (except \t), &, <, >, "
+		ctx_start_symbol = 4,	  // Any symbol > 127, a-z, A-Z, _
+		ctx_digit = 8,			  // 0-9
+		ctx_symbol = 16			  // Any symbol > 127, a-z, A-Z, 0-9, _, -, .
+	};
+	
+	static const unsigned char chartypex_table[256] =
+	{
+		3,  3,  3,  3,  3,  3,  3,  3,     3,  0,  2,  3,  3,  2,  3,  3,     // 0-15
+		3,  3,  3,  3,  3,  3,  3,  3,     3,  3,  3,  3,  3,  3,  3,  3,     // 16-31
+		0,  0,  2,  0,  0,  0,  3,  0,     0,  0,  0,  0,  0, 16, 16,  0,     // 32-47
+		24, 24, 24, 24, 24, 24, 24, 24,    24, 24, 0,  0,  3,  0,  3,  0,     // 48-63
+
+		0,  20, 20, 20, 20, 20, 20, 20,    20, 20, 20, 20, 20, 20, 20, 20,    // 64-79
+		20, 20, 20, 20, 20, 20, 20, 20,    20, 20, 20, 0,  0,  0,  0,  20,    // 80-95
+		0,  20, 20, 20, 20, 20, 20, 20,    20, 20, 20, 20, 20, 20, 20, 20,    // 96-111
+		20, 20, 20, 20, 20, 20, 20, 20,    20, 20, 20, 0,  0,  0,  0,  0,     // 112-127
+
+		20, 20, 20, 20, 20, 20, 20, 20,    20, 20, 20, 20, 20, 20, 20, 20,    // 128+
+		20, 20, 20, 20, 20, 20, 20, 20,    20, 20, 20, 20, 20, 20, 20, 20,
+		20, 20, 20, 20, 20, 20, 20, 20,    20, 20, 20, 20, 20, 20, 20, 20,
+		20, 20, 20, 20, 20, 20, 20, 20,    20, 20, 20, 20, 20, 20, 20, 20,
+		20, 20, 20, 20, 20, 20, 20, 20,    20, 20, 20, 20, 20, 20, 20, 20,
+		20, 20, 20, 20, 20, 20, 20, 20,    20, 20, 20, 20, 20, 20, 20, 20,
+		20, 20, 20, 20, 20, 20, 20, 20,    20, 20, 20, 20, 20, 20, 20, 20,
+		20, 20, 20, 20, 20, 20, 20, 20,    20, 20, 20, 20, 20, 20, 20, 20
+	};
+	
+#ifdef PUGIXML_WCHAR_MODE
+	#define PUGI__IS_CHARTYPE_IMPL(c, ct, table) ((static_cast<unsigned int>(c) < 128 ? table[static_cast<unsigned int>(c)] : table[128]) & (ct))
+#else
+	#define PUGI__IS_CHARTYPE_IMPL(c, ct, table) (table[static_cast<unsigned char>(c)] & (ct))
+#endif
+
+	#define PUGI__IS_CHARTYPE(c, ct) PUGI__IS_CHARTYPE_IMPL(c, ct, chartype_table)
+	#define PUGI__IS_CHARTYPEX(c, ct) PUGI__IS_CHARTYPE_IMPL(c, ct, chartypex_table)
+
+	PUGI__FN bool is_little_endian()
+	{
+		unsigned int ui = 1;
+
+		return *reinterpret_cast<unsigned char*>(&ui) == 1;
+	}
+
+	PUGI__FN xml_encoding get_wchar_encoding()
+	{
+		PUGI__STATIC_ASSERT(sizeof(wchar_t) == 2 || sizeof(wchar_t) == 4);
+
+		if (sizeof(wchar_t) == 2)
+			return is_little_endian() ? encoding_utf16_le : encoding_utf16_be;
+		else 
+			return is_little_endian() ? encoding_utf32_le : encoding_utf32_be;
+	}
+
+	PUGI__FN xml_encoding guess_buffer_encoding(uint8_t d0, uint8_t d1, uint8_t d2, uint8_t d3)
+	{
+		// look for BOM in first few bytes
+		if (d0 == 0 && d1 == 0 && d2 == 0xfe && d3 == 0xff) return encoding_utf32_be;
+		if (d0 == 0xff && d1 == 0xfe && d2 == 0 && d3 == 0) return encoding_utf32_le;
+		if (d0 == 0xfe && d1 == 0xff) return encoding_utf16_be;
+		if (d0 == 0xff && d1 == 0xfe) return encoding_utf16_le;
+		if (d0 == 0xef && d1 == 0xbb && d2 == 0xbf) return encoding_utf8;
+
+		// look for <, <? or <?xm in various encodings
+		if (d0 == 0 && d1 == 0 && d2 == 0 && d3 == 0x3c) return encoding_utf32_be;
+		if (d0 == 0x3c && d1 == 0 && d2 == 0 && d3 == 0) return encoding_utf32_le;
+		if (d0 == 0 && d1 == 0x3c && d2 == 0 && d3 == 0x3f) return encoding_utf16_be;
+		if (d0 == 0x3c && d1 == 0 && d2 == 0x3f && d3 == 0) return encoding_utf16_le;
+		if (d0 == 0x3c && d1 == 0x3f && d2 == 0x78 && d3 == 0x6d) return encoding_utf8;
+
+		// look for utf16 < followed by node name (this may fail, but is better than utf8 since it's zero terminated so early)
+		if (d0 == 0 && d1 == 0x3c) return encoding_utf16_be;
+		if (d0 == 0x3c && d1 == 0) return encoding_utf16_le;
+
+		// no known BOM detected, assume utf8
+		return encoding_utf8;
+	}
+
+	PUGI__FN xml_encoding get_buffer_encoding(xml_encoding encoding, const void* contents, size_t size)
+	{
+		// replace wchar encoding with utf implementation
+		if (encoding == encoding_wchar) return get_wchar_encoding();
+
+		// replace utf16 encoding with utf16 with specific endianness
+		if (encoding == encoding_utf16) return is_little_endian() ? encoding_utf16_le : encoding_utf16_be;
+
+		// replace utf32 encoding with utf32 with specific endianness
+		if (encoding == encoding_utf32) return is_little_endian() ? encoding_utf32_le : encoding_utf32_be;
+
+		// only do autodetection if no explicit encoding is requested
+		if (encoding != encoding_auto) return encoding;
+
+		// skip encoding autodetection if input buffer is too small
+		if (size < 4) return encoding_utf8;
+
+		// try to guess encoding (based on XML specification, Appendix F.1)
+		const uint8_t* data = static_cast<const uint8_t*>(contents);
+
+		PUGI__DMC_VOLATILE uint8_t d0 = data[0], d1 = data[1], d2 = data[2], d3 = data[3];
+
+		return guess_buffer_encoding(d0, d1, d2, d3);
+	}
+
+	PUGI__FN bool get_mutable_buffer(char_t*& out_buffer, size_t& out_length, const void* contents, size_t size, bool is_mutable)
+	{
+		size_t length = size / sizeof(char_t);
+
+		if (is_mutable)
+		{
+			out_buffer = static_cast<char_t*>(const_cast<void*>(contents));
+			out_length = length;
+		}
+		else
+		{
+			char_t* buffer = static_cast<char_t*>(xml_memory::allocate((length + 1) * sizeof(char_t)));
+			if (!buffer) return false;
+
+			if (contents)
+				memcpy(buffer, contents, length * sizeof(char_t));
+			else
+				assert(length == 0);
+
+			buffer[length] = 0;
+
+			out_buffer = buffer;
+			out_length = length + 1;
+		}
+
+		return true;
+	}
+
+#ifdef PUGIXML_WCHAR_MODE
+	PUGI__FN bool need_endian_swap_utf(xml_encoding le, xml_encoding re)
+	{
+		return (le == encoding_utf16_be && re == encoding_utf16_le) || (le == encoding_utf16_le && re == encoding_utf16_be) ||
+			   (le == encoding_utf32_be && re == encoding_utf32_le) || (le == encoding_utf32_le && re == encoding_utf32_be);
+	}
+
+	PUGI__FN bool convert_buffer_endian_swap(char_t*& out_buffer, size_t& out_length, const void* contents, size_t size, bool is_mutable)
+	{
+		const char_t* data = static_cast<const char_t*>(contents);
+		size_t length = size / sizeof(char_t);
+
+		if (is_mutable)
+		{
+			char_t* buffer = const_cast<char_t*>(data);
+
+			convert_wchar_endian_swap(buffer, data, length);
+
+			out_buffer = buffer;
+			out_length = length;
+		}
+		else
+		{
+			char_t* buffer = static_cast<char_t*>(xml_memory::allocate((length + 1) * sizeof(char_t)));
+			if (!buffer) return false;
+
+			convert_wchar_endian_swap(buffer, data, length);
+			buffer[length] = 0;
+
+			out_buffer = buffer;
+			out_length = length + 1;
+		}
+
+		return true;
+	}
+
+	PUGI__FN bool convert_buffer_utf8(char_t*& out_buffer, size_t& out_length, const void* contents, size_t size)
+	{
+		const uint8_t* data = static_cast<const uint8_t*>(contents);
+		size_t data_length = size;
+
+		// first pass: get length in wchar_t units
+		size_t length = utf_decoder<wchar_counter>::decode_utf8_block(data, data_length, 0);
+
+		// allocate buffer of suitable length
+		char_t* buffer = static_cast<char_t*>(xml_memory::allocate((length + 1) * sizeof(char_t)));
+		if (!buffer) return false;
+
+		// second pass: convert utf8 input to wchar_t
+		wchar_writer::value_type obegin = reinterpret_cast<wchar_writer::value_type>(buffer);
+		wchar_writer::value_type oend = utf_decoder<wchar_writer>::decode_utf8_block(data, data_length, obegin);
+
+		assert(oend == obegin + length);
+		*oend = 0;
+
+		out_buffer = buffer;
+		out_length = length + 1;
+
+		return true;
+	}
+
+	template <typename opt_swap> PUGI__FN bool convert_buffer_utf16(char_t*& out_buffer, size_t& out_length, const void* contents, size_t size, opt_swap)
+	{
+		const uint16_t* data = static_cast<const uint16_t*>(contents);
+		size_t data_length = size / sizeof(uint16_t);
+
+		// first pass: get length in wchar_t units
+		size_t length = utf_decoder<wchar_counter, opt_swap>::decode_utf16_block(data, data_length, 0);
+
+		// allocate buffer of suitable length
+		char_t* buffer = static_cast<char_t*>(xml_memory::allocate((length + 1) * sizeof(char_t)));
+		if (!buffer) return false;
+
+		// second pass: convert utf16 input to wchar_t
+		wchar_writer::value_type obegin = reinterpret_cast<wchar_writer::value_type>(buffer);
+		wchar_writer::value_type oend = utf_decoder<wchar_writer, opt_swap>::decode_utf16_block(data, data_length, obegin);
+
+		assert(oend == obegin + length);
+		*oend = 0;
+
+		out_buffer = buffer;
+		out_length = length + 1;
+
+		return true;
+	}
+
+	template <typename opt_swap> PUGI__FN bool convert_buffer_utf32(char_t*& out_buffer, size_t& out_length, const void* contents, size_t size, opt_swap)
+	{
+		const uint32_t* data = static_cast<const uint32_t*>(contents);
+		size_t data_length = size / sizeof(uint32_t);
+
+		// first pass: get length in wchar_t units
+		size_t length = utf_decoder<wchar_counter, opt_swap>::decode_utf32_block(data, data_length, 0);
+
+		// allocate buffer of suitable length
+		char_t* buffer = static_cast<char_t*>(xml_memory::allocate((length + 1) * sizeof(char_t)));
+		if (!buffer) return false;
+
+		// second pass: convert utf32 input to wchar_t
+		wchar_writer::value_type obegin = reinterpret_cast<wchar_writer::value_type>(buffer);
+		wchar_writer::value_type oend = utf_decoder<wchar_writer, opt_swap>::decode_utf32_block(data, data_length, obegin);
+
+		assert(oend == obegin + length);
+		*oend = 0;
+
+		out_buffer = buffer;
+		out_length = length + 1;
+
+		return true;
+	}
+
+	PUGI__FN bool convert_buffer_latin1(char_t*& out_buffer, size_t& out_length, const void* contents, size_t size)
+	{
+		const uint8_t* data = static_cast<const uint8_t*>(contents);
+		size_t data_length = size;
+
+		// get length in wchar_t units
+		size_t length = data_length;
+
+		// allocate buffer of suitable length
+		char_t* buffer = static_cast<char_t*>(xml_memory::allocate((length + 1) * sizeof(char_t)));
+		if (!buffer) return false;
+
+		// convert latin1 input to wchar_t
+		wchar_writer::value_type obegin = reinterpret_cast<wchar_writer::value_type>(buffer);
+		wchar_writer::value_type oend = utf_decoder<wchar_writer>::decode_latin1_block(data, data_length, obegin);
+
+		assert(oend == obegin + length);
+		*oend = 0;
+
+		out_buffer = buffer;
+		out_length = length + 1;
+
+		return true;
+	}
+
+	PUGI__FN bool convert_buffer(char_t*& out_buffer, size_t& out_length, xml_encoding encoding, const void* contents, size_t size, bool is_mutable)
+	{
+		// get native encoding
+		xml_encoding wchar_encoding = get_wchar_encoding();
+
+		// fast path: no conversion required
+		if (encoding == wchar_encoding) return get_mutable_buffer(out_buffer, out_length, contents, size, is_mutable);
+
+		// only endian-swapping is required
+		if (need_endian_swap_utf(encoding, wchar_encoding)) return convert_buffer_endian_swap(out_buffer, out_length, contents, size, is_mutable);
+
+		// source encoding is utf8
+		if (encoding == encoding_utf8) return convert_buffer_utf8(out_buffer, out_length, contents, size);
+
+		// source encoding is utf16
+		if (encoding == encoding_utf16_be || encoding == encoding_utf16_le)
+		{
+			xml_encoding native_encoding = is_little_endian() ? encoding_utf16_le : encoding_utf16_be;
+
+			return (native_encoding == encoding) ?
+				convert_buffer_utf16(out_buffer, out_length, contents, size, opt_false()) :
+				convert_buffer_utf16(out_buffer, out_length, contents, size, opt_true());
+		}
+
+		// source encoding is utf32
+		if (encoding == encoding_utf32_be || encoding == encoding_utf32_le)
+		{
+			xml_encoding native_encoding = is_little_endian() ? encoding_utf32_le : encoding_utf32_be;
+
+			return (native_encoding == encoding) ?
+				convert_buffer_utf32(out_buffer, out_length, contents, size, opt_false()) :
+				convert_buffer_utf32(out_buffer, out_length, contents, size, opt_true());
+		}
+
+		// source encoding is latin1
+		if (encoding == encoding_latin1) return convert_buffer_latin1(out_buffer, out_length, contents, size);
+
+		assert(!"Invalid encoding");
+		return false;
+	}
+#else
+	template <typename opt_swap> PUGI__FN bool convert_buffer_utf16(char_t*& out_buffer, size_t& out_length, const void* contents, size_t size, opt_swap)
+	{
+		const uint16_t* data = static_cast<const uint16_t*>(contents);
+		size_t data_length = size / sizeof(uint16_t);
+
+		// first pass: get length in utf8 units
+		size_t length = utf_decoder<utf8_counter, opt_swap>::decode_utf16_block(data, data_length, 0);
+
+		// allocate buffer of suitable length
+		char_t* buffer = static_cast<char_t*>(xml_memory::allocate((length + 1) * sizeof(char_t)));
+		if (!buffer) return false;
+
+		// second pass: convert utf16 input to utf8
+		uint8_t* obegin = reinterpret_cast<uint8_t*>(buffer);
+		uint8_t* oend = utf_decoder<utf8_writer, opt_swap>::decode_utf16_block(data, data_length, obegin);
+
+		assert(oend == obegin + length);
+		*oend = 0;
+
+		out_buffer = buffer;
+		out_length = length + 1;
+
+		return true;
+	}
+
+	template <typename opt_swap> PUGI__FN bool convert_buffer_utf32(char_t*& out_buffer, size_t& out_length, const void* contents, size_t size, opt_swap)
+	{
+		const uint32_t* data = static_cast<const uint32_t*>(contents);
+		size_t data_length = size / sizeof(uint32_t);
+
+		// first pass: get length in utf8 units
+		size_t length = utf_decoder<utf8_counter, opt_swap>::decode_utf32_block(data, data_length, 0);
+
+		// allocate buffer of suitable length
+		char_t* buffer = static_cast<char_t*>(xml_memory::allocate((length + 1) * sizeof(char_t)));
+		if (!buffer) return false;
+
+		// second pass: convert utf32 input to utf8
+		uint8_t* obegin = reinterpret_cast<uint8_t*>(buffer);
+		uint8_t* oend = utf_decoder<utf8_writer, opt_swap>::decode_utf32_block(data, data_length, obegin);
+
+		assert(oend == obegin + length);
+		*oend = 0;
+
+		out_buffer = buffer;
+		out_length = length + 1;
+
+		return true;
+	}
+
+	PUGI__FN size_t get_latin1_7bit_prefix_length(const uint8_t* data, size_t size)
+	{
+		for (size_t i = 0; i < size; ++i)
+			if (data[i] > 127)
+				return i;
+
+		return size;
+	}
+
+	PUGI__FN bool convert_buffer_latin1(char_t*& out_buffer, size_t& out_length, const void* contents, size_t size, bool is_mutable)
+	{
+		const uint8_t* data = static_cast<const uint8_t*>(contents);
+		size_t data_length = size;
+
+		// get size of prefix that does not need utf8 conversion
+		size_t prefix_length = get_latin1_7bit_prefix_length(data, data_length);
+		assert(prefix_length <= data_length);
+
+		const uint8_t* postfix = data + prefix_length;
+		size_t postfix_length = data_length - prefix_length;
+
+		// if no conversion is needed, just return the original buffer
+		if (postfix_length == 0) return get_mutable_buffer(out_buffer, out_length, contents, size, is_mutable);
+
+		// first pass: get length in utf8 units
+		size_t length = prefix_length + utf_decoder<utf8_counter>::decode_latin1_block(postfix, postfix_length, 0);
+
+		// allocate buffer of suitable length
+		char_t* buffer = static_cast<char_t*>(xml_memory::allocate((length + 1) * sizeof(char_t)));
+		if (!buffer) return false;
+
+		// second pass: convert latin1 input to utf8
+		memcpy(buffer, data, prefix_length);
+
+		uint8_t* obegin = reinterpret_cast<uint8_t*>(buffer);
+		uint8_t* oend = utf_decoder<utf8_writer>::decode_latin1_block(postfix, postfix_length, obegin + prefix_length);
+
+		assert(oend == obegin + length);
+		*oend = 0;
+
+		out_buffer = buffer;
+		out_length = length + 1;
+
+		return true;
+	}
+
+	PUGI__FN bool convert_buffer(char_t*& out_buffer, size_t& out_length, xml_encoding encoding, const void* contents, size_t size, bool is_mutable)
+	{
+		// fast path: no conversion required
+		if (encoding == encoding_utf8) return get_mutable_buffer(out_buffer, out_length, contents, size, is_mutable);
+
+		// source encoding is utf16
+		if (encoding == encoding_utf16_be || encoding == encoding_utf16_le)
+		{
+			xml_encoding native_encoding = is_little_endian() ? encoding_utf16_le : encoding_utf16_be;
+
+			return (native_encoding == encoding) ?
+				convert_buffer_utf16(out_buffer, out_length, contents, size, opt_false()) :
+				convert_buffer_utf16(out_buffer, out_length, contents, size, opt_true());
+		}
+
+		// source encoding is utf32
+		if (encoding == encoding_utf32_be || encoding == encoding_utf32_le)
+		{
+			xml_encoding native_encoding = is_little_endian() ? encoding_utf32_le : encoding_utf32_be;
+
+			return (native_encoding == encoding) ?
+				convert_buffer_utf32(out_buffer, out_length, contents, size, opt_false()) :
+				convert_buffer_utf32(out_buffer, out_length, contents, size, opt_true());
+		}
+
+		// source encoding is latin1
+		if (encoding == encoding_latin1) return convert_buffer_latin1(out_buffer, out_length, contents, size, is_mutable);
+
+		assert(!"Invalid encoding");
+		return false;
+	}
+#endif
+
+	PUGI__FN size_t as_utf8_begin(const wchar_t* str, size_t length)
+	{
+		// get length in utf8 characters
+		return utf_decoder<utf8_counter>::decode_wchar_block(str, length, 0);
+	}
+
+	PUGI__FN void as_utf8_end(char* buffer, size_t size, const wchar_t* str, size_t length)
+	{
+		// convert to utf8
+		uint8_t* begin = reinterpret_cast<uint8_t*>(buffer);
+		uint8_t* end = utf_decoder<utf8_writer>::decode_wchar_block(str, length, begin);
+	
+		assert(begin + size == end);
+		(void)!end;
+		(void)!size;
+	}
+	
+#ifndef PUGIXML_NO_STL
+	PUGI__FN std::string as_utf8_impl(const wchar_t* str, size_t length)
+	{
+		// first pass: get length in utf8 characters
+		size_t size = as_utf8_begin(str, length);
+
+		// allocate resulting string
+		std::string result;
+		result.resize(size);
+
+		// second pass: convert to utf8
+		if (size > 0) as_utf8_end(&result[0], size, str, length);
+
+		return result;
+	}
+
+	PUGI__FN std::basic_string<wchar_t> as_wide_impl(const char* str, size_t size)
+	{
+		const uint8_t* data = reinterpret_cast<const uint8_t*>(str);
+
+		// first pass: get length in wchar_t units
+		size_t length = utf_decoder<wchar_counter>::decode_utf8_block(data, size, 0);
+
+		// allocate resulting string
+		std::basic_string<wchar_t> result;
+		result.resize(length);
+
+		// second pass: convert to wchar_t
+		if (length > 0)
+		{
+			wchar_writer::value_type begin = reinterpret_cast<wchar_writer::value_type>(&result[0]);
+			wchar_writer::value_type end = utf_decoder<wchar_writer>::decode_utf8_block(data, size, begin);
+
+			assert(begin + length == end);
+			(void)!end;
+		}
+
+		return result;
+	}
+#endif
+
+	template <typename Header>
+	inline bool strcpy_insitu_allow(size_t length, const Header& header, uintptr_t header_mask, char_t* target)
+	{
+		// never reuse shared memory
+		if (header & xml_memory_page_contents_shared_mask) return false;
+
+		size_t target_length = strlength(target);
+
+		// always reuse document buffer memory if possible
+		if ((header & header_mask) == 0) return target_length >= length;
+
+		// reuse heap memory if waste is not too great
+		const size_t reuse_threshold = 32;
+
+		return target_length >= length && (target_length < reuse_threshold || target_length - length < target_length / 2);
+	}
+
+	template <typename String, typename Header>
+	PUGI__FN bool strcpy_insitu(String& dest, Header& header, uintptr_t header_mask, const char_t* source)
+	{
+		size_t source_length = strlength(source);
+
+		if (source_length == 0)
+		{
+			// empty string and null pointer are equivalent, so just deallocate old memory
+			xml_allocator* alloc = PUGI__GETPAGE_IMPL(header)->allocator;
+
+			if (header & header_mask) alloc->deallocate_string(dest);
+			
+			// mark the string as not allocated
+			dest = 0;
+			header &= ~header_mask;
+
+			return true;
+		}
+		else if (dest && strcpy_insitu_allow(source_length, header, header_mask, dest))
+		{
+			// we can reuse old buffer, so just copy the new data (including zero terminator)
+			memcpy(dest, source, (source_length + 1) * sizeof(char_t));
+			
+			return true;
+		}
+		else
+		{
+			xml_allocator* alloc = PUGI__GETPAGE_IMPL(header)->allocator;
+
+			if (!alloc->reserve()) return false;
+
+			// allocate new buffer
+			char_t* buf = alloc->allocate_string(source_length + 1);
+			if (!buf) return false;
+
+			// copy the string (including zero terminator)
+			memcpy(buf, source, (source_length + 1) * sizeof(char_t));
+
+			// deallocate old buffer (*after* the above to protect against overlapping memory and/or allocation failures)
+			if (header & header_mask) alloc->deallocate_string(dest);
+			
+			// the string is now allocated, so set the flag
+			dest = buf;
+			header |= header_mask;
+
+			return true;
+		}
+	}
+
+	struct gap
+	{
+		char_t* end;
+		size_t size;
+			
+		gap(): end(0), size(0)
+		{
+		}
+			
+		// Push new gap, move s count bytes further (skipping the gap).
+		// Collapse previous gap.
+		void push(char_t*& s, size_t count)
+		{
+			if (end) // there was a gap already; collapse it
+			{
+				// Move [old_gap_end, new_gap_start) to [old_gap_start, ...)
+				assert(s >= end);
+				memmove(end - size, end, reinterpret_cast<char*>(s) - reinterpret_cast<char*>(end));
+			}
+				
+			s += count; // end of current gap
+				
+			// "merge" two gaps
+			end = s;
+			size += count;
+		}
+			
+		// Collapse all gaps, return past-the-end pointer
+		char_t* flush(char_t* s)
+		{
+			if (end)
+			{
+				// Move [old_gap_end, current_pos) to [old_gap_start, ...)
+				assert(s >= end);
+				memmove(end - size, end, reinterpret_cast<char*>(s) - reinterpret_cast<char*>(end));
+
+				return s - size;
+			}
+			else return s;
+		}
+	};
+	
+	PUGI__FN char_t* strconv_escape(char_t* s, gap& g)
+	{
+		char_t* stre = s + 1;
+
+		switch (*stre)
+		{
+			case '#':	// &#...
+			{
+				unsigned int ucsc = 0;
+
+				if (stre[1] == 'x') // &#x... (hex code)
+				{
+					stre += 2;
+
+					char_t ch = *stre;
+
+					if (ch == ';') return stre;
+
+					for (;;)
+					{
+						if (static_cast<unsigned int>(ch - '0') <= 9)
+							ucsc = 16 * ucsc + (ch - '0');
+						else if (static_cast<unsigned int>((ch | ' ') - 'a') <= 5)
+							ucsc = 16 * ucsc + ((ch | ' ') - 'a' + 10);
+						else if (ch == ';')
+							break;
+						else // cancel
+							return stre;
+
+						ch = *++stre;
+					}
+					
+					++stre;
+				}
+				else	// &#... (dec code)
+				{
+					char_t ch = *++stre;
+
+					if (ch == ';') return stre;
+
+					for (;;)
+					{
+						if (static_cast<unsigned int>(static_cast<unsigned int>(ch) - '0') <= 9)
+							ucsc = 10 * ucsc + (ch - '0');
+						else if (ch == ';')
+							break;
+						else // cancel
+							return stre;
+
+						ch = *++stre;
+					}
+					
+					++stre;
+				}
+
+			#ifdef PUGIXML_WCHAR_MODE
+				s = reinterpret_cast<char_t*>(wchar_writer::any(reinterpret_cast<wchar_writer::value_type>(s), ucsc));
+			#else
+				s = reinterpret_cast<char_t*>(utf8_writer::any(reinterpret_cast<uint8_t*>(s), ucsc));
+			#endif
+					
+				g.push(s, stre - s);
+				return stre;
+			}
+
+			case 'a':	// &a
+			{
+				++stre;
+
+				if (*stre == 'm') // &am
+				{
+					if (*++stre == 'p' && *++stre == ';') // &amp;
+					{
+						*s++ = '&';
+						++stre;
+							
+						g.push(s, stre - s);
+						return stre;
+					}
+				}
+				else if (*stre == 'p') // &ap
+				{
+					if (*++stre == 'o' && *++stre == 's' && *++stre == ';') // &apos;
+					{
+						*s++ = '\'';
+						++stre;
+
+						g.push(s, stre - s);
+						return stre;
+					}
+				}
+				break;
+			}
+
+			case 'g': // &g
+			{
+				if (*++stre == 't' && *++stre == ';') // &gt;
+				{
+					*s++ = '>';
+					++stre;
+					
+					g.push(s, stre - s);
+					return stre;
+				}
+				break;
+			}
+
+			case 'l': // &l
+			{
+				if (*++stre == 't' && *++stre == ';') // &lt;
+				{
+					*s++ = '<';
+					++stre;
+						
+					g.push(s, stre - s);
+					return stre;
+				}
+				break;
+			}
+
+			case 'q': // &q
+			{
+				if (*++stre == 'u' && *++stre == 'o' && *++stre == 't' && *++stre == ';') // &quot;
+				{
+					*s++ = '"';
+					++stre;
+					
+					g.push(s, stre - s);
+					return stre;
+				}
+				break;
+			}
+
+			default:
+				break;
+		}
+		
+		return stre;
+	}
+
+	// Parser utilities
+	#define PUGI__ENDSWITH(c, e)        ((c) == (e) || ((c) == 0 && endch == (e)))
+	#define PUGI__SKIPWS()              { while (PUGI__IS_CHARTYPE(*s, ct_space)) ++s; }
+	#define PUGI__OPTSET(OPT)           ( optmsk & (OPT) )
+	#define PUGI__PUSHNODE(TYPE)        { cursor = append_new_node(cursor, alloc, TYPE); if (!cursor) PUGI__THROW_ERROR(status_out_of_memory, s); }
+	#define PUGI__POPNODE()             { cursor = cursor->parent; }
+	#define PUGI__SCANFOR(X)            { while (*s != 0 && !(X)) ++s; }
+	#define PUGI__SCANWHILE(X)          { while (X) ++s; }
+	#define PUGI__SCANWHILE_UNROLL(X)   { for (;;) { char_t ss = s[0]; if (PUGI__UNLIKELY(!(X))) { break; } ss = s[1]; if (PUGI__UNLIKELY(!(X))) { s += 1; break; } ss = s[2]; if (PUGI__UNLIKELY(!(X))) { s += 2; break; } ss = s[3]; if (PUGI__UNLIKELY(!(X))) { s += 3; break; } s += 4; } }
+	#define PUGI__ENDSEG()              { ch = *s; *s = 0; ++s; }
+	#define PUGI__THROW_ERROR(err, m)   return error_offset = m, error_status = err, static_cast<char_t*>(0)
+	#define PUGI__CHECK_ERROR(err, m)   { if (*s == 0) PUGI__THROW_ERROR(err, m); }
+
+	PUGI__FN char_t* strconv_comment(char_t* s, char_t endch)
+	{
+		gap g;
+		
+		while (true)
+		{
+			PUGI__SCANWHILE_UNROLL(!PUGI__IS_CHARTYPE(ss, ct_parse_comment));
+		
+			if (*s == '\r') // Either a single 0x0d or 0x0d 0x0a pair
+			{
+				*s++ = '\n'; // replace first one with 0x0a
+				
+				if (*s == '\n') g.push(s, 1);
+			}
+			else if (s[0] == '-' && s[1] == '-' && PUGI__ENDSWITH(s[2], '>')) // comment ends here
+			{
+				*g.flush(s) = 0;
+				
+				return s + (s[2] == '>' ? 3 : 2);
+			}
+			else if (*s == 0)
+			{
+				return 0;
+			}
+			else ++s;
+		}
+	}
+
+	PUGI__FN char_t* strconv_cdata(char_t* s, char_t endch)
+	{
+		gap g;
+			
+		while (true)
+		{
+			PUGI__SCANWHILE_UNROLL(!PUGI__IS_CHARTYPE(ss, ct_parse_cdata));
+			
+			if (*s == '\r') // Either a single 0x0d or 0x0d 0x0a pair
+			{
+				*s++ = '\n'; // replace first one with 0x0a
+				
+				if (*s == '\n') g.push(s, 1);
+			}
+			else if (s[0] == ']' && s[1] == ']' && PUGI__ENDSWITH(s[2], '>')) // CDATA ends here
+			{
+				*g.flush(s) = 0;
+				
+				return s + 1;
+			}
+			else if (*s == 0)
+			{
+				return 0;
+			}
+			else ++s;
+		}
+	}
+	
+	typedef char_t* (*strconv_pcdata_t)(char_t*);
+		
+	template <typename opt_trim, typename opt_eol, typename opt_escape> struct strconv_pcdata_impl
+	{
+		static char_t* parse(char_t* s)
+		{
+			gap g;
+
+			char_t* begin = s;
+
+			while (true)
+			{
+				PUGI__SCANWHILE_UNROLL(!PUGI__IS_CHARTYPE(ss, ct_parse_pcdata));
+
+				if (*s == '<') // PCDATA ends here
+				{
+					char_t* end = g.flush(s);
+
+					if (opt_trim::value)
+						while (end > begin && PUGI__IS_CHARTYPE(end[-1], ct_space))
+							--end;
+
+					*end = 0;
+					
+					return s + 1;
+				}
+				else if (opt_eol::value && *s == '\r') // Either a single 0x0d or 0x0d 0x0a pair
+				{
+					*s++ = '\n'; // replace first one with 0x0a
+					
+					if (*s == '\n') g.push(s, 1);
+				}
+				else if (opt_escape::value && *s == '&')
+				{
+					s = strconv_escape(s, g);
+				}
+				else if (*s == 0)
+				{
+					char_t* end = g.flush(s);
+
+					if (opt_trim::value)
+						while (end > begin && PUGI__IS_CHARTYPE(end[-1], ct_space))
+							--end;
+
+					*end = 0;
+
+					return s;
+				}
+				else ++s;
+			}
+		}
+	};
+	
+	PUGI__FN strconv_pcdata_t get_strconv_pcdata(unsigned int optmask)
+	{
+		PUGI__STATIC_ASSERT(parse_escapes == 0x10 && parse_eol == 0x20 && parse_trim_pcdata == 0x0800);
+
+		switch (((optmask >> 4) & 3) | ((optmask >> 9) & 4)) // get bitmask for flags (eol escapes trim)
+		{
+		case 0: return strconv_pcdata_impl<opt_false, opt_false, opt_false>::parse;
+		case 1: return strconv_pcdata_impl<opt_false, opt_false, opt_true>::parse;
+		case 2: return strconv_pcdata_impl<opt_false, opt_true, opt_false>::parse;
+		case 3: return strconv_pcdata_impl<opt_false, opt_true, opt_true>::parse;
+		case 4: return strconv_pcdata_impl<opt_true, opt_false, opt_false>::parse;
+		case 5: return strconv_pcdata_impl<opt_true, opt_false, opt_true>::parse;
+		case 6: return strconv_pcdata_impl<opt_true, opt_true, opt_false>::parse;
+		case 7: return strconv_pcdata_impl<opt_true, opt_true, opt_true>::parse;
+		default: assert(false); return 0; // should not get here
+		}
+	}
+
+	typedef char_t* (*strconv_attribute_t)(char_t*, char_t);
+	
+	template <typename opt_escape> struct strconv_attribute_impl
+	{
+		static char_t* parse_wnorm(char_t* s, char_t end_quote)
+		{
+			gap g;
+
+			// trim leading whitespaces
+			if (PUGI__IS_CHARTYPE(*s, ct_space))
+			{
+				char_t* str = s;
+				
+				do ++str;
+				while (PUGI__IS_CHARTYPE(*str, ct_space));
+				
+				g.push(s, str - s);
+			}
+
+			while (true)
+			{
+				PUGI__SCANWHILE_UNROLL(!PUGI__IS_CHARTYPE(ss, ct_parse_attr_ws | ct_space));
+				
+				if (*s == end_quote)
+				{
+					char_t* str = g.flush(s);
+					
+					do *str-- = 0;
+					while (PUGI__IS_CHARTYPE(*str, ct_space));
+				
+					return s + 1;
+				}
+				else if (PUGI__IS_CHARTYPE(*s, ct_space))
+				{
+					*s++ = ' ';
+		
+					if (PUGI__IS_CHARTYPE(*s, ct_space))
+					{
+						char_t* str = s + 1;
+						while (PUGI__IS_CHARTYPE(*str, ct_space)) ++str;
+						
+						g.push(s, str - s);
+					}
+				}
+				else if (opt_escape::value && *s == '&')
+				{
+					s = strconv_escape(s, g);
+				}
+				else if (!*s)
+				{
+					return 0;
+				}
+				else ++s;
+			}
+		}
+
+		static char_t* parse_wconv(char_t* s, char_t end_quote)
+		{
+			gap g;
+
+			while (true)
+			{
+				PUGI__SCANWHILE_UNROLL(!PUGI__IS_CHARTYPE(ss, ct_parse_attr_ws));
+				
+				if (*s == end_quote)
+				{
+					*g.flush(s) = 0;
+				
+					return s + 1;
+				}
+				else if (PUGI__IS_CHARTYPE(*s, ct_space))
+				{
+					if (*s == '\r')
+					{
+						*s++ = ' ';
+				
+						if (*s == '\n') g.push(s, 1);
+					}
+					else *s++ = ' ';
+				}
+				else if (opt_escape::value && *s == '&')
+				{
+					s = strconv_escape(s, g);
+				}
+				else if (!*s)
+				{
+					return 0;
+				}
+				else ++s;
+			}
+		}
+
+		static char_t* parse_eol(char_t* s, char_t end_quote)
+		{
+			gap g;
+
+			while (true)
+			{
+				PUGI__SCANWHILE_UNROLL(!PUGI__IS_CHARTYPE(ss, ct_parse_attr));
+				
+				if (*s == end_quote)
+				{
+					*g.flush(s) = 0;
+				
+					return s + 1;
+				}
+				else if (*s == '\r')
+				{
+					*s++ = '\n';
+					
+					if (*s == '\n') g.push(s, 1);
+				}
+				else if (opt_escape::value && *s == '&')
+				{
+					s = strconv_escape(s, g);
+				}
+				else if (!*s)
+				{
+					return 0;
+				}
+				else ++s;
+			}
+		}
+
+		static char_t* parse_simple(char_t* s, char_t end_quote)
+		{
+			gap g;
+
+			while (true)
+			{
+				PUGI__SCANWHILE_UNROLL(!PUGI__IS_CHARTYPE(ss, ct_parse_attr));
+				
+				if (*s == end_quote)
+				{
+					*g.flush(s) = 0;
+				
+					return s + 1;
+				}
+				else if (opt_escape::value && *s == '&')
+				{
+					s = strconv_escape(s, g);
+				}
+				else if (!*s)
+				{
+					return 0;
+				}
+				else ++s;
+			}
+		}
+	};
+
+	PUGI__FN strconv_attribute_t get_strconv_attribute(unsigned int optmask)
+	{
+		PUGI__STATIC_ASSERT(parse_escapes == 0x10 && parse_eol == 0x20 && parse_wconv_attribute == 0x40 && parse_wnorm_attribute == 0x80);
+		
+		switch ((optmask >> 4) & 15) // get bitmask for flags (wconv wnorm eol escapes)
+		{
+		case 0:  return strconv_attribute_impl<opt_false>::parse_simple;
+		case 1:  return strconv_attribute_impl<opt_true>::parse_simple;
+		case 2:  return strconv_attribute_impl<opt_false>::parse_eol;
+		case 3:  return strconv_attribute_impl<opt_true>::parse_eol;
+		case 4:  return strconv_attribute_impl<opt_false>::parse_wconv;
+		case 5:  return strconv_attribute_impl<opt_true>::parse_wconv;
+		case 6:  return strconv_attribute_impl<opt_false>::parse_wconv;
+		case 7:  return strconv_attribute_impl<opt_true>::parse_wconv;
+		case 8:  return strconv_attribute_impl<opt_false>::parse_wnorm;
+		case 9:  return strconv_attribute_impl<opt_true>::parse_wnorm;
+		case 10: return strconv_attribute_impl<opt_false>::parse_wnorm;
+		case 11: return strconv_attribute_impl<opt_true>::parse_wnorm;
+		case 12: return strconv_attribute_impl<opt_false>::parse_wnorm;
+		case 13: return strconv_attribute_impl<opt_true>::parse_wnorm;
+		case 14: return strconv_attribute_impl<opt_false>::parse_wnorm;
+		case 15: return strconv_attribute_impl<opt_true>::parse_wnorm;
+		default: assert(false); return 0; // should not get here
+		}
+	}
+
+	inline xml_parse_result make_parse_result(xml_parse_status status, ptrdiff_t offset = 0)
+	{
+		xml_parse_result result;
+		result.status = status;
+		result.offset = offset;
+
+		return result;
+	}
+
+	struct xml_parser
+	{
+		xml_allocator alloc;
+		xml_allocator* alloc_state;
+		char_t* error_offset;
+		xml_parse_status error_status;
+		
+		xml_parser(xml_allocator* alloc_): alloc(*alloc_), alloc_state(alloc_), error_offset(0), error_status(status_ok)
+		{
+		}
+
+		~xml_parser()
+		{
+			*alloc_state = alloc;
+		}
+
+		// DOCTYPE consists of nested sections of the following possible types:
+		// <!-- ... -->, <? ... ?>, "...", '...'
+		// <![...]]>
+		// <!...>
+		// First group can not contain nested groups
+		// Second group can contain nested groups of the same type
+		// Third group can contain all other groups
+		char_t* parse_doctype_primitive(char_t* s)
+		{
+			if (*s == '"' || *s == '\'')
+			{
+				// quoted string
+				char_t ch = *s++;
+				PUGI__SCANFOR(*s == ch);
+				if (!*s) PUGI__THROW_ERROR(status_bad_doctype, s);
+
+				s++;
+			}
+			else if (s[0] == '<' && s[1] == '?')
+			{
+				// <? ... ?>
+				s += 2;
+				PUGI__SCANFOR(s[0] == '?' && s[1] == '>'); // no need for ENDSWITH because ?> can't terminate proper doctype
+				if (!*s) PUGI__THROW_ERROR(status_bad_doctype, s);
+
+				s += 2;
+			}
+			else if (s[0] == '<' && s[1] == '!' && s[2] == '-' && s[3] == '-')
+			{
+				s += 4;
+				PUGI__SCANFOR(s[0] == '-' && s[1] == '-' && s[2] == '>'); // no need for ENDSWITH because --> can't terminate proper doctype
+				if (!*s) PUGI__THROW_ERROR(status_bad_doctype, s);
+
+				s += 3;
+			}
+			else PUGI__THROW_ERROR(status_bad_doctype, s);
+
+			return s;
+		}
+
+		char_t* parse_doctype_ignore(char_t* s)
+		{
+			size_t depth = 0;
+
+			assert(s[0] == '<' && s[1] == '!' && s[2] == '[');
+			s += 3;
+
+			while (*s)
+			{
+				if (s[0] == '<' && s[1] == '!' && s[2] == '[')
+				{
+					// nested ignore section
+					s += 3;
+					depth++;
+				}
+				else if (s[0] == ']' && s[1] == ']' && s[2] == '>')
+				{
+					// ignore section end
+					s += 3;
+
+					if (depth == 0)
+						return s;
+
+					depth--;
+				}
+				else s++;
+			}
+
+			PUGI__THROW_ERROR(status_bad_doctype, s);
+		}
+
+		char_t* parse_doctype_group(char_t* s, char_t endch)
+		{
+			size_t depth = 0;
+
+			assert((s[0] == '<' || s[0] == 0) && s[1] == '!');
+			s += 2;
+
+			while (*s)
+			{
+				if (s[0] == '<' && s[1] == '!' && s[2] != '-')
+				{
+					if (s[2] == '[')
+					{
+						// ignore
+						s = parse_doctype_ignore(s);
+						if (!s) return s;
+					}
+					else
+					{
+						// some control group
+						s += 2;
+						depth++;
+					}
+				}
+				else if (s[0] == '<' || s[0] == '"' || s[0] == '\'')
+				{
+					// unknown tag (forbidden), or some primitive group
+					s = parse_doctype_primitive(s);
+					if (!s) return s;
+				}
+				else if (*s == '>')
+				{
+					if (depth == 0)
+						return s;
+
+					depth--;
+					s++;
+				}
+				else s++;
+			}
+
+			if (depth != 0 || endch != '>') PUGI__THROW_ERROR(status_bad_doctype, s);
+
+			return s;
+		}
+
+		char_t* parse_exclamation(char_t* s, xml_node_struct* cursor, unsigned int optmsk, char_t endch)
+		{
+			// parse node contents, starting with exclamation mark
+			++s;
+
+			if (*s == '-') // '<!-...'
+			{
+				++s;
+
+				if (*s == '-') // '<!--...'
+				{
+					++s;
+
+					if (PUGI__OPTSET(parse_comments))
+					{
+						PUGI__PUSHNODE(node_comment); // Append a new node on the tree.
+						cursor->value = s; // Save the offset.
+					}
+
+					if (PUGI__OPTSET(parse_eol) && PUGI__OPTSET(parse_comments))
+					{
+						s = strconv_comment(s, endch);
+
+						if (!s) PUGI__THROW_ERROR(status_bad_comment, cursor->value);
+					}
+					else
+					{
+						// Scan for terminating '-->'.
+						PUGI__SCANFOR(s[0] == '-' && s[1] == '-' && PUGI__ENDSWITH(s[2], '>'));
+						PUGI__CHECK_ERROR(status_bad_comment, s);
+
+						if (PUGI__OPTSET(parse_comments))
+							*s = 0; // Zero-terminate this segment at the first terminating '-'.
+
+						s += (s[2] == '>' ? 3 : 2); // Step over the '\0->'.
+					}
+				}
+				else PUGI__THROW_ERROR(status_bad_comment, s);
+			}
+			else if (*s == '[')
+			{
+				// '<![CDATA[...'
+				if (*++s=='C' && *++s=='D' && *++s=='A' && *++s=='T' && *++s=='A' && *++s == '[')
+				{
+					++s;
+
+					if (PUGI__OPTSET(parse_cdata))
+					{
+						PUGI__PUSHNODE(node_cdata); // Append a new node on the tree.
+						cursor->value = s; // Save the offset.
+
+						if (PUGI__OPTSET(parse_eol))
+						{
+							s = strconv_cdata(s, endch);
+
+							if (!s) PUGI__THROW_ERROR(status_bad_cdata, cursor->value);
+						}
+						else
+						{
+							// Scan for terminating ']]>'.
+							PUGI__SCANFOR(s[0] == ']' && s[1] == ']' && PUGI__ENDSWITH(s[2], '>'));
+							PUGI__CHECK_ERROR(status_bad_cdata, s);
+
+							*s++ = 0; // Zero-terminate this segment.
+						}
+					}
+					else // Flagged for discard, but we still have to scan for the terminator.
+					{
+						// Scan for terminating ']]>'.
+						PUGI__SCANFOR(s[0] == ']' && s[1] == ']' && PUGI__ENDSWITH(s[2], '>'));
+						PUGI__CHECK_ERROR(status_bad_cdata, s);
+
+						++s;
+					}
+
+					s += (s[1] == '>' ? 2 : 1); // Step over the last ']>'.
+				}
+				else PUGI__THROW_ERROR(status_bad_cdata, s);
+			}
+			else if (s[0] == 'D' && s[1] == 'O' && s[2] == 'C' && s[3] == 'T' && s[4] == 'Y' && s[5] == 'P' && PUGI__ENDSWITH(s[6], 'E'))
+			{
+				s -= 2;
+
+				if (cursor->parent) PUGI__THROW_ERROR(status_bad_doctype, s);
+
+				char_t* mark = s + 9;
+
+				s = parse_doctype_group(s, endch);
+				if (!s) return s;
+
+				assert((*s == 0 && endch == '>') || *s == '>');
+				if (*s) *s++ = 0;
+
+				if (PUGI__OPTSET(parse_doctype))
+				{
+					while (PUGI__IS_CHARTYPE(*mark, ct_space)) ++mark;
+
+					PUGI__PUSHNODE(node_doctype);
+
+					cursor->value = mark;
+				}
+			}
+			else if (*s == 0 && endch == '-') PUGI__THROW_ERROR(status_bad_comment, s);
+			else if (*s == 0 && endch == '[') PUGI__THROW_ERROR(status_bad_cdata, s);
+			else PUGI__THROW_ERROR(status_unrecognized_tag, s);
+
+			return s;
+		}
+
+		char_t* parse_question(char_t* s, xml_node_struct*& ref_cursor, unsigned int optmsk, char_t endch)
+		{
+			// load into registers
+			xml_node_struct* cursor = ref_cursor;
+			char_t ch = 0;
+
+			// parse node contents, starting with question mark
+			++s;
+
+			// read PI target
+			char_t* target = s;
+
+			if (!PUGI__IS_CHARTYPE(*s, ct_start_symbol)) PUGI__THROW_ERROR(status_bad_pi, s);
+
+			PUGI__SCANWHILE(PUGI__IS_CHARTYPE(*s, ct_symbol));
+			PUGI__CHECK_ERROR(status_bad_pi, s);
+
+			// determine node type; stricmp / strcasecmp is not portable
+			bool declaration = (target[0] | ' ') == 'x' && (target[1] | ' ') == 'm' && (target[2] | ' ') == 'l' && target + 3 == s;
+
+			if (declaration ? PUGI__OPTSET(parse_declaration) : PUGI__OPTSET(parse_pi))
+			{
+				if (declaration)
+				{
+					// disallow non top-level declarations
+					if (cursor->parent) PUGI__THROW_ERROR(status_bad_pi, s);
+
+					PUGI__PUSHNODE(node_declaration);
+				}
+				else
+				{
+					PUGI__PUSHNODE(node_pi);
+				}
+
+				cursor->name = target;
+
+				PUGI__ENDSEG();
+
+				// parse value/attributes
+				if (ch == '?')
+				{
+					// empty node
+					if (!PUGI__ENDSWITH(*s, '>')) PUGI__THROW_ERROR(status_bad_pi, s);
+					s += (*s == '>');
+
+					PUGI__POPNODE();
+				}
+				else if (PUGI__IS_CHARTYPE(ch, ct_space))
+				{
+					PUGI__SKIPWS();
+
+					// scan for tag end
+					char_t* value = s;
+
+					PUGI__SCANFOR(s[0] == '?' && PUGI__ENDSWITH(s[1], '>'));
+					PUGI__CHECK_ERROR(status_bad_pi, s);
+
+					if (declaration)
+					{
+						// replace ending ? with / so that 'element' terminates properly
+						*s = '/';
+
+						// we exit from this function with cursor at node_declaration, which is a signal to parse() to go to LOC_ATTRIBUTES
+						s = value;
+					}
+					else
+					{
+						// store value and step over >
+						cursor->value = value;
+
+						PUGI__POPNODE();
+
+						PUGI__ENDSEG();
+
+						s += (*s == '>');
+					}
+				}
+				else PUGI__THROW_ERROR(status_bad_pi, s);
+			}
+			else
+			{
+				// scan for tag end
+				PUGI__SCANFOR(s[0] == '?' && PUGI__ENDSWITH(s[1], '>'));
+				PUGI__CHECK_ERROR(status_bad_pi, s);
+
+				s += (s[1] == '>' ? 2 : 1);
+			}
+
+			// store from registers
+			ref_cursor = cursor;
+
+			return s;
+		}
+
+		char_t* parse_tree(char_t* s, xml_node_struct* root, unsigned int optmsk, char_t endch)
+		{
+			strconv_attribute_t strconv_attribute = get_strconv_attribute(optmsk);
+			strconv_pcdata_t strconv_pcdata = get_strconv_pcdata(optmsk);
+			
+			char_t ch = 0;
+			xml_node_struct* cursor = root;
+			char_t* mark = s;
+
+			while (*s != 0)
+			{
+				if (*s == '<')
+				{
+					++s;
+
+				LOC_TAG:
+					if (PUGI__IS_CHARTYPE(*s, ct_start_symbol)) // '<#...'
+					{
+						PUGI__PUSHNODE(node_element); // Append a new node to the tree.
+
+						cursor->name = s;
+
+						PUGI__SCANWHILE_UNROLL(PUGI__IS_CHARTYPE(ss, ct_symbol)); // Scan for a terminator.
+						PUGI__ENDSEG(); // Save char in 'ch', terminate & step over.
+
+						if (ch == '>')
+						{
+							// end of tag
+						}
+						else if (PUGI__IS_CHARTYPE(ch, ct_space))
+						{
+						LOC_ATTRIBUTES:
+							while (true)
+							{
+								PUGI__SKIPWS(); // Eat any whitespace.
+						
+								if (PUGI__IS_CHARTYPE(*s, ct_start_symbol)) // <... #...
+								{
+									xml_attribute_struct* a = append_new_attribute(cursor, alloc); // Make space for this attribute.
+									if (!a) PUGI__THROW_ERROR(status_out_of_memory, s);
+
+									a->name = s; // Save the offset.
+
+									PUGI__SCANWHILE_UNROLL(PUGI__IS_CHARTYPE(ss, ct_symbol)); // Scan for a terminator.
+									PUGI__ENDSEG(); // Save char in 'ch', terminate & step over.
+
+									if (PUGI__IS_CHARTYPE(ch, ct_space))
+									{
+										PUGI__SKIPWS(); // Eat any whitespace.
+
+										ch = *s;
+										++s;
+									}
+									
+									if (ch == '=') // '<... #=...'
+									{
+										PUGI__SKIPWS(); // Eat any whitespace.
+
+										if (*s == '"' || *s == '\'') // '<... #="...'
+										{
+											ch = *s; // Save quote char to avoid breaking on "''" -or- '""'.
+											++s; // Step over the quote.
+											a->value = s; // Save the offset.
+
+											s = strconv_attribute(s, ch);
+										
+											if (!s) PUGI__THROW_ERROR(status_bad_attribute, a->value);
+
+											// After this line the loop continues from the start;
+											// Whitespaces, / and > are ok, symbols and EOF are wrong,
+											// everything else will be detected
+											if (PUGI__IS_CHARTYPE(*s, ct_start_symbol)) PUGI__THROW_ERROR(status_bad_attribute, s);
+										}
+										else PUGI__THROW_ERROR(status_bad_attribute, s);
+									}
+									else PUGI__THROW_ERROR(status_bad_attribute, s);
+								}
+								else if (*s == '/')
+								{
+									++s;
+									
+									if (*s == '>')
+									{
+										PUGI__POPNODE();
+										s++;
+										break;
+									}
+									else if (*s == 0 && endch == '>')
+									{
+										PUGI__POPNODE();
+										break;
+									}
+									else PUGI__THROW_ERROR(status_bad_start_element, s);
+								}
+								else if (*s == '>')
+								{
+									++s;
+
+									break;
+								}
+								else if (*s == 0 && endch == '>')
+								{
+									break;
+								}
+								else PUGI__THROW_ERROR(status_bad_start_element, s);
+							}
+
+							// !!!
+						}
+						else if (ch == '/') // '<#.../'
+						{
+							if (!PUGI__ENDSWITH(*s, '>')) PUGI__THROW_ERROR(status_bad_start_element, s);
+
+							PUGI__POPNODE(); // Pop.
+
+							s += (*s == '>');
+						}
+						else if (ch == 0)
+						{
+							// we stepped over null terminator, backtrack & handle closing tag
+							--s;
+							
+							if (endch != '>') PUGI__THROW_ERROR(status_bad_start_element, s);
+						}
+						else PUGI__THROW_ERROR(status_bad_start_element, s);
+					}
+					else if (*s == '/')
+					{
+						++s;
+
+						char_t* name = cursor->name;
+						if (!name) PUGI__THROW_ERROR(status_end_element_mismatch, s);
+						
+						while (PUGI__IS_CHARTYPE(*s, ct_symbol))
+						{
+							if (*s++ != *name++) PUGI__THROW_ERROR(status_end_element_mismatch, s);
+						}
+
+						if (*name)
+						{
+							if (*s == 0 && name[0] == endch && name[1] == 0) PUGI__THROW_ERROR(status_bad_end_element, s);
+							else PUGI__THROW_ERROR(status_end_element_mismatch, s);
+						}
+							
+						PUGI__POPNODE(); // Pop.
+
+						PUGI__SKIPWS();
+
+						if (*s == 0)
+						{
+							if (endch != '>') PUGI__THROW_ERROR(status_bad_end_element, s);
+						}
+						else
+						{
+							if (*s != '>') PUGI__THROW_ERROR(status_bad_end_element, s);
+							++s;
+						}
+					}
+					else if (*s == '?') // '<?...'
+					{
+						s = parse_question(s, cursor, optmsk, endch);
+						if (!s) return s;
+
+						assert(cursor);
+						if (PUGI__NODETYPE(cursor) == node_declaration) goto LOC_ATTRIBUTES;
+					}
+					else if (*s == '!') // '<!...'
+					{
+						s = parse_exclamation(s, cursor, optmsk, endch);
+						if (!s) return s;
+					}
+					else if (*s == 0 && endch == '?') PUGI__THROW_ERROR(status_bad_pi, s);
+					else PUGI__THROW_ERROR(status_unrecognized_tag, s);
+				}
+				else
+				{
+					mark = s; // Save this offset while searching for a terminator.
+
+					PUGI__SKIPWS(); // Eat whitespace if no genuine PCDATA here.
+
+					if (*s == '<' || !*s)
+					{
+						// We skipped some whitespace characters because otherwise we would take the tag branch instead of PCDATA one
+						assert(mark != s);
+
+						if (!PUGI__OPTSET(parse_ws_pcdata | parse_ws_pcdata_single) || PUGI__OPTSET(parse_trim_pcdata))
+						{
+							continue;
+						}
+						else if (PUGI__OPTSET(parse_ws_pcdata_single))
+						{
+							if (s[0] != '<' || s[1] != '/' || cursor->first_child) continue;
+						}
+					}
+
+					if (!PUGI__OPTSET(parse_trim_pcdata))
+						s = mark;
+							
+					if (cursor->parent || PUGI__OPTSET(parse_fragment))
+					{
+						PUGI__PUSHNODE(node_pcdata); // Append a new node on the tree.
+						cursor->value = s; // Save the offset.
+
+						s = strconv_pcdata(s);
+								
+						PUGI__POPNODE(); // Pop since this is a standalone.
+						
+						if (!*s) break;
+					}
+					else
+					{
+						PUGI__SCANFOR(*s == '<'); // '...<'
+						if (!*s) break;
+						
+						++s;
+					}
+
+					// We're after '<'
+					goto LOC_TAG;
+				}
+			}
+
+			// check that last tag is closed
+			if (cursor != root) PUGI__THROW_ERROR(status_end_element_mismatch, s);
+
+			return s;
+		}
+
+	#ifdef PUGIXML_WCHAR_MODE
+		static char_t* parse_skip_bom(char_t* s)
+		{
+			unsigned int bom = 0xfeff;
+			return (s[0] == static_cast<wchar_t>(bom)) ? s + 1 : s;
+		}
+	#else
+		static char_t* parse_skip_bom(char_t* s)
+		{
+			return (s[0] == '\xef' && s[1] == '\xbb' && s[2] == '\xbf') ? s + 3 : s;
+		}
+	#endif
+
+		static bool has_element_node_siblings(xml_node_struct* node)
+		{
+			while (node)
+			{
+				if (PUGI__NODETYPE(node) == node_element) return true;
+
+				node = node->next_sibling;
+			}
+
+			return false;
+		}
+
+		static xml_parse_result parse(char_t* buffer, size_t length, xml_document_struct* xmldoc, xml_node_struct* root, unsigned int optmsk)
+		{
+			// early-out for empty documents
+			if (length == 0)
+				return make_parse_result(PUGI__OPTSET(parse_fragment) ? status_ok : status_no_document_element);
+
+			// get last child of the root before parsing
+			xml_node_struct* last_root_child = root->first_child ? root->first_child->prev_sibling_c + 0 : 0;
+	
+			// create parser on stack
+			xml_parser parser(static_cast<xml_allocator*>(xmldoc));
+
+			// save last character and make buffer zero-terminated (speeds up parsing)
+			char_t endch = buffer[length - 1];
+			buffer[length - 1] = 0;
+			
+			// skip BOM to make sure it does not end up as part of parse output
+			char_t* buffer_data = parse_skip_bom(buffer);
+
+			// perform actual parsing
+			parser.parse_tree(buffer_data, root, optmsk, endch);
+
+			xml_parse_result result = make_parse_result(parser.error_status, parser.error_offset ? parser.error_offset - buffer : 0);
+			assert(result.offset >= 0 && static_cast<size_t>(result.offset) <= length);
+
+			if (result)
+			{
+				// since we removed last character, we have to handle the only possible false positive (stray <)
+				if (endch == '<')
+					return make_parse_result(status_unrecognized_tag, length - 1);
+
+				// check if there are any element nodes parsed
+				xml_node_struct* first_root_child_parsed = last_root_child ? last_root_child->next_sibling + 0 : root->first_child;
+
+				if (!PUGI__OPTSET(parse_fragment) && !has_element_node_siblings(first_root_child_parsed))
+					return make_parse_result(status_no_document_element, length - 1);
+			}
+			else
+			{
+				// roll back offset if it occurs on a null terminator in the source buffer
+				if (result.offset > 0 && static_cast<size_t>(result.offset) == length - 1 && endch == 0)
+					result.offset--;
+			}
+
+			return result;
+		}
+	};
+
+	// Output facilities
+	PUGI__FN xml_encoding get_write_native_encoding()
+	{
+	#ifdef PUGIXML_WCHAR_MODE
+		return get_wchar_encoding();
+	#else
+		return encoding_utf8;
+	#endif
+	}
+
+	PUGI__FN xml_encoding get_write_encoding(xml_encoding encoding)
+	{
+		// replace wchar encoding with utf implementation
+		if (encoding == encoding_wchar) return get_wchar_encoding();
+
+		// replace utf16 encoding with utf16 with specific endianness
+		if (encoding == encoding_utf16) return is_little_endian() ? encoding_utf16_le : encoding_utf16_be;
+
+		// replace utf32 encoding with utf32 with specific endianness
+		if (encoding == encoding_utf32) return is_little_endian() ? encoding_utf32_le : encoding_utf32_be;
+
+		// only do autodetection if no explicit encoding is requested
+		if (encoding != encoding_auto) return encoding;
+
+		// assume utf8 encoding
+		return encoding_utf8;
+	}
+
+#ifdef PUGIXML_WCHAR_MODE
+	PUGI__FN size_t get_valid_length(const char_t* data, size_t length)
+	{
+		if (length < 1) return 0;
+
+		// discard last character if it's the lead of a surrogate pair 
+		return (sizeof(wchar_t) == 2 && static_cast<unsigned int>(static_cast<uint16_t>(data[length - 1]) - 0xD800) < 0x400) ? length - 1 : length;
+	}
+
+	PUGI__FN size_t convert_buffer_output(char_t* r_char, uint8_t* r_u8, uint16_t* r_u16, uint32_t* r_u32, const char_t* data, size_t length, xml_encoding encoding)
+	{
+		// only endian-swapping is required
+		if (need_endian_swap_utf(encoding, get_wchar_encoding()))
+		{
+			convert_wchar_endian_swap(r_char, data, length);
+
+			return length * sizeof(char_t);
+		}
+	
+		// convert to utf8
+		if (encoding == encoding_utf8)
+		{
+			uint8_t* dest = r_u8;
+			uint8_t* end = utf_decoder<utf8_writer>::decode_wchar_block(data, length, dest);
+
+			return static_cast<size_t>(end - dest);
+		}
+
+		// convert to utf16
+		if (encoding == encoding_utf16_be || encoding == encoding_utf16_le)
+		{
+			uint16_t* dest = r_u16;
+
+			// convert to native utf16
+			uint16_t* end = utf_decoder<utf16_writer>::decode_wchar_block(data, length, dest);
+
+			// swap if necessary
+			xml_encoding native_encoding = is_little_endian() ? encoding_utf16_le : encoding_utf16_be;
+
+			if (native_encoding != encoding) convert_utf_endian_swap(dest, dest, static_cast<size_t>(end - dest));
+
+			return static_cast<size_t>(end - dest) * sizeof(uint16_t);
+		}
+
+		// convert to utf32
+		if (encoding == encoding_utf32_be || encoding == encoding_utf32_le)
+		{
+			uint32_t* dest = r_u32;
+
+			// convert to native utf32
+			uint32_t* end = utf_decoder<utf32_writer>::decode_wchar_block(data, length, dest);
+
+			// swap if necessary
+			xml_encoding native_encoding = is_little_endian() ? encoding_utf32_le : encoding_utf32_be;
+
+			if (native_encoding != encoding) convert_utf_endian_swap(dest, dest, static_cast<size_t>(end - dest));
+
+			return static_cast<size_t>(end - dest) * sizeof(uint32_t);
+		}
+
+		// convert to latin1
+		if (encoding == encoding_latin1)
+		{
+			uint8_t* dest = r_u8;
+			uint8_t* end = utf_decoder<latin1_writer>::decode_wchar_block(data, length, dest);
+
+			return static_cast<size_t>(end - dest);
+		}
+
+		assert(!"Invalid encoding");
+		return 0;
+	}
+#else
+	PUGI__FN size_t get_valid_length(const char_t* data, size_t length)
+	{
+		if (length < 5) return 0;
+
+		for (size_t i = 1; i <= 4; ++i)
+		{
+			uint8_t ch = static_cast<uint8_t>(data[length - i]);
+
+			// either a standalone character or a leading one
+			if ((ch & 0xc0) != 0x80) return length - i;
+		}
+
+		// there are four non-leading characters at the end, sequence tail is broken so might as well process the whole chunk
+		return length;
+	}
+
+	PUGI__FN size_t convert_buffer_output(char_t* /* r_char */, uint8_t* r_u8, uint16_t* r_u16, uint32_t* r_u32, const char_t* data, size_t length, xml_encoding encoding)
+	{
+		if (encoding == encoding_utf16_be || encoding == encoding_utf16_le)
+		{
+			uint16_t* dest = r_u16;
+
+			// convert to native utf16
+			uint16_t* end = utf_decoder<utf16_writer>::decode_utf8_block(reinterpret_cast<const uint8_t*>(data), length, dest);
+
+			// swap if necessary
+			xml_encoding native_encoding = is_little_endian() ? encoding_utf16_le : encoding_utf16_be;
+
+			if (native_encoding != encoding) convert_utf_endian_swap(dest, dest, static_cast<size_t>(end - dest));
+
+			return static_cast<size_t>(end - dest) * sizeof(uint16_t);
+		}
+
+		if (encoding == encoding_utf32_be || encoding == encoding_utf32_le)
+		{
+			uint32_t* dest = r_u32;
+
+			// convert to native utf32
+			uint32_t* end = utf_decoder<utf32_writer>::decode_utf8_block(reinterpret_cast<const uint8_t*>(data), length, dest);
+
+			// swap if necessary
+			xml_encoding native_encoding = is_little_endian() ? encoding_utf32_le : encoding_utf32_be;
+
+			if (native_encoding != encoding) convert_utf_endian_swap(dest, dest, static_cast<size_t>(end - dest));
+
+			return static_cast<size_t>(end - dest) * sizeof(uint32_t);
+		}
+
+		if (encoding == encoding_latin1)
+		{
+			uint8_t* dest = r_u8;
+			uint8_t* end = utf_decoder<latin1_writer>::decode_utf8_block(reinterpret_cast<const uint8_t*>(data), length, dest);
+
+			return static_cast<size_t>(end - dest);
+		}
+
+		assert(!"Invalid encoding");
+		return 0;
+	}
+#endif
+
+	class xml_buffered_writer
+	{
+		xml_buffered_writer(const xml_buffered_writer&);
+		xml_buffered_writer& operator=(const xml_buffered_writer&);
+
+	public:
+		xml_buffered_writer(xml_writer& writer_, xml_encoding user_encoding): writer(writer_), bufsize(0), encoding(get_write_encoding(user_encoding))
+		{
+			PUGI__STATIC_ASSERT(bufcapacity >= 8);
+		}
+
+		size_t flush()
+		{
+			flush(buffer, bufsize);
+			bufsize = 0;
+			return 0;
+		}
+
+		void flush(const char_t* data, size_t size)
+		{
+			if (size == 0) return;
+
+			// fast path, just write data
+			if (encoding == get_write_native_encoding())
+				writer.write(data, size * sizeof(char_t));
+			else
+			{
+				// convert chunk
+				size_t result = convert_buffer_output(scratch.data_char, scratch.data_u8, scratch.data_u16, scratch.data_u32, data, size, encoding);
+				assert(result <= sizeof(scratch));
+
+				// write data
+				writer.write(scratch.data_u8, result);
+			}
+		}
+
+		void write_direct(const char_t* data, size_t length)
+		{
+			// flush the remaining buffer contents
+			flush();
+
+			// handle large chunks
+			if (length > bufcapacity)
+			{
+				if (encoding == get_write_native_encoding())
+				{
+					// fast path, can just write data chunk
+					writer.write(data, length * sizeof(char_t));
+					return;
+				}
+
+				// need to convert in suitable chunks
+				while (length > bufcapacity)
+				{
+					// get chunk size by selecting such number of characters that are guaranteed to fit into scratch buffer
+					// and form a complete codepoint sequence (i.e. discard start of last codepoint if necessary)
+					size_t chunk_size = get_valid_length(data, bufcapacity);
+					assert(chunk_size);
+
+					// convert chunk and write
+					flush(data, chunk_size);
+
+					// iterate
+					data += chunk_size;
+					length -= chunk_size;
+				}
+
+				// small tail is copied below
+				bufsize = 0;
+			}
+
+			memcpy(buffer + bufsize, data, length * sizeof(char_t));
+			bufsize += length;
+		}
+
+		void write_buffer(const char_t* data, size_t length)
+		{
+			size_t offset = bufsize;
+
+			if (offset + length <= bufcapacity)
+			{
+				memcpy(buffer + offset, data, length * sizeof(char_t));
+				bufsize = offset + length;
+			}
+			else
+			{
+				write_direct(data, length);
+			}
+		}
+
+		void write_string(const char_t* data)
+		{
+			// write the part of the string that fits in the buffer
+			size_t offset = bufsize;
+
+			while (*data && offset < bufcapacity)
+				buffer[offset++] = *data++;
+
+			// write the rest
+			if (offset < bufcapacity)
+			{
+				bufsize = offset;
+			}
+			else
+			{
+				// backtrack a bit if we have split the codepoint
+				size_t length = offset - bufsize;
+				size_t extra = length - get_valid_length(data - length, length);
+
+				bufsize = offset - extra;
+
+				write_direct(data - extra, strlength(data) + extra);
+			}
+		}
+
+		void write(char_t d0)
+		{
+			size_t offset = bufsize;
+			if (offset > bufcapacity - 1) offset = flush();
+
+			buffer[offset + 0] = d0;
+			bufsize = offset + 1;
+		}
+
+		void write(char_t d0, char_t d1)
+		{
+			size_t offset = bufsize;
+			if (offset > bufcapacity - 2) offset = flush();
+
+			buffer[offset + 0] = d0;
+			buffer[offset + 1] = d1;
+			bufsize = offset + 2;
+		}
+
+		void write(char_t d0, char_t d1, char_t d2)
+		{
+			size_t offset = bufsize;
+			if (offset > bufcapacity - 3) offset = flush();
+
+			buffer[offset + 0] = d0;
+			buffer[offset + 1] = d1;
+			buffer[offset + 2] = d2;
+			bufsize = offset + 3;
+		}
+
+		void write(char_t d0, char_t d1, char_t d2, char_t d3)
+		{
+			size_t offset = bufsize;
+			if (offset > bufcapacity - 4) offset = flush();
+
+			buffer[offset + 0] = d0;
+			buffer[offset + 1] = d1;
+			buffer[offset + 2] = d2;
+			buffer[offset + 3] = d3;
+			bufsize = offset + 4;
+		}
+
+		void write(char_t d0, char_t d1, char_t d2, char_t d3, char_t d4)
+		{
+			size_t offset = bufsize;
+			if (offset > bufcapacity - 5) offset = flush();
+
+			buffer[offset + 0] = d0;
+			buffer[offset + 1] = d1;
+			buffer[offset + 2] = d2;
+			buffer[offset + 3] = d3;
+			buffer[offset + 4] = d4;
+			bufsize = offset + 5;
+		}
+
+		void write(char_t d0, char_t d1, char_t d2, char_t d3, char_t d4, char_t d5)
+		{
+			size_t offset = bufsize;
+			if (offset > bufcapacity - 6) offset = flush();
+
+			buffer[offset + 0] = d0;
+			buffer[offset + 1] = d1;
+			buffer[offset + 2] = d2;
+			buffer[offset + 3] = d3;
+			buffer[offset + 4] = d4;
+			buffer[offset + 5] = d5;
+			bufsize = offset + 6;
+		}
+
+		// utf8 maximum expansion: x4 (-> utf32)
+		// utf16 maximum expansion: x2 (-> utf32)
+		// utf32 maximum expansion: x1
+		enum
+		{
+			bufcapacitybytes =
+			#ifdef PUGIXML_MEMORY_OUTPUT_STACK
+				PUGIXML_MEMORY_OUTPUT_STACK
+			#else
+				10240
+			#endif
+			,
+			bufcapacity = bufcapacitybytes / (sizeof(char_t) + 4)
+		};
+
+		char_t buffer[bufcapacity];
+
+		union
+		{
+			uint8_t data_u8[4 * bufcapacity];
+			uint16_t data_u16[2 * bufcapacity];
+			uint32_t data_u32[bufcapacity];
+			char_t data_char[bufcapacity];
+		} scratch;
+
+		xml_writer& writer;
+		size_t bufsize;
+		xml_encoding encoding;
+	};
+
+	PUGI__FN void text_output_escaped(xml_buffered_writer& writer, const char_t* s, chartypex_t type)
+	{
+		while (*s)
+		{
+			const char_t* prev = s;
+			
+			// While *s is a usual symbol
+			PUGI__SCANWHILE_UNROLL(!PUGI__IS_CHARTYPEX(ss, type));
+		
+			writer.write_buffer(prev, static_cast<size_t>(s - prev));
+
+			switch (*s)
+			{
+				case 0: break;
+				case '&':
+					writer.write('&', 'a', 'm', 'p', ';');
+					++s;
+					break;
+				case '<':
+					writer.write('&', 'l', 't', ';');
+					++s;
+					break;
+				case '>':
+					writer.write('&', 'g', 't', ';');
+					++s;
+					break;
+				case '"':
+					writer.write('&', 'q', 'u', 'o', 't', ';');
+					++s;
+					break;
+				default: // s is not a usual symbol
+				{
+					unsigned int ch = static_cast<unsigned int>(*s++);
+					assert(ch < 32);
+
+					writer.write('&', '#', static_cast<char_t>((ch / 10) + '0'), static_cast<char_t>((ch % 10) + '0'), ';');
+				}
+			}
+		}
+	}
+
+	PUGI__FN void text_output(xml_buffered_writer& writer, const char_t* s, chartypex_t type, unsigned int flags)
+	{
+		if (flags & format_no_escapes)
+			writer.write_string(s);
+		else
+			text_output_escaped(writer, s, type);
+	}
+
+	PUGI__FN void text_output_cdata(xml_buffered_writer& writer, const char_t* s)
+	{
+		do
+		{
+			writer.write('<', '!', '[', 'C', 'D');
+			writer.write('A', 'T', 'A', '[');
+
+			const char_t* prev = s;
+
+			// look for ]]> sequence - we can't output it as is since it terminates CDATA
+			while (*s && !(s[0] == ']' && s[1] == ']' && s[2] == '>')) ++s;
+
+			// skip ]] if we stopped at ]]>, > will go to the next CDATA section
+			if (*s) s += 2;
+
+			writer.write_buffer(prev, static_cast<size_t>(s - prev));
+
+			writer.write(']', ']', '>');
+		}
+		while (*s);
+	}
+
+	PUGI__FN void text_output_indent(xml_buffered_writer& writer, const char_t* indent, size_t indent_length, unsigned int depth)
+	{
+		switch (indent_length)
+		{
+		case 1:
+		{
+			for (unsigned int i = 0; i < depth; ++i)
+				writer.write(indent[0]);
+			break;
+		}
+
+		case 2:
+		{
+			for (unsigned int i = 0; i < depth; ++i)
+				writer.write(indent[0], indent[1]);
+			break;
+		}
+
+		case 3:
+		{
+			for (unsigned int i = 0; i < depth; ++i)
+				writer.write(indent[0], indent[1], indent[2]);
+			break;
+		}
+
+		case 4:
+		{
+			for (unsigned int i = 0; i < depth; ++i)
+				writer.write(indent[0], indent[1], indent[2], indent[3]);
+			break;
+		}
+
+		default:
+		{
+			for (unsigned int i = 0; i < depth; ++i)
+				writer.write_buffer(indent, indent_length);
+		}
+		}
+	}
+
+	PUGI__FN void node_output_comment(xml_buffered_writer& writer, const char_t* s)
+	{
+		writer.write('<', '!', '-', '-');
+
+		while (*s)
+		{
+			const char_t* prev = s;
+
+			// look for -\0 or -- sequence - we can't output it since -- is illegal in comment body
+			while (*s && !(s[0] == '-' && (s[1] == '-' || s[1] == 0))) ++s;
+
+			writer.write_buffer(prev, static_cast<size_t>(s - prev));
+
+			if (*s)
+			{
+				assert(*s == '-');
+
+				writer.write('-', ' ');
+				++s;
+			}
+		}
+
+		writer.write('-', '-', '>');
+	}
+
+	PUGI__FN void node_output_pi_value(xml_buffered_writer& writer, const char_t* s)
+	{
+		while (*s)
+		{
+			const char_t* prev = s;
+
+			// look for ?> sequence - we can't output it since ?> terminates PI
+			while (*s && !(s[0] == '?' && s[1] == '>')) ++s;
+
+			writer.write_buffer(prev, static_cast<size_t>(s - prev));
+
+			if (*s)
+			{
+				assert(s[0] == '?' && s[1] == '>');
+
+				writer.write('?', ' ', '>');
+				s += 2;
+			}
+		}
+	}
+
+	PUGI__FN void node_output_attributes(xml_buffered_writer& writer, xml_node_struct* node, const char_t* indent, size_t indent_length, unsigned int flags, unsigned int depth)
+	{
+		const char_t* default_name = PUGIXML_TEXT(":anonymous");
+
+		for (xml_attribute_struct* a = node->first_attribute; a; a = a->next_attribute)
+		{
+			if ((flags & (format_indent_attributes | format_raw)) == format_indent_attributes)
+			{
+				writer.write('\n');
+
+				text_output_indent(writer, indent, indent_length, depth + 1);
+			}
+			else
+			{
+				writer.write(' ');
+			}
+
+			writer.write_string(a->name ? a->name : default_name);
+			writer.write('=', '"');
+
+			if (a->value)
+				text_output(writer, a->value, ctx_special_attr, flags);
+
+			writer.write('"');
+		}
+	}
+
+	PUGI__FN bool node_output_start(xml_buffered_writer& writer, xml_node_struct* node, const char_t* indent, size_t indent_length, unsigned int flags, unsigned int depth)
+	{
+		const char_t* default_name = PUGIXML_TEXT(":anonymous");
+		const char_t* name = node->name ? node->name : default_name;
+
+		writer.write('<');
+		writer.write_string(name);
+
+		if (node->first_attribute)
+			node_output_attributes(writer, node, indent, indent_length, flags, depth);
+
+		if (!node->first_child)
+		{
+			writer.write(' ', '/', '>');
+
+			return false;
+		}
+		else
+		{
+			writer.write('>');
+
+			return true;
+		}
+	}
+
+	PUGI__FN void node_output_end(xml_buffered_writer& writer, xml_node_struct* node)
+	{
+		const char_t* default_name = PUGIXML_TEXT(":anonymous");
+		const char_t* name = node->name ? node->name : default_name;
+
+		writer.write('<', '/');
+		writer.write_string(name);
+		writer.write('>');
+	}
+
+	PUGI__FN void node_output_simple(xml_buffered_writer& writer, xml_node_struct* node, unsigned int flags)
+	{
+		const char_t* default_name = PUGIXML_TEXT(":anonymous");
+
+		switch (PUGI__NODETYPE(node))
+		{
+			case node_pcdata:
+				text_output(writer, node->value ? node->value + 0 : PUGIXML_TEXT(""), ctx_special_pcdata, flags);
+				break;
+
+			case node_cdata:
+				text_output_cdata(writer, node->value ? node->value + 0 : PUGIXML_TEXT(""));
+				break;
+
+			case node_comment:
+				node_output_comment(writer, node->value ? node->value + 0 : PUGIXML_TEXT(""));
+				break;
+
+			case node_pi:
+				writer.write('<', '?');
+				writer.write_string(node->name ? node->name : default_name);
+
+				if (node->value)
+				{
+					writer.write(' ');
+					node_output_pi_value(writer, node->value);
+				}
+
+				writer.write('?', '>');
+				break;
+
+			case node_declaration:
+				writer.write('<', '?');
+				writer.write_string(node->name ? node->name : default_name);
+				node_output_attributes(writer, node, PUGIXML_TEXT(""), 0, flags | format_raw, 0);
+				writer.write('?', '>');
+				break;
+
+			case node_doctype:
+				writer.write('<', '!', 'D', 'O', 'C');
+				writer.write('T', 'Y', 'P', 'E');
+
+				if (node->value)
+				{
+					writer.write(' ');
+					writer.write_string(node->value);
+				}
+
+				writer.write('>');
+				break;
+
+			default:
+				assert(!"Invalid node type");
+		}
+	}
+
+	enum indent_flags_t
+	{
+		indent_newline = 1,
+		indent_indent = 2
+	};
+
+	PUGI__FN void node_output(xml_buffered_writer& writer, xml_node_struct* root, const char_t* indent, unsigned int flags, unsigned int depth)
+	{
+		size_t indent_length = ((flags & (format_indent | format_indent_attributes)) && (flags & format_raw) == 0) ? strlength(indent) : 0;
+		unsigned int indent_flags = indent_indent;
+
+		xml_node_struct* node = root;
+
+		do
+		{
+			assert(node);
+
+			// begin writing current node
+			if (PUGI__NODETYPE(node) == node_pcdata || PUGI__NODETYPE(node) == node_cdata)
+			{
+				node_output_simple(writer, node, flags);
+
+				indent_flags = 0;
+			}
+			else
+			{
+				if ((indent_flags & indent_newline) && (flags & format_raw) == 0)
+					writer.write('\n');
+
+				if ((indent_flags & indent_indent) && indent_length)
+					text_output_indent(writer, indent, indent_length, depth);
+
+				if (PUGI__NODETYPE(node) == node_element)
+				{
+					indent_flags = indent_newline | indent_indent;
+
+					if (node_output_start(writer, node, indent, indent_length, flags, depth))
+					{
+						node = node->first_child;
+						depth++;
+						continue;
+					}
+				}
+				else if (PUGI__NODETYPE(node) == node_document)
+				{
+					indent_flags = indent_indent;
+
+					if (node->first_child)
+					{
+						node = node->first_child;
+						continue;
+					}
+				}
+				else
+				{
+					node_output_simple(writer, node, flags);
+
+					indent_flags = indent_newline | indent_indent;
+				}
+			}
+
+			// continue to the next node
+			while (node != root)
+			{
+				if (node->next_sibling)
+				{
+					node = node->next_sibling;
+					break;
+				}
+
+				node = node->parent;
+
+				// write closing node
+				if (PUGI__NODETYPE(node) == node_element)
+				{
+					depth--;
+
+					if ((indent_flags & indent_newline) && (flags & format_raw) == 0)
+						writer.write('\n');
+
+					if ((indent_flags & indent_indent) && indent_length)
+						text_output_indent(writer, indent, indent_length, depth);
+
+					node_output_end(writer, node);
+
+					indent_flags = indent_newline | indent_indent;
+				}
+			}
+		}
+		while (node != root);
+
+		if ((indent_flags & indent_newline) && (flags & format_raw) == 0)
+			writer.write('\n');
+	}
+
+	PUGI__FN bool has_declaration(xml_node_struct* node)
+	{
+		for (xml_node_struct* child = node->first_child; child; child = child->next_sibling)
+		{
+			xml_node_type type = PUGI__NODETYPE(child);
+
+			if (type == node_declaration) return true;
+			if (type == node_element) return false;
+		}
+
+		return false;
+	}
+
+	PUGI__FN bool is_attribute_of(xml_attribute_struct* attr, xml_node_struct* node)
+	{
+		for (xml_attribute_struct* a = node->first_attribute; a; a = a->next_attribute)
+			if (a == attr)
+				return true;
+
+		return false;
+	}
+
+	PUGI__FN bool allow_insert_attribute(xml_node_type parent)
+	{
+		return parent == node_element || parent == node_declaration;
+	}
+
+	PUGI__FN bool allow_insert_child(xml_node_type parent, xml_node_type child)
+	{
+		if (parent != node_document && parent != node_element) return false;
+		if (child == node_document || child == node_null) return false;
+		if (parent != node_document && (child == node_declaration || child == node_doctype)) return false;
+
+		return true;
+	}
+
+	PUGI__FN bool allow_move(xml_node parent, xml_node child)
+	{
+		// check that child can be a child of parent
+		if (!allow_insert_child(parent.type(), child.type()))
+			return false;
+
+		// check that node is not moved between documents
+		if (parent.root() != child.root())
+			return false;
+
+		// check that new parent is not in the child subtree
+		xml_node cur = parent;
+
+		while (cur)
+		{
+			if (cur == child)
+				return false;
+
+			cur = cur.parent();
+		}
+
+		return true;
+	}
+
+	template <typename String, typename Header>
+	PUGI__FN void node_copy_string(String& dest, Header& header, uintptr_t header_mask, char_t* source, Header& source_header, xml_allocator* alloc)
+	{
+		assert(!dest && (header & header_mask) == 0);
+
+		if (source)
+		{
+			if (alloc && (source_header & header_mask) == 0)
+			{
+				dest = source;
+
+				// since strcpy_insitu can reuse document buffer memory we need to mark both source and dest as shared
+				header |= xml_memory_page_contents_shared_mask;
+				source_header |= xml_memory_page_contents_shared_mask;
+			}
+			else
+				strcpy_insitu(dest, header, header_mask, source);
+		}
+	}
+
+	PUGI__FN void node_copy_contents(xml_node_struct* dn, xml_node_struct* sn, xml_allocator* shared_alloc)
+	{
+		node_copy_string(dn->name, dn->header, xml_memory_page_name_allocated_mask, sn->name, sn->header, shared_alloc);
+		node_copy_string(dn->value, dn->header, xml_memory_page_value_allocated_mask, sn->value, sn->header, shared_alloc);
+
+		for (xml_attribute_struct* sa = sn->first_attribute; sa; sa = sa->next_attribute)
+		{
+			xml_attribute_struct* da = append_new_attribute(dn, get_allocator(dn));
+
+			if (da)
+			{
+				node_copy_string(da->name, da->header, xml_memory_page_name_allocated_mask, sa->name, sa->header, shared_alloc);
+				node_copy_string(da->value, da->header, xml_memory_page_value_allocated_mask, sa->value, sa->header, shared_alloc);
+			}
+		}
+	}
+
+	PUGI__FN void node_copy_tree(xml_node_struct* dn, xml_node_struct* sn)
+	{
+		xml_allocator& alloc = get_allocator(dn);
+		xml_allocator* shared_alloc = (&alloc == &get_allocator(sn)) ? &alloc : 0;
+
+		node_copy_contents(dn, sn, shared_alloc);
+
+		xml_node_struct* dit = dn;
+		xml_node_struct* sit = sn->first_child;
+
+		while (sit && sit != sn)
+		{
+			if (sit != dn)
+			{
+				xml_node_struct* copy = append_new_node(dit, alloc, PUGI__NODETYPE(sit));
+
+				if (copy)
+				{
+					node_copy_contents(copy, sit, shared_alloc);
+
+					if (sit->first_child)
+					{
+						dit = copy;
+						sit = sit->first_child;
+						continue;
+					}
+				}
+			}
+
+			// continue to the next node
+			do
+			{
+				if (sit->next_sibling)
+				{
+					sit = sit->next_sibling;
+					break;
+				}
+
+				sit = sit->parent;
+				dit = dit->parent;
+			}
+			while (sit != sn);
+		}
+	}
+
+	PUGI__FN void node_copy_attribute(xml_attribute_struct* da, xml_attribute_struct* sa)
+	{
+		xml_allocator& alloc = get_allocator(da);
+		xml_allocator* shared_alloc = (&alloc == &get_allocator(sa)) ? &alloc : 0;
+
+		node_copy_string(da->name, da->header, xml_memory_page_name_allocated_mask, sa->name, sa->header, shared_alloc);
+		node_copy_string(da->value, da->header, xml_memory_page_value_allocated_mask, sa->value, sa->header, shared_alloc);
+	}
+
+	inline bool is_text_node(xml_node_struct* node)
+	{
+		xml_node_type type = PUGI__NODETYPE(node);
+
+		return type == node_pcdata || type == node_cdata;
+	}
+
+	// get value with conversion functions
+	PUGI__FN int get_integer_base(const char_t* value)
+	{
+		const char_t* s = value;
+
+		while (PUGI__IS_CHARTYPE(*s, ct_space))
+			s++;
+
+		if (*s == '-')
+			s++;
+
+		return (s[0] == '0' && (s[1] == 'x' || s[1] == 'X')) ? 16 : 10;
+	}
+
+	PUGI__FN int get_value_int(const char_t* value)
+	{
+		int base = get_integer_base(value);
+
+	#ifdef PUGIXML_WCHAR_MODE
+		return static_cast<int>(wcstol(value, 0, base));
+	#else
+		return static_cast<int>(strtol(value, 0, base));
+	#endif
+	}
+
+	PUGI__FN unsigned int get_value_uint(const char_t* value)
+	{
+		int base = get_integer_base(value);
+
+	#ifdef PUGIXML_WCHAR_MODE
+		return static_cast<unsigned int>(wcstoul(value, 0, base));
+	#else
+		return static_cast<unsigned int>(strtoul(value, 0, base));
+	#endif
+	}
+
+	PUGI__FN double get_value_double(const char_t* value)
+	{
+	#ifdef PUGIXML_WCHAR_MODE
+		return wcstod(value, 0);
+	#else
+		return strtod(value, 0);
+	#endif
+	}
+
+	PUGI__FN float get_value_float(const char_t* value)
+	{
+	#ifdef PUGIXML_WCHAR_MODE
+		return static_cast<float>(wcstod(value, 0));
+	#else
+		return static_cast<float>(strtod(value, 0));
+	#endif
+	}
+
+	PUGI__FN bool get_value_bool(const char_t* value)
+	{
+		// only look at first char
+		char_t first = *value;
+
+		// 1*, t* (true), T* (True), y* (yes), Y* (YES)
+		return (first == '1' || first == 't' || first == 'T' || first == 'y' || first == 'Y');
+	}
+
+#ifdef PUGIXML_HAS_LONG_LONG
+	PUGI__FN long long get_value_llong(const char_t* value)
+	{
+		int base = get_integer_base(value);
+
+	#ifdef PUGIXML_WCHAR_MODE
+		#ifdef PUGI__MSVC_CRT_VERSION
+			return _wcstoi64(value, 0, base);
+		#else
+			return wcstoll(value, 0, base);
+		#endif
+	#else
+		#ifdef PUGI__MSVC_CRT_VERSION
+			return _strtoi64(value, 0, base);
+		#else
+			return strtoll(value, 0, base);
+		#endif
+	#endif
+	}
+
+	PUGI__FN unsigned long long get_value_ullong(const char_t* value)
+	{
+		int base = get_integer_base(value);
+
+	#ifdef PUGIXML_WCHAR_MODE
+		#ifdef PUGI__MSVC_CRT_VERSION
+			return _wcstoui64(value, 0, base);
+		#else
+			return wcstoull(value, 0, base);
+		#endif
+	#else
+		#ifdef PUGI__MSVC_CRT_VERSION
+			return _strtoui64(value, 0, base);
+		#else
+			return strtoull(value, 0, base);
+		#endif
+	#endif
+	}
+#endif
+
+	// set value with conversion functions
+	template <typename String, typename Header>
+	PUGI__FN bool set_value_buffer(String& dest, Header& header, uintptr_t header_mask, char (&buf)[128])
+	{
+	#ifdef PUGIXML_WCHAR_MODE
+		char_t wbuf[128];
+		impl::widen_ascii(wbuf, buf);
+
+		return strcpy_insitu(dest, header, header_mask, wbuf);
+	#else
+		return strcpy_insitu(dest, header, header_mask, buf);
+	#endif
+	}
+
+	template <typename String, typename Header>
+	PUGI__FN bool set_value_convert(String& dest, Header& header, uintptr_t header_mask, int value)
+	{
+		char buf[128];
+		sprintf(buf, "%d", value);
+	
+		return set_value_buffer(dest, header, header_mask, buf);
+	}
+
+	template <typename String, typename Header>
+	PUGI__FN bool set_value_convert(String& dest, Header& header, uintptr_t header_mask, unsigned int value)
+	{
+		char buf[128];
+		sprintf(buf, "%u", value);
+
+		return set_value_buffer(dest, header, header_mask, buf);
+	}
+
+	template <typename String, typename Header>
+	PUGI__FN bool set_value_convert(String& dest, Header& header, uintptr_t header_mask, float value)
+	{
+		char buf[128];
+		sprintf(buf, "%.9g", value);
+
+		return set_value_buffer(dest, header, header_mask, buf);
+	}
+
+	template <typename String, typename Header>
+	PUGI__FN bool set_value_convert(String& dest, Header& header, uintptr_t header_mask, double value)
+	{
+		char buf[128];
+		sprintf(buf, "%.17g", value);
+
+		return set_value_buffer(dest, header, header_mask, buf);
+	}
+	
+	template <typename String, typename Header>
+	PUGI__FN bool set_value_convert(String& dest, Header& header, uintptr_t header_mask, bool value)
+	{
+		return strcpy_insitu(dest, header, header_mask, value ? PUGIXML_TEXT("true") : PUGIXML_TEXT("false"));
+	}
+
+#ifdef PUGIXML_HAS_LONG_LONG
+	template <typename String, typename Header>
+	PUGI__FN bool set_value_convert(String& dest, Header& header, uintptr_t header_mask, long long value)
+	{
+		char buf[128];
+		sprintf(buf, "%lld", value);
+	
+		return set_value_buffer(dest, header, header_mask, buf);
+	}
+
+	template <typename String, typename Header>
+	PUGI__FN bool set_value_convert(String& dest, Header& header, uintptr_t header_mask, unsigned long long value)
+	{
+		char buf[128];
+		sprintf(buf, "%llu", value);
+	
+		return set_value_buffer(dest, header, header_mask, buf);
+	}
+#endif
+
+	// we need to get length of entire file to load it in memory; the only (relatively) sane way to do it is via seek/tell trick
+	PUGI__FN xml_parse_status get_file_size(FILE* file, size_t& out_result)
+	{
+	#if defined(PUGI__MSVC_CRT_VERSION) && PUGI__MSVC_CRT_VERSION >= 1400 && !defined(_WIN32_WCE)
+		// there are 64-bit versions of fseek/ftell, let's use them
+		typedef __int64 length_type;
+
+		_fseeki64(file, 0, SEEK_END);
+		length_type length = _ftelli64(file);
+		_fseeki64(file, 0, SEEK_SET);
+	#elif defined(__MINGW32__) && !defined(__NO_MINGW_LFS) && (!defined(__STRICT_ANSI__) || defined(__MINGW64_VERSION_MAJOR))
+		// there are 64-bit versions of fseek/ftell, let's use them
+		typedef off64_t length_type;
+
+		fseeko64(file, 0, SEEK_END);
+		length_type length = ftello64(file);
+		fseeko64(file, 0, SEEK_SET);
+	#else
+		// if this is a 32-bit OS, long is enough; if this is a unix system, long is 64-bit, which is enough; otherwise we can't do anything anyway.
+		typedef long length_type;
+
+		fseek(file, 0, SEEK_END);
+		length_type length = ftell(file);
+		fseek(file, 0, SEEK_SET);
+	#endif
+
+		// check for I/O errors
+		if (length < 0) return status_io_error;
+		
+		// check for overflow
+		size_t result = static_cast<size_t>(length);
+
+		if (static_cast<length_type>(result) != length) return status_out_of_memory;
+
+		// finalize
+		out_result = result;
+
+		return status_ok;
+	}
+
+	// This function assumes that buffer has extra sizeof(char_t) writable bytes after size
+	PUGI__FN size_t zero_terminate_buffer(void* buffer, size_t size, xml_encoding encoding) 
+	{
+		// We only need to zero-terminate if encoding conversion does not do it for us
+	#ifdef PUGIXML_WCHAR_MODE
+		xml_encoding wchar_encoding = get_wchar_encoding();
+
+		if (encoding == wchar_encoding || need_endian_swap_utf(encoding, wchar_encoding))
+		{
+			size_t length = size / sizeof(char_t);
+
+			static_cast<char_t*>(buffer)[length] = 0;
+			return (length + 1) * sizeof(char_t);
+		}
+	#else
+		if (encoding == encoding_utf8)
+		{
+			static_cast<char*>(buffer)[size] = 0;
+			return size + 1;
+		}
+	#endif
+
+		return size;
+	}
+
+	PUGI__FN xml_parse_result load_file_impl(xml_document& doc, FILE* file, unsigned int options, xml_encoding encoding)
+	{
+		if (!file) return make_parse_result(status_file_not_found);
+
+		// get file size (can result in I/O errors)
+		size_t size = 0;
+		xml_parse_status size_status = get_file_size(file, size);
+		if (size_status != status_ok) return make_parse_result(size_status);
+		
+		size_t max_suffix_size = sizeof(char_t);
+
+		// allocate buffer for the whole file
+		char* contents = static_cast<char*>(xml_memory::allocate(size + max_suffix_size));
+		if (!contents) return make_parse_result(status_out_of_memory);
+
+		// read file in memory
+		size_t read_size = fread(contents, 1, size, file);
+
+		if (read_size != size)
+		{
+			xml_memory::deallocate(contents);
+			return make_parse_result(status_io_error);
+		}
+
+		xml_encoding real_encoding = get_buffer_encoding(encoding, contents, size);
+		
+		return doc.load_buffer_inplace_own(contents, zero_terminate_buffer(contents, size, real_encoding), options, real_encoding);
+	}
+
+#ifndef PUGIXML_NO_STL
+	template <typename T> struct xml_stream_chunk
+	{
+		static xml_stream_chunk* create()
+		{
+			void* memory = xml_memory::allocate(sizeof(xml_stream_chunk));
+			if (!memory) return 0;
+			
+			return new (memory) xml_stream_chunk();
+		}
+
+		static void destroy(xml_stream_chunk* chunk)
+		{
+			// free chunk chain
+			while (chunk)
+			{
+				xml_stream_chunk* next_ = chunk->next;
+
+				xml_memory::deallocate(chunk);
+
+				chunk = next_;
+			}
+		}
+
+		xml_stream_chunk(): next(0), size(0)
+		{
+		}
+
+		xml_stream_chunk* next;
+		size_t size;
+
+		T data[xml_memory_page_size / sizeof(T)];
+	};
+
+	template <typename T> PUGI__FN xml_parse_status load_stream_data_noseek(std::basic_istream<T>& stream, void** out_buffer, size_t* out_size)
+	{
+		auto_deleter<xml_stream_chunk<T> > chunks(0, xml_stream_chunk<T>::destroy);
+
+		// read file to a chunk list
+		size_t total = 0;
+		xml_stream_chunk<T>* last = 0;
+
+		while (!stream.eof())
+		{
+			// allocate new chunk
+			xml_stream_chunk<T>* chunk = xml_stream_chunk<T>::create();
+			if (!chunk) return status_out_of_memory;
+
+			// append chunk to list
+			if (last) last = last->next = chunk;
+			else chunks.data = last = chunk;
+
+			// read data to chunk
+			stream.read(chunk->data, static_cast<std::streamsize>(sizeof(chunk->data) / sizeof(T)));
+			chunk->size = static_cast<size_t>(stream.gcount()) * sizeof(T);
+
+			// read may set failbit | eofbit in case gcount() is less than read length, so check for other I/O errors
+			if (stream.bad() || (!stream.eof() && stream.fail())) return status_io_error;
+
+			// guard against huge files (chunk size is small enough to make this overflow check work)
+			if (total + chunk->size < total) return status_out_of_memory;
+			total += chunk->size;
+		}
+
+		size_t max_suffix_size = sizeof(char_t);
+
+		// copy chunk list to a contiguous buffer
+		char* buffer = static_cast<char*>(xml_memory::allocate(total + max_suffix_size));
+		if (!buffer) return status_out_of_memory;
+
+		char* write = buffer;
+
+		for (xml_stream_chunk<T>* chunk = chunks.data; chunk; chunk = chunk->next)
+		{
+			assert(write + chunk->size <= buffer + total);
+			memcpy(write, chunk->data, chunk->size);
+			write += chunk->size;
+		}
+
+		assert(write == buffer + total);
+
+		// return buffer
+		*out_buffer = buffer;
+		*out_size = total;
+
+		return status_ok;
+	}
+
+	template <typename T> PUGI__FN xml_parse_status load_stream_data_seek(std::basic_istream<T>& stream, void** out_buffer, size_t* out_size)
+	{
+		// get length of remaining data in stream
+		typename std::basic_istream<T>::pos_type pos = stream.tellg();
+		stream.seekg(0, std::ios::end);
+		std::streamoff length = stream.tellg() - pos;
+		stream.seekg(pos);
+
+		if (stream.fail() || pos < 0) return status_io_error;
+
+		// guard against huge files
+		size_t read_length = static_cast<size_t>(length);
+
+		if (static_cast<std::streamsize>(read_length) != length || length < 0) return status_out_of_memory;
+
+		size_t max_suffix_size = sizeof(char_t);
+
+		// read stream data into memory (guard against stream exceptions with buffer holder)
+		auto_deleter<void> buffer(xml_memory::allocate(read_length * sizeof(T) + max_suffix_size), xml_memory::deallocate);
+		if (!buffer.data) return status_out_of_memory;
+
+		stream.read(static_cast<T*>(buffer.data), static_cast<std::streamsize>(read_length));
+
+		// read may set failbit | eofbit in case gcount() is less than read_length (i.e. line ending conversion), so check for other I/O errors
+		if (stream.bad() || (!stream.eof() && stream.fail())) return status_io_error;
+
+		// return buffer
+		size_t actual_length = static_cast<size_t>(stream.gcount());
+		assert(actual_length <= read_length);
+		
+		*out_buffer = buffer.release();
+		*out_size = actual_length * sizeof(T);
+
+		return status_ok;
+	}
+
+	template <typename T> PUGI__FN xml_parse_result load_stream_impl(xml_document& doc, std::basic_istream<T>& stream, unsigned int options, xml_encoding encoding)
+	{
+		void* buffer = 0;
+		size_t size = 0;
+		xml_parse_status status = status_ok;
+
+		// if stream has an error bit set, bail out (otherwise tellg() can fail and we'll clear error bits)
+		if (stream.fail()) return make_parse_result(status_io_error);
+
+		// load stream to memory (using seek-based implementation if possible, since it's faster and takes less memory)
+		if (stream.tellg() < 0)
+		{
+			stream.clear(); // clear error flags that could be set by a failing tellg
+			status = load_stream_data_noseek(stream, &buffer, &size);
+		}
+		else
+			status = load_stream_data_seek(stream, &buffer, &size);
+
+		if (status != status_ok) return make_parse_result(status);
+
+		xml_encoding real_encoding = get_buffer_encoding(encoding, buffer, size);
+		
+		return doc.load_buffer_inplace_own(buffer, zero_terminate_buffer(buffer, size, real_encoding), options, real_encoding);
+	}
+#endif
+
+#if defined(PUGI__MSVC_CRT_VERSION) || defined(__BORLANDC__) || (defined(__MINGW32__) && (!defined(__STRICT_ANSI__) || defined(__MINGW64_VERSION_MAJOR)))
+	PUGI__FN FILE* open_file_wide(const wchar_t* path, const wchar_t* mode)
+	{
+		return _wfopen(path, mode);
+	}
+#else
+	PUGI__FN char* convert_path_heap(const wchar_t* str)
+	{
+		assert(str);
+
+		// first pass: get length in utf8 characters
+		size_t length = strlength_wide(str);
+		size_t size = as_utf8_begin(str, length);
+
+		// allocate resulting string
+		char* result = static_cast<char*>(xml_memory::allocate(size + 1));
+		if (!result) return 0;
+
+		// second pass: convert to utf8
+		as_utf8_end(result, size, str, length);
+
+		// zero-terminate
+		result[size] = 0;
+
+		return result;
+	}
+
+	PUGI__FN FILE* open_file_wide(const wchar_t* path, const wchar_t* mode)
+	{
+		// there is no standard function to open wide paths, so our best bet is to try utf8 path
+		char* path_utf8 = convert_path_heap(path);
+		if (!path_utf8) return 0;
+
+		// convert mode to ASCII (we mirror _wfopen interface)
+		char mode_ascii[4] = {0};
+		for (size_t i = 0; mode[i]; ++i) mode_ascii[i] = static_cast<char>(mode[i]);
+
+		// try to open the utf8 path
+		FILE* result = fopen(path_utf8, mode_ascii);
+
+		// free dummy buffer
+		xml_memory::deallocate(path_utf8);
+
+		return result;
+	}
+#endif
+
+	PUGI__FN bool save_file_impl(const xml_document& doc, FILE* file, const char_t* indent, unsigned int flags, xml_encoding encoding)
+	{
+		if (!file) return false;
+
+		xml_writer_file writer(file);
+		doc.save(writer, indent, flags, encoding);
+
+		return ferror(file) == 0;
+	}
+
+	PUGI__FN xml_parse_result load_buffer_impl(xml_document_struct* doc, xml_node_struct* root, void* contents, size_t size, unsigned int options, xml_encoding encoding, bool is_mutable, bool own, char_t** out_buffer)
+	{
+		// check input buffer
+		if (!contents && size) return make_parse_result(status_io_error);
+
+		// get actual encoding
+		xml_encoding buffer_encoding = impl::get_buffer_encoding(encoding, contents, size);
+
+		// get private buffer
+		char_t* buffer = 0;
+		size_t length = 0;
+
+		if (!impl::convert_buffer(buffer, length, buffer_encoding, contents, size, is_mutable)) return impl::make_parse_result(status_out_of_memory);
+		
+		// delete original buffer if we performed a conversion
+		if (own && buffer != contents && contents) impl::xml_memory::deallocate(contents);
+
+		// grab onto buffer if it's our buffer, user is responsible for deallocating contents himself
+		if (own || buffer != contents) *out_buffer = buffer;
+
+		// store buffer for offset_debug
+		doc->buffer = buffer;
+
+		// parse
+		xml_parse_result res = impl::xml_parser::parse(buffer, length, doc, root, options);
+
+		// remember encoding
+		res.encoding = buffer_encoding;
+
+		return res;
+	}
+PUGI__NS_END
+
+namespace pugi
+{
+	PUGI__FN xml_writer_file::xml_writer_file(void* file_): file(file_)
+	{
+	}
+
+	PUGI__FN void xml_writer_file::write(const void* data, size_t size)
+	{
+		size_t result = fwrite(data, 1, size, static_cast<FILE*>(file));
+		(void)!result; // unfortunately we can't do proper error handling here
+	}
+
+#ifndef PUGIXML_NO_STL
+	PUGI__FN xml_writer_stream::xml_writer_stream(std::basic_ostream<char, std::char_traits<char> >& stream): narrow_stream(&stream), wide_stream(0)
+	{
+	}
+
+	PUGI__FN xml_writer_stream::xml_writer_stream(std::basic_ostream<wchar_t, std::char_traits<wchar_t> >& stream): narrow_stream(0), wide_stream(&stream)
+	{
+	}
+
+	PUGI__FN void xml_writer_stream::write(const void* data, size_t size)
+	{
+		if (narrow_stream)
+		{
+			assert(!wide_stream);
+			narrow_stream->write(reinterpret_cast<const char*>(data), static_cast<std::streamsize>(size));
+		}
+		else
+		{
+			assert(wide_stream);
+			assert(size % sizeof(wchar_t) == 0);
+
+			wide_stream->write(reinterpret_cast<const wchar_t*>(data), static_cast<std::streamsize>(size / sizeof(wchar_t)));
+		}
+	}
+#endif
+
+	PUGI__FN xml_tree_walker::xml_tree_walker(): _depth(0)
+	{
+	}
+	
+	PUGI__FN xml_tree_walker::~xml_tree_walker()
+	{
+	}
+
+	PUGI__FN int xml_tree_walker::depth() const
+	{
+		return _depth;
+	}
+
+	PUGI__FN bool xml_tree_walker::begin(xml_node&)
+	{
+		return true;
+	}
+
+	PUGI__FN bool xml_tree_walker::end(xml_node&)
+	{
+		return true;
+	}
+
+	PUGI__FN xml_attribute::xml_attribute(): _attr(0)
+	{
+	}
+
+	PUGI__FN xml_attribute::xml_attribute(xml_attribute_struct* attr): _attr(attr)
+	{
+	}
+
+	PUGI__FN static void unspecified_bool_xml_attribute(xml_attribute***)
+	{
+	}
+
+	PUGI__FN xml_attribute::operator xml_attribute::unspecified_bool_type() const
+	{
+		return _attr ? unspecified_bool_xml_attribute : 0;
+	}
+
+	PUGI__FN bool xml_attribute::operator!() const
+	{
+		return !_attr;
+	}
+
+	PUGI__FN bool xml_attribute::operator==(const xml_attribute& r) const
+	{
+		return (_attr == r._attr);
+	}
+	
+	PUGI__FN bool xml_attribute::operator!=(const xml_attribute& r) const
+	{
+		return (_attr != r._attr);
+	}
+
+	PUGI__FN bool xml_attribute::operator<(const xml_attribute& r) const
+	{
+		return (_attr < r._attr);
+	}
+	
+	PUGI__FN bool xml_attribute::operator>(const xml_attribute& r) const
+	{
+		return (_attr > r._attr);
+	}
+	
+	PUGI__FN bool xml_attribute::operator<=(const xml_attribute& r) const
+	{
+		return (_attr <= r._attr);
+	}
+	
+	PUGI__FN bool xml_attribute::operator>=(const xml_attribute& r) const
+	{
+		return (_attr >= r._attr);
+	}
+
+	PUGI__FN xml_attribute xml_attribute::next_attribute() const
+	{
+		return _attr ? xml_attribute(_attr->next_attribute) : xml_attribute();
+	}
+
+	PUGI__FN xml_attribute xml_attribute::previous_attribute() const
+	{
+		return _attr && _attr->prev_attribute_c->next_attribute ? xml_attribute(_attr->prev_attribute_c) : xml_attribute();
+	}
+
+	PUGI__FN const char_t* xml_attribute::as_string(const char_t* def) const
+	{
+		return (_attr && _attr->value) ? _attr->value : def;
+	}
+
+	PUGI__FN int xml_attribute::as_int(int def) const
+	{
+		return (_attr && _attr->value) ? impl::get_value_int(_attr->value) : def;
+	}
+
+	PUGI__FN unsigned int xml_attribute::as_uint(unsigned int def) const
+	{
+		return (_attr && _attr->value) ? impl::get_value_uint(_attr->value) : def;
+	}
+
+	PUGI__FN double xml_attribute::as_double(double def) const
+	{
+		return (_attr && _attr->value) ? impl::get_value_double(_attr->value) : def;
+	}
+
+	PUGI__FN float xml_attribute::as_float(float def) const
+	{
+		return (_attr && _attr->value) ? impl::get_value_float(_attr->value) : def;
+	}
+
+	PUGI__FN bool xml_attribute::as_bool(bool def) const
+	{
+		return (_attr && _attr->value) ? impl::get_value_bool(_attr->value) : def;
+	}
+
+#ifdef PUGIXML_HAS_LONG_LONG
+	PUGI__FN long long xml_attribute::as_llong(long long def) const
+	{
+		return (_attr && _attr->value) ? impl::get_value_llong(_attr->value) : def;
+	}
+
+	PUGI__FN unsigned long long xml_attribute::as_ullong(unsigned long long def) const
+	{
+		return (_attr && _attr->value) ? impl::get_value_ullong(_attr->value) : def;
+	}
+#endif
+
+	PUGI__FN bool xml_attribute::empty() const
+	{
+		return !_attr;
+	}
+
+	PUGI__FN const char_t* xml_attribute::name() const
+	{
+		return (_attr && _attr->name) ? _attr->name + 0 : PUGIXML_TEXT("");
+	}
+
+	PUGI__FN const char_t* xml_attribute::value() const
+	{
+		return (_attr && _attr->value) ? _attr->value + 0 : PUGIXML_TEXT("");
+	}
+
+	PUGI__FN size_t xml_attribute::hash_value() const
+	{
+		return static_cast<size_t>(reinterpret_cast<uintptr_t>(_attr) / sizeof(xml_attribute_struct));
+	}
+
+	PUGI__FN xml_attribute_struct* xml_attribute::internal_object() const
+	{
+		return _attr;
+	}
+
+	PUGI__FN xml_attribute& xml_attribute::operator=(const char_t* rhs)
+	{
+		set_value(rhs);
+		return *this;
+	}
+	
+	PUGI__FN xml_attribute& xml_attribute::operator=(int rhs)
+	{
+		set_value(rhs);
+		return *this;
+	}
+
+	PUGI__FN xml_attribute& xml_attribute::operator=(unsigned int rhs)
+	{
+		set_value(rhs);
+		return *this;
+	}
+
+	PUGI__FN xml_attribute& xml_attribute::operator=(double rhs)
+	{
+		set_value(rhs);
+		return *this;
+	}
+	
+	PUGI__FN xml_attribute& xml_attribute::operator=(float rhs)
+	{
+		set_value(rhs);
+		return *this;
+	}
+
+	PUGI__FN xml_attribute& xml_attribute::operator=(bool rhs)
+	{
+		set_value(rhs);
+		return *this;
+	}
+
+#ifdef PUGIXML_HAS_LONG_LONG
+	PUGI__FN xml_attribute& xml_attribute::operator=(long long rhs)
+	{
+		set_value(rhs);
+		return *this;
+	}
+
+	PUGI__FN xml_attribute& xml_attribute::operator=(unsigned long long rhs)
+	{
+		set_value(rhs);
+		return *this;
+	}
+#endif
+
+	PUGI__FN bool xml_attribute::set_name(const char_t* rhs)
+	{
+		if (!_attr) return false;
+		
+		return impl::strcpy_insitu(_attr->name, _attr->header, impl::xml_memory_page_name_allocated_mask, rhs);
+	}
+		
+	PUGI__FN bool xml_attribute::set_value(const char_t* rhs)
+	{
+		if (!_attr) return false;
+
+		return impl::strcpy_insitu(_attr->value, _attr->header, impl::xml_memory_page_value_allocated_mask, rhs);
+	}
+
+	PUGI__FN bool xml_attribute::set_value(int rhs)
+	{
+		if (!_attr) return false;
+
+		return impl::set_value_convert(_attr->value, _attr->header, impl::xml_memory_page_value_allocated_mask, rhs);
+	}
+
+	PUGI__FN bool xml_attribute::set_value(unsigned int rhs)
+	{
+		if (!_attr) return false;
+
+		return impl::set_value_convert(_attr->value, _attr->header, impl::xml_memory_page_value_allocated_mask, rhs);
+	}
+
+	PUGI__FN bool xml_attribute::set_value(double rhs)
+	{
+		if (!_attr) return false;
+
+		return impl::set_value_convert(_attr->value, _attr->header, impl::xml_memory_page_value_allocated_mask, rhs);
+	}
+	
+	PUGI__FN bool xml_attribute::set_value(float rhs)
+	{
+		if (!_attr) return false;
+
+		return impl::set_value_convert(_attr->value, _attr->header, impl::xml_memory_page_value_allocated_mask, rhs);
+	}
+
+	PUGI__FN bool xml_attribute::set_value(bool rhs)
+	{
+		if (!_attr) return false;
+
+		return impl::set_value_convert(_attr->value, _attr->header, impl::xml_memory_page_value_allocated_mask, rhs);
+	}
+
+#ifdef PUGIXML_HAS_LONG_LONG
+	PUGI__FN bool xml_attribute::set_value(long long rhs)
+	{
+		if (!_attr) return false;
+
+		return impl::set_value_convert(_attr->value, _attr->header, impl::xml_memory_page_value_allocated_mask, rhs);
+	}
+
+	PUGI__FN bool xml_attribute::set_value(unsigned long long rhs)
+	{
+		if (!_attr) return false;
+
+		return impl::set_value_convert(_attr->value, _attr->header, impl::xml_memory_page_value_allocated_mask, rhs);
+	}
+#endif
+
+#ifdef __BORLANDC__
+	PUGI__FN bool operator&&(const xml_attribute& lhs, bool rhs)
+	{
+		return (bool)lhs && rhs;
+	}
+
+	PUGI__FN bool operator||(const xml_attribute& lhs, bool rhs)
+	{
+		return (bool)lhs || rhs;
+	}
+#endif
+
+	PUGI__FN xml_node::xml_node(): _root(0)
+	{
+	}
+
+	PUGI__FN xml_node::xml_node(xml_node_struct* p): _root(p)
+	{
+	}
+	
+	PUGI__FN static void unspecified_bool_xml_node(xml_node***)
+	{
+	}
+
+	PUGI__FN xml_node::operator xml_node::unspecified_bool_type() const
+	{
+		return _root ? unspecified_bool_xml_node : 0;
+	}
+
+	PUGI__FN bool xml_node::operator!() const
+	{
+		return !_root;
+	}
+
+	PUGI__FN xml_node::iterator xml_node::begin() const
+	{
+		return iterator(_root ? _root->first_child + 0 : 0, _root);
+	}
+
+	PUGI__FN xml_node::iterator xml_node::end() const
+	{
+		return iterator(0, _root);
+	}
+	
+	PUGI__FN xml_node::attribute_iterator xml_node::attributes_begin() const
+	{
+		return attribute_iterator(_root ? _root->first_attribute + 0 : 0, _root);
+	}
+
+	PUGI__FN xml_node::attribute_iterator xml_node::attributes_end() const
+	{
+		return attribute_iterator(0, _root);
+	}
+	
+	PUGI__FN xml_object_range<xml_node_iterator> xml_node::children() const
+	{
+		return xml_object_range<xml_node_iterator>(begin(), end());
+	}
+
+	PUGI__FN xml_object_range<xml_named_node_iterator> xml_node::children(const char_t* name_) const
+	{
+		return xml_object_range<xml_named_node_iterator>(xml_named_node_iterator(child(name_)._root, _root, name_), xml_named_node_iterator(0, _root, name_));
+	}
+
+	PUGI__FN xml_object_range<xml_attribute_iterator> xml_node::attributes() const
+	{
+		return xml_object_range<xml_attribute_iterator>(attributes_begin(), attributes_end());
+	}
+
+	PUGI__FN bool xml_node::operator==(const xml_node& r) const
+	{
+		return (_root == r._root);
+	}
+
+	PUGI__FN bool xml_node::operator!=(const xml_node& r) const
+	{
+		return (_root != r._root);
+	}
+
+	PUGI__FN bool xml_node::operator<(const xml_node& r) const
+	{
+		return (_root < r._root);
+	}
+	
+	PUGI__FN bool xml_node::operator>(const xml_node& r) const
+	{
+		return (_root > r._root);
+	}
+	
+	PUGI__FN bool xml_node::operator<=(const xml_node& r) const
+	{
+		return (_root <= r._root);
+	}
+	
+	PUGI__FN bool xml_node::operator>=(const xml_node& r) const
+	{
+		return (_root >= r._root);
+	}
+
+	PUGI__FN bool xml_node::empty() const
+	{
+		return !_root;
+	}
+	
+	PUGI__FN const char_t* xml_node::name() const
+	{
+		return (_root && _root->name) ? _root->name + 0 : PUGIXML_TEXT("");
+	}
+
+	PUGI__FN xml_node_type xml_node::type() const
+	{
+		return _root ? PUGI__NODETYPE(_root) : node_null;
+	}
+	
+	PUGI__FN const char_t* xml_node::value() const
+	{
+		return (_root && _root->value) ? _root->value + 0 : PUGIXML_TEXT("");
+	}
+	
+	PUGI__FN xml_node xml_node::child(const char_t* name_) const
+	{
+		if (!_root) return xml_node();
+
+		for (xml_node_struct* i = _root->first_child; i; i = i->next_sibling)
+			if (i->name && impl::strequal(name_, i->name)) return xml_node(i);
+
+		return xml_node();
+	}
+
+	PUGI__FN xml_attribute xml_node::attribute(const char_t* name_) const
+	{
+		if (!_root) return xml_attribute();
+
+		for (xml_attribute_struct* i = _root->first_attribute; i; i = i->next_attribute)
+			if (i->name && impl::strequal(name_, i->name))
+				return xml_attribute(i);
+		
+		return xml_attribute();
+	}
+	
+	PUGI__FN xml_node xml_node::next_sibling(const char_t* name_) const
+	{
+		if (!_root) return xml_node();
+		
+		for (xml_node_struct* i = _root->next_sibling; i; i = i->next_sibling)
+			if (i->name && impl::strequal(name_, i->name)) return xml_node(i);
+
+		return xml_node();
+	}
+
+	PUGI__FN xml_node xml_node::next_sibling() const
+	{
+		return _root ? xml_node(_root->next_sibling) : xml_node();
+	}
+
+	PUGI__FN xml_node xml_node::previous_sibling(const char_t* name_) const
+	{
+		if (!_root) return xml_node();
+		
+		for (xml_node_struct* i = _root->prev_sibling_c; i->next_sibling; i = i->prev_sibling_c)
+			if (i->name && impl::strequal(name_, i->name)) return xml_node(i);
+
+		return xml_node();
+	}
+
+	PUGI__FN xml_attribute xml_node::attribute(const char_t* name_, xml_attribute& hint_) const
+	{
+		xml_attribute_struct* hint = hint_._attr;
+
+		// if hint is not an attribute of node, behavior is not defined
+		assert(!hint || (_root && impl::is_attribute_of(hint, _root)));
+
+		if (!_root) return xml_attribute();
+
+		// optimistically search from hint up until the end
+		for (xml_attribute_struct* i = hint; i; i = i->next_attribute)
+			if (i->name && impl::strequal(name_, i->name))
+			{
+				// update hint to maximize efficiency of searching for consecutive attributes
+				hint_._attr = i->next_attribute;
+
+				return xml_attribute(i);
+			}
+
+		// wrap around and search from the first attribute until the hint
+		// 'j' null pointer check is technically redundant, but it prevents a crash in case the assertion above fails
+		for (xml_attribute_struct* j = _root->first_attribute; j && j != hint; j = j->next_attribute)
+			if (j->name && impl::strequal(name_, j->name))
+			{
+				// update hint to maximize efficiency of searching for consecutive attributes
+				hint_._attr = j->next_attribute;
+
+				return xml_attribute(j);
+			}
+
+		return xml_attribute();
+	}
+
+	PUGI__FN xml_node xml_node::previous_sibling() const
+	{
+		if (!_root) return xml_node();
+		
+		if (_root->prev_sibling_c->next_sibling) return xml_node(_root->prev_sibling_c);
+		else return xml_node();
+	}
+
+	PUGI__FN xml_node xml_node::parent() const
+	{
+		return _root ? xml_node(_root->parent) : xml_node();
+	}
+
+	PUGI__FN xml_node xml_node::root() const
+	{
+		return _root ? xml_node(&impl::get_document(_root)) : xml_node();
+	}
+
+	PUGI__FN xml_text xml_node::text() const
+	{
+		return xml_text(_root);
+	}
+
+	PUGI__FN const char_t* xml_node::child_value() const
+	{
+		if (!_root) return PUGIXML_TEXT("");
+		
+		for (xml_node_struct* i = _root->first_child; i; i = i->next_sibling)
+			if (impl::is_text_node(i) && i->value)
+				return i->value;
+
+		return PUGIXML_TEXT("");
+	}
+
+	PUGI__FN const char_t* xml_node::child_value(const char_t* name_) const
+	{
+		return child(name_).child_value();
+	}
+
+	PUGI__FN xml_attribute xml_node::first_attribute() const
+	{
+		return _root ? xml_attribute(_root->first_attribute) : xml_attribute();
+	}
+
+	PUGI__FN xml_attribute xml_node::last_attribute() const
+	{
+		return _root && _root->first_attribute ? xml_attribute(_root->first_attribute->prev_attribute_c) : xml_attribute();
+	}
+
+	PUGI__FN xml_node xml_node::first_child() const
+	{
+		return _root ? xml_node(_root->first_child) : xml_node();
+	}
+
+	PUGI__FN xml_node xml_node::last_child() const
+	{
+		return _root && _root->first_child ? xml_node(_root->first_child->prev_sibling_c) : xml_node();
+	}
+
+	PUGI__FN bool xml_node::set_name(const char_t* rhs)
+	{
+		static const bool has_name[] = { false, false, true, false, false, false, true, true, false };
+
+		if (!_root || !has_name[PUGI__NODETYPE(_root)])
+			return false;
+
+		return impl::strcpy_insitu(_root->name, _root->header, impl::xml_memory_page_name_allocated_mask, rhs);
+	}
+		
+	PUGI__FN bool xml_node::set_value(const char_t* rhs)
+	{
+		static const bool has_value[] = { false, false, false, true, true, true, true, false, true };
+
+		if (!_root || !has_value[PUGI__NODETYPE(_root)])
+			return false;
+
+		return impl::strcpy_insitu(_root->value, _root->header, impl::xml_memory_page_value_allocated_mask, rhs);
+	}
+
+	PUGI__FN xml_attribute xml_node::append_attribute(const char_t* name_)
+	{
+		if (!impl::allow_insert_attribute(type())) return xml_attribute();
+		
+		impl::xml_allocator& alloc = impl::get_allocator(_root);
+		if (!alloc.reserve()) return xml_attribute();
+
+		xml_attribute a(impl::allocate_attribute(alloc));
+		if (!a) return xml_attribute();
+
+		impl::append_attribute(a._attr, _root);
+
+		a.set_name(name_);
+		
+		return a;
+	}
+
+	PUGI__FN xml_attribute xml_node::prepend_attribute(const char_t* name_)
+	{
+		if (!impl::allow_insert_attribute(type())) return xml_attribute();
+		
+		impl::xml_allocator& alloc = impl::get_allocator(_root);
+		if (!alloc.reserve()) return xml_attribute();
+
+		xml_attribute a(impl::allocate_attribute(alloc));
+		if (!a) return xml_attribute();
+
+		impl::prepend_attribute(a._attr, _root);
+
+		a.set_name(name_);
+
+		return a;
+	}
+
+	PUGI__FN xml_attribute xml_node::insert_attribute_after(const char_t* name_, const xml_attribute& attr)
+	{
+		if (!impl::allow_insert_attribute(type())) return xml_attribute();
+		if (!attr || !impl::is_attribute_of(attr._attr, _root)) return xml_attribute();
+		
+		impl::xml_allocator& alloc = impl::get_allocator(_root);
+		if (!alloc.reserve()) return xml_attribute();
+
+		xml_attribute a(impl::allocate_attribute(alloc));
+		if (!a) return xml_attribute();
+
+		impl::insert_attribute_after(a._attr, attr._attr, _root);
+
+		a.set_name(name_);
+
+		return a;
+	}
+
+	PUGI__FN xml_attribute xml_node::insert_attribute_before(const char_t* name_, const xml_attribute& attr)
+	{
+		if (!impl::allow_insert_attribute(type())) return xml_attribute();
+		if (!attr || !impl::is_attribute_of(attr._attr, _root)) return xml_attribute();
+		
+		impl::xml_allocator& alloc = impl::get_allocator(_root);
+		if (!alloc.reserve()) return xml_attribute();
+
+		xml_attribute a(impl::allocate_attribute(alloc));
+		if (!a) return xml_attribute();
+
+		impl::insert_attribute_before(a._attr, attr._attr, _root);
+
+		a.set_name(name_);
+
+		return a;
+	}
+
+	PUGI__FN xml_attribute xml_node::append_copy(const xml_attribute& proto)
+	{
+		if (!proto) return xml_attribute();
+		if (!impl::allow_insert_attribute(type())) return xml_attribute();
+
+		impl::xml_allocator& alloc = impl::get_allocator(_root);
+		if (!alloc.reserve()) return xml_attribute();
+
+		xml_attribute a(impl::allocate_attribute(alloc));
+		if (!a) return xml_attribute();
+
+		impl::append_attribute(a._attr, _root);
+		impl::node_copy_attribute(a._attr, proto._attr);
+
+		return a;
+	}
+
+	PUGI__FN xml_attribute xml_node::prepend_copy(const xml_attribute& proto)
+	{
+		if (!proto) return xml_attribute();
+		if (!impl::allow_insert_attribute(type())) return xml_attribute();
+
+		impl::xml_allocator& alloc = impl::get_allocator(_root);
+		if (!alloc.reserve()) return xml_attribute();
+
+		xml_attribute a(impl::allocate_attribute(alloc));
+		if (!a) return xml_attribute();
+
+		impl::prepend_attribute(a._attr, _root);
+		impl::node_copy_attribute(a._attr, proto._attr);
+
+		return a;
+	}
+
+	PUGI__FN xml_attribute xml_node::insert_copy_after(const xml_attribute& proto, const xml_attribute& attr)
+	{
+		if (!proto) return xml_attribute();
+		if (!impl::allow_insert_attribute(type())) return xml_attribute();
+		if (!attr || !impl::is_attribute_of(attr._attr, _root)) return xml_attribute();
+
+		impl::xml_allocator& alloc = impl::get_allocator(_root);
+		if (!alloc.reserve()) return xml_attribute();
+
+		xml_attribute a(impl::allocate_attribute(alloc));
+		if (!a) return xml_attribute();
+
+		impl::insert_attribute_after(a._attr, attr._attr, _root);
+		impl::node_copy_attribute(a._attr, proto._attr);
+
+		return a;
+	}
+
+	PUGI__FN xml_attribute xml_node::insert_copy_before(const xml_attribute& proto, const xml_attribute& attr)
+	{
+		if (!proto) return xml_attribute();
+		if (!impl::allow_insert_attribute(type())) return xml_attribute();
+		if (!attr || !impl::is_attribute_of(attr._attr, _root)) return xml_attribute();
+
+		impl::xml_allocator& alloc = impl::get_allocator(_root);
+		if (!alloc.reserve()) return xml_attribute();
+
+		xml_attribute a(impl::allocate_attribute(alloc));
+		if (!a) return xml_attribute();
+
+		impl::insert_attribute_before(a._attr, attr._attr, _root);
+		impl::node_copy_attribute(a._attr, proto._attr);
+
+		return a;
+	}
+
+	PUGI__FN xml_node xml_node::append_child(xml_node_type type_)
+	{
+		if (!impl::allow_insert_child(type(), type_)) return xml_node();
+		
+		impl::xml_allocator& alloc = impl::get_allocator(_root);
+		if (!alloc.reserve()) return xml_node();
+
+		xml_node n(impl::allocate_node(alloc, type_));
+		if (!n) return xml_node();
+
+		impl::append_node(n._root, _root);
+
+		if (type_ == node_declaration) n.set_name(PUGIXML_TEXT("xml"));
+
+		return n;
+	}
+
+	PUGI__FN xml_node xml_node::prepend_child(xml_node_type type_)
+	{
+		if (!impl::allow_insert_child(type(), type_)) return xml_node();
+
+		impl::xml_allocator& alloc = impl::get_allocator(_root);
+		if (!alloc.reserve()) return xml_node();
+		
+		xml_node n(impl::allocate_node(alloc, type_));
+		if (!n) return xml_node();
+
+		impl::prepend_node(n._root, _root);
+				
+		if (type_ == node_declaration) n.set_name(PUGIXML_TEXT("xml"));
+
+		return n;
+	}
+
+	PUGI__FN xml_node xml_node::insert_child_before(xml_node_type type_, const xml_node& node)
+	{
+		if (!impl::allow_insert_child(type(), type_)) return xml_node();
+		if (!node._root || node._root->parent != _root) return xml_node();
+
+		impl::xml_allocator& alloc = impl::get_allocator(_root);
+		if (!alloc.reserve()) return xml_node();
+	
+		xml_node n(impl::allocate_node(alloc, type_));
+		if (!n) return xml_node();
+
+		impl::insert_node_before(n._root, node._root);
+
+		if (type_ == node_declaration) n.set_name(PUGIXML_TEXT("xml"));
+
+		return n;
+	}
+
+	PUGI__FN xml_node xml_node::insert_child_after(xml_node_type type_, const xml_node& node)
+	{
+		if (!impl::allow_insert_child(type(), type_)) return xml_node();
+		if (!node._root || node._root->parent != _root) return xml_node();
+
+		impl::xml_allocator& alloc = impl::get_allocator(_root);
+		if (!alloc.reserve()) return xml_node();
+	
+		xml_node n(impl::allocate_node(alloc, type_));
+		if (!n) return xml_node();
+
+		impl::insert_node_after(n._root, node._root);
+
+		if (type_ == node_declaration) n.set_name(PUGIXML_TEXT("xml"));
+
+		return n;
+	}
+
+	PUGI__FN xml_node xml_node::append_child(const char_t* name_)
+	{
+		xml_node result = append_child(node_element);
+
+		result.set_name(name_);
+
+		return result;
+	}
+
+	PUGI__FN xml_node xml_node::prepend_child(const char_t* name_)
+	{
+		xml_node result = prepend_child(node_element);
+
+		result.set_name(name_);
+
+		return result;
+	}
+
+	PUGI__FN xml_node xml_node::insert_child_after(const char_t* name_, const xml_node& node)
+	{
+		xml_node result = insert_child_after(node_element, node);
+
+		result.set_name(name_);
+
+		return result;
+	}
+
+	PUGI__FN xml_node xml_node::insert_child_before(const char_t* name_, const xml_node& node)
+	{
+		xml_node result = insert_child_before(node_element, node);
+
+		result.set_name(name_);
+
+		return result;
+	}
+
+	PUGI__FN xml_node xml_node::append_copy(const xml_node& proto)
+	{
+		xml_node_type type_ = proto.type();
+		if (!impl::allow_insert_child(type(), type_)) return xml_node();
+
+		impl::xml_allocator& alloc = impl::get_allocator(_root);
+		if (!alloc.reserve()) return xml_node();
+
+		xml_node n(impl::allocate_node(alloc, type_));
+		if (!n) return xml_node();
+
+		impl::append_node(n._root, _root);
+		impl::node_copy_tree(n._root, proto._root);
+
+		return n;
+	}
+
+	PUGI__FN xml_node xml_node::prepend_copy(const xml_node& proto)
+	{
+		xml_node_type type_ = proto.type();
+		if (!impl::allow_insert_child(type(), type_)) return xml_node();
+
+		impl::xml_allocator& alloc = impl::get_allocator(_root);
+		if (!alloc.reserve()) return xml_node();
+
+		xml_node n(impl::allocate_node(alloc, type_));
+		if (!n) return xml_node();
+
+		impl::prepend_node(n._root, _root);
+		impl::node_copy_tree(n._root, proto._root);
+
+		return n;
+	}
+
+	PUGI__FN xml_node xml_node::insert_copy_after(const xml_node& proto, const xml_node& node)
+	{
+		xml_node_type type_ = proto.type();
+		if (!impl::allow_insert_child(type(), type_)) return xml_node();
+		if (!node._root || node._root->parent != _root) return xml_node();
+
+		impl::xml_allocator& alloc = impl::get_allocator(_root);
+		if (!alloc.reserve()) return xml_node();
+
+		xml_node n(impl::allocate_node(alloc, type_));
+		if (!n) return xml_node();
+
+		impl::insert_node_after(n._root, node._root);
+		impl::node_copy_tree(n._root, proto._root);
+
+		return n;
+	}
+
+	PUGI__FN xml_node xml_node::insert_copy_before(const xml_node& proto, const xml_node& node)
+	{
+		xml_node_type type_ = proto.type();
+		if (!impl::allow_insert_child(type(), type_)) return xml_node();
+		if (!node._root || node._root->parent != _root) return xml_node();
+
+		impl::xml_allocator& alloc = impl::get_allocator(_root);
+		if (!alloc.reserve()) return xml_node();
+
+		xml_node n(impl::allocate_node(alloc, type_));
+		if (!n) return xml_node();
+
+		impl::insert_node_before(n._root, node._root);
+		impl::node_copy_tree(n._root, proto._root);
+
+		return n;
+	}
+
+	PUGI__FN xml_node xml_node::append_move(const xml_node& moved)
+	{
+		if (!impl::allow_move(*this, moved)) return xml_node();
+
+		impl::xml_allocator& alloc = impl::get_allocator(_root);
+		if (!alloc.reserve()) return xml_node();
+
+		// disable document_buffer_order optimization since moving nodes around changes document order without changing buffer pointers
+		impl::get_document(_root).header |= impl::xml_memory_page_contents_shared_mask;
+
+		impl::remove_node(moved._root);
+		impl::append_node(moved._root, _root);
+
+		return moved;
+	}
+
+	PUGI__FN xml_node xml_node::prepend_move(const xml_node& moved)
+	{
+		if (!impl::allow_move(*this, moved)) return xml_node();
+
+		impl::xml_allocator& alloc = impl::get_allocator(_root);
+		if (!alloc.reserve()) return xml_node();
+
+		// disable document_buffer_order optimization since moving nodes around changes document order without changing buffer pointers
+		impl::get_document(_root).header |= impl::xml_memory_page_contents_shared_mask;
+
+		impl::remove_node(moved._root);
+		impl::prepend_node(moved._root, _root);
+
+		return moved;
+	}
+
+	PUGI__FN xml_node xml_node::insert_move_after(const xml_node& moved, const xml_node& node)
+	{
+		if (!impl::allow_move(*this, moved)) return xml_node();
+		if (!node._root || node._root->parent != _root) return xml_node();
+		if (moved._root == node._root) return xml_node();
+
+		impl::xml_allocator& alloc = impl::get_allocator(_root);
+		if (!alloc.reserve()) return xml_node();
+
+		// disable document_buffer_order optimization since moving nodes around changes document order without changing buffer pointers
+		impl::get_document(_root).header |= impl::xml_memory_page_contents_shared_mask;
+
+		impl::remove_node(moved._root);
+		impl::insert_node_after(moved._root, node._root);
+
+		return moved;
+	}
+
+	PUGI__FN xml_node xml_node::insert_move_before(const xml_node& moved, const xml_node& node)
+	{
+		if (!impl::allow_move(*this, moved)) return xml_node();
+		if (!node._root || node._root->parent != _root) return xml_node();
+		if (moved._root == node._root) return xml_node();
+
+		impl::xml_allocator& alloc = impl::get_allocator(_root);
+		if (!alloc.reserve()) return xml_node();
+
+		// disable document_buffer_order optimization since moving nodes around changes document order without changing buffer pointers
+		impl::get_document(_root).header |= impl::xml_memory_page_contents_shared_mask;
+
+		impl::remove_node(moved._root);
+		impl::insert_node_before(moved._root, node._root);
+
+		return moved;
+	}
+
+	PUGI__FN bool xml_node::remove_attribute(const char_t* name_)
+	{
+		return remove_attribute(attribute(name_));
+	}
+
+	PUGI__FN bool xml_node::remove_attribute(const xml_attribute& a)
+	{
+		if (!_root || !a._attr) return false;
+		if (!impl::is_attribute_of(a._attr, _root)) return false;
+
+		impl::xml_allocator& alloc = impl::get_allocator(_root);
+		if (!alloc.reserve()) return false;
+
+		impl::remove_attribute(a._attr, _root);
+		impl::destroy_attribute(a._attr, alloc);
+
+		return true;
+	}
+
+	PUGI__FN bool xml_node::remove_child(const char_t* name_)
+	{
+		return remove_child(child(name_));
+	}
+
+	PUGI__FN bool xml_node::remove_child(const xml_node& n)
+	{
+		if (!_root || !n._root || n._root->parent != _root) return false;
+
+		impl::xml_allocator& alloc = impl::get_allocator(_root);
+		if (!alloc.reserve()) return false;
+
+		impl::remove_node(n._root);
+		impl::destroy_node(n._root, alloc);
+
+		return true;
+	}
+
+	PUGI__FN xml_parse_result xml_node::append_buffer(const void* contents, size_t size, unsigned int options, xml_encoding encoding)
+	{
+		// append_buffer is only valid for elements/documents
+		if (!impl::allow_insert_child(type(), node_element)) return impl::make_parse_result(status_append_invalid_root);
+
+		// get document node
+		impl::xml_document_struct* doc = &impl::get_document(_root);
+
+		// disable document_buffer_order optimization since in a document with multiple buffers comparing buffer pointers does not make sense
+		doc->header |= impl::xml_memory_page_contents_shared_mask;
+		
+		// get extra buffer element (we'll store the document fragment buffer there so that we can deallocate it later)
+		impl::xml_memory_page* page = 0;
+		impl::xml_extra_buffer* extra = static_cast<impl::xml_extra_buffer*>(doc->allocate_memory(sizeof(impl::xml_extra_buffer), page));
+		(void)page;
+
+		if (!extra) return impl::make_parse_result(status_out_of_memory);
+
+		// add extra buffer to the list
+		extra->buffer = 0;
+		extra->next = doc->extra_buffers;
+		doc->extra_buffers = extra;
+
+		// name of the root has to be NULL before parsing - otherwise closing node mismatches will not be detected at the top level
+		struct name_sentry
+		{
+			xml_node_struct* node;
+			char_t* name;
+
+			~name_sentry() { node->name = name; }
+		};
+
+		name_sentry sentry = { _root, _root->name };
+
+		sentry.node->name = 0;
+
+		return impl::load_buffer_impl(doc, _root, const_cast<void*>(contents), size, options, encoding, false, false, &extra->buffer);
+	}
+
+	PUGI__FN xml_node xml_node::find_child_by_attribute(const char_t* name_, const char_t* attr_name, const char_t* attr_value) const
+	{
+		if (!_root) return xml_node();
+		
+		for (xml_node_struct* i = _root->first_child; i; i = i->next_sibling)
+			if (i->name && impl::strequal(name_, i->name))
+			{
+				for (xml_attribute_struct* a = i->first_attribute; a; a = a->next_attribute)
+					if (a->name && impl::strequal(attr_name, a->name) && impl::strequal(attr_value, a->value ? a->value + 0 : PUGIXML_TEXT("")))
+						return xml_node(i);
+			}
+
+		return xml_node();
+	}
+
+	PUGI__FN xml_node xml_node::find_child_by_attribute(const char_t* attr_name, const char_t* attr_value) const
+	{
+		if (!_root) return xml_node();
+		
+		for (xml_node_struct* i = _root->first_child; i; i = i->next_sibling)
+			for (xml_attribute_struct* a = i->first_attribute; a; a = a->next_attribute)
+				if (a->name && impl::strequal(attr_name, a->name) && impl::strequal(attr_value, a->value ? a->value + 0 : PUGIXML_TEXT("")))
+					return xml_node(i);
+
+		return xml_node();
+	}
+
+#ifndef PUGIXML_NO_STL
+	PUGI__FN string_t xml_node::path(char_t delimiter) const
+	{
+		if (!_root) return string_t();
+
+		size_t offset = 0;
+
+		for (xml_node_struct* i = _root; i; i = i->parent)
+		{
+			offset += (i != _root);
+			offset += i->name ? impl::strlength(i->name) : 0;
+		}
+
+		string_t result;
+		result.resize(offset);
+
+		for (xml_node_struct* j = _root; j; j = j->parent)
+		{
+			if (j != _root)
+				result[--offset] = delimiter;
+
+			if (j->name && *j->name)
+			{
+				size_t length = impl::strlength(j->name);
+
+				offset -= length;
+				memcpy(&result[offset], j->name, length * sizeof(char_t));
+			}
+		}
+
+		assert(offset == 0);
+
+		return result;
+	}
+#endif
+
+	PUGI__FN xml_node xml_node::first_element_by_path(const char_t* path_, char_t delimiter) const
+	{
+		xml_node found = *this; // Current search context.
+
+		if (!_root || !path_ || !path_[0]) return found;
+
+		if (path_[0] == delimiter)
+		{
+			// Absolute path; e.g. '/foo/bar'
+			found = found.root();
+			++path_;
+		}
+
+		const char_t* path_segment = path_;
+
+		while (*path_segment == delimiter) ++path_segment;
+
+		const char_t* path_segment_end = path_segment;
+
+		while (*path_segment_end && *path_segment_end != delimiter) ++path_segment_end;
+
+		if (path_segment == path_segment_end) return found;
+
+		const char_t* next_segment = path_segment_end;
+
+		while (*next_segment == delimiter) ++next_segment;
+
+		if (*path_segment == '.' && path_segment + 1 == path_segment_end)
+			return found.first_element_by_path(next_segment, delimiter);
+		else if (*path_segment == '.' && *(path_segment+1) == '.' && path_segment + 2 == path_segment_end)
+			return found.parent().first_element_by_path(next_segment, delimiter);
+		else
+		{
+			for (xml_node_struct* j = found._root->first_child; j; j = j->next_sibling)
+			{
+				if (j->name && impl::strequalrange(j->name, path_segment, static_cast<size_t>(path_segment_end - path_segment)))
+				{
+					xml_node subsearch = xml_node(j).first_element_by_path(next_segment, delimiter);
+
+					if (subsearch) return subsearch;
+				}
+			}
+
+			return xml_node();
+		}
+	}
+
+	PUGI__FN bool xml_node::traverse(xml_tree_walker& walker)
+	{
+		walker._depth = -1;
+		
+		xml_node arg_begin = *this;
+		if (!walker.begin(arg_begin)) return false;
+
+		xml_node cur = first_child();
+				
+		if (cur)
+		{
+			++walker._depth;
+
+			do 
+			{
+				xml_node arg_for_each = cur;
+				if (!walker.for_each(arg_for_each))
+					return false;
+						
+				if (cur.first_child())
+				{
+					++walker._depth;
+					cur = cur.first_child();
+				}
+				else if (cur.next_sibling())
+					cur = cur.next_sibling();
+				else
+				{
+					// Borland C++ workaround
+					while (!cur.next_sibling() && cur != *this && !cur.parent().empty())
+					{
+						--walker._depth;
+						cur = cur.parent();
+					}
+						
+					if (cur != *this)
+						cur = cur.next_sibling();
+				}
+			}
+			while (cur && cur != *this);
+		}
+
+		assert(walker._depth == -1);
+
+		xml_node arg_end = *this;
+		return walker.end(arg_end);
+	}
+
+	PUGI__FN size_t xml_node::hash_value() const
+	{
+		return static_cast<size_t>(reinterpret_cast<uintptr_t>(_root) / sizeof(xml_node_struct));
+	}
+
+	PUGI__FN xml_node_struct* xml_node::internal_object() const
+	{
+		return _root;
+	}
+
+	PUGI__FN void xml_node::print(xml_writer& writer, const char_t* indent, unsigned int flags, xml_encoding encoding, unsigned int depth) const
+	{
+		if (!_root) return;
+
+		impl::xml_buffered_writer buffered_writer(writer, encoding);
+
+		impl::node_output(buffered_writer, _root, indent, flags, depth);
+
+		buffered_writer.flush();
+	}
+
+#ifndef PUGIXML_NO_STL
+	PUGI__FN void xml_node::print(std::basic_ostream<char, std::char_traits<char> >& stream, const char_t* indent, unsigned int flags, xml_encoding encoding, unsigned int depth) const
+	{
+		xml_writer_stream writer(stream);
+
+		print(writer, indent, flags, encoding, depth);
+	}
+
+	PUGI__FN void xml_node::print(std::basic_ostream<wchar_t, std::char_traits<wchar_t> >& stream, const char_t* indent, unsigned int flags, unsigned int depth) const
+	{
+		xml_writer_stream writer(stream);
+
+		print(writer, indent, flags, encoding_wchar, depth);
+	}
+#endif
+
+	PUGI__FN ptrdiff_t xml_node::offset_debug() const
+	{
+		if (!_root) return -1;
+
+		impl::xml_document_struct& doc = impl::get_document(_root);
+
+		// we can determine the offset reliably only if there is exactly once parse buffer
+		if (!doc.buffer || doc.extra_buffers) return -1;
+
+		switch (type())
+		{
+		case node_document:
+			return 0;
+
+		case node_element:
+		case node_declaration:
+		case node_pi:
+			return _root->name && (_root->header & impl::xml_memory_page_name_allocated_or_shared_mask) == 0 ? _root->name - doc.buffer : -1;
+
+		case node_pcdata:
+		case node_cdata:
+		case node_comment:
+		case node_doctype:
+			return _root->value && (_root->header & impl::xml_memory_page_value_allocated_or_shared_mask) == 0 ? _root->value - doc.buffer : -1;
+
+		default:
+			return -1;
+		}
+	}
+
+#ifdef __BORLANDC__
+	PUGI__FN bool operator&&(const xml_node& lhs, bool rhs)
+	{
+		return (bool)lhs && rhs;
+	}
+
+	PUGI__FN bool operator||(const xml_node& lhs, bool rhs)
+	{
+		return (bool)lhs || rhs;
+	}
+#endif
+
+	PUGI__FN xml_text::xml_text(xml_node_struct* root): _root(root)
+	{
+	}
+
+	PUGI__FN xml_node_struct* xml_text::_data() const
+	{
+		if (!_root || impl::is_text_node(_root)) return _root;
+
+		for (xml_node_struct* node = _root->first_child; node; node = node->next_sibling)
+			if (impl::is_text_node(node))
+				return node;
+
+		return 0;
+	}
+
+	PUGI__FN xml_node_struct* xml_text::_data_new()
+	{
+		xml_node_struct* d = _data();
+		if (d) return d;
+
+		return xml_node(_root).append_child(node_pcdata).internal_object();
+	}
+
+	PUGI__FN xml_text::xml_text(): _root(0)
+	{
+	}
+
+	PUGI__FN static void unspecified_bool_xml_text(xml_text***)
+	{
+	}
+
+	PUGI__FN xml_text::operator xml_text::unspecified_bool_type() const
+	{
+		return _data() ? unspecified_bool_xml_text : 0;
+	}
+
+	PUGI__FN bool xml_text::operator!() const
+	{
+		return !_data();
+	}
+
+	PUGI__FN bool xml_text::empty() const
+	{
+		return _data() == 0;
+	}
+
+	PUGI__FN const char_t* xml_text::get() const
+	{
+		xml_node_struct* d = _data();
+
+		return (d && d->value) ? d->value + 0 : PUGIXML_TEXT("");
+	}
+
+	PUGI__FN const char_t* xml_text::as_string(const char_t* def) const
+	{
+		xml_node_struct* d = _data();
+
+		return (d && d->value) ? d->value : def;
+	}
+
+	PUGI__FN int xml_text::as_int(int def) const
+	{
+		xml_node_struct* d = _data();
+
+		return (d && d->value) ? impl::get_value_int(d->value) : def;
+	}
+
+	PUGI__FN unsigned int xml_text::as_uint(unsigned int def) const
+	{
+		xml_node_struct* d = _data();
+
+		return (d && d->value) ? impl::get_value_uint(d->value) : def;
+	}
+
+	PUGI__FN double xml_text::as_double(double def) const
+	{
+		xml_node_struct* d = _data();
+
+		return (d && d->value) ? impl::get_value_double(d->value) : def;
+	}
+
+	PUGI__FN float xml_text::as_float(float def) const
+	{
+		xml_node_struct* d = _data();
+
+		return (d && d->value) ? impl::get_value_float(d->value) : def;
+	}
+
+	PUGI__FN bool xml_text::as_bool(bool def) const
+	{
+		xml_node_struct* d = _data();
+
+		return (d && d->value) ? impl::get_value_bool(d->value) : def;
+	}
+
+#ifdef PUGIXML_HAS_LONG_LONG
+	PUGI__FN long long xml_text::as_llong(long long def) const
+	{
+		xml_node_struct* d = _data();
+
+		return (d && d->value) ? impl::get_value_llong(d->value) : def;
+	}
+
+	PUGI__FN unsigned long long xml_text::as_ullong(unsigned long long def) const
+	{
+		xml_node_struct* d = _data();
+
+		return (d && d->value) ? impl::get_value_ullong(d->value) : def;
+	}
+#endif
+
+	PUGI__FN bool xml_text::set(const char_t* rhs)
+	{
+		xml_node_struct* dn = _data_new();
+
+		return dn ? impl::strcpy_insitu(dn->value, dn->header, impl::xml_memory_page_value_allocated_mask, rhs) : false;
+	}
+
+	PUGI__FN bool xml_text::set(int rhs)
+	{
+		xml_node_struct* dn = _data_new();
+
+		return dn ? impl::set_value_convert(dn->value, dn->header, impl::xml_memory_page_value_allocated_mask, rhs) : false;
+	}
+
+	PUGI__FN bool xml_text::set(unsigned int rhs)
+	{
+		xml_node_struct* dn = _data_new();
+
+		return dn ? impl::set_value_convert(dn->value, dn->header, impl::xml_memory_page_value_allocated_mask, rhs) : false;
+	}
+
+	PUGI__FN bool xml_text::set(float rhs)
+	{
+		xml_node_struct* dn = _data_new();
+
+		return dn ? impl::set_value_convert(dn->value, dn->header, impl::xml_memory_page_value_allocated_mask, rhs) : false;
+	}
+
+	PUGI__FN bool xml_text::set(double rhs)
+	{
+		xml_node_struct* dn = _data_new();
+
+		return dn ? impl::set_value_convert(dn->value, dn->header, impl::xml_memory_page_value_allocated_mask, rhs) : false;
+	}
+
+	PUGI__FN bool xml_text::set(bool rhs)
+	{
+		xml_node_struct* dn = _data_new();
+
+		return dn ? impl::set_value_convert(dn->value, dn->header, impl::xml_memory_page_value_allocated_mask, rhs) : false;
+	}
+
+#ifdef PUGIXML_HAS_LONG_LONG
+	PUGI__FN bool xml_text::set(long long rhs)
+	{
+		xml_node_struct* dn = _data_new();
+
+		return dn ? impl::set_value_convert(dn->value, dn->header, impl::xml_memory_page_value_allocated_mask, rhs) : false;
+	}
+
+	PUGI__FN bool xml_text::set(unsigned long long rhs)
+	{
+		xml_node_struct* dn = _data_new();
+
+		return dn ? impl::set_value_convert(dn->value, dn->header, impl::xml_memory_page_value_allocated_mask, rhs) : false;
+	}
+#endif
+
+	PUGI__FN xml_text& xml_text::operator=(const char_t* rhs)
+	{
+		set(rhs);
+		return *this;
+	}
+
+	PUGI__FN xml_text& xml_text::operator=(int rhs)
+	{
+		set(rhs);
+		return *this;
+	}
+
+	PUGI__FN xml_text& xml_text::operator=(unsigned int rhs)
+	{
+		set(rhs);
+		return *this;
+	}
+
+	PUGI__FN xml_text& xml_text::operator=(double rhs)
+	{
+		set(rhs);
+		return *this;
+	}
+
+	PUGI__FN xml_text& xml_text::operator=(float rhs)
+	{
+		set(rhs);
+		return *this;
+	}
+
+	PUGI__FN xml_text& xml_text::operator=(bool rhs)
+	{
+		set(rhs);
+		return *this;
+	}
+
+#ifdef PUGIXML_HAS_LONG_LONG
+	PUGI__FN xml_text& xml_text::operator=(long long rhs)
+	{
+		set(rhs);
+		return *this;
+	}
+
+	PUGI__FN xml_text& xml_text::operator=(unsigned long long rhs)
+	{
+		set(rhs);
+		return *this;
+	}
+#endif
+
+	PUGI__FN xml_node xml_text::data() const
+	{
+		return xml_node(_data());
+	}
+
+#ifdef __BORLANDC__
+	PUGI__FN bool operator&&(const xml_text& lhs, bool rhs)
+	{
+		return (bool)lhs && rhs;
+	}
+
+	PUGI__FN bool operator||(const xml_text& lhs, bool rhs)
+	{
+		return (bool)lhs || rhs;
+	}
+#endif
+
+	PUGI__FN xml_node_iterator::xml_node_iterator()
+	{
+	}
+
+	PUGI__FN xml_node_iterator::xml_node_iterator(const xml_node& node): _wrap(node), _parent(node.parent())
+	{
+	}
+
+	PUGI__FN xml_node_iterator::xml_node_iterator(xml_node_struct* ref, xml_node_struct* parent): _wrap(ref), _parent(parent)
+	{
+	}
+
+	PUGI__FN bool xml_node_iterator::operator==(const xml_node_iterator& rhs) const
+	{
+		return _wrap._root == rhs._wrap._root && _parent._root == rhs._parent._root;
+	}
+	
+	PUGI__FN bool xml_node_iterator::operator!=(const xml_node_iterator& rhs) const
+	{
+		return _wrap._root != rhs._wrap._root || _parent._root != rhs._parent._root;
+	}
+
+	PUGI__FN xml_node& xml_node_iterator::operator*() const
+	{
+		assert(_wrap._root);
+		return _wrap;
+	}
+
+	PUGI__FN xml_node* xml_node_iterator::operator->() const
+	{
+		assert(_wrap._root);
+		return const_cast<xml_node*>(&_wrap); // BCC32 workaround
+	}
+
+	PUGI__FN const xml_node_iterator& xml_node_iterator::operator++()
+	{
+		assert(_wrap._root);
+		_wrap._root = _wrap._root->next_sibling;
+		return *this;
+	}
+
+	PUGI__FN xml_node_iterator xml_node_iterator::operator++(int)
+	{
+		xml_node_iterator temp = *this;
+		++*this;
+		return temp;
+	}
+
+	PUGI__FN const xml_node_iterator& xml_node_iterator::operator--()
+	{
+		_wrap = _wrap._root ? _wrap.previous_sibling() : _parent.last_child();
+		return *this;
+	}
+
+	PUGI__FN xml_node_iterator xml_node_iterator::operator--(int)
+	{
+		xml_node_iterator temp = *this;
+		--*this;
+		return temp;
+	}
+
+	PUGI__FN xml_attribute_iterator::xml_attribute_iterator()
+	{
+	}
+
+	PUGI__FN xml_attribute_iterator::xml_attribute_iterator(const xml_attribute& attr, const xml_node& parent): _wrap(attr), _parent(parent)
+	{
+	}
+
+	PUGI__FN xml_attribute_iterator::xml_attribute_iterator(xml_attribute_struct* ref, xml_node_struct* parent): _wrap(ref), _parent(parent)
+	{
+	}
+
+	PUGI__FN bool xml_attribute_iterator::operator==(const xml_attribute_iterator& rhs) const
+	{
+		return _wrap._attr == rhs._wrap._attr && _parent._root == rhs._parent._root;
+	}
+	
+	PUGI__FN bool xml_attribute_iterator::operator!=(const xml_attribute_iterator& rhs) const
+	{
+		return _wrap._attr != rhs._wrap._attr || _parent._root != rhs._parent._root;
+	}
+
+	PUGI__FN xml_attribute& xml_attribute_iterator::operator*() const
+	{
+		assert(_wrap._attr);
+		return _wrap;
+	}
+
+	PUGI__FN xml_attribute* xml_attribute_iterator::operator->() const
+	{
+		assert(_wrap._attr);
+		return const_cast<xml_attribute*>(&_wrap); // BCC32 workaround
+	}
+
+	PUGI__FN const xml_attribute_iterator& xml_attribute_iterator::operator++()
+	{
+		assert(_wrap._attr);
+		_wrap._attr = _wrap._attr->next_attribute;
+		return *this;
+	}
+
+	PUGI__FN xml_attribute_iterator xml_attribute_iterator::operator++(int)
+	{
+		xml_attribute_iterator temp = *this;
+		++*this;
+		return temp;
+	}
+
+	PUGI__FN const xml_attribute_iterator& xml_attribute_iterator::operator--()
+	{
+		_wrap = _wrap._attr ? _wrap.previous_attribute() : _parent.last_attribute();
+		return *this;
+	}
+
+	PUGI__FN xml_attribute_iterator xml_attribute_iterator::operator--(int)
+	{
+		xml_attribute_iterator temp = *this;
+		--*this;
+		return temp;
+	}
+
+	PUGI__FN xml_named_node_iterator::xml_named_node_iterator(): _name(0)
+	{
+	}
+
+	PUGI__FN xml_named_node_iterator::xml_named_node_iterator(const xml_node& node, const char_t* name): _wrap(node), _parent(node.parent()), _name(name)
+	{
+	}
+
+	PUGI__FN xml_named_node_iterator::xml_named_node_iterator(xml_node_struct* ref, xml_node_struct* parent, const char_t* name): _wrap(ref), _parent(parent), _name(name)
+	{
+	}
+
+	PUGI__FN bool xml_named_node_iterator::operator==(const xml_named_node_iterator& rhs) const
+	{
+		return _wrap._root == rhs._wrap._root && _parent._root == rhs._parent._root;
+	}
+
+	PUGI__FN bool xml_named_node_iterator::operator!=(const xml_named_node_iterator& rhs) const
+	{
+		return _wrap._root != rhs._wrap._root || _parent._root != rhs._parent._root;
+	}
+
+	PUGI__FN xml_node& xml_named_node_iterator::operator*() const
+	{
+		assert(_wrap._root);
+		return _wrap;
+	}
+
+	PUGI__FN xml_node* xml_named_node_iterator::operator->() const
+	{
+		assert(_wrap._root);
+		return const_cast<xml_node*>(&_wrap); // BCC32 workaround
+	}
+
+	PUGI__FN const xml_named_node_iterator& xml_named_node_iterator::operator++()
+	{
+		assert(_wrap._root);
+		_wrap = _wrap.next_sibling(_name);
+		return *this;
+	}
+
+	PUGI__FN xml_named_node_iterator xml_named_node_iterator::operator++(int)
+	{
+		xml_named_node_iterator temp = *this;
+		++*this;
+		return temp;
+	}
+
+	PUGI__FN const xml_named_node_iterator& xml_named_node_iterator::operator--()
+	{
+		if (_wrap._root)
+			_wrap = _wrap.previous_sibling(_name);
+		else
+		{
+			_wrap = _parent.last_child();
+
+			if (!impl::strequal(_wrap.name(), _name))
+				_wrap = _wrap.previous_sibling(_name);
+		}
+
+		return *this;
+	}
+
+	PUGI__FN xml_named_node_iterator xml_named_node_iterator::operator--(int)
+	{
+		xml_named_node_iterator temp = *this;
+		--*this;
+		return temp;
+	}
+
+	PUGI__FN xml_parse_result::xml_parse_result(): status(status_internal_error), offset(0), encoding(encoding_auto)
+	{
+	}
+
+	PUGI__FN xml_parse_result::operator bool() const
+	{
+		return status == status_ok;
+	}
+
+	PUGI__FN const char* xml_parse_result::description() const
+	{
+		switch (status)
+		{
+		case status_ok: return "No error";
+
+		case status_file_not_found: return "File was not found";
+		case status_io_error: return "Error reading from file/stream";
+		case status_out_of_memory: return "Could not allocate memory";
+		case status_internal_error: return "Internal error occurred";
+
+		case status_unrecognized_tag: return "Could not determine tag type";
+
+		case status_bad_pi: return "Error parsing document declaration/processing instruction";
+		case status_bad_comment: return "Error parsing comment";
+		case status_bad_cdata: return "Error parsing CDATA section";
+		case status_bad_doctype: return "Error parsing document type declaration";
+		case status_bad_pcdata: return "Error parsing PCDATA section";
+		case status_bad_start_element: return "Error parsing start element tag";
+		case status_bad_attribute: return "Error parsing element attribute";
+		case status_bad_end_element: return "Error parsing end element tag";
+		case status_end_element_mismatch: return "Start-end tags mismatch";
+
+		case status_append_invalid_root: return "Unable to append nodes: root is not an element or document";
+
+		case status_no_document_element: return "No document element found";
+
+		default: return "Unknown error";
+		}
+	}
+
+	PUGI__FN xml_document::xml_document(): _buffer(0)
+	{
+		create();
+	}
+
+	PUGI__FN xml_document::~xml_document()
+	{
+		destroy();
+	}
+
+	PUGI__FN void xml_document::reset()
+	{
+		destroy();
+		create();
+	}
+
+	PUGI__FN void xml_document::reset(const xml_document& proto)
+	{
+		reset();
+
+		for (xml_node cur = proto.first_child(); cur; cur = cur.next_sibling())
+			append_copy(cur);
+	}
+
+	PUGI__FN void xml_document::create()
+	{
+		assert(!_root);
+
+	#ifdef PUGIXML_COMPACT
+		const size_t page_offset = sizeof(uint32_t);
+	#else
+		const size_t page_offset = 0;
+	#endif
+
+		// initialize sentinel page
+		PUGI__STATIC_ASSERT(sizeof(impl::xml_memory_page) + sizeof(impl::xml_document_struct) + impl::xml_memory_page_alignment - sizeof(void*) + page_offset <= sizeof(_memory));
+
+		// align upwards to page boundary
+		void* page_memory = reinterpret_cast<void*>((reinterpret_cast<uintptr_t>(_memory) + (impl::xml_memory_page_alignment - 1)) & ~(impl::xml_memory_page_alignment - 1));
+
+		// prepare page structure
+		impl::xml_memory_page* page = impl::xml_memory_page::construct(page_memory);
+		assert(page);
+
+		page->busy_size = impl::xml_memory_page_size;
+
+		// setup first page marker
+	#ifdef PUGIXML_COMPACT
+		page->compact_page_marker = reinterpret_cast<uint32_t*>(reinterpret_cast<char*>(page) + sizeof(impl::xml_memory_page));
+		*page->compact_page_marker = sizeof(impl::xml_memory_page);
+	#endif
+
+		// allocate new root
+		_root = new (reinterpret_cast<char*>(page) + sizeof(impl::xml_memory_page) + page_offset) impl::xml_document_struct(page);
+		_root->prev_sibling_c = _root;
+
+		// setup sentinel page
+		page->allocator = static_cast<impl::xml_document_struct*>(_root);
+
+		// verify the document allocation
+		assert(reinterpret_cast<char*>(_root) + sizeof(impl::xml_document_struct) <= _memory + sizeof(_memory));
+	}
+
+	PUGI__FN void xml_document::destroy()
+	{
+		assert(_root);
+
+		// destroy static storage
+		if (_buffer)
+		{
+			impl::xml_memory::deallocate(_buffer);
+			_buffer = 0;
+		}
+
+		// destroy extra buffers (note: no need to destroy linked list nodes, they're allocated using document allocator)
+		for (impl::xml_extra_buffer* extra = static_cast<impl::xml_document_struct*>(_root)->extra_buffers; extra; extra = extra->next)
+		{
+			if (extra->buffer) impl::xml_memory::deallocate(extra->buffer);
+		}
+
+		// destroy dynamic storage, leave sentinel page (it's in static memory)
+		impl::xml_memory_page* root_page = PUGI__GETPAGE(_root);
+		assert(root_page && !root_page->prev);
+		assert(reinterpret_cast<char*>(root_page) >= _memory && reinterpret_cast<char*>(root_page) < _memory + sizeof(_memory));
+
+		for (impl::xml_memory_page* page = root_page->next; page; )
+		{
+			impl::xml_memory_page* next = page->next;
+
+			impl::xml_allocator::deallocate_page(page);
+
+			page = next;
+		}
+
+	#ifdef PUGIXML_COMPACT
+		// destroy hash table
+		static_cast<impl::xml_document_struct*>(_root)->hash.clear();
+	#endif
+
+		_root = 0;
+	}
+
+#ifndef PUGIXML_NO_STL
+	PUGI__FN xml_parse_result xml_document::load(std::basic_istream<char, std::char_traits<char> >& stream, unsigned int options, xml_encoding encoding)
+	{
+		reset();
+
+		return impl::load_stream_impl(*this, stream, options, encoding);
+	}
+
+	PUGI__FN xml_parse_result xml_document::load(std::basic_istream<wchar_t, std::char_traits<wchar_t> >& stream, unsigned int options)
+	{
+		reset();
+
+		return impl::load_stream_impl(*this, stream, options, encoding_wchar);
+	}
+#endif
+
+	PUGI__FN xml_parse_result xml_document::load_string(const char_t* contents, unsigned int options)
+	{
+		// Force native encoding (skip autodetection)
+	#ifdef PUGIXML_WCHAR_MODE
+		xml_encoding encoding = encoding_wchar;
+	#else
+		xml_encoding encoding = encoding_utf8;
+	#endif
+
+		return load_buffer(contents, impl::strlength(contents) * sizeof(char_t), options, encoding);
+	}
+
+	PUGI__FN xml_parse_result xml_document::load(const char_t* contents, unsigned int options)
+	{
+		return load_string(contents, options);
+	}
+
+	PUGI__FN xml_parse_result xml_document::load_file(const char* path_, unsigned int options, xml_encoding encoding)
+	{
+		reset();
+
+		using impl::auto_deleter; // MSVC7 workaround
+		auto_deleter<FILE, int(*)(FILE*)> file(fopen(path_, "rb"), fclose);
+
+		return impl::load_file_impl(*this, file.data, options, encoding);
+	}
+
+	PUGI__FN xml_parse_result xml_document::load_file(const wchar_t* path_, unsigned int options, xml_encoding encoding)
+	{
+		reset();
+
+		using impl::auto_deleter; // MSVC7 workaround
+		auto_deleter<FILE, int(*)(FILE*)> file(impl::open_file_wide(path_, L"rb"), fclose);
+
+		return impl::load_file_impl(*this, file.data, options, encoding);
+	}
+
+	PUGI__FN xml_parse_result xml_document::load_buffer(const void* contents, size_t size, unsigned int options, xml_encoding encoding)
+	{
+		reset();
+
+		return impl::load_buffer_impl(static_cast<impl::xml_document_struct*>(_root), _root, const_cast<void*>(contents), size, options, encoding, false, false, &_buffer);
+	}
+
+	PUGI__FN xml_parse_result xml_document::load_buffer_inplace(void* contents, size_t size, unsigned int options, xml_encoding encoding)
+	{
+		reset();
+
+		return impl::load_buffer_impl(static_cast<impl::xml_document_struct*>(_root), _root, contents, size, options, encoding, true, false, &_buffer);
+	}
+
+	PUGI__FN xml_parse_result xml_document::load_buffer_inplace_own(void* contents, size_t size, unsigned int options, xml_encoding encoding)
+	{
+		reset();
+
+		return impl::load_buffer_impl(static_cast<impl::xml_document_struct*>(_root), _root, contents, size, options, encoding, true, true, &_buffer);
+	}
+
+	PUGI__FN void xml_document::save(xml_writer& writer, const char_t* indent, unsigned int flags, xml_encoding encoding) const
+	{
+		impl::xml_buffered_writer buffered_writer(writer, encoding);
+
+		if ((flags & format_write_bom) && encoding != encoding_latin1)
+		{
+			// BOM always represents the codepoint U+FEFF, so just write it in native encoding
+		#ifdef PUGIXML_WCHAR_MODE
+			unsigned int bom = 0xfeff;
+			buffered_writer.write(static_cast<wchar_t>(bom));
+		#else
+			buffered_writer.write('\xef', '\xbb', '\xbf');
+		#endif
+		}
+
+		if (!(flags & format_no_declaration) && !impl::has_declaration(_root))
+		{
+			buffered_writer.write_string(PUGIXML_TEXT("<?xml version=\"1.0\""));
+			if (encoding == encoding_latin1) buffered_writer.write_string(PUGIXML_TEXT(" encoding=\"ISO-8859-1\""));
+			buffered_writer.write('?', '>');
+			if (!(flags & format_raw)) buffered_writer.write('\n');
+		}
+
+		impl::node_output(buffered_writer, _root, indent, flags, 0);
+
+		buffered_writer.flush();
+	}
+
+#ifndef PUGIXML_NO_STL
+	PUGI__FN void xml_document::save(std::basic_ostream<char, std::char_traits<char> >& stream, const char_t* indent, unsigned int flags, xml_encoding encoding) const
+	{
+		xml_writer_stream writer(stream);
+
+		save(writer, indent, flags, encoding);
+	}
+
+	PUGI__FN void xml_document::save(std::basic_ostream<wchar_t, std::char_traits<wchar_t> >& stream, const char_t* indent, unsigned int flags) const
+	{
+		xml_writer_stream writer(stream);
+
+		save(writer, indent, flags, encoding_wchar);
+	}
+#endif
+
+	PUGI__FN bool xml_document::save_file(const char* path_, const char_t* indent, unsigned int flags, xml_encoding encoding) const
+	{
+		using impl::auto_deleter; // MSVC7 workaround
+		auto_deleter<FILE, int(*)(FILE*)> file(fopen(path_, (flags & format_save_file_text) ? "w" : "wb"), fclose);
+
+		return impl::save_file_impl(*this, file.data, indent, flags, encoding);
+	}
+
+	PUGI__FN bool xml_document::save_file(const wchar_t* path_, const char_t* indent, unsigned int flags, xml_encoding encoding) const
+	{
+		using impl::auto_deleter; // MSVC7 workaround
+		auto_deleter<FILE, int(*)(FILE*)> file(impl::open_file_wide(path_, (flags & format_save_file_text) ? L"w" : L"wb"), fclose);
+
+		return impl::save_file_impl(*this, file.data, indent, flags, encoding);
+	}
+
+	PUGI__FN xml_node xml_document::document_element() const
+	{
+		assert(_root);
+
+		for (xml_node_struct* i = _root->first_child; i; i = i->next_sibling)
+			if (PUGI__NODETYPE(i) == node_element)
+				return xml_node(i);
+
+		return xml_node();
+	}
+
+#ifndef PUGIXML_NO_STL
+	PUGI__FN std::string PUGIXML_FUNCTION as_utf8(const wchar_t* str)
+	{
+		assert(str);
+
+		return impl::as_utf8_impl(str, impl::strlength_wide(str));
+	}
+
+	PUGI__FN std::string PUGIXML_FUNCTION as_utf8(const std::basic_string<wchar_t>& str)
+	{
+		return impl::as_utf8_impl(str.c_str(), str.size());
+	}
+	
+	PUGI__FN std::basic_string<wchar_t> PUGIXML_FUNCTION as_wide(const char* str)
+	{
+		assert(str);
+
+		return impl::as_wide_impl(str, strlen(str));
+	}
+	
+	PUGI__FN std::basic_string<wchar_t> PUGIXML_FUNCTION as_wide(const std::string& str)
+	{
+		return impl::as_wide_impl(str.c_str(), str.size());
+	}
+#endif
+
+	PUGI__FN void PUGIXML_FUNCTION set_memory_management_functions(allocation_function allocate, deallocation_function deallocate)
+	{
+		impl::xml_memory::allocate = allocate;
+		impl::xml_memory::deallocate = deallocate;
+	}
+
+	PUGI__FN allocation_function PUGIXML_FUNCTION get_memory_allocation_function()
+	{
+		return impl::xml_memory::allocate;
+	}
+
+	PUGI__FN deallocation_function PUGIXML_FUNCTION get_memory_deallocation_function()
+	{
+		return impl::xml_memory::deallocate;
+	}
+}
+
+#if !defined(PUGIXML_NO_STL) && (defined(_MSC_VER) || defined(__ICC))
+namespace std
+{
+	// Workarounds for (non-standard) iterator category detection for older versions (MSVC7/IC8 and earlier)
+	PUGI__FN std::bidirectional_iterator_tag _Iter_cat(const pugi::xml_node_iterator&)
+	{
+		return std::bidirectional_iterator_tag();
+	}
+
+	PUGI__FN std::bidirectional_iterator_tag _Iter_cat(const pugi::xml_attribute_iterator&)
+	{
+		return std::bidirectional_iterator_tag();
+	}
+
+	PUGI__FN std::bidirectional_iterator_tag _Iter_cat(const pugi::xml_named_node_iterator&)
+	{
+		return std::bidirectional_iterator_tag();
+	}
+}
+#endif
+
+#if !defined(PUGIXML_NO_STL) && defined(__SUNPRO_CC)
+namespace std
+{
+	// Workarounds for (non-standard) iterator category detection
+	PUGI__FN std::bidirectional_iterator_tag __iterator_category(const pugi::xml_node_iterator&)
+	{
+		return std::bidirectional_iterator_tag();
+	}
+
+	PUGI__FN std::bidirectional_iterator_tag __iterator_category(const pugi::xml_attribute_iterator&)
+	{
+		return std::bidirectional_iterator_tag();
+	}
+
+	PUGI__FN std::bidirectional_iterator_tag __iterator_category(const pugi::xml_named_node_iterator&)
+	{
+		return std::bidirectional_iterator_tag();
+	}
+}
+#endif
+
+#ifndef PUGIXML_NO_XPATH
+// STL replacements
+PUGI__NS_BEGIN
+	struct equal_to
+	{
+		template <typename T> bool operator()(const T& lhs, const T& rhs) const
+		{
+			return lhs == rhs;
+		}
+	};
+
+	struct not_equal_to
+	{
+		template <typename T> bool operator()(const T& lhs, const T& rhs) const
+		{
+			return lhs != rhs;
+		}
+	};
+
+	struct less
+	{
+		template <typename T> bool operator()(const T& lhs, const T& rhs) const
+		{
+			return lhs < rhs;
+		}
+	};
+
+	struct less_equal
+	{
+		template <typename T> bool operator()(const T& lhs, const T& rhs) const
+		{
+			return lhs <= rhs;
+		}
+	};
+
+	template <typename T> void swap(T& lhs, T& rhs)
+	{
+		T temp = lhs;
+		lhs = rhs;
+		rhs = temp;
+	}
+
+	template <typename I, typename Pred> I min_element(I begin, I end, const Pred& pred)
+	{
+		I result = begin;
+
+		for (I it = begin + 1; it != end; ++it)
+			if (pred(*it, *result))
+				result = it;
+
+		return result;
+	}
+
+	template <typename I> void reverse(I begin, I end)
+	{
+		while (end - begin > 1) swap(*begin++, *--end);
+	}
+
+	template <typename I> I unique(I begin, I end)
+	{
+		// fast skip head
+		while (end - begin > 1 && *begin != *(begin + 1)) begin++;
+
+		if (begin == end) return begin;
+
+		// last written element
+		I write = begin++; 
+
+		// merge unique elements
+		while (begin != end)
+		{
+			if (*begin != *write)
+				*++write = *begin++;
+			else
+				begin++;
+		}
+
+		// past-the-end (write points to live element)
+		return write + 1;
+	}
+
+	template <typename I> void copy_backwards(I begin, I end, I target)
+	{
+		while (begin != end) *--target = *--end;
+	}
+
+	template <typename I, typename Pred, typename T> void insertion_sort(I begin, I end, const Pred& pred, T*)
+	{
+		assert(begin != end);
+
+		for (I it = begin + 1; it != end; ++it)
+		{
+			T val = *it;
+
+			if (pred(val, *begin))
+			{
+				// move to front
+				copy_backwards(begin, it, it + 1);
+				*begin = val;
+			}
+			else
+			{
+				I hole = it;
+
+				// move hole backwards
+				while (pred(val, *(hole - 1)))
+				{
+					*hole = *(hole - 1);
+					hole--;
+				}
+
+				// fill hole with element
+				*hole = val;
+			}
+		}
+	}
+
+	// std variant for elements with ==
+	template <typename I, typename Pred> void partition(I begin, I middle, I end, const Pred& pred, I* out_eqbeg, I* out_eqend)
+	{
+		I eqbeg = middle, eqend = middle + 1;
+
+		// expand equal range
+		while (eqbeg != begin && *(eqbeg - 1) == *eqbeg) --eqbeg;
+		while (eqend != end && *eqend == *eqbeg) ++eqend;
+
+		// process outer elements
+		I ltend = eqbeg, gtbeg = eqend;
+
+		for (;;)
+		{
+			// find the element from the right side that belongs to the left one
+			for (; gtbeg != end; ++gtbeg)
+				if (!pred(*eqbeg, *gtbeg))
+				{
+					if (*gtbeg == *eqbeg) swap(*gtbeg, *eqend++);
+					else break;
+				}
+
+			// find the element from the left side that belongs to the right one
+			for (; ltend != begin; --ltend)
+				if (!pred(*(ltend - 1), *eqbeg))
+				{
+					if (*eqbeg == *(ltend - 1)) swap(*(ltend - 1), *--eqbeg);
+					else break;
+				}
+
+			// scanned all elements
+			if (gtbeg == end && ltend == begin)
+			{
+				*out_eqbeg = eqbeg;
+				*out_eqend = eqend;
+				return;
+			}
+
+			// make room for elements by moving equal area
+			if (gtbeg == end)
+			{
+				if (--ltend != --eqbeg) swap(*ltend, *eqbeg);
+				swap(*eqbeg, *--eqend);
+			}
+			else if (ltend == begin)
+			{
+				if (eqend != gtbeg) swap(*eqbeg, *eqend);
+				++eqend;
+				swap(*gtbeg++, *eqbeg++);
+			}
+			else swap(*gtbeg++, *--ltend);
+		}
+	}
+
+	template <typename I, typename Pred> void median3(I first, I middle, I last, const Pred& pred)
+	{
+		if (pred(*middle, *first)) swap(*middle, *first);
+		if (pred(*last, *middle)) swap(*last, *middle);
+		if (pred(*middle, *first)) swap(*middle, *first);
+	}
+
+	template <typename I, typename Pred> void median(I first, I middle, I last, const Pred& pred)
+	{
+		if (last - first <= 40)
+		{
+			// median of three for small chunks
+			median3(first, middle, last, pred);
+		}
+		else
+		{
+			// median of nine
+			size_t step = (last - first + 1) / 8;
+
+			median3(first, first + step, first + 2 * step, pred);
+			median3(middle - step, middle, middle + step, pred);
+			median3(last - 2 * step, last - step, last, pred);
+			median3(first + step, middle, last - step, pred);
+		}
+	}
+
+	template <typename I, typename Pred> void sort(I begin, I end, const Pred& pred)
+	{
+		// sort large chunks
+		while (end - begin > 32)
+		{
+			// find median element
+			I middle = begin + (end - begin) / 2;
+			median(begin, middle, end - 1, pred);
+
+			// partition in three chunks (< = >)
+			I eqbeg, eqend;
+			partition(begin, middle, end, pred, &eqbeg, &eqend);
+
+			// loop on larger half
+			if (eqbeg - begin > end - eqend)
+			{
+				sort(eqend, end, pred);
+				end = eqbeg;
+			}
+			else
+			{
+				sort(begin, eqbeg, pred);
+				begin = eqend;
+			}
+		}
+
+		// insertion sort small chunk
+		if (begin != end) insertion_sort(begin, end, pred, &*begin);
+	}
+PUGI__NS_END
+
+// Allocator used for AST and evaluation stacks
+PUGI__NS_BEGIN
+	static const size_t xpath_memory_page_size =
+	#ifdef PUGIXML_MEMORY_XPATH_PAGE_SIZE
+		PUGIXML_MEMORY_XPATH_PAGE_SIZE
+	#else
+		4096
+	#endif
+		;
+
+	static const uintptr_t xpath_memory_block_alignment = sizeof(double) > sizeof(void*) ? sizeof(double) : sizeof(void*);
+
+	struct xpath_memory_block
+	{	
+		xpath_memory_block* next;
+		size_t capacity;
+
+		union
+		{
+			char data[xpath_memory_page_size];
+			double alignment;
+		};
+	};
+		
+	class xpath_allocator
+	{
+		xpath_memory_block* _root;
+		size_t _root_size;
+
+	public:
+	#ifdef PUGIXML_NO_EXCEPTIONS
+		jmp_buf* error_handler;
+	#endif
+
+		xpath_allocator(xpath_memory_block* root, size_t root_size = 0): _root(root), _root_size(root_size)
+		{
+		#ifdef PUGIXML_NO_EXCEPTIONS
+			error_handler = 0;
+		#endif
+		}
+		
+		void* allocate_nothrow(size_t size)
+		{
+			// round size up to block alignment boundary
+			size = (size + xpath_memory_block_alignment - 1) & ~(xpath_memory_block_alignment - 1);
+
+			if (_root_size + size <= _root->capacity)
+			{
+				void* buf = &_root->data[0] + _root_size;
+				_root_size += size;
+				return buf;
+			}
+			else
+			{
+				// make sure we have at least 1/4th of the page free after allocation to satisfy subsequent allocation requests
+				size_t block_capacity_base = sizeof(_root->data);
+				size_t block_capacity_req = size + block_capacity_base / 4;
+				size_t block_capacity = (block_capacity_base > block_capacity_req) ? block_capacity_base : block_capacity_req;
+
+				size_t block_size = block_capacity + offsetof(xpath_memory_block, data);
+
+				xpath_memory_block* block = static_cast<xpath_memory_block*>(xml_memory::allocate(block_size));
+				if (!block) return 0;
+				
+				block->next = _root;
+				block->capacity = block_capacity;
+				
+				_root = block;
+				_root_size = size;
+				
+				return block->data;
+			}
+		}
+
+		void* allocate(size_t size)
+		{
+			void* result = allocate_nothrow(size);
+
+			if (!result)
+			{
+			#ifdef PUGIXML_NO_EXCEPTIONS
+				assert(error_handler);
+				longjmp(*error_handler, 1);
+			#else
+				throw std::bad_alloc();
+			#endif
+			}
+
+			return result;
+		}
+
+		void* reallocate(void* ptr, size_t old_size, size_t new_size)
+		{
+			// round size up to block alignment boundary
+			old_size = (old_size + xpath_memory_block_alignment - 1) & ~(xpath_memory_block_alignment - 1);
+			new_size = (new_size + xpath_memory_block_alignment - 1) & ~(xpath_memory_block_alignment - 1);
+
+			// we can only reallocate the last object
+			assert(ptr == 0 || static_cast<char*>(ptr) + old_size == &_root->data[0] + _root_size);
+
+			// adjust root size so that we have not allocated the object at all
+			bool only_object = (_root_size == old_size);
+
+			if (ptr) _root_size -= old_size;
+
+			// allocate a new version (this will obviously reuse the memory if possible)
+			void* result = allocate(new_size);
+			assert(result);
+
+			// we have a new block
+			if (result != ptr && ptr)
+			{
+				// copy old data
+				assert(new_size >= old_size);
+				memcpy(result, ptr, old_size);
+
+				// free the previous page if it had no other objects
+				if (only_object)
+				{
+					assert(_root->data == result);
+					assert(_root->next);
+
+					xpath_memory_block* next = _root->next->next;
+
+					if (next)
+					{
+						// deallocate the whole page, unless it was the first one
+						xml_memory::deallocate(_root->next);
+						_root->next = next;
+					}
+				}
+			}
+
+			return result;
+		}
+
+		void revert(const xpath_allocator& state)
+		{
+			// free all new pages
+			xpath_memory_block* cur = _root;
+
+			while (cur != state._root)
+			{
+				xpath_memory_block* next = cur->next;
+
+				xml_memory::deallocate(cur);
+
+				cur = next;
+			}
+
+			// restore state
+			_root = state._root;
+			_root_size = state._root_size;
+		}
+
+		void release()
+		{
+			xpath_memory_block* cur = _root;
+			assert(cur);
+
+			while (cur->next)
+			{
+				xpath_memory_block* next = cur->next;
+
+				xml_memory::deallocate(cur);
+
+				cur = next;
+			}
+		}
+	};
+
+	struct xpath_allocator_capture
+	{
+		xpath_allocator_capture(xpath_allocator* alloc): _target(alloc), _state(*alloc)
+		{
+		}
+
+		~xpath_allocator_capture()
+		{
+			_target->revert(_state);
+		}
+
+		xpath_allocator* _target;
+		xpath_allocator _state;
+	};
+
+	struct xpath_stack
+	{
+		xpath_allocator* result;
+		xpath_allocator* temp;
+	};
+
+	struct xpath_stack_data
+	{
+		xpath_memory_block blocks[2];
+		xpath_allocator result;
+		xpath_allocator temp;
+		xpath_stack stack;
+
+	#ifdef PUGIXML_NO_EXCEPTIONS
+		jmp_buf error_handler;
+	#endif
+
+		xpath_stack_data(): result(blocks + 0), temp(blocks + 1)
+		{
+			blocks[0].next = blocks[1].next = 0;
+			blocks[0].capacity = blocks[1].capacity = sizeof(blocks[0].data);
+
+			stack.result = &result;
+			stack.temp = &temp;
+
+		#ifdef PUGIXML_NO_EXCEPTIONS
+			result.error_handler = temp.error_handler = &error_handler;
+		#endif
+		}
+
+		~xpath_stack_data()
+		{
+			result.release();
+			temp.release();
+		}
+	};
+PUGI__NS_END
+
+// String class
+PUGI__NS_BEGIN
+	class xpath_string
+	{
+		const char_t* _buffer;
+		bool _uses_heap;
+		size_t _length_heap;
+
+		static char_t* duplicate_string(const char_t* string, size_t length, xpath_allocator* alloc)
+		{
+			char_t* result = static_cast<char_t*>(alloc->allocate((length + 1) * sizeof(char_t)));
+			assert(result);
+
+			memcpy(result, string, length * sizeof(char_t));
+			result[length] = 0;
+
+			return result;
+		}
+
+		xpath_string(const char_t* buffer, bool uses_heap_, size_t length_heap): _buffer(buffer), _uses_heap(uses_heap_), _length_heap(length_heap)
+		{
+		}
+
+	public:
+		static xpath_string from_const(const char_t* str)
+		{
+			return xpath_string(str, false, 0);
+		}
+
+		static xpath_string from_heap_preallocated(const char_t* begin, const char_t* end)
+		{
+			assert(begin <= end && *end == 0);
+
+			return xpath_string(begin, true, static_cast<size_t>(end - begin));
+		}
+
+		static xpath_string from_heap(const char_t* begin, const char_t* end, xpath_allocator* alloc)
+		{
+			assert(begin <= end);
+
+			size_t length = static_cast<size_t>(end - begin);
+
+			return length == 0 ? xpath_string() : xpath_string(duplicate_string(begin, length, alloc), true, length);
+		}
+
+		xpath_string(): _buffer(PUGIXML_TEXT("")), _uses_heap(false), _length_heap(0)
+		{
+		}
+
+		void append(const xpath_string& o, xpath_allocator* alloc)
+		{
+			// skip empty sources
+			if (!*o._buffer) return;
+
+			// fast append for constant empty target and constant source
+			if (!*_buffer && !_uses_heap && !o._uses_heap)
+			{
+				_buffer = o._buffer;
+			}
+			else
+			{
+				// need to make heap copy
+				size_t target_length = length();
+				size_t source_length = o.length();
+				size_t result_length = target_length + source_length;
+
+				// allocate new buffer
+				char_t* result = static_cast<char_t*>(alloc->reallocate(_uses_heap ? const_cast<char_t*>(_buffer) : 0, (target_length + 1) * sizeof(char_t), (result_length + 1) * sizeof(char_t)));
+				assert(result);
+
+				// append first string to the new buffer in case there was no reallocation
+				if (!_uses_heap) memcpy(result, _buffer, target_length * sizeof(char_t));
+
+				// append second string to the new buffer
+				memcpy(result + target_length, o._buffer, source_length * sizeof(char_t));
+				result[result_length] = 0;
+
+				// finalize
+				_buffer = result;
+				_uses_heap = true;
+				_length_heap = result_length;
+			}
+		}
+
+		const char_t* c_str() const
+		{
+			return _buffer;
+		}
+
+		size_t length() const
+		{
+			return _uses_heap ? _length_heap : strlength(_buffer);
+		}
+		
+		char_t* data(xpath_allocator* alloc)
+		{
+			// make private heap copy
+			if (!_uses_heap)
+			{
+				size_t length_ = strlength(_buffer);
+
+				_buffer = duplicate_string(_buffer, length_, alloc);
+				_uses_heap = true;
+				_length_heap = length_;
+			}
+
+			return const_cast<char_t*>(_buffer);
+		}
+
+		bool empty() const
+		{
+			return *_buffer == 0;
+		}
+
+		bool operator==(const xpath_string& o) const
+		{
+			return strequal(_buffer, o._buffer);
+		}
+
+		bool operator!=(const xpath_string& o) const
+		{
+			return !strequal(_buffer, o._buffer);
+		}
+
+		bool uses_heap() const
+		{
+			return _uses_heap;
+		}
+	};
+PUGI__NS_END
+
+PUGI__NS_BEGIN
+	PUGI__FN bool starts_with(const char_t* string, const char_t* pattern)
+	{
+		while (*pattern && *string == *pattern)
+		{
+			string++;
+			pattern++;
+		}
+
+		return *pattern == 0;
+	}
+
+	PUGI__FN const char_t* find_char(const char_t* s, char_t c)
+	{
+	#ifdef PUGIXML_WCHAR_MODE
+		return wcschr(s, c);
+	#else
+		return strchr(s, c);
+	#endif
+	}
+
+	PUGI__FN const char_t* find_substring(const char_t* s, const char_t* p)
+	{
+	#ifdef PUGIXML_WCHAR_MODE
+		// MSVC6 wcsstr bug workaround (if s is empty it always returns 0)
+		return (*p == 0) ? s : wcsstr(s, p);
+	#else
+		return strstr(s, p);
+	#endif
+	}
+
+	// Converts symbol to lower case, if it is an ASCII one
+	PUGI__FN char_t tolower_ascii(char_t ch)
+	{
+		return static_cast<unsigned int>(ch - 'A') < 26 ? static_cast<char_t>(ch | ' ') : ch;
+	}
+
+	PUGI__FN xpath_string string_value(const xpath_node& na, xpath_allocator* alloc)
+	{
+		if (na.attribute())
+			return xpath_string::from_const(na.attribute().value());
+		else
+		{
+			xml_node n = na.node();
+
+			switch (n.type())
+			{
+			case node_pcdata:
+			case node_cdata:
+			case node_comment:
+			case node_pi:
+				return xpath_string::from_const(n.value());
+			
+			case node_document:
+			case node_element:
+			{
+				xpath_string result;
+
+				xml_node cur = n.first_child();
+				
+				while (cur && cur != n)
+				{
+					if (cur.type() == node_pcdata || cur.type() == node_cdata)
+						result.append(xpath_string::from_const(cur.value()), alloc);
+
+					if (cur.first_child())
+						cur = cur.first_child();
+					else if (cur.next_sibling())
+						cur = cur.next_sibling();
+					else
+					{
+						while (!cur.next_sibling() && cur != n)
+							cur = cur.parent();
+
+						if (cur != n) cur = cur.next_sibling();
+					}
+				}
+				
+				return result;
+			}
+			
+			default:
+				return xpath_string();
+			}
+		}
+	}
+	
+	PUGI__FN bool node_is_before_sibling(xml_node_struct* ln, xml_node_struct* rn)
+	{
+		assert(ln->parent == rn->parent);
+
+		// there is no common ancestor (the shared parent is null), nodes are from different documents
+		if (!ln->parent) return ln < rn;
+
+		// determine sibling order
+		xml_node_struct* ls = ln;
+		xml_node_struct* rs = rn;
+
+		while (ls && rs)
+		{
+			if (ls == rn) return true;
+			if (rs == ln) return false;
+
+			ls = ls->next_sibling;
+			rs = rs->next_sibling;
+		}
+
+		// if rn sibling chain ended ln must be before rn
+		return !rs;
+	}
+	
+	PUGI__FN bool node_is_before(xml_node_struct* ln, xml_node_struct* rn)
+	{
+		// find common ancestor at the same depth, if any
+		xml_node_struct* lp = ln;
+		xml_node_struct* rp = rn;
+
+		while (lp && rp && lp->parent != rp->parent)
+		{
+			lp = lp->parent;
+			rp = rp->parent;
+		}
+
+		// parents are the same!
+		if (lp && rp) return node_is_before_sibling(lp, rp);
+
+		// nodes are at different depths, need to normalize heights
+		bool left_higher = !lp;
+
+		while (lp)
+		{
+			lp = lp->parent;
+			ln = ln->parent;
+		}
+
+		while (rp)
+		{
+			rp = rp->parent;
+			rn = rn->parent;
+		}
+
+		// one node is the ancestor of the other
+		if (ln == rn) return left_higher;
+
+		// find common ancestor... again
+		while (ln->parent != rn->parent)
+		{
+			ln = ln->parent;
+			rn = rn->parent;
+		}
+
+		return node_is_before_sibling(ln, rn);
+	}
+
+	PUGI__FN bool node_is_ancestor(xml_node_struct* parent, xml_node_struct* node)
+	{
+		while (node && node != parent) node = node->parent;
+
+		return parent && node == parent;
+	}
+
+	PUGI__FN const void* document_buffer_order(const xpath_node& xnode)
+	{
+		xml_node_struct* node = xnode.node().internal_object();
+
+		if (node)
+		{
+			if ((get_document(node).header & xml_memory_page_contents_shared_mask) == 0)
+			{
+				if (node->name && (node->header & impl::xml_memory_page_name_allocated_or_shared_mask) == 0) return node->name;
+				if (node->value && (node->header & impl::xml_memory_page_value_allocated_or_shared_mask) == 0) return node->value;
+			}
+
+			return 0;
+		}
+
+		xml_attribute_struct* attr = xnode.attribute().internal_object();
+
+		if (attr)
+		{
+			if ((get_document(attr).header & xml_memory_page_contents_shared_mask) == 0)
+			{
+				if ((attr->header & impl::xml_memory_page_name_allocated_or_shared_mask) == 0) return attr->name;
+				if ((attr->header & impl::xml_memory_page_value_allocated_or_shared_mask) == 0) return attr->value;
+			}
+
+			return 0;
+		}
+
+		return 0;
+	}
+	
+	struct document_order_comparator
+	{
+		bool operator()(const xpath_node& lhs, const xpath_node& rhs) const
+		{
+			// optimized document order based check
+			const void* lo = document_buffer_order(lhs);
+			const void* ro = document_buffer_order(rhs);
+
+			if (lo && ro) return lo < ro;
+
+			// slow comparison
+			xml_node ln = lhs.node(), rn = rhs.node();
+
+			// compare attributes
+			if (lhs.attribute() && rhs.attribute())
+			{
+				// shared parent
+				if (lhs.parent() == rhs.parent())
+				{
+					// determine sibling order
+					for (xml_attribute a = lhs.attribute(); a; a = a.next_attribute())
+						if (a == rhs.attribute())
+							return true;
+					
+					return false;
+				}
+				
+				// compare attribute parents
+				ln = lhs.parent();
+				rn = rhs.parent();
+			}
+			else if (lhs.attribute())
+			{
+				// attributes go after the parent element
+				if (lhs.parent() == rhs.node()) return false;
+				
+				ln = lhs.parent();
+			}
+			else if (rhs.attribute())
+			{
+				// attributes go after the parent element
+				if (rhs.parent() == lhs.node()) return true;
+				
+				rn = rhs.parent();
+			}
+
+			if (ln == rn) return false;
+
+			if (!ln || !rn) return ln < rn;
+			
+			return node_is_before(ln.internal_object(), rn.internal_object());
+		}
+	};
+
+	struct duplicate_comparator
+	{
+		bool operator()(const xpath_node& lhs, const xpath_node& rhs) const
+		{
+			if (lhs.attribute()) return rhs.attribute() ? lhs.attribute() < rhs.attribute() : true;
+			else return rhs.attribute() ? false : lhs.node() < rhs.node();
+		}
+	};
+	
+	PUGI__FN double gen_nan()
+	{
+	#if defined(__STDC_IEC_559__) || ((FLT_RADIX - 0 == 2) && (FLT_MAX_EXP - 0 == 128) && (FLT_MANT_DIG - 0 == 24))
+		union { float f; uint32_t i; } u[sizeof(float) == sizeof(uint32_t) ? 1 : -1];
+		u[0].i = 0x7fc00000;
+		return u[0].f;
+	#else
+		// fallback
+		const volatile double zero = 0.0;
+		return zero / zero;
+	#endif
+	}
+	
+	PUGI__FN bool is_nan(double value)
+	{
+	#if defined(PUGI__MSVC_CRT_VERSION) || defined(__BORLANDC__)
+		return !!_isnan(value);
+	#elif defined(fpclassify) && defined(FP_NAN)
+		return fpclassify(value) == FP_NAN;
+	#else
+		// fallback
+		const volatile double v = value;
+		return v != v;
+	#endif
+	}
+	
+	PUGI__FN const char_t* convert_number_to_string_special(double value)
+	{
+	#if defined(PUGI__MSVC_CRT_VERSION) || defined(__BORLANDC__)
+		if (_finite(value)) return (value == 0) ? PUGIXML_TEXT("0") : 0;
+		if (_isnan(value)) return PUGIXML_TEXT("NaN");
+		return value > 0 ? PUGIXML_TEXT("Infinity") : PUGIXML_TEXT("-Infinity");
+	#elif defined(fpclassify) && defined(FP_NAN) && defined(FP_INFINITE) && defined(FP_ZERO)
+		switch (fpclassify(value))
+		{
+		case FP_NAN:
+			return PUGIXML_TEXT("NaN");
+
+		case FP_INFINITE:
+			return value > 0 ? PUGIXML_TEXT("Infinity") : PUGIXML_TEXT("-Infinity");
+
+		case FP_ZERO:
+			return PUGIXML_TEXT("0");
+
+		default:
+			return 0;
+		}
+	#else
+		// fallback
+		const volatile double v = value;
+
+		if (v == 0) return PUGIXML_TEXT("0");
+		if (v != v) return PUGIXML_TEXT("NaN");
+		if (v * 2 == v) return value > 0 ? PUGIXML_TEXT("Infinity") : PUGIXML_TEXT("-Infinity");
+		return 0;
+	#endif
+	}
+	
+	PUGI__FN bool convert_number_to_boolean(double value)
+	{
+		return (value != 0 && !is_nan(value));
+	}
+	
+	PUGI__FN void truncate_zeros(char* begin, char* end)
+	{
+		while (begin != end && end[-1] == '0') end--;
+
+		*end = 0;
+	}
+
+	// gets mantissa digits in the form of 0.xxxxx with 0. implied and the exponent
+#if defined(PUGI__MSVC_CRT_VERSION) && PUGI__MSVC_CRT_VERSION >= 1400 && !defined(_WIN32_WCE)
+	PUGI__FN void convert_number_to_mantissa_exponent(double value, char* buffer, size_t buffer_size, char** out_mantissa, int* out_exponent)
+	{
+		// get base values
+		int sign, exponent;
+		_ecvt_s(buffer, buffer_size, value, DBL_DIG + 1, &exponent, &sign);
+
+		// truncate redundant zeros
+		truncate_zeros(buffer, buffer + strlen(buffer));
+
+		// fill results
+		*out_mantissa = buffer;
+		*out_exponent = exponent;
+	}
+#else
+	PUGI__FN void convert_number_to_mantissa_exponent(double value, char* buffer, size_t buffer_size, char** out_mantissa, int* out_exponent)
+	{
+		// get a scientific notation value with IEEE DBL_DIG decimals
+		sprintf(buffer, "%.*e", DBL_DIG, value);
+		assert(strlen(buffer) < buffer_size);
+		(void)!buffer_size;
+
+		// get the exponent (possibly negative)
+		char* exponent_string = strchr(buffer, 'e');
+		assert(exponent_string);
+
+		int exponent = atoi(exponent_string + 1);
+
+		// extract mantissa string: skip sign
+		char* mantissa = buffer[0] == '-' ? buffer + 1 : buffer;
+		assert(mantissa[0] != '0' && mantissa[1] == '.');
+
+		// divide mantissa by 10 to eliminate integer part
+		mantissa[1] = mantissa[0];
+		mantissa++;
+		exponent++;
+
+		// remove extra mantissa digits and zero-terminate mantissa
+		truncate_zeros(mantissa, exponent_string);
+
+		// fill results
+		*out_mantissa = mantissa;
+		*out_exponent = exponent;
+	}
+#endif
+
+	PUGI__FN xpath_string convert_number_to_string(double value, xpath_allocator* alloc)
+	{
+		// try special number conversion
+		const char_t* special = convert_number_to_string_special(value);
+		if (special) return xpath_string::from_const(special);
+
+		// get mantissa + exponent form
+		char mantissa_buffer[32];
+
+		char* mantissa;
+		int exponent;
+		convert_number_to_mantissa_exponent(value, mantissa_buffer, sizeof(mantissa_buffer), &mantissa, &exponent);
+
+		// allocate a buffer of suitable length for the number
+		size_t result_size = strlen(mantissa_buffer) + (exponent > 0 ? exponent : -exponent) + 4;
+		char_t* result = static_cast<char_t*>(alloc->allocate(sizeof(char_t) * result_size));
+		assert(result);
+
+		// make the number!
+		char_t* s = result;
+
+		// sign
+		if (value < 0) *s++ = '-';
+
+		// integer part
+		if (exponent <= 0)
+		{
+			*s++ = '0';
+		}
+		else
+		{
+			while (exponent > 0)
+			{
+				assert(*mantissa == 0 || static_cast<unsigned int>(static_cast<unsigned int>(*mantissa) - '0') <= 9);
+				*s++ = *mantissa ? *mantissa++ : '0';
+				exponent--;
+			}
+		}
+
+		// fractional part
+		if (*mantissa)
+		{
+			// decimal point
+			*s++ = '.';
+
+			// extra zeroes from negative exponent
+			while (exponent < 0)
+			{
+				*s++ = '0';
+				exponent++;
+			}
+
+			// extra mantissa digits
+			while (*mantissa)
+			{
+				assert(static_cast<unsigned int>(*mantissa - '0') <= 9);
+				*s++ = *mantissa++;
+			}
+		}
+
+		// zero-terminate
+		assert(s < result + result_size);
+		*s = 0;
+
+		return xpath_string::from_heap_preallocated(result, s);
+	}
+	
+	PUGI__FN bool check_string_to_number_format(const char_t* string)
+	{
+		// parse leading whitespace
+		while (PUGI__IS_CHARTYPE(*string, ct_space)) ++string;
+
+		// parse sign
+		if (*string == '-') ++string;
+
+		if (!*string) return false;
+
+		// if there is no integer part, there should be a decimal part with at least one digit
+		if (!PUGI__IS_CHARTYPEX(string[0], ctx_digit) && (string[0] != '.' || !PUGI__IS_CHARTYPEX(string[1], ctx_digit))) return false;
+
+		// parse integer part
+		while (PUGI__IS_CHARTYPEX(*string, ctx_digit)) ++string;
+
+		// parse decimal part
+		if (*string == '.')
+		{
+			++string;
+
+			while (PUGI__IS_CHARTYPEX(*string, ctx_digit)) ++string;
+		}
+
+		// parse trailing whitespace
+		while (PUGI__IS_CHARTYPE(*string, ct_space)) ++string;
+
+		return *string == 0;
+	}
+
+	PUGI__FN double convert_string_to_number(const char_t* string)
+	{
+		// check string format
+		if (!check_string_to_number_format(string)) return gen_nan();
+
+		// parse string
+	#ifdef PUGIXML_WCHAR_MODE
+		return wcstod(string, 0);
+	#else
+		return strtod(string, 0);
+	#endif
+	}
+
+	PUGI__FN bool convert_string_to_number_scratch(char_t (&buffer)[32], const char_t* begin, const char_t* end, double* out_result)
+	{
+		size_t length = static_cast<size_t>(end - begin);
+		char_t* scratch = buffer;
+
+		if (length >= sizeof(buffer) / sizeof(buffer[0]))
+		{
+			// need to make dummy on-heap copy
+			scratch = static_cast<char_t*>(xml_memory::allocate((length + 1) * sizeof(char_t)));
+			if (!scratch) return false;
+		}
+
+		// copy string to zero-terminated buffer and perform conversion
+		memcpy(scratch, begin, length * sizeof(char_t));
+		scratch[length] = 0;
+
+		*out_result = convert_string_to_number(scratch);
+
+		// free dummy buffer
+		if (scratch != buffer) xml_memory::deallocate(scratch);
+
+		return true;
+	}
+	
+	PUGI__FN double round_nearest(double value)
+	{
+		return floor(value + 0.5);
+	}
+
+	PUGI__FN double round_nearest_nzero(double value)
+	{
+		// same as round_nearest, but returns -0 for [-0.5, -0]
+		// ceil is used to differentiate between +0 and -0 (we return -0 for [-0.5, -0] and +0 for +0)
+		return (value >= -0.5 && value <= 0) ? ceil(value) : floor(value + 0.5);
+	}
+	
+	PUGI__FN const char_t* qualified_name(const xpath_node& node)
+	{
+		return node.attribute() ? node.attribute().name() : node.node().name();
+	}
+	
+	PUGI__FN const char_t* local_name(const xpath_node& node)
+	{
+		const char_t* name = qualified_name(node);
+		const char_t* p = find_char(name, ':');
+		
+		return p ? p + 1 : name;
+	}
+
+	struct namespace_uri_predicate
+	{
+		const char_t* prefix;
+		size_t prefix_length;
+
+		namespace_uri_predicate(const char_t* name)
+		{
+			const char_t* pos = find_char(name, ':');
+
+			prefix = pos ? name : 0;
+			prefix_length = pos ? static_cast<size_t>(pos - name) : 0;
+		}
+
+		bool operator()(xml_attribute a) const
+		{
+			const char_t* name = a.name();
+
+			if (!starts_with(name, PUGIXML_TEXT("xmlns"))) return false;
+
+			return prefix ? name[5] == ':' && strequalrange(name + 6, prefix, prefix_length) : name[5] == 0;
+		}
+	};
+
+	PUGI__FN const char_t* namespace_uri(xml_node node)
+	{
+		namespace_uri_predicate pred = node.name();
+		
+		xml_node p = node;
+		
+		while (p)
+		{
+			xml_attribute a = p.find_attribute(pred);
+			
+			if (a) return a.value();
+			
+			p = p.parent();
+		}
+		
+		return PUGIXML_TEXT("");
+	}
+
+	PUGI__FN const char_t* namespace_uri(xml_attribute attr, xml_node parent)
+	{
+		namespace_uri_predicate pred = attr.name();
+		
+		// Default namespace does not apply to attributes
+		if (!pred.prefix) return PUGIXML_TEXT("");
+		
+		xml_node p = parent;
+		
+		while (p)
+		{
+			xml_attribute a = p.find_attribute(pred);
+			
+			if (a) return a.value();
+			
+			p = p.parent();
+		}
+		
+		return PUGIXML_TEXT("");
+	}
+
+	PUGI__FN const char_t* namespace_uri(const xpath_node& node)
+	{
+		return node.attribute() ? namespace_uri(node.attribute(), node.parent()) : namespace_uri(node.node());
+	}
+
+	PUGI__FN char_t* normalize_space(char_t* buffer)
+	{
+		char_t* write = buffer;
+
+		for (char_t* it = buffer; *it; )
+		{
+			char_t ch = *it++;
+
+			if (PUGI__IS_CHARTYPE(ch, ct_space))
+			{
+				// replace whitespace sequence with single space
+				while (PUGI__IS_CHARTYPE(*it, ct_space)) it++;
+
+				// avoid leading spaces
+				if (write != buffer) *write++ = ' ';
+			}
+			else *write++ = ch;
+		}
+
+		// remove trailing space
+		if (write != buffer && PUGI__IS_CHARTYPE(write[-1], ct_space)) write--;
+
+		// zero-terminate
+		*write = 0;
+
+		return write;
+	}
+
+	PUGI__FN char_t* translate(char_t* buffer, const char_t* from, const char_t* to, size_t to_length)
+	{
+		char_t* write = buffer;
+
+		while (*buffer)
+		{
+			PUGI__DMC_VOLATILE char_t ch = *buffer++;
+
+			const char_t* pos = find_char(from, ch);
+
+			if (!pos)
+				*write++ = ch; // do not process
+			else if (static_cast<size_t>(pos - from) < to_length)
+				*write++ = to[pos - from]; // replace
+		}
+
+		// zero-terminate
+		*write = 0;
+
+		return write;
+	}
+
+	PUGI__FN unsigned char* translate_table_generate(xpath_allocator* alloc, const char_t* from, const char_t* to)
+	{
+		unsigned char table[128] = {0};
+
+		while (*from)
+		{
+			unsigned int fc = static_cast<unsigned int>(*from);
+			unsigned int tc = static_cast<unsigned int>(*to);
+
+			if (fc >= 128 || tc >= 128)
+				return 0;
+
+			// code=128 means "skip character"
+			if (!table[fc])
+				table[fc] = static_cast<unsigned char>(tc ? tc : 128);
+
+			from++;
+			if (tc) to++;
+		}
+
+		for (int i = 0; i < 128; ++i)
+			if (!table[i])
+				table[i] = static_cast<unsigned char>(i);
+
+		void* result = alloc->allocate_nothrow(sizeof(table));
+
+		if (result)
+		{
+			memcpy(result, table, sizeof(table));
+		}
+
+		return static_cast<unsigned char*>(result);
+	}
+
+	PUGI__FN char_t* translate_table(char_t* buffer, const unsigned char* table)
+	{
+		char_t* write = buffer;
+
+		while (*buffer)
+		{
+			char_t ch = *buffer++;
+			unsigned int index = static_cast<unsigned int>(ch);
+
+			if (index < 128)
+			{
+				unsigned char code = table[index];
+
+				// code=128 means "skip character" (table size is 128 so 128 can be a special value)
+				// this code skips these characters without extra branches
+				*write = static_cast<char_t>(code);
+				write += 1 - (code >> 7);
+			}
+			else
+			{
+				*write++ = ch;
+			}
+		}
+
+		// zero-terminate
+		*write = 0;
+
+		return write;
+	}
+
+	inline bool is_xpath_attribute(const char_t* name)
+	{
+		return !(starts_with(name, PUGIXML_TEXT("xmlns")) && (name[5] == 0 || name[5] == ':'));
+	}
+
+	struct xpath_variable_boolean: xpath_variable
+	{
+		xpath_variable_boolean(): xpath_variable(xpath_type_boolean), value(false)
+		{
+		}
+
+		bool value;
+		char_t name[1];
+	};
+
+	struct xpath_variable_number: xpath_variable
+	{
+		xpath_variable_number(): xpath_variable(xpath_type_number), value(0)
+		{
+		}
+
+		double value;
+		char_t name[1];
+	};
+
+	struct xpath_variable_string: xpath_variable
+	{
+		xpath_variable_string(): xpath_variable(xpath_type_string), value(0)
+		{
+		}
+
+		~xpath_variable_string()
+		{
+			if (value) xml_memory::deallocate(value);
+		}
+
+		char_t* value;
+		char_t name[1];
+	};
+
+	struct xpath_variable_node_set: xpath_variable
+	{
+		xpath_variable_node_set(): xpath_variable(xpath_type_node_set)
+		{
+		}
+
+		xpath_node_set value;
+		char_t name[1];
+	};
+
+	static const xpath_node_set dummy_node_set;
+
+	PUGI__FN unsigned int hash_string(const char_t* str)
+	{
+		// Jenkins one-at-a-time hash (http://en.wikipedia.org/wiki/Jenkins_hash_function#one-at-a-time)
+		unsigned int result = 0;
+
+		while (*str)
+		{
+			result += static_cast<unsigned int>(*str++);
+			result += result << 10;
+			result ^= result >> 6;
+		}
+	
+		result += result << 3;
+		result ^= result >> 11;
+		result += result << 15;
+	
+		return result;
+	}
+
+	template <typename T> PUGI__FN T* new_xpath_variable(const char_t* name)
+	{
+		size_t length = strlength(name);
+		if (length == 0) return 0; // empty variable names are invalid
+
+		// $$ we can't use offsetof(T, name) because T is non-POD, so we just allocate additional length characters
+		void* memory = xml_memory::allocate(sizeof(T) + length * sizeof(char_t));
+		if (!memory) return 0;
+
+		T* result = new (memory) T();
+
+		memcpy(result->name, name, (length + 1) * sizeof(char_t));
+
+		return result;
+	}
+
+	PUGI__FN xpath_variable* new_xpath_variable(xpath_value_type type, const char_t* name)
+	{
+		switch (type)
+		{
+		case xpath_type_node_set:
+			return new_xpath_variable<xpath_variable_node_set>(name);
+
+		case xpath_type_number:
+			return new_xpath_variable<xpath_variable_number>(name);
+
+		case xpath_type_string:
+			return new_xpath_variable<xpath_variable_string>(name);
+
+		case xpath_type_boolean:
+			return new_xpath_variable<xpath_variable_boolean>(name);
+
+		default:
+			return 0;
+		}
+	}
+
+	template <typename T> PUGI__FN void delete_xpath_variable(T* var)
+	{
+		var->~T();
+		xml_memory::deallocate(var);
+	}
+
+	PUGI__FN void delete_xpath_variable(xpath_value_type type, xpath_variable* var)
+	{
+		switch (type)
+		{
+		case xpath_type_node_set:
+			delete_xpath_variable(static_cast<xpath_variable_node_set*>(var));
+			break;
+
+		case xpath_type_number:
+			delete_xpath_variable(static_cast<xpath_variable_number*>(var));
+			break;
+
+		case xpath_type_string:
+			delete_xpath_variable(static_cast<xpath_variable_string*>(var));
+			break;
+
+		case xpath_type_boolean:
+			delete_xpath_variable(static_cast<xpath_variable_boolean*>(var));
+			break;
+
+		default:
+			assert(!"Invalid variable type");
+		}
+	}
+
+	PUGI__FN bool copy_xpath_variable(xpath_variable* lhs, const xpath_variable* rhs)
+	{
+		switch (rhs->type())
+		{
+		case xpath_type_node_set:
+			return lhs->set(static_cast<const xpath_variable_node_set*>(rhs)->value);
+
+		case xpath_type_number:
+			return lhs->set(static_cast<const xpath_variable_number*>(rhs)->value);
+
+		case xpath_type_string:
+			return lhs->set(static_cast<const xpath_variable_string*>(rhs)->value);
+
+		case xpath_type_boolean:
+			return lhs->set(static_cast<const xpath_variable_boolean*>(rhs)->value);
+
+		default:
+			assert(!"Invalid variable type");
+			return false;
+		}
+	}
+
+	PUGI__FN bool get_variable_scratch(char_t (&buffer)[32], xpath_variable_set* set, const char_t* begin, const char_t* end, xpath_variable** out_result)
+	{
+		size_t length = static_cast<size_t>(end - begin);
+		char_t* scratch = buffer;
+
+		if (length >= sizeof(buffer) / sizeof(buffer[0]))
+		{
+			// need to make dummy on-heap copy
+			scratch = static_cast<char_t*>(xml_memory::allocate((length + 1) * sizeof(char_t)));
+			if (!scratch) return false;
+		}
+
+		// copy string to zero-terminated buffer and perform lookup
+		memcpy(scratch, begin, length * sizeof(char_t));
+		scratch[length] = 0;
+
+		*out_result = set->get(scratch);
+
+		// free dummy buffer
+		if (scratch != buffer) xml_memory::deallocate(scratch);
+
+		return true;
+	}
+PUGI__NS_END
+
+// Internal node set class
+PUGI__NS_BEGIN
+	PUGI__FN xpath_node_set::type_t xpath_get_order(const xpath_node* begin, const xpath_node* end)
+	{
+		if (end - begin < 2)
+			return xpath_node_set::type_sorted;
+
+		document_order_comparator cmp;
+
+		bool first = cmp(begin[0], begin[1]);
+
+		for (const xpath_node* it = begin + 1; it + 1 < end; ++it)
+			if (cmp(it[0], it[1]) != first)
+				return xpath_node_set::type_unsorted;
+
+		return first ? xpath_node_set::type_sorted : xpath_node_set::type_sorted_reverse;
+	}
+
+	PUGI__FN xpath_node_set::type_t xpath_sort(xpath_node* begin, xpath_node* end, xpath_node_set::type_t type, bool rev)
+	{
+		xpath_node_set::type_t order = rev ? xpath_node_set::type_sorted_reverse : xpath_node_set::type_sorted;
+
+		if (type == xpath_node_set::type_unsorted)
+		{
+			xpath_node_set::type_t sorted = xpath_get_order(begin, end);
+
+			if (sorted == xpath_node_set::type_unsorted)
+			{
+				sort(begin, end, document_order_comparator());
+
+				type = xpath_node_set::type_sorted;
+			}
+			else
+				type = sorted;
+		}
+		
+		if (type != order) reverse(begin, end);
+			
+		return order;
+	}
+
+	PUGI__FN xpath_node xpath_first(const xpath_node* begin, const xpath_node* end, xpath_node_set::type_t type)
+	{
+		if (begin == end) return xpath_node();
+
+		switch (type)
+		{
+		case xpath_node_set::type_sorted:
+			return *begin;
+
+		case xpath_node_set::type_sorted_reverse:
+			return *(end - 1);
+
+		case xpath_node_set::type_unsorted:
+			return *min_element(begin, end, document_order_comparator());
+
+		default:
+			assert(!"Invalid node set type");
+			return xpath_node();
+		}
+	}
+
+	class xpath_node_set_raw
+	{
+		xpath_node_set::type_t _type;
+
+		xpath_node* _begin;
+		xpath_node* _end;
+		xpath_node* _eos;
+
+	public:
+		xpath_node_set_raw(): _type(xpath_node_set::type_unsorted), _begin(0), _end(0), _eos(0)
+		{
+		}
+
+		xpath_node* begin() const
+		{
+			return _begin;
+		}
+
+		xpath_node* end() const
+		{
+			return _end;
+		}
+
+		bool empty() const
+		{
+			return _begin == _end;
+		}
+
+		size_t size() const
+		{
+			return static_cast<size_t>(_end - _begin);
+		}
+
+		xpath_node first() const
+		{
+			return xpath_first(_begin, _end, _type);
+		}
+
+		void push_back_grow(const xpath_node& node, xpath_allocator* alloc);
+
+		void push_back(const xpath_node& node, xpath_allocator* alloc)
+		{
+			if (_end != _eos)
+				*_end++ = node;
+			else
+				push_back_grow(node, alloc);
+		}
+
+		void append(const xpath_node* begin_, const xpath_node* end_, xpath_allocator* alloc)
+		{
+			if (begin_ == end_) return;
+
+			size_t size_ = static_cast<size_t>(_end - _begin);
+			size_t capacity = static_cast<size_t>(_eos - _begin);
+			size_t count = static_cast<size_t>(end_ - begin_);
+
+			if (size_ + count > capacity)
+			{
+				// reallocate the old array or allocate a new one
+				xpath_node* data = static_cast<xpath_node*>(alloc->reallocate(_begin, capacity * sizeof(xpath_node), (size_ + count) * sizeof(xpath_node)));
+				assert(data);
+
+				// finalize
+				_begin = data;
+				_end = data + size_;
+				_eos = data + size_ + count;
+			}
+
+			memcpy(_end, begin_, count * sizeof(xpath_node));
+			_end += count;
+		}
+
+		void sort_do()
+		{
+			_type = xpath_sort(_begin, _end, _type, false);
+		}
+
+		void truncate(xpath_node* pos)
+		{
+			assert(_begin <= pos && pos <= _end);
+
+			_end = pos;
+		}
+
+		void remove_duplicates()
+		{
+			if (_type == xpath_node_set::type_unsorted)
+				sort(_begin, _end, duplicate_comparator());
+		
+			_end = unique(_begin, _end);
+		}
+
+		xpath_node_set::type_t type() const
+		{
+			return _type;
+		}
+
+		void set_type(xpath_node_set::type_t value)
+		{
+			_type = value;
+		}
+	};
+
+	PUGI__FN_NO_INLINE void xpath_node_set_raw::push_back_grow(const xpath_node& node, xpath_allocator* alloc)
+	{
+		size_t capacity = static_cast<size_t>(_eos - _begin);
+
+		// get new capacity (1.5x rule)
+		size_t new_capacity = capacity + capacity / 2 + 1;
+
+		// reallocate the old array or allocate a new one
+		xpath_node* data = static_cast<xpath_node*>(alloc->reallocate(_begin, capacity * sizeof(xpath_node), new_capacity * sizeof(xpath_node)));
+		assert(data);
+
+		// finalize
+		_begin = data;
+		_end = data + capacity;
+		_eos = data + new_capacity;
+
+		// push
+		*_end++ = node;
+	}
+PUGI__NS_END
+
+PUGI__NS_BEGIN
+	struct xpath_context
+	{
+		xpath_node n;
+		size_t position, size;
+
+		xpath_context(const xpath_node& n_, size_t position_, size_t size_): n(n_), position(position_), size(size_)
+		{
+		}
+	};
+
+	enum lexeme_t
+	{
+		lex_none = 0,
+		lex_equal,
+		lex_not_equal,
+		lex_less,
+		lex_greater,
+		lex_less_or_equal,
+		lex_greater_or_equal,
+		lex_plus,
+		lex_minus,
+		lex_multiply,
+		lex_union,
+		lex_var_ref,
+		lex_open_brace,
+		lex_close_brace,
+		lex_quoted_string,
+		lex_number,
+		lex_slash,
+		lex_double_slash,
+		lex_open_square_brace,
+		lex_close_square_brace,
+		lex_string,
+		lex_comma,
+		lex_axis_attribute,
+		lex_dot,
+		lex_double_dot,
+		lex_double_colon,
+		lex_eof
+	};
+
+	struct xpath_lexer_string
+	{
+		const char_t* begin;
+		const char_t* end;
+
+		xpath_lexer_string(): begin(0), end(0)
+		{
+		}
+
+		bool operator==(const char_t* other) const
+		{
+			size_t length = static_cast<size_t>(end - begin);
+
+			return strequalrange(other, begin, length);
+		}
+	};
+
+	class xpath_lexer
+	{
+		const char_t* _cur;
+		const char_t* _cur_lexeme_pos;
+		xpath_lexer_string _cur_lexeme_contents;
+
+		lexeme_t _cur_lexeme;
+
+	public:
+		explicit xpath_lexer(const char_t* query): _cur(query)
+		{
+			next();
+		}
+		
+		const char_t* state() const
+		{
+			return _cur;
+		}
+		
+		void next()
+		{
+			const char_t* cur = _cur;
+
+			while (PUGI__IS_CHARTYPE(*cur, ct_space)) ++cur;
+
+			// save lexeme position for error reporting
+			_cur_lexeme_pos = cur;
+
+			switch (*cur)
+			{
+			case 0:
+				_cur_lexeme = lex_eof;
+				break;
+			
+			case '>':
+				if (*(cur+1) == '=')
+				{
+					cur += 2;
+					_cur_lexeme = lex_greater_or_equal;
+				}
+				else
+				{
+					cur += 1;
+					_cur_lexeme = lex_greater;
+				}
+				break;
+
+			case '<':
+				if (*(cur+1) == '=')
+				{
+					cur += 2;
+					_cur_lexeme = lex_less_or_equal;
+				}
+				else
+				{
+					cur += 1;
+					_cur_lexeme = lex_less;
+				}
+				break;
+
+			case '!':
+				if (*(cur+1) == '=')
+				{
+					cur += 2;
+					_cur_lexeme = lex_not_equal;
+				}
+				else
+				{
+					_cur_lexeme = lex_none;
+				}
+				break;
+
+			case '=':
+				cur += 1;
+				_cur_lexeme = lex_equal;
+
+				break;
+			
+			case '+':
+				cur += 1;
+				_cur_lexeme = lex_plus;
+
+				break;
+
+			case '-':
+				cur += 1;
+				_cur_lexeme = lex_minus;
+
+				break;
+
+			case '*':
+				cur += 1;
+				_cur_lexeme = lex_multiply;
+
+				break;
+
+			case '|':
+				cur += 1;
+				_cur_lexeme = lex_union;
+
+				break;
+			
+			case '$':
+				cur += 1;
+
+				if (PUGI__IS_CHARTYPEX(*cur, ctx_start_symbol))
+				{
+					_cur_lexeme_contents.begin = cur;
+
+					while (PUGI__IS_CHARTYPEX(*cur, ctx_symbol)) cur++;
+
+					if (cur[0] == ':' && PUGI__IS_CHARTYPEX(cur[1], ctx_symbol)) // qname
+					{
+						cur++; // :
+
+						while (PUGI__IS_CHARTYPEX(*cur, ctx_symbol)) cur++;
+					}
+
+					_cur_lexeme_contents.end = cur;
+				
+					_cur_lexeme = lex_var_ref;
+				}
+				else
+				{
+					_cur_lexeme = lex_none;
+				}
+
+				break;
+
+			case '(':
+				cur += 1;
+				_cur_lexeme = lex_open_brace;
+
+				break;
+
+			case ')':
+				cur += 1;
+				_cur_lexeme = lex_close_brace;
+
+				break;
+			
+			case '[':
+				cur += 1;
+				_cur_lexeme = lex_open_square_brace;
+
+				break;
+
+			case ']':
+				cur += 1;
+				_cur_lexeme = lex_close_square_brace;
+
+				break;
+
+			case ',':
+				cur += 1;
+				_cur_lexeme = lex_comma;
+
+				break;
+
+			case '/':
+				if (*(cur+1) == '/')
+				{
+					cur += 2;
+					_cur_lexeme = lex_double_slash;
+				}
+				else
+				{
+					cur += 1;
+					_cur_lexeme = lex_slash;
+				}
+				break;
+		
+			case '.':
+				if (*(cur+1) == '.')
+				{
+					cur += 2;
+					_cur_lexeme = lex_double_dot;
+				}
+				else if (PUGI__IS_CHARTYPEX(*(cur+1), ctx_digit))
+				{
+					_cur_lexeme_contents.begin = cur; // .
+
+					++cur;
+
+					while (PUGI__IS_CHARTYPEX(*cur, ctx_digit)) cur++;
+
+					_cur_lexeme_contents.end = cur;
+					
+					_cur_lexeme = lex_number;
+				}
+				else
+				{
+					cur += 1;
+					_cur_lexeme = lex_dot;
+				}
+				break;
+
+			case '@':
+				cur += 1;
+				_cur_lexeme = lex_axis_attribute;
+
+				break;
+
+			case '"':
+			case '\'':
+			{
+				char_t terminator = *cur;
+
+				++cur;
+
+				_cur_lexeme_contents.begin = cur;
+				while (*cur && *cur != terminator) cur++;
+				_cur_lexeme_contents.end = cur;
+				
+				if (!*cur)
+					_cur_lexeme = lex_none;
+				else
+				{
+					cur += 1;
+					_cur_lexeme = lex_quoted_string;
+				}
+
+				break;
+			}
+
+			case ':':
+				if (*(cur+1) == ':')
+				{
+					cur += 2;
+					_cur_lexeme = lex_double_colon;
+				}
+				else
+				{
+					_cur_lexeme = lex_none;
+				}
+				break;
+
+			default:
+				if (PUGI__IS_CHARTYPEX(*cur, ctx_digit))
+				{
+					_cur_lexeme_contents.begin = cur;
+
+					while (PUGI__IS_CHARTYPEX(*cur, ctx_digit)) cur++;
+				
+					if (*cur == '.')
+					{
+						cur++;
+
+						while (PUGI__IS_CHARTYPEX(*cur, ctx_digit)) cur++;
+					}
+
+					_cur_lexeme_contents.end = cur;
+
+					_cur_lexeme = lex_number;
+				}
+				else if (PUGI__IS_CHARTYPEX(*cur, ctx_start_symbol))
+				{
+					_cur_lexeme_contents.begin = cur;
+
+					while (PUGI__IS_CHARTYPEX(*cur, ctx_symbol)) cur++;
+
+					if (cur[0] == ':')
+					{
+						if (cur[1] == '*') // namespace test ncname:*
+						{
+							cur += 2; // :*
+						}
+						else if (PUGI__IS_CHARTYPEX(cur[1], ctx_symbol)) // namespace test qname
+						{
+							cur++; // :
+
+							while (PUGI__IS_CHARTYPEX(*cur, ctx_symbol)) cur++;
+						}
+					}
+
+					_cur_lexeme_contents.end = cur;
+				
+					_cur_lexeme = lex_string;
+				}
+				else
+				{
+					_cur_lexeme = lex_none;
+				}
+			}
+
+			_cur = cur;
+		}
+
+		lexeme_t current() const
+		{
+			return _cur_lexeme;
+		}
+
+		const char_t* current_pos() const
+		{
+			return _cur_lexeme_pos;
+		}
+
+		const xpath_lexer_string& contents() const
+		{
+			assert(_cur_lexeme == lex_var_ref || _cur_lexeme == lex_number || _cur_lexeme == lex_string || _cur_lexeme == lex_quoted_string);
+
+			return _cur_lexeme_contents;
+		}
+	};
+
+	enum ast_type_t
+	{
+		ast_unknown,
+		ast_op_or,						// left or right
+		ast_op_and,						// left and right
+		ast_op_equal,					// left = right
+		ast_op_not_equal,				// left != right
+		ast_op_less,					// left < right
+		ast_op_greater,					// left > right
+		ast_op_less_or_equal,			// left <= right
+		ast_op_greater_or_equal,		// left >= right
+		ast_op_add,						// left + right
+		ast_op_subtract,				// left - right
+		ast_op_multiply,				// left * right
+		ast_op_divide,					// left / right
+		ast_op_mod,						// left % right
+		ast_op_negate,					// left - right
+		ast_op_union,					// left | right
+		ast_predicate,					// apply predicate to set; next points to next predicate
+		ast_filter,						// select * from left where right
+		ast_string_constant,			// string constant
+		ast_number_constant,			// number constant
+		ast_variable,					// variable
+		ast_func_last,					// last()
+		ast_func_position,				// position()
+		ast_func_count,					// count(left)
+		ast_func_id,					// id(left)
+		ast_func_local_name_0,			// local-name()
+		ast_func_local_name_1,			// local-name(left)
+		ast_func_namespace_uri_0,		// namespace-uri()
+		ast_func_namespace_uri_1,		// namespace-uri(left)
+		ast_func_name_0,				// name()
+		ast_func_name_1,				// name(left)
+		ast_func_string_0,				// string()
+		ast_func_string_1,				// string(left)
+		ast_func_concat,				// concat(left, right, siblings)
+		ast_func_starts_with,			// starts_with(left, right)
+		ast_func_contains,				// contains(left, right)
+		ast_func_substring_before,		// substring-before(left, right)
+		ast_func_substring_after,		// substring-after(left, right)
+		ast_func_substring_2,			// substring(left, right)
+		ast_func_substring_3,			// substring(left, right, third)
+		ast_func_string_length_0,		// string-length()
+		ast_func_string_length_1,		// string-length(left)
+		ast_func_normalize_space_0,		// normalize-space()
+		ast_func_normalize_space_1,		// normalize-space(left)
+		ast_func_translate,				// translate(left, right, third)
+		ast_func_boolean,				// boolean(left)
+		ast_func_not,					// not(left)
+		ast_func_true,					// true()
+		ast_func_false,					// false()
+		ast_func_lang,					// lang(left)
+		ast_func_number_0,				// number()
+		ast_func_number_1,				// number(left)
+		ast_func_sum,					// sum(left)
+		ast_func_floor,					// floor(left)
+		ast_func_ceiling,				// ceiling(left)
+		ast_func_round,					// round(left)
+		ast_step,						// process set left with step
+		ast_step_root,					// select root node
+
+		ast_opt_translate_table,		// translate(left, right, third) where right/third are constants
+		ast_opt_compare_attribute		// @name = 'string'
+	};
+
+	enum axis_t
+	{
+		axis_ancestor,
+		axis_ancestor_or_self,
+		axis_attribute,
+		axis_child,
+		axis_descendant,
+		axis_descendant_or_self,
+		axis_following,
+		axis_following_sibling,
+		axis_namespace,
+		axis_parent,
+		axis_preceding,
+		axis_preceding_sibling,
+		axis_self
+	};
+	
+	enum nodetest_t
+	{
+		nodetest_none,
+		nodetest_name,
+		nodetest_type_node,
+		nodetest_type_comment,
+		nodetest_type_pi,
+		nodetest_type_text,
+		nodetest_pi,
+		nodetest_all,
+		nodetest_all_in_namespace
+	};
+
+	enum predicate_t
+	{
+		predicate_default,
+		predicate_posinv,
+		predicate_constant,
+		predicate_constant_one
+	};
+
+	enum nodeset_eval_t
+	{
+		nodeset_eval_all,
+		nodeset_eval_any,
+		nodeset_eval_first
+	};
+
+	template <axis_t N> struct axis_to_type
+	{
+		static const axis_t axis;
+	};
+
+	template <axis_t N> const axis_t axis_to_type<N>::axis = N;
+		
+	class xpath_ast_node
+	{
+	private:
+		// node type
+		char _type;
+		char _rettype;
+
+		// for ast_step
+		char _axis;
+
+		// for ast_step/ast_predicate/ast_filter
+		char _test;
+
+		// tree node structure
+		xpath_ast_node* _left;
+		xpath_ast_node* _right;
+		xpath_ast_node* _next;
+
+		union
+		{
+			// value for ast_string_constant
+			const char_t* string;
+			// value for ast_number_constant
+			double number;
+			// variable for ast_variable
+			xpath_variable* variable;
+			// node test for ast_step (node name/namespace/node type/pi target)
+			const char_t* nodetest;
+			// table for ast_opt_translate_table
+			const unsigned char* table;
+		} _data;
+
+		xpath_ast_node(const xpath_ast_node&);
+		xpath_ast_node& operator=(const xpath_ast_node&);
+
+		template <class Comp> static bool compare_eq(xpath_ast_node* lhs, xpath_ast_node* rhs, const xpath_context& c, const xpath_stack& stack, const Comp& comp)
+		{
+			xpath_value_type lt = lhs->rettype(), rt = rhs->rettype();
+
+			if (lt != xpath_type_node_set && rt != xpath_type_node_set)
+			{
+				if (lt == xpath_type_boolean || rt == xpath_type_boolean)
+					return comp(lhs->eval_boolean(c, stack), rhs->eval_boolean(c, stack));
+				else if (lt == xpath_type_number || rt == xpath_type_number)
+					return comp(lhs->eval_number(c, stack), rhs->eval_number(c, stack));
+				else if (lt == xpath_type_string || rt == xpath_type_string)
+				{
+					xpath_allocator_capture cr(stack.result);
+
+					xpath_string ls = lhs->eval_string(c, stack);
+					xpath_string rs = rhs->eval_string(c, stack);
+
+					return comp(ls, rs);
+				}
+			}
+			else if (lt == xpath_type_node_set && rt == xpath_type_node_set)
+			{
+				xpath_allocator_capture cr(stack.result);
+
+				xpath_node_set_raw ls = lhs->eval_node_set(c, stack, nodeset_eval_all);
+				xpath_node_set_raw rs = rhs->eval_node_set(c, stack, nodeset_eval_all);
+
+				for (const xpath_node* li = ls.begin(); li != ls.end(); ++li)
+					for (const xpath_node* ri = rs.begin(); ri != rs.end(); ++ri)
+					{
+						xpath_allocator_capture cri(stack.result);
+
+						if (comp(string_value(*li, stack.result), string_value(*ri, stack.result)))
+							return true;
+					}
+
+				return false;
+			}
+			else
+			{
+				if (lt == xpath_type_node_set)
+				{
+					swap(lhs, rhs);
+					swap(lt, rt);
+				}
+
+				if (lt == xpath_type_boolean)
+					return comp(lhs->eval_boolean(c, stack), rhs->eval_boolean(c, stack));
+				else if (lt == xpath_type_number)
+				{
+					xpath_allocator_capture cr(stack.result);
+
+					double l = lhs->eval_number(c, stack);
+					xpath_node_set_raw rs = rhs->eval_node_set(c, stack, nodeset_eval_all);
+
+					for (const xpath_node* ri = rs.begin(); ri != rs.end(); ++ri)
+					{
+						xpath_allocator_capture cri(stack.result);
+
+						if (comp(l, convert_string_to_number(string_value(*ri, stack.result).c_str())))
+							return true;
+					}
+
+					return false;
+				}
+				else if (lt == xpath_type_string)
+				{
+					xpath_allocator_capture cr(stack.result);
+
+					xpath_string l = lhs->eval_string(c, stack);
+					xpath_node_set_raw rs = rhs->eval_node_set(c, stack, nodeset_eval_all);
+
+					for (const xpath_node* ri = rs.begin(); ri != rs.end(); ++ri)
+					{
+						xpath_allocator_capture cri(stack.result);
+
+						if (comp(l, string_value(*ri, stack.result)))
+							return true;
+					}
+
+					return false;
+				}
+			}
+
+			assert(!"Wrong types");
+			return false;
+		}
+
+		static bool eval_once(xpath_node_set::type_t type, nodeset_eval_t eval)
+		{
+			return type == xpath_node_set::type_sorted ? eval != nodeset_eval_all : eval == nodeset_eval_any;
+		}
+
+		template <class Comp> static bool compare_rel(xpath_ast_node* lhs, xpath_ast_node* rhs, const xpath_context& c, const xpath_stack& stack, const Comp& comp)
+		{
+			xpath_value_type lt = lhs->rettype(), rt = rhs->rettype();
+
+			if (lt != xpath_type_node_set && rt != xpath_type_node_set)
+				return comp(lhs->eval_number(c, stack), rhs->eval_number(c, stack));
+			else if (lt == xpath_type_node_set && rt == xpath_type_node_set)
+			{
+				xpath_allocator_capture cr(stack.result);
+
+				xpath_node_set_raw ls = lhs->eval_node_set(c, stack, nodeset_eval_all);
+				xpath_node_set_raw rs = rhs->eval_node_set(c, stack, nodeset_eval_all);
+
+				for (const xpath_node* li = ls.begin(); li != ls.end(); ++li)
+				{
+					xpath_allocator_capture cri(stack.result);
+
+					double l = convert_string_to_number(string_value(*li, stack.result).c_str());
+
+					for (const xpath_node* ri = rs.begin(); ri != rs.end(); ++ri)
+					{
+						xpath_allocator_capture crii(stack.result);
+
+						if (comp(l, convert_string_to_number(string_value(*ri, stack.result).c_str())))
+							return true;
+					}
+				}
+
+				return false;
+			}
+			else if (lt != xpath_type_node_set && rt == xpath_type_node_set)
+			{
+				xpath_allocator_capture cr(stack.result);
+
+				double l = lhs->eval_number(c, stack);
+				xpath_node_set_raw rs = rhs->eval_node_set(c, stack, nodeset_eval_all);
+
+				for (const xpath_node* ri = rs.begin(); ri != rs.end(); ++ri)
+				{
+					xpath_allocator_capture cri(stack.result);
+
+					if (comp(l, convert_string_to_number(string_value(*ri, stack.result).c_str())))
+						return true;
+				}
+
+				return false;
+			}
+			else if (lt == xpath_type_node_set && rt != xpath_type_node_set)
+			{
+				xpath_allocator_capture cr(stack.result);
+
+				xpath_node_set_raw ls = lhs->eval_node_set(c, stack, nodeset_eval_all);
+				double r = rhs->eval_number(c, stack);
+
+				for (const xpath_node* li = ls.begin(); li != ls.end(); ++li)
+				{
+					xpath_allocator_capture cri(stack.result);
+
+					if (comp(convert_string_to_number(string_value(*li, stack.result).c_str()), r))
+						return true;
+				}
+
+				return false;
+			}
+			else
+			{
+				assert(!"Wrong types");
+				return false;
+			}
+		}
+
+		static void apply_predicate_boolean(xpath_node_set_raw& ns, size_t first, xpath_ast_node* expr, const xpath_stack& stack, bool once)
+		{
+			assert(ns.size() >= first);
+			assert(expr->rettype() != xpath_type_number);
+
+			size_t i = 1;
+			size_t size = ns.size() - first;
+
+			xpath_node* last = ns.begin() + first;
+
+			// remove_if... or well, sort of
+			for (xpath_node* it = last; it != ns.end(); ++it, ++i)
+			{
+				xpath_context c(*it, i, size);
+
+				if (expr->eval_boolean(c, stack))
+				{
+					*last++ = *it;
+
+					if (once) break;
+				}
+			}
+
+			ns.truncate(last);
+		}
+
+		static void apply_predicate_number(xpath_node_set_raw& ns, size_t first, xpath_ast_node* expr, const xpath_stack& stack, bool once)
+		{
+			assert(ns.size() >= first);
+			assert(expr->rettype() == xpath_type_number);
+
+			size_t i = 1;
+			size_t size = ns.size() - first;
+
+			xpath_node* last = ns.begin() + first;
+
+			// remove_if... or well, sort of
+			for (xpath_node* it = last; it != ns.end(); ++it, ++i)
+			{
+				xpath_context c(*it, i, size);
+
+				if (expr->eval_number(c, stack) == i)
+				{
+					*last++ = *it;
+
+					if (once) break;
+				}
+			}
+
+			ns.truncate(last);
+		}
+
+		static void apply_predicate_number_const(xpath_node_set_raw& ns, size_t first, xpath_ast_node* expr, const xpath_stack& stack)
+		{
+			assert(ns.size() >= first);
+			assert(expr->rettype() == xpath_type_number);
+
+			size_t size = ns.size() - first;
+
+			xpath_node* last = ns.begin() + first;
+
+			xpath_context c(xpath_node(), 1, size);
+
+			double er = expr->eval_number(c, stack);
+
+			if (er >= 1.0 && er <= size)
+			{
+				size_t eri = static_cast<size_t>(er);
+
+				if (er == eri)
+				{
+					xpath_node r = last[eri - 1];
+
+					*last++ = r;
+				}
+			}
+
+			ns.truncate(last);
+		}
+
+		void apply_predicate(xpath_node_set_raw& ns, size_t first, const xpath_stack& stack, bool once)
+		{
+			if (ns.size() == first) return;
+
+			assert(_type == ast_filter || _type == ast_predicate);
+
+			if (_test == predicate_constant || _test == predicate_constant_one)
+				apply_predicate_number_const(ns, first, _right, stack);
+			else if (_right->rettype() == xpath_type_number)
+				apply_predicate_number(ns, first, _right, stack, once);
+			else
+				apply_predicate_boolean(ns, first, _right, stack, once);
+		}
+
+		void apply_predicates(xpath_node_set_raw& ns, size_t first, const xpath_stack& stack, nodeset_eval_t eval)
+		{
+			if (ns.size() == first) return;
+
+			bool last_once = eval_once(ns.type(), eval);
+
+			for (xpath_ast_node* pred = _right; pred; pred = pred->_next)
+				pred->apply_predicate(ns, first, stack, !pred->_next && last_once);
+		}
+
+		bool step_push(xpath_node_set_raw& ns, xml_attribute_struct* a, xml_node_struct* parent, xpath_allocator* alloc)
+		{
+			assert(a);
+
+			const char_t* name = a->name ? a->name + 0 : PUGIXML_TEXT("");
+
+			switch (_test)
+			{
+			case nodetest_name:
+				if (strequal(name, _data.nodetest) && is_xpath_attribute(name))
+				{
+					ns.push_back(xpath_node(xml_attribute(a), xml_node(parent)), alloc);
+					return true;
+				}
+				break;
+				
+			case nodetest_type_node:
+			case nodetest_all:
+				if (is_xpath_attribute(name))
+				{
+					ns.push_back(xpath_node(xml_attribute(a), xml_node(parent)), alloc);
+					return true;
+				}
+				break;
+				
+			case nodetest_all_in_namespace:
+				if (starts_with(name, _data.nodetest) && is_xpath_attribute(name))
+				{
+					ns.push_back(xpath_node(xml_attribute(a), xml_node(parent)), alloc);
+					return true;
+				}
+				break;
+			
+			default:
+				;
+			}
+
+			return false;
+		}
+		
+		bool step_push(xpath_node_set_raw& ns, xml_node_struct* n, xpath_allocator* alloc)
+		{
+			assert(n);
+
+			xml_node_type type = PUGI__NODETYPE(n);
+
+			switch (_test)
+			{
+			case nodetest_name:
+				if (type == node_element && n->name && strequal(n->name, _data.nodetest))
+				{
+					ns.push_back(xml_node(n), alloc);
+					return true;
+				}
+				break;
+				
+			case nodetest_type_node:
+				ns.push_back(xml_node(n), alloc);
+				return true;
+				
+			case nodetest_type_comment:
+				if (type == node_comment)
+				{
+					ns.push_back(xml_node(n), alloc);
+					return true;
+				}
+				break;
+				
+			case nodetest_type_text:
+				if (type == node_pcdata || type == node_cdata)
+				{
+					ns.push_back(xml_node(n), alloc);
+					return true;
+				}
+				break;
+				
+			case nodetest_type_pi:
+				if (type == node_pi)
+				{
+					ns.push_back(xml_node(n), alloc);
+					return true;
+				}
+				break;
+									
+			case nodetest_pi:
+				if (type == node_pi && n->name && strequal(n->name, _data.nodetest))
+				{
+					ns.push_back(xml_node(n), alloc);
+					return true;
+				}
+				break;
+				
+			case nodetest_all:
+				if (type == node_element)
+				{
+					ns.push_back(xml_node(n), alloc);
+					return true;
+				}
+				break;
+				
+			case nodetest_all_in_namespace:
+				if (type == node_element && n->name && starts_with(n->name, _data.nodetest))
+				{
+					ns.push_back(xml_node(n), alloc);
+					return true;
+				}
+				break;
+
+			default:
+				assert(!"Unknown axis");
+			}
+
+			return false;
+		}
+
+		template <class T> void step_fill(xpath_node_set_raw& ns, xml_node_struct* n, xpath_allocator* alloc, bool once, T)
+		{
+			const axis_t axis = T::axis;
+
+			switch (axis)
+			{
+			case axis_attribute:
+			{
+				for (xml_attribute_struct* a = n->first_attribute; a; a = a->next_attribute)
+					if (step_push(ns, a, n, alloc) & once)
+						return;
+				
+				break;
+			}
+			
+			case axis_child:
+			{
+				for (xml_node_struct* c = n->first_child; c; c = c->next_sibling)
+					if (step_push(ns, c, alloc) & once)
+						return;
+					
+				break;
+			}
+			
+			case axis_descendant:
+			case axis_descendant_or_self:
+			{
+				if (axis == axis_descendant_or_self)
+					if (step_push(ns, n, alloc) & once)
+						return;
+					
+				xml_node_struct* cur = n->first_child;
+				
+				while (cur)
+				{
+					if (step_push(ns, cur, alloc) & once)
+						return;
+					
+					if (cur->first_child)
+						cur = cur->first_child;
+					else
+					{
+						while (!cur->next_sibling)
+						{
+							cur = cur->parent;
+
+							if (cur == n) return;
+						}
+					
+						cur = cur->next_sibling;
+					}
+				}
+				
+				break;
+			}
+			
+			case axis_following_sibling:
+			{
+				for (xml_node_struct* c = n->next_sibling; c; c = c->next_sibling)
+					if (step_push(ns, c, alloc) & once)
+						return;
+				
+				break;
+			}
+			
+			case axis_preceding_sibling:
+			{
+				for (xml_node_struct* c = n->prev_sibling_c; c->next_sibling; c = c->prev_sibling_c)
+					if (step_push(ns, c, alloc) & once)
+						return;
+				
+				break;
+			}
+			
+			case axis_following:
+			{
+				xml_node_struct* cur = n;
+
+				// exit from this node so that we don't include descendants
+				while (!cur->next_sibling)
+				{
+					cur = cur->parent;
+
+					if (!cur) return;
+				}
+
+				cur = cur->next_sibling;
+
+				while (cur)
+				{
+					if (step_push(ns, cur, alloc) & once)
+						return;
+
+					if (cur->first_child)
+						cur = cur->first_child;
+					else
+					{
+						while (!cur->next_sibling)
+						{
+							cur = cur->parent;
+
+							if (!cur) return;
+						}
+
+						cur = cur->next_sibling;
+					}
+				}
+
+				break;
+			}
+
+			case axis_preceding:
+			{
+				xml_node_struct* cur = n;
+
+				// exit from this node so that we don't include descendants
+				while (!cur->prev_sibling_c->next_sibling)
+				{
+					cur = cur->parent;
+
+					if (!cur) return;
+				}
+
+				cur = cur->prev_sibling_c;
+
+				while (cur)
+				{
+					if (cur->first_child)
+						cur = cur->first_child->prev_sibling_c;
+					else
+					{
+						// leaf node, can't be ancestor
+						if (step_push(ns, cur, alloc) & once)
+							return;
+
+						while (!cur->prev_sibling_c->next_sibling)
+						{
+							cur = cur->parent;
+
+							if (!cur) return;
+
+							if (!node_is_ancestor(cur, n))
+								if (step_push(ns, cur, alloc) & once)
+									return;
+						}
+
+						cur = cur->prev_sibling_c;
+					}
+				}
+
+				break;
+			}
+			
+			case axis_ancestor:
+			case axis_ancestor_or_self:
+			{
+				if (axis == axis_ancestor_or_self)
+					if (step_push(ns, n, alloc) & once)
+						return;
+
+				xml_node_struct* cur = n->parent;
+				
+				while (cur)
+				{
+					if (step_push(ns, cur, alloc) & once)
+						return;
+					
+					cur = cur->parent;
+				}
+				
+				break;
+			}
+
+			case axis_self:
+			{
+				step_push(ns, n, alloc);
+
+				break;
+			}
+
+			case axis_parent:
+			{
+				if (n->parent)
+					step_push(ns, n->parent, alloc);
+
+				break;
+			}
+				
+			default:
+				assert(!"Unimplemented axis");
+			}
+		}
+		
+		template <class T> void step_fill(xpath_node_set_raw& ns, xml_attribute_struct* a, xml_node_struct* p, xpath_allocator* alloc, bool once, T v)
+		{
+			const axis_t axis = T::axis;
+
+			switch (axis)
+			{
+			case axis_ancestor:
+			case axis_ancestor_or_self:
+			{
+				if (axis == axis_ancestor_or_self && _test == nodetest_type_node) // reject attributes based on principal node type test
+					if (step_push(ns, a, p, alloc) & once)
+						return;
+
+				xml_node_struct* cur = p;
+				
+				while (cur)
+				{
+					if (step_push(ns, cur, alloc) & once)
+						return;
+					
+					cur = cur->parent;
+				}
+				
+				break;
+			}
+
+			case axis_descendant_or_self:
+			case axis_self:
+			{
+				if (_test == nodetest_type_node) // reject attributes based on principal node type test
+					step_push(ns, a, p, alloc);
+
+				break;
+			}
+
+			case axis_following:
+			{
+				xml_node_struct* cur = p;
+				
+				while (cur)
+				{
+					if (cur->first_child)
+						cur = cur->first_child;
+					else
+					{
+						while (!cur->next_sibling)
+						{
+							cur = cur->parent;
+
+							if (!cur) return;
+						}
+
+						cur = cur->next_sibling;
+					}
+
+					if (step_push(ns, cur, alloc) & once)
+						return;
+				}
+
+				break;
+			}
+
+			case axis_parent:
+			{
+				step_push(ns, p, alloc);
+
+				break;
+			}
+
+			case axis_preceding:
+			{
+				// preceding:: axis does not include attribute nodes and attribute ancestors (they are the same as parent's ancestors), so we can reuse node preceding
+				step_fill(ns, p, alloc, once, v);
+				break;
+			}
+			
+			default:
+				assert(!"Unimplemented axis");
+			}
+		}
+
+		template <class T> void step_fill(xpath_node_set_raw& ns, const xpath_node& xn, xpath_allocator* alloc, bool once, T v)
+		{
+			const axis_t axis = T::axis;
+			const bool axis_has_attributes = (axis == axis_ancestor || axis == axis_ancestor_or_self || axis == axis_descendant_or_self || axis == axis_following || axis == axis_parent || axis == axis_preceding || axis == axis_self);
+
+			if (xn.node())
+				step_fill(ns, xn.node().internal_object(), alloc, once, v);
+			else if (axis_has_attributes && xn.attribute() && xn.parent())
+				step_fill(ns, xn.attribute().internal_object(), xn.parent().internal_object(), alloc, once, v);
+		}
+
+		template <class T> xpath_node_set_raw step_do(const xpath_context& c, const xpath_stack& stack, nodeset_eval_t eval, T v)
+		{
+			const axis_t axis = T::axis;
+			const bool axis_reverse = (axis == axis_ancestor || axis == axis_ancestor_or_self || axis == axis_preceding || axis == axis_preceding_sibling);
+			const xpath_node_set::type_t axis_type = axis_reverse ? xpath_node_set::type_sorted_reverse : xpath_node_set::type_sorted;
+
+			bool once =
+				(axis == axis_attribute && _test == nodetest_name) ||
+				(!_right && eval_once(axis_type, eval)) ||
+				(_right && !_right->_next && _right->_test == predicate_constant_one);
+
+			xpath_node_set_raw ns;
+			ns.set_type(axis_type);
+
+			if (_left)
+			{
+				xpath_node_set_raw s = _left->eval_node_set(c, stack, nodeset_eval_all);
+
+				// self axis preserves the original order
+				if (axis == axis_self) ns.set_type(s.type());
+
+				for (const xpath_node* it = s.begin(); it != s.end(); ++it)
+				{
+					size_t size = ns.size();
+
+					// in general, all axes generate elements in a particular order, but there is no order guarantee if axis is applied to two nodes
+					if (axis != axis_self && size != 0) ns.set_type(xpath_node_set::type_unsorted);
+					
+					step_fill(ns, *it, stack.result, once, v);
+					if (_right) apply_predicates(ns, size, stack, eval);
+				}
+			}
+			else
+			{
+				step_fill(ns, c.n, stack.result, once, v);
+				if (_right) apply_predicates(ns, 0, stack, eval);
+			}
+
+			// child, attribute and self axes always generate unique set of nodes
+			// for other axis, if the set stayed sorted, it stayed unique because the traversal algorithms do not visit the same node twice
+			if (axis != axis_child && axis != axis_attribute && axis != axis_self && ns.type() == xpath_node_set::type_unsorted)
+				ns.remove_duplicates();
+
+			return ns;
+		}
+		
+	public:
+		xpath_ast_node(ast_type_t type, xpath_value_type rettype_, const char_t* value):
+			_type(static_cast<char>(type)), _rettype(static_cast<char>(rettype_)), _axis(0), _test(0), _left(0), _right(0), _next(0)
+		{
+			assert(type == ast_string_constant);
+			_data.string = value;
+		}
+
+		xpath_ast_node(ast_type_t type, xpath_value_type rettype_, double value):
+			_type(static_cast<char>(type)), _rettype(static_cast<char>(rettype_)), _axis(0), _test(0), _left(0), _right(0), _next(0)
+		{
+			assert(type == ast_number_constant);
+			_data.number = value;
+		}
+		
+		xpath_ast_node(ast_type_t type, xpath_value_type rettype_, xpath_variable* value):
+			_type(static_cast<char>(type)), _rettype(static_cast<char>(rettype_)), _axis(0), _test(0), _left(0), _right(0), _next(0)
+		{
+			assert(type == ast_variable);
+			_data.variable = value;
+		}
+		
+		xpath_ast_node(ast_type_t type, xpath_value_type rettype_, xpath_ast_node* left = 0, xpath_ast_node* right = 0):
+			_type(static_cast<char>(type)), _rettype(static_cast<char>(rettype_)), _axis(0), _test(0), _left(left), _right(right), _next(0)
+		{
+		}
+
+		xpath_ast_node(ast_type_t type, xpath_ast_node* left, axis_t axis, nodetest_t test, const char_t* contents):
+			_type(static_cast<char>(type)), _rettype(xpath_type_node_set), _axis(static_cast<char>(axis)), _test(static_cast<char>(test)), _left(left), _right(0), _next(0)
+		{
+			assert(type == ast_step);
+			_data.nodetest = contents;
+		}
+
+		xpath_ast_node(ast_type_t type, xpath_ast_node* left, xpath_ast_node* right, predicate_t test):
+			_type(static_cast<char>(type)), _rettype(xpath_type_node_set), _axis(0), _test(static_cast<char>(test)), _left(left), _right(right), _next(0)
+		{
+			assert(type == ast_filter || type == ast_predicate);
+		}
+
+		void set_next(xpath_ast_node* value)
+		{
+			_next = value;
+		}
+
+		void set_right(xpath_ast_node* value)
+		{
+			_right = value;
+		}
+
+		bool eval_boolean(const xpath_context& c, const xpath_stack& stack)
+		{
+			switch (_type)
+			{
+			case ast_op_or:
+				return _left->eval_boolean(c, stack) || _right->eval_boolean(c, stack);
+				
+			case ast_op_and:
+				return _left->eval_boolean(c, stack) && _right->eval_boolean(c, stack);
+				
+			case ast_op_equal:
+				return compare_eq(_left, _right, c, stack, equal_to());
+
+			case ast_op_not_equal:
+				return compare_eq(_left, _right, c, stack, not_equal_to());
+	
+			case ast_op_less:
+				return compare_rel(_left, _right, c, stack, less());
+			
+			case ast_op_greater:
+				return compare_rel(_right, _left, c, stack, less());
+
+			case ast_op_less_or_equal:
+				return compare_rel(_left, _right, c, stack, less_equal());
+			
+			case ast_op_greater_or_equal:
+				return compare_rel(_right, _left, c, stack, less_equal());
+
+			case ast_func_starts_with:
+			{
+				xpath_allocator_capture cr(stack.result);
+
+				xpath_string lr = _left->eval_string(c, stack);
+				xpath_string rr = _right->eval_string(c, stack);
+
+				return starts_with(lr.c_str(), rr.c_str());
+			}
+
+			case ast_func_contains:
+			{
+				xpath_allocator_capture cr(stack.result);
+
+				xpath_string lr = _left->eval_string(c, stack);
+				xpath_string rr = _right->eval_string(c, stack);
+
+				return find_substring(lr.c_str(), rr.c_str()) != 0;
+			}
+
+			case ast_func_boolean:
+				return _left->eval_boolean(c, stack);
+				
+			case ast_func_not:
+				return !_left->eval_boolean(c, stack);
+				
+			case ast_func_true:
+				return true;
+				
+			case ast_func_false:
+				return false;
+
+			case ast_func_lang:
+			{
+				if (c.n.attribute()) return false;
+				
+				xpath_allocator_capture cr(stack.result);
+
+				xpath_string lang = _left->eval_string(c, stack);
+				
+				for (xml_node n = c.n.node(); n; n = n.parent())
+				{
+					xml_attribute a = n.attribute(PUGIXML_TEXT("xml:lang"));
+					
+					if (a)
+					{
+						const char_t* value = a.value();
+						
+						// strnicmp / strncasecmp is not portable
+						for (const char_t* lit = lang.c_str(); *lit; ++lit)
+						{
+							if (tolower_ascii(*lit) != tolower_ascii(*value)) return false;
+							++value;
+						}
+						
+						return *value == 0 || *value == '-';
+					}
+				}
+				
+				return false;
+			}
+
+			case ast_opt_compare_attribute:
+			{
+				const char_t* value = (_right->_type == ast_string_constant) ? _right->_data.string : _right->_data.variable->get_string();
+
+				xml_attribute attr = c.n.node().attribute(_left->_data.nodetest);
+
+				return attr && strequal(attr.value(), value) && is_xpath_attribute(attr.name());
+			}
+
+			case ast_variable:
+			{
+				assert(_rettype == _data.variable->type());
+
+				if (_rettype == xpath_type_boolean)
+					return _data.variable->get_boolean();
+
+				// fallthrough to type conversion
+			}
+
+			default:
+			{
+				switch (_rettype)
+				{
+				case xpath_type_number:
+					return convert_number_to_boolean(eval_number(c, stack));
+					
+				case xpath_type_string:
+				{
+					xpath_allocator_capture cr(stack.result);
+
+					return !eval_string(c, stack).empty();
+				}
+					
+				case xpath_type_node_set:				
+				{
+					xpath_allocator_capture cr(stack.result);
+
+					return !eval_node_set(c, stack, nodeset_eval_any).empty();
+				}
+
+				default:
+					assert(!"Wrong expression for return type boolean");
+					return false;
+				}
+			}
+			}
+		}
+
+		double eval_number(const xpath_context& c, const xpath_stack& stack)
+		{
+			switch (_type)
+			{
+			case ast_op_add:
+				return _left->eval_number(c, stack) + _right->eval_number(c, stack);
+				
+			case ast_op_subtract:
+				return _left->eval_number(c, stack) - _right->eval_number(c, stack);
+
+			case ast_op_multiply:
+				return _left->eval_number(c, stack) * _right->eval_number(c, stack);
+
+			case ast_op_divide:
+				return _left->eval_number(c, stack) / _right->eval_number(c, stack);
+
+			case ast_op_mod:
+				return fmod(_left->eval_number(c, stack), _right->eval_number(c, stack));
+
+			case ast_op_negate:
+				return -_left->eval_number(c, stack);
+
+			case ast_number_constant:
+				return _data.number;
+
+			case ast_func_last:
+				return static_cast<double>(c.size);
+			
+			case ast_func_position:
+				return static_cast<double>(c.position);
+
+			case ast_func_count:
+			{
+				xpath_allocator_capture cr(stack.result);
+
+				return static_cast<double>(_left->eval_node_set(c, stack, nodeset_eval_all).size());
+			}
+			
+			case ast_func_string_length_0:
+			{
+				xpath_allocator_capture cr(stack.result);
+
+				return static_cast<double>(string_value(c.n, stack.result).length());
+			}
+			
+			case ast_func_string_length_1:
+			{
+				xpath_allocator_capture cr(stack.result);
+
+				return static_cast<double>(_left->eval_string(c, stack).length());
+			}
+			
+			case ast_func_number_0:
+			{
+				xpath_allocator_capture cr(stack.result);
+
+				return convert_string_to_number(string_value(c.n, stack.result).c_str());
+			}
+			
+			case ast_func_number_1:
+				return _left->eval_number(c, stack);
+
+			case ast_func_sum:
+			{
+				xpath_allocator_capture cr(stack.result);
+
+				double r = 0;
+				
+				xpath_node_set_raw ns = _left->eval_node_set(c, stack, nodeset_eval_all);
+				
+				for (const xpath_node* it = ns.begin(); it != ns.end(); ++it)
+				{
+					xpath_allocator_capture cri(stack.result);
+
+					r += convert_string_to_number(string_value(*it, stack.result).c_str());
+				}
+			
+				return r;
+			}
+
+			case ast_func_floor:
+			{
+				double r = _left->eval_number(c, stack);
+				
+				return r == r ? floor(r) : r;
+			}
+
+			case ast_func_ceiling:
+			{
+				double r = _left->eval_number(c, stack);
+				
+				return r == r ? ceil(r) : r;
+			}
+
+			case ast_func_round:
+				return round_nearest_nzero(_left->eval_number(c, stack));
+			
+			case ast_variable:
+			{
+				assert(_rettype == _data.variable->type());
+
+				if (_rettype == xpath_type_number)
+					return _data.variable->get_number();
+
+				// fallthrough to type conversion
+			}
+
+			default:
+			{
+				switch (_rettype)
+				{
+				case xpath_type_boolean:
+					return eval_boolean(c, stack) ? 1 : 0;
+					
+				case xpath_type_string:
+				{
+					xpath_allocator_capture cr(stack.result);
+
+					return convert_string_to_number(eval_string(c, stack).c_str());
+				}
+					
+				case xpath_type_node_set:
+				{
+					xpath_allocator_capture cr(stack.result);
+
+					return convert_string_to_number(eval_string(c, stack).c_str());
+				}
+					
+				default:
+					assert(!"Wrong expression for return type number");
+					return 0;
+				}
+				
+			}
+			}
+		}
+		
+		xpath_string eval_string_concat(const xpath_context& c, const xpath_stack& stack)
+		{
+			assert(_type == ast_func_concat);
+
+			xpath_allocator_capture ct(stack.temp);
+
+			// count the string number
+			size_t count = 1;
+			for (xpath_ast_node* nc = _right; nc; nc = nc->_next) count++;
+
+			// gather all strings
+			xpath_string static_buffer[4];
+			xpath_string* buffer = static_buffer;
+
+			// allocate on-heap for large concats
+			if (count > sizeof(static_buffer) / sizeof(static_buffer[0]))
+			{
+				buffer = static_cast<xpath_string*>(stack.temp->allocate(count * sizeof(xpath_string)));
+				assert(buffer);
+			}
+
+			// evaluate all strings to temporary stack
+			xpath_stack swapped_stack = {stack.temp, stack.result};
+
+			buffer[0] = _left->eval_string(c, swapped_stack);
+
+			size_t pos = 1;
+			for (xpath_ast_node* n = _right; n; n = n->_next, ++pos) buffer[pos] = n->eval_string(c, swapped_stack);
+			assert(pos == count);
+
+			// get total length
+			size_t length = 0;
+			for (size_t i = 0; i < count; ++i) length += buffer[i].length();
+
+			// create final string
+			char_t* result = static_cast<char_t*>(stack.result->allocate((length + 1) * sizeof(char_t)));
+			assert(result);
+
+			char_t* ri = result;
+
+			for (size_t j = 0; j < count; ++j)
+				for (const char_t* bi = buffer[j].c_str(); *bi; ++bi)
+					*ri++ = *bi;
+
+			*ri = 0;
+
+			return xpath_string::from_heap_preallocated(result, ri);
+		}
+
+		xpath_string eval_string(const xpath_context& c, const xpath_stack& stack)
+		{
+			switch (_type)
+			{
+			case ast_string_constant:
+				return xpath_string::from_const(_data.string);
+			
+			case ast_func_local_name_0:
+			{
+				xpath_node na = c.n;
+				
+				return xpath_string::from_const(local_name(na));
+			}
+
+			case ast_func_local_name_1:
+			{
+				xpath_allocator_capture cr(stack.result);
+
+				xpath_node_set_raw ns = _left->eval_node_set(c, stack, nodeset_eval_first);
+				xpath_node na = ns.first();
+				
+				return xpath_string::from_const(local_name(na));
+			}
+
+			case ast_func_name_0:
+			{
+				xpath_node na = c.n;
+				
+				return xpath_string::from_const(qualified_name(na));
+			}
+
+			case ast_func_name_1:
+			{
+				xpath_allocator_capture cr(stack.result);
+
+				xpath_node_set_raw ns = _left->eval_node_set(c, stack, nodeset_eval_first);
+				xpath_node na = ns.first();
+				
+				return xpath_string::from_const(qualified_name(na));
+			}
+
+			case ast_func_namespace_uri_0:
+			{
+				xpath_node na = c.n;
+				
+				return xpath_string::from_const(namespace_uri(na));
+			}
+
+			case ast_func_namespace_uri_1:
+			{
+				xpath_allocator_capture cr(stack.result);
+
+				xpath_node_set_raw ns = _left->eval_node_set(c, stack, nodeset_eval_first);
+				xpath_node na = ns.first();
+				
+				return xpath_string::from_const(namespace_uri(na));
+			}
+
+			case ast_func_string_0:
+				return string_value(c.n, stack.result);
+
+			case ast_func_string_1:
+				return _left->eval_string(c, stack);
+
+			case ast_func_concat:
+				return eval_string_concat(c, stack);
+
+			case ast_func_substring_before:
+			{
+				xpath_allocator_capture cr(stack.temp);
+
+				xpath_stack swapped_stack = {stack.temp, stack.result};
+
+				xpath_string s = _left->eval_string(c, swapped_stack);
+				xpath_string p = _right->eval_string(c, swapped_stack);
+
+				const char_t* pos = find_substring(s.c_str(), p.c_str());
+				
+				return pos ? xpath_string::from_heap(s.c_str(), pos, stack.result) : xpath_string();
+			}
+			
+			case ast_func_substring_after:
+			{
+				xpath_allocator_capture cr(stack.temp);
+
+				xpath_stack swapped_stack = {stack.temp, stack.result};
+
+				xpath_string s = _left->eval_string(c, swapped_stack);
+				xpath_string p = _right->eval_string(c, swapped_stack);
+				
+				const char_t* pos = find_substring(s.c_str(), p.c_str());
+				if (!pos) return xpath_string();
+
+				const char_t* rbegin = pos + p.length();
+				const char_t* rend = s.c_str() + s.length();
+
+				return s.uses_heap() ? xpath_string::from_heap(rbegin, rend, stack.result) : xpath_string::from_const(rbegin);
+			}
+
+			case ast_func_substring_2:
+			{
+				xpath_allocator_capture cr(stack.temp);
+
+				xpath_stack swapped_stack = {stack.temp, stack.result};
+
+				xpath_string s = _left->eval_string(c, swapped_stack);
+				size_t s_length = s.length();
+
+				double first = round_nearest(_right->eval_number(c, stack));
+				
+				if (is_nan(first)) return xpath_string(); // NaN
+				else if (first >= s_length + 1) return xpath_string();
+				
+				size_t pos = first < 1 ? 1 : static_cast<size_t>(first);
+				assert(1 <= pos && pos <= s_length + 1);
+
+				const char_t* rbegin = s.c_str() + (pos - 1);
+				const char_t* rend = s.c_str() + s.length();
+				
+				return s.uses_heap() ? xpath_string::from_heap(rbegin, rend, stack.result) : xpath_string::from_const(rbegin);
+			}
+			
+			case ast_func_substring_3:
+			{
+				xpath_allocator_capture cr(stack.temp);
+
+				xpath_stack swapped_stack = {stack.temp, stack.result};
+
+				xpath_string s = _left->eval_string(c, swapped_stack);
+				size_t s_length = s.length();
+
+				double first = round_nearest(_right->eval_number(c, stack));
+				double last = first + round_nearest(_right->_next->eval_number(c, stack));
+				
+				if (is_nan(first) || is_nan(last)) return xpath_string();
+				else if (first >= s_length + 1) return xpath_string();
+				else if (first >= last) return xpath_string();
+				else if (last < 1) return xpath_string();
+				
+				size_t pos = first < 1 ? 1 : static_cast<size_t>(first);
+				size_t end = last >= s_length + 1 ? s_length + 1 : static_cast<size_t>(last);
+
+				assert(1 <= pos && pos <= end && end <= s_length + 1);
+				const char_t* rbegin = s.c_str() + (pos - 1);
+				const char_t* rend = s.c_str() + (end - 1);
+
+				return (end == s_length + 1 && !s.uses_heap()) ? xpath_string::from_const(rbegin) : xpath_string::from_heap(rbegin, rend, stack.result);
+			}
+
+			case ast_func_normalize_space_0:
+			{
+				xpath_string s = string_value(c.n, stack.result);
+
+				char_t* begin = s.data(stack.result);
+				char_t* end = normalize_space(begin);
+
+				return xpath_string::from_heap_preallocated(begin, end);
+			}
+
+			case ast_func_normalize_space_1:
+			{
+				xpath_string s = _left->eval_string(c, stack);
+
+				char_t* begin = s.data(stack.result);
+				char_t* end = normalize_space(begin);
+			
+				return xpath_string::from_heap_preallocated(begin, end);
+			}
+
+			case ast_func_translate:
+			{
+				xpath_allocator_capture cr(stack.temp);
+
+				xpath_stack swapped_stack = {stack.temp, stack.result};
+
+				xpath_string s = _left->eval_string(c, stack);
+				xpath_string from = _right->eval_string(c, swapped_stack);
+				xpath_string to = _right->_next->eval_string(c, swapped_stack);
+
+				char_t* begin = s.data(stack.result);
+				char_t* end = translate(begin, from.c_str(), to.c_str(), to.length());
+
+				return xpath_string::from_heap_preallocated(begin, end);
+			}
+
+			case ast_opt_translate_table:
+			{
+				xpath_string s = _left->eval_string(c, stack);
+
+				char_t* begin = s.data(stack.result);
+				char_t* end = translate_table(begin, _data.table);
+
+				return xpath_string::from_heap_preallocated(begin, end);
+			}
+
+			case ast_variable:
+			{
+				assert(_rettype == _data.variable->type());
+
+				if (_rettype == xpath_type_string)
+					return xpath_string::from_const(_data.variable->get_string());
+
+				// fallthrough to type conversion
+			}
+
+			default:
+			{
+				switch (_rettype)
+				{
+				case xpath_type_boolean:
+					return xpath_string::from_const(eval_boolean(c, stack) ? PUGIXML_TEXT("true") : PUGIXML_TEXT("false"));
+					
+				case xpath_type_number:
+					return convert_number_to_string(eval_number(c, stack), stack.result);
+					
+				case xpath_type_node_set:
+				{
+					xpath_allocator_capture cr(stack.temp);
+
+					xpath_stack swapped_stack = {stack.temp, stack.result};
+
+					xpath_node_set_raw ns = eval_node_set(c, swapped_stack, nodeset_eval_first);
+					return ns.empty() ? xpath_string() : string_value(ns.first(), stack.result);
+				}
+				
+				default:
+					assert(!"Wrong expression for return type string");
+					return xpath_string();
+				}
+			}
+			}
+		}
+
+		xpath_node_set_raw eval_node_set(const xpath_context& c, const xpath_stack& stack, nodeset_eval_t eval)
+		{
+			switch (_type)
+			{
+			case ast_op_union:
+			{
+				xpath_allocator_capture cr(stack.temp);
+
+				xpath_stack swapped_stack = {stack.temp, stack.result};
+
+				xpath_node_set_raw ls = _left->eval_node_set(c, swapped_stack, eval);
+				xpath_node_set_raw rs = _right->eval_node_set(c, stack, eval);
+
+				// we can optimize merging two sorted sets, but this is a very rare operation, so don't bother
+				rs.set_type(xpath_node_set::type_unsorted);
+
+				rs.append(ls.begin(), ls.end(), stack.result);
+				rs.remove_duplicates();
+
+				return rs;
+			}
+
+			case ast_filter:
+			{
+				xpath_node_set_raw set = _left->eval_node_set(c, stack, _test == predicate_constant_one ? nodeset_eval_first : nodeset_eval_all);
+
+				// either expression is a number or it contains position() call; sort by document order
+				if (_test != predicate_posinv) set.sort_do();
+
+				bool once = eval_once(set.type(), eval);
+
+				apply_predicate(set, 0, stack, once);
+			
+				return set;
+			}
+			
+			case ast_func_id:
+				return xpath_node_set_raw();
+			
+			case ast_step:
+			{
+				switch (_axis)
+				{
+				case axis_ancestor:
+					return step_do(c, stack, eval, axis_to_type<axis_ancestor>());
+					
+				case axis_ancestor_or_self:
+					return step_do(c, stack, eval, axis_to_type<axis_ancestor_or_self>());
+
+				case axis_attribute:
+					return step_do(c, stack, eval, axis_to_type<axis_attribute>());
+
+				case axis_child:
+					return step_do(c, stack, eval, axis_to_type<axis_child>());
+				
+				case axis_descendant:
+					return step_do(c, stack, eval, axis_to_type<axis_descendant>());
+
+				case axis_descendant_or_self:
+					return step_do(c, stack, eval, axis_to_type<axis_descendant_or_self>());
+
+				case axis_following:
+					return step_do(c, stack, eval, axis_to_type<axis_following>());
+				
+				case axis_following_sibling:
+					return step_do(c, stack, eval, axis_to_type<axis_following_sibling>());
+				
+				case axis_namespace:
+					// namespaced axis is not supported
+					return xpath_node_set_raw();
+				
+				case axis_parent:
+					return step_do(c, stack, eval, axis_to_type<axis_parent>());
+				
+				case axis_preceding:
+					return step_do(c, stack, eval, axis_to_type<axis_preceding>());
+
+				case axis_preceding_sibling:
+					return step_do(c, stack, eval, axis_to_type<axis_preceding_sibling>());
+				
+				case axis_self:
+					return step_do(c, stack, eval, axis_to_type<axis_self>());
+
+				default:
+					assert(!"Unknown axis");
+					return xpath_node_set_raw();
+				}
+			}
+
+			case ast_step_root:
+			{
+				assert(!_right); // root step can't have any predicates
+
+				xpath_node_set_raw ns;
+
+				ns.set_type(xpath_node_set::type_sorted);
+
+				if (c.n.node()) ns.push_back(c.n.node().root(), stack.result);
+				else if (c.n.attribute()) ns.push_back(c.n.parent().root(), stack.result);
+
+				return ns;
+			}
+
+			case ast_variable:
+			{
+				assert(_rettype == _data.variable->type());
+
+				if (_rettype == xpath_type_node_set)
+				{
+					const xpath_node_set& s = _data.variable->get_node_set();
+
+					xpath_node_set_raw ns;
+
+					ns.set_type(s.type());
+					ns.append(s.begin(), s.end(), stack.result);
+
+					return ns;
+				}
+
+				// fallthrough to type conversion
+			}
+
+			default:
+				assert(!"Wrong expression for return type node set");
+				return xpath_node_set_raw();
+			}
+		}
+
+		void optimize(xpath_allocator* alloc)
+		{
+			if (_left) _left->optimize(alloc);
+			if (_right) _right->optimize(alloc);
+			if (_next) _next->optimize(alloc);
+
+			optimize_self(alloc);
+		}
+
+		void optimize_self(xpath_allocator* alloc)
+		{
+			// Rewrite [position()=expr] with [expr]
+			// Note that this step has to go before classification to recognize [position()=1]
+			if ((_type == ast_filter || _type == ast_predicate) &&
+				_right->_type == ast_op_equal && _right->_left->_type == ast_func_position && _right->_right->_rettype == xpath_type_number)
+			{
+				_right = _right->_right;
+			}
+
+			// Classify filter/predicate ops to perform various optimizations during evaluation
+			if (_type == ast_filter || _type == ast_predicate)
+			{
+				assert(_test == predicate_default);
+
+				if (_right->_type == ast_number_constant && _right->_data.number == 1.0)
+					_test = predicate_constant_one;
+				else if (_right->_rettype == xpath_type_number && (_right->_type == ast_number_constant || _right->_type == ast_variable || _right->_type == ast_func_last))
+					_test = predicate_constant;
+				else if (_right->_rettype != xpath_type_number && _right->is_posinv_expr())
+					_test = predicate_posinv;
+			}
+
+			// Rewrite descendant-or-self::node()/child::foo with descendant::foo
+			// The former is a full form of //foo, the latter is much faster since it executes the node test immediately
+			// Do a similar kind of rewrite for self/descendant/descendant-or-self axes
+			// Note that we only rewrite positionally invariant steps (//foo[1] != /descendant::foo[1])
+			if (_type == ast_step && (_axis == axis_child || _axis == axis_self || _axis == axis_descendant || _axis == axis_descendant_or_self) && _left &&
+				_left->_type == ast_step && _left->_axis == axis_descendant_or_self && _left->_test == nodetest_type_node && !_left->_right &&
+				is_posinv_step())
+			{
+				if (_axis == axis_child || _axis == axis_descendant)
+					_axis = axis_descendant;
+				else
+					_axis = axis_descendant_or_self;
+
+				_left = _left->_left;
+			}
+
+			// Use optimized lookup table implementation for translate() with constant arguments
+			if (_type == ast_func_translate && _right->_type == ast_string_constant && _right->_next->_type == ast_string_constant)
+			{
+				unsigned char* table = translate_table_generate(alloc, _right->_data.string, _right->_next->_data.string);
+
+				if (table)
+				{
+					_type = ast_opt_translate_table;
+					_data.table = table;
+				}
+			}
+
+			// Use optimized path for @attr = 'value' or @attr = $value
+			if (_type == ast_op_equal &&
+				_left->_type == ast_step && _left->_axis == axis_attribute && _left->_test == nodetest_name && !_left->_left && !_left->_right &&
+				(_right->_type == ast_string_constant || (_right->_type == ast_variable && _right->_rettype == xpath_type_string)))
+			{
+				_type = ast_opt_compare_attribute;
+			}
+		}
+		
+		bool is_posinv_expr() const
+		{
+			switch (_type)
+			{
+			case ast_func_position:
+			case ast_func_last:
+				return false;
+
+			case ast_string_constant:
+			case ast_number_constant:
+			case ast_variable:
+				return true;
+
+			case ast_step:
+			case ast_step_root:
+				return true;
+
+			case ast_predicate:
+			case ast_filter:
+				return true;
+
+			default:
+				if (_left && !_left->is_posinv_expr()) return false;
+				
+				for (xpath_ast_node* n = _right; n; n = n->_next)
+					if (!n->is_posinv_expr()) return false;
+					
+				return true;
+			}
+		}
+
+		bool is_posinv_step() const
+		{
+			assert(_type == ast_step);
+
+			for (xpath_ast_node* n = _right; n; n = n->_next)
+			{
+				assert(n->_type == ast_predicate);
+
+				if (n->_test != predicate_posinv)
+					return false;
+			}
+
+			return true;
+		}
+
+		xpath_value_type rettype() const
+		{
+			return static_cast<xpath_value_type>(_rettype);
+		}
+	};
+
+	struct xpath_parser
+	{
+		xpath_allocator* _alloc;
+		xpath_lexer _lexer;
+
+		const char_t* _query;
+		xpath_variable_set* _variables;
+
+		xpath_parse_result* _result;
+
+		char_t _scratch[32];
+
+	#ifdef PUGIXML_NO_EXCEPTIONS
+		jmp_buf _error_handler;
+	#endif
+
+		void throw_error(const char* message)
+		{
+			_result->error = message;
+			_result->offset = _lexer.current_pos() - _query;
+
+		#ifdef PUGIXML_NO_EXCEPTIONS
+			longjmp(_error_handler, 1);
+		#else
+			throw xpath_exception(*_result);
+		#endif
+		}
+
+		void throw_error_oom()
+		{
+		#ifdef PUGIXML_NO_EXCEPTIONS
+			throw_error("Out of memory");
+		#else
+			throw std::bad_alloc();
+		#endif
+		}
+
+		void* alloc_node()
+		{
+			void* result = _alloc->allocate_nothrow(sizeof(xpath_ast_node));
+
+			if (!result) throw_error_oom();
+
+			return result;
+		}
+
+		const char_t* alloc_string(const xpath_lexer_string& value)
+		{
+			if (value.begin)
+			{
+				size_t length = static_cast<size_t>(value.end - value.begin);
+
+				char_t* c = static_cast<char_t*>(_alloc->allocate_nothrow((length + 1) * sizeof(char_t)));
+				if (!c) throw_error_oom();
+				assert(c); // workaround for clang static analysis
+
+				memcpy(c, value.begin, length * sizeof(char_t));
+				c[length] = 0;
+
+				return c;
+			}
+			else return 0;
+		}
+
+		xpath_ast_node* parse_function_helper(ast_type_t type0, ast_type_t type1, size_t argc, xpath_ast_node* args[2])
+		{
+			assert(argc <= 1);
+
+			if (argc == 1 && args[0]->rettype() != xpath_type_node_set) throw_error("Function has to be applied to node set");
+
+			return new (alloc_node()) xpath_ast_node(argc == 0 ? type0 : type1, xpath_type_string, args[0]);
+		}
+
+		xpath_ast_node* parse_function(const xpath_lexer_string& name, size_t argc, xpath_ast_node* args[2])
+		{
+			switch (name.begin[0])
+			{
+			case 'b':
+				if (name == PUGIXML_TEXT("boolean") && argc == 1)
+					return new (alloc_node()) xpath_ast_node(ast_func_boolean, xpath_type_boolean, args[0]);
+					
+				break;
+			
+			case 'c':
+				if (name == PUGIXML_TEXT("count") && argc == 1)
+				{
+					if (args[0]->rettype() != xpath_type_node_set) throw_error("Function has to be applied to node set");
+					return new (alloc_node()) xpath_ast_node(ast_func_count, xpath_type_number, args[0]);
+				}
+				else if (name == PUGIXML_TEXT("contains") && argc == 2)
+					return new (alloc_node()) xpath_ast_node(ast_func_contains, xpath_type_boolean, args[0], args[1]);
+				else if (name == PUGIXML_TEXT("concat") && argc >= 2)
+					return new (alloc_node()) xpath_ast_node(ast_func_concat, xpath_type_string, args[0], args[1]);
+				else if (name == PUGIXML_TEXT("ceiling") && argc == 1)
+					return new (alloc_node()) xpath_ast_node(ast_func_ceiling, xpath_type_number, args[0]);
+					
+				break;
+			
+			case 'f':
+				if (name == PUGIXML_TEXT("false") && argc == 0)
+					return new (alloc_node()) xpath_ast_node(ast_func_false, xpath_type_boolean);
+				else if (name == PUGIXML_TEXT("floor") && argc == 1)
+					return new (alloc_node()) xpath_ast_node(ast_func_floor, xpath_type_number, args[0]);
+					
+				break;
+			
+			case 'i':
+				if (name == PUGIXML_TEXT("id") && argc == 1)
+					return new (alloc_node()) xpath_ast_node(ast_func_id, xpath_type_node_set, args[0]);
+					
+				break;
+			
+			case 'l':
+				if (name == PUGIXML_TEXT("last") && argc == 0)
+					return new (alloc_node()) xpath_ast_node(ast_func_last, xpath_type_number);
+				else if (name == PUGIXML_TEXT("lang") && argc == 1)
+					return new (alloc_node()) xpath_ast_node(ast_func_lang, xpath_type_boolean, args[0]);
+				else if (name == PUGIXML_TEXT("local-name") && argc <= 1)
+					return parse_function_helper(ast_func_local_name_0, ast_func_local_name_1, argc, args);
+			
+				break;
+			
+			case 'n':
+				if (name == PUGIXML_TEXT("name") && argc <= 1)
+					return parse_function_helper(ast_func_name_0, ast_func_name_1, argc, args);
+				else if (name == PUGIXML_TEXT("namespace-uri") && argc <= 1)
+					return parse_function_helper(ast_func_namespace_uri_0, ast_func_namespace_uri_1, argc, args);
+				else if (name == PUGIXML_TEXT("normalize-space") && argc <= 1)
+					return new (alloc_node()) xpath_ast_node(argc == 0 ? ast_func_normalize_space_0 : ast_func_normalize_space_1, xpath_type_string, args[0], args[1]);
+				else if (name == PUGIXML_TEXT("not") && argc == 1)
+					return new (alloc_node()) xpath_ast_node(ast_func_not, xpath_type_boolean, args[0]);
+				else if (name == PUGIXML_TEXT("number") && argc <= 1)
+					return new (alloc_node()) xpath_ast_node(argc == 0 ? ast_func_number_0 : ast_func_number_1, xpath_type_number, args[0]);
+			
+				break;
+			
+			case 'p':
+				if (name == PUGIXML_TEXT("position") && argc == 0)
+					return new (alloc_node()) xpath_ast_node(ast_func_position, xpath_type_number);
+				
+				break;
+			
+			case 'r':
+				if (name == PUGIXML_TEXT("round") && argc == 1)
+					return new (alloc_node()) xpath_ast_node(ast_func_round, xpath_type_number, args[0]);
+
+				break;
+			
+			case 's':
+				if (name == PUGIXML_TEXT("string") && argc <= 1)
+					return new (alloc_node()) xpath_ast_node(argc == 0 ? ast_func_string_0 : ast_func_string_1, xpath_type_string, args[0]);
+				else if (name == PUGIXML_TEXT("string-length") && argc <= 1)
+					return new (alloc_node()) xpath_ast_node(argc == 0 ? ast_func_string_length_0 : ast_func_string_length_1, xpath_type_number, args[0]);
+				else if (name == PUGIXML_TEXT("starts-with") && argc == 2)
+					return new (alloc_node()) xpath_ast_node(ast_func_starts_with, xpath_type_boolean, args[0], args[1]);
+				else if (name == PUGIXML_TEXT("substring-before") && argc == 2)
+					return new (alloc_node()) xpath_ast_node(ast_func_substring_before, xpath_type_string, args[0], args[1]);
+				else if (name == PUGIXML_TEXT("substring-after") && argc == 2)
+					return new (alloc_node()) xpath_ast_node(ast_func_substring_after, xpath_type_string, args[0], args[1]);
+				else if (name == PUGIXML_TEXT("substring") && (argc == 2 || argc == 3))
+					return new (alloc_node()) xpath_ast_node(argc == 2 ? ast_func_substring_2 : ast_func_substring_3, xpath_type_string, args[0], args[1]);
+				else if (name == PUGIXML_TEXT("sum") && argc == 1)
+				{
+					if (args[0]->rettype() != xpath_type_node_set) throw_error("Function has to be applied to node set");
+					return new (alloc_node()) xpath_ast_node(ast_func_sum, xpath_type_number, args[0]);
+				}
+
+				break;
+			
+			case 't':
+				if (name == PUGIXML_TEXT("translate") && argc == 3)
+					return new (alloc_node()) xpath_ast_node(ast_func_translate, xpath_type_string, args[0], args[1]);
+				else if (name == PUGIXML_TEXT("true") && argc == 0)
+					return new (alloc_node()) xpath_ast_node(ast_func_true, xpath_type_boolean);
+					
+				break;
+
+			default:
+				break;
+			}
+
+			throw_error("Unrecognized function or wrong parameter count");
+
+			return 0;
+		}
+
+		axis_t parse_axis_name(const xpath_lexer_string& name, bool& specified)
+		{
+			specified = true;
+
+			switch (name.begin[0])
+			{
+			case 'a':
+				if (name == PUGIXML_TEXT("ancestor"))
+					return axis_ancestor;
+				else if (name == PUGIXML_TEXT("ancestor-or-self"))
+					return axis_ancestor_or_self;
+				else if (name == PUGIXML_TEXT("attribute"))
+					return axis_attribute;
+				
+				break;
+			
+			case 'c':
+				if (name == PUGIXML_TEXT("child"))
+					return axis_child;
+				
+				break;
+			
+			case 'd':
+				if (name == PUGIXML_TEXT("descendant"))
+					return axis_descendant;
+				else if (name == PUGIXML_TEXT("descendant-or-self"))
+					return axis_descendant_or_self;
+				
+				break;
+			
+			case 'f':
+				if (name == PUGIXML_TEXT("following"))
+					return axis_following;
+				else if (name == PUGIXML_TEXT("following-sibling"))
+					return axis_following_sibling;
+				
+				break;
+			
+			case 'n':
+				if (name == PUGIXML_TEXT("namespace"))
+					return axis_namespace;
+				
+				break;
+			
+			case 'p':
+				if (name == PUGIXML_TEXT("parent"))
+					return axis_parent;
+				else if (name == PUGIXML_TEXT("preceding"))
+					return axis_preceding;
+				else if (name == PUGIXML_TEXT("preceding-sibling"))
+					return axis_preceding_sibling;
+				
+				break;
+			
+			case 's':
+				if (name == PUGIXML_TEXT("self"))
+					return axis_self;
+				
+				break;
+
+			default:
+				break;
+			}
+
+			specified = false;
+			return axis_child;
+		}
+
+		nodetest_t parse_node_test_type(const xpath_lexer_string& name)
+		{
+			switch (name.begin[0])
+			{
+			case 'c':
+				if (name == PUGIXML_TEXT("comment"))
+					return nodetest_type_comment;
+
+				break;
+
+			case 'n':
+				if (name == PUGIXML_TEXT("node"))
+					return nodetest_type_node;
+
+				break;
+
+			case 'p':
+				if (name == PUGIXML_TEXT("processing-instruction"))
+					return nodetest_type_pi;
+
+				break;
+
+			case 't':
+				if (name == PUGIXML_TEXT("text"))
+					return nodetest_type_text;
+
+				break;
+			
+			default:
+				break;
+			}
+
+			return nodetest_none;
+		}
+
+		// PrimaryExpr ::= VariableReference | '(' Expr ')' | Literal | Number | FunctionCall
+		xpath_ast_node* parse_primary_expression()
+		{
+			switch (_lexer.current())
+			{
+			case lex_var_ref:
+			{
+				xpath_lexer_string name = _lexer.contents();
+
+				if (!_variables)
+					throw_error("Unknown variable: variable set is not provided");
+
+				xpath_variable* var = 0;
+				if (!get_variable_scratch(_scratch, _variables, name.begin, name.end, &var))
+					throw_error_oom();
+
+				if (!var)
+					throw_error("Unknown variable: variable set does not contain the given name");
+
+				_lexer.next();
+
+				return new (alloc_node()) xpath_ast_node(ast_variable, var->type(), var);
+			}
+
+			case lex_open_brace:
+			{
+				_lexer.next();
+
+				xpath_ast_node* n = parse_expression();
+
+				if (_lexer.current() != lex_close_brace)
+					throw_error("Unmatched braces");
+
+				_lexer.next();
+
+				return n;
+			}
+
+			case lex_quoted_string:
+			{
+				const char_t* value = alloc_string(_lexer.contents());
+
+				xpath_ast_node* n = new (alloc_node()) xpath_ast_node(ast_string_constant, xpath_type_string, value);
+				_lexer.next();
+
+				return n;
+			}
+
+			case lex_number:
+			{
+				double value = 0;
+
+				if (!convert_string_to_number_scratch(_scratch, _lexer.contents().begin, _lexer.contents().end, &value))
+					throw_error_oom();
+
+				xpath_ast_node* n = new (alloc_node()) xpath_ast_node(ast_number_constant, xpath_type_number, value);
+				_lexer.next();
+
+				return n;
+			}
+
+			case lex_string:
+			{
+				xpath_ast_node* args[2] = {0};
+				size_t argc = 0;
+				
+				xpath_lexer_string function = _lexer.contents();
+				_lexer.next();
+				
+				xpath_ast_node* last_arg = 0;
+				
+				if (_lexer.current() != lex_open_brace)
+					throw_error("Unrecognized function call");
+				_lexer.next();
+
+				if (_lexer.current() != lex_close_brace)
+					args[argc++] = parse_expression();
+
+				while (_lexer.current() != lex_close_brace)
+				{
+					if (_lexer.current() != lex_comma)
+						throw_error("No comma between function arguments");
+					_lexer.next();
+					
+					xpath_ast_node* n = parse_expression();
+					
+					if (argc < 2) args[argc] = n;
+					else last_arg->set_next(n);
+
+					argc++;
+					last_arg = n;
+				}
+				
+				_lexer.next();
+
+				return parse_function(function, argc, args);
+			}
+
+			default:
+				throw_error("Unrecognizable primary expression");
+
+				return 0;
+			}
+		}
+		
+		// FilterExpr ::= PrimaryExpr | FilterExpr Predicate
+		// Predicate ::= '[' PredicateExpr ']'
+		// PredicateExpr ::= Expr
+		xpath_ast_node* parse_filter_expression()
+		{
+			xpath_ast_node* n = parse_primary_expression();
+
+			while (_lexer.current() == lex_open_square_brace)
+			{
+				_lexer.next();
+
+				xpath_ast_node* expr = parse_expression();
+
+				if (n->rettype() != xpath_type_node_set) throw_error("Predicate has to be applied to node set");
+
+				n = new (alloc_node()) xpath_ast_node(ast_filter, n, expr, predicate_default);
+
+				if (_lexer.current() != lex_close_square_brace)
+					throw_error("Unmatched square brace");
+			
+				_lexer.next();
+			}
+			
+			return n;
+		}
+		
+		// Step ::= AxisSpecifier NodeTest Predicate* | AbbreviatedStep
+		// AxisSpecifier ::= AxisName '::' | '@'?
+		// NodeTest ::= NameTest | NodeType '(' ')' | 'processing-instruction' '(' Literal ')'
+		// NameTest ::= '*' | NCName ':' '*' | QName
+		// AbbreviatedStep ::= '.' | '..'
+		xpath_ast_node* parse_step(xpath_ast_node* set)
+		{
+			if (set && set->rettype() != xpath_type_node_set)
+				throw_error("Step has to be applied to node set");
+
+			bool axis_specified = false;
+			axis_t axis = axis_child; // implied child axis
+
+			if (_lexer.current() == lex_axis_attribute)
+			{
+				axis = axis_attribute;
+				axis_specified = true;
+				
+				_lexer.next();
+			}
+			else if (_lexer.current() == lex_dot)
+			{
+				_lexer.next();
+				
+				return new (alloc_node()) xpath_ast_node(ast_step, set, axis_self, nodetest_type_node, 0);
+			}
+			else if (_lexer.current() == lex_double_dot)
+			{
+				_lexer.next();
+				
+				return new (alloc_node()) xpath_ast_node(ast_step, set, axis_parent, nodetest_type_node, 0);
+			}
+		
+			nodetest_t nt_type = nodetest_none;
+			xpath_lexer_string nt_name;
+			
+			if (_lexer.current() == lex_string)
+			{
+				// node name test
+				nt_name = _lexer.contents();
+				_lexer.next();
+
+				// was it an axis name?
+				if (_lexer.current() == lex_double_colon)
+				{
+					// parse axis name
+					if (axis_specified) throw_error("Two axis specifiers in one step");
+
+					axis = parse_axis_name(nt_name, axis_specified);
+
+					if (!axis_specified) throw_error("Unknown axis");
+
+					// read actual node test
+					_lexer.next();
+
+					if (_lexer.current() == lex_multiply)
+					{
+						nt_type = nodetest_all;
+						nt_name = xpath_lexer_string();
+						_lexer.next();
+					}
+					else if (_lexer.current() == lex_string)
+					{
+						nt_name = _lexer.contents();
+						_lexer.next();
+					}
+					else throw_error("Unrecognized node test");
+				}
+				
+				if (nt_type == nodetest_none)
+				{
+					// node type test or processing-instruction
+					if (_lexer.current() == lex_open_brace)
+					{
+						_lexer.next();
+						
+						if (_lexer.current() == lex_close_brace)
+						{
+							_lexer.next();
+
+							nt_type = parse_node_test_type(nt_name);
+
+							if (nt_type == nodetest_none) throw_error("Unrecognized node type");
+							
+							nt_name = xpath_lexer_string();
+						}
+						else if (nt_name == PUGIXML_TEXT("processing-instruction"))
+						{
+							if (_lexer.current() != lex_quoted_string)
+								throw_error("Only literals are allowed as arguments to processing-instruction()");
+						
+							nt_type = nodetest_pi;
+							nt_name = _lexer.contents();
+							_lexer.next();
+							
+							if (_lexer.current() != lex_close_brace)
+								throw_error("Unmatched brace near processing-instruction()");
+							_lexer.next();
+						}
+						else
+							throw_error("Unmatched brace near node type test");
+
+					}
+					// QName or NCName:*
+					else
+					{
+						if (nt_name.end - nt_name.begin > 2 && nt_name.end[-2] == ':' && nt_name.end[-1] == '*') // NCName:*
+						{
+							nt_name.end--; // erase *
+							
+							nt_type = nodetest_all_in_namespace;
+						}
+						else nt_type = nodetest_name;
+					}
+				}
+			}
+			else if (_lexer.current() == lex_multiply)
+			{
+				nt_type = nodetest_all;
+				_lexer.next();
+			}
+			else throw_error("Unrecognized node test");
+			
+			xpath_ast_node* n = new (alloc_node()) xpath_ast_node(ast_step, set, axis, nt_type, alloc_string(nt_name));
+			
+			xpath_ast_node* last = 0;
+			
+			while (_lexer.current() == lex_open_square_brace)
+			{
+				_lexer.next();
+				
+				xpath_ast_node* expr = parse_expression();
+
+				xpath_ast_node* pred = new (alloc_node()) xpath_ast_node(ast_predicate, 0, expr, predicate_default);
+				
+				if (_lexer.current() != lex_close_square_brace)
+					throw_error("Unmatched square brace");
+				_lexer.next();
+				
+				if (last) last->set_next(pred);
+				else n->set_right(pred);
+				
+				last = pred;
+			}
+
+			return n;
+		}
+		
+		// RelativeLocationPath ::= Step | RelativeLocationPath '/' Step | RelativeLocationPath '//' Step
+		xpath_ast_node* parse_relative_location_path(xpath_ast_node* set)
+		{
+			xpath_ast_node* n = parse_step(set);
+			
+			while (_lexer.current() == lex_slash || _lexer.current() == lex_double_slash)
+			{
+				lexeme_t l = _lexer.current();
+				_lexer.next();
+
+				if (l == lex_double_slash)
+					n = new (alloc_node()) xpath_ast_node(ast_step, n, axis_descendant_or_self, nodetest_type_node, 0);
+				
+				n = parse_step(n);
+			}
+			
+			return n;
+		}
+		
+		// LocationPath ::= RelativeLocationPath | AbsoluteLocationPath
+		// AbsoluteLocationPath ::= '/' RelativeLocationPath? | '//' RelativeLocationPath
+		xpath_ast_node* parse_location_path()
+		{
+			if (_lexer.current() == lex_slash)
+			{
+				_lexer.next();
+				
+				xpath_ast_node* n = new (alloc_node()) xpath_ast_node(ast_step_root, xpath_type_node_set);
+
+				// relative location path can start from axis_attribute, dot, double_dot, multiply and string lexemes; any other lexeme means standalone root path
+				lexeme_t l = _lexer.current();
+
+				if (l == lex_string || l == lex_axis_attribute || l == lex_dot || l == lex_double_dot || l == lex_multiply)
+					return parse_relative_location_path(n);
+				else
+					return n;
+			}
+			else if (_lexer.current() == lex_double_slash)
+			{
+				_lexer.next();
+				
+				xpath_ast_node* n = new (alloc_node()) xpath_ast_node(ast_step_root, xpath_type_node_set);
+				n = new (alloc_node()) xpath_ast_node(ast_step, n, axis_descendant_or_self, nodetest_type_node, 0);
+				
+				return parse_relative_location_path(n);
+			}
+
+			// else clause moved outside of if because of bogus warning 'control may reach end of non-void function being inlined' in gcc 4.0.1
+			return parse_relative_location_path(0);
+		}
+		
+		// PathExpr ::= LocationPath
+		//				| FilterExpr
+		//				| FilterExpr '/' RelativeLocationPath
+		//				| FilterExpr '//' RelativeLocationPath
+		// UnionExpr ::= PathExpr | UnionExpr '|' PathExpr
+		// UnaryExpr ::= UnionExpr | '-' UnaryExpr
+		xpath_ast_node* parse_path_or_unary_expression()
+		{
+			// Clarification.
+			// PathExpr begins with either LocationPath or FilterExpr.
+			// FilterExpr begins with PrimaryExpr
+			// PrimaryExpr begins with '$' in case of it being a variable reference,
+			// '(' in case of it being an expression, string literal, number constant or
+			// function call.
+
+			if (_lexer.current() == lex_var_ref || _lexer.current() == lex_open_brace || 
+				_lexer.current() == lex_quoted_string || _lexer.current() == lex_number ||
+				_lexer.current() == lex_string)
+			{
+				if (_lexer.current() == lex_string)
+				{
+					// This is either a function call, or not - if not, we shall proceed with location path
+					const char_t* state = _lexer.state();
+					
+					while (PUGI__IS_CHARTYPE(*state, ct_space)) ++state;
+					
+					if (*state != '(') return parse_location_path();
+
+					// This looks like a function call; however this still can be a node-test. Check it.
+					if (parse_node_test_type(_lexer.contents()) != nodetest_none) return parse_location_path();
+				}
+				
+				xpath_ast_node* n = parse_filter_expression();
+
+				if (_lexer.current() == lex_slash || _lexer.current() == lex_double_slash)
+				{
+					lexeme_t l = _lexer.current();
+					_lexer.next();
+					
+					if (l == lex_double_slash)
+					{
+						if (n->rettype() != xpath_type_node_set) throw_error("Step has to be applied to node set");
+
+						n = new (alloc_node()) xpath_ast_node(ast_step, n, axis_descendant_or_self, nodetest_type_node, 0);
+					}
+	
+					// select from location path
+					return parse_relative_location_path(n);
+				}
+
+				return n;
+			}
+			else if (_lexer.current() == lex_minus)
+			{
+				_lexer.next();
+
+				// precedence 7+ - only parses union expressions
+				xpath_ast_node* expr = parse_expression_rec(parse_path_or_unary_expression(), 7);
+
+				return new (alloc_node()) xpath_ast_node(ast_op_negate, xpath_type_number, expr);
+			}
+			else
+				return parse_location_path();
+		}
+
+		struct binary_op_t
+		{
+			ast_type_t asttype;
+			xpath_value_type rettype;
+			int precedence;
+
+			binary_op_t(): asttype(ast_unknown), rettype(xpath_type_none), precedence(0)
+			{
+			}
+
+			binary_op_t(ast_type_t asttype_, xpath_value_type rettype_, int precedence_): asttype(asttype_), rettype(rettype_), precedence(precedence_)
+			{
+			}
+
+			static binary_op_t parse(xpath_lexer& lexer)
+			{
+				switch (lexer.current())
+				{
+				case lex_string:
+					if (lexer.contents() == PUGIXML_TEXT("or"))
+						return binary_op_t(ast_op_or, xpath_type_boolean, 1);
+					else if (lexer.contents() == PUGIXML_TEXT("and"))
+						return binary_op_t(ast_op_and, xpath_type_boolean, 2);
+					else if (lexer.contents() == PUGIXML_TEXT("div"))
+						return binary_op_t(ast_op_divide, xpath_type_number, 6);
+					else if (lexer.contents() == PUGIXML_TEXT("mod"))
+						return binary_op_t(ast_op_mod, xpath_type_number, 6);
+					else
+						return binary_op_t();
+
+				case lex_equal:
+					return binary_op_t(ast_op_equal, xpath_type_boolean, 3);
+
+				case lex_not_equal:
+					return binary_op_t(ast_op_not_equal, xpath_type_boolean, 3);
+
+				case lex_less:
+					return binary_op_t(ast_op_less, xpath_type_boolean, 4);
+
+				case lex_greater:
+					return binary_op_t(ast_op_greater, xpath_type_boolean, 4);
+
+				case lex_less_or_equal:
+					return binary_op_t(ast_op_less_or_equal, xpath_type_boolean, 4);
+
+				case lex_greater_or_equal:
+					return binary_op_t(ast_op_greater_or_equal, xpath_type_boolean, 4);
+
+				case lex_plus:
+					return binary_op_t(ast_op_add, xpath_type_number, 5);
+
+				case lex_minus:
+					return binary_op_t(ast_op_subtract, xpath_type_number, 5);
+
+				case lex_multiply:
+					return binary_op_t(ast_op_multiply, xpath_type_number, 6);
+
+				case lex_union:
+					return binary_op_t(ast_op_union, xpath_type_node_set, 7);
+
+				default:
+					return binary_op_t();
+				}
+			}
+		};
+
+		xpath_ast_node* parse_expression_rec(xpath_ast_node* lhs, int limit)
+		{
+			binary_op_t op = binary_op_t::parse(_lexer);
+
+			while (op.asttype != ast_unknown && op.precedence >= limit)
+			{
+				_lexer.next();
+
+				xpath_ast_node* rhs = parse_path_or_unary_expression();
+
+				binary_op_t nextop = binary_op_t::parse(_lexer);
+
+				while (nextop.asttype != ast_unknown && nextop.precedence > op.precedence)
+				{
+					rhs = parse_expression_rec(rhs, nextop.precedence);
+
+					nextop = binary_op_t::parse(_lexer);
+				}
+
+				if (op.asttype == ast_op_union && (lhs->rettype() != xpath_type_node_set || rhs->rettype() != xpath_type_node_set))
+					throw_error("Union operator has to be applied to node sets");
+
+				lhs = new (alloc_node()) xpath_ast_node(op.asttype, op.rettype, lhs, rhs);
+
+				op = binary_op_t::parse(_lexer);
+			}
+
+			return lhs;
+		}
+
+		// Expr ::= OrExpr
+		// OrExpr ::= AndExpr | OrExpr 'or' AndExpr
+		// AndExpr ::= EqualityExpr | AndExpr 'and' EqualityExpr
+		// EqualityExpr ::= RelationalExpr
+		//					| EqualityExpr '=' RelationalExpr
+		//					| EqualityExpr '!=' RelationalExpr
+		// RelationalExpr ::= AdditiveExpr
+		//					  | RelationalExpr '<' AdditiveExpr
+		//					  | RelationalExpr '>' AdditiveExpr
+		//					  | RelationalExpr '<=' AdditiveExpr
+		//					  | RelationalExpr '>=' AdditiveExpr
+		// AdditiveExpr ::= MultiplicativeExpr
+		//					| AdditiveExpr '+' MultiplicativeExpr
+		//					| AdditiveExpr '-' MultiplicativeExpr
+		// MultiplicativeExpr ::= UnaryExpr
+		//						  | MultiplicativeExpr '*' UnaryExpr
+		//						  | MultiplicativeExpr 'div' UnaryExpr
+		//						  | MultiplicativeExpr 'mod' UnaryExpr
+		xpath_ast_node* parse_expression()
+		{
+			return parse_expression_rec(parse_path_or_unary_expression(), 0);
+		}
+
+		xpath_parser(const char_t* query, xpath_variable_set* variables, xpath_allocator* alloc, xpath_parse_result* result): _alloc(alloc), _lexer(query), _query(query), _variables(variables), _result(result)
+		{
+		}
+
+		xpath_ast_node* parse()
+		{
+			xpath_ast_node* result = parse_expression();
+			
+			if (_lexer.current() != lex_eof)
+			{
+				// there are still unparsed tokens left, error
+				throw_error("Incorrect query");
+			}
+			
+			return result;
+		}
+
+		static xpath_ast_node* parse(const char_t* query, xpath_variable_set* variables, xpath_allocator* alloc, xpath_parse_result* result)
+		{
+			xpath_parser parser(query, variables, alloc, result);
+
+		#ifdef PUGIXML_NO_EXCEPTIONS
+			int error = setjmp(parser._error_handler);
+
+			return (error == 0) ? parser.parse() : 0;
+		#else
+			return parser.parse();
+		#endif
+		}
+	};
+
+	struct xpath_query_impl
+	{
+		static xpath_query_impl* create()
+		{
+			void* memory = xml_memory::allocate(sizeof(xpath_query_impl));
+			if (!memory) return 0;
+
+			return new (memory) xpath_query_impl();
+		}
+
+		static void destroy(xpath_query_impl* impl)
+		{
+			// free all allocated pages
+			impl->alloc.release();
+
+			// free allocator memory (with the first page)
+			xml_memory::deallocate(impl);
+		}
+
+		xpath_query_impl(): root(0), alloc(&block)
+		{
+			block.next = 0;
+			block.capacity = sizeof(block.data);
+		}
+
+		xpath_ast_node* root;
+		xpath_allocator alloc;
+		xpath_memory_block block;
+	};
+
+	PUGI__FN xpath_string evaluate_string_impl(xpath_query_impl* impl, const xpath_node& n, xpath_stack_data& sd)
+	{
+		if (!impl) return xpath_string();
+
+	#ifdef PUGIXML_NO_EXCEPTIONS
+		if (setjmp(sd.error_handler)) return xpath_string();
+	#endif
+
+		xpath_context c(n, 1, 1);
+
+		return impl->root->eval_string(c, sd.stack);
+	}
+
+	PUGI__FN impl::xpath_ast_node* evaluate_node_set_prepare(xpath_query_impl* impl)
+	{
+		if (!impl) return 0;
+
+		if (impl->root->rettype() != xpath_type_node_set)
+		{
+		#ifdef PUGIXML_NO_EXCEPTIONS
+			return 0;
+		#else
+			xpath_parse_result res;
+			res.error = "Expression does not evaluate to node set";
+
+			throw xpath_exception(res);
+		#endif
+		}
+
+		return impl->root;
+	}
+PUGI__NS_END
+
+namespace pugi
+{
+#ifndef PUGIXML_NO_EXCEPTIONS
+	PUGI__FN xpath_exception::xpath_exception(const xpath_parse_result& result_): _result(result_)
+	{
+		assert(_result.error);
+	}
+	
+	PUGI__FN const char* xpath_exception::what() const throw()
+	{
+		return _result.error;
+	}
+
+	PUGI__FN const xpath_parse_result& xpath_exception::result() const
+	{
+		return _result;
+	}
+#endif
+	
+	PUGI__FN xpath_node::xpath_node()
+	{
+	}
+		
+	PUGI__FN xpath_node::xpath_node(const xml_node& node_): _node(node_)
+	{
+	}
+		
+	PUGI__FN xpath_node::xpath_node(const xml_attribute& attribute_, const xml_node& parent_): _node(attribute_ ? parent_ : xml_node()), _attribute(attribute_)
+	{
+	}
+
+	PUGI__FN xml_node xpath_node::node() const
+	{
+		return _attribute ? xml_node() : _node;
+	}
+		
+	PUGI__FN xml_attribute xpath_node::attribute() const
+	{
+		return _attribute;
+	}
+	
+	PUGI__FN xml_node xpath_node::parent() const
+	{
+		return _attribute ? _node : _node.parent();
+	}
+
+	PUGI__FN static void unspecified_bool_xpath_node(xpath_node***)
+	{
+	}
+
+	PUGI__FN xpath_node::operator xpath_node::unspecified_bool_type() const
+	{
+		return (_node || _attribute) ? unspecified_bool_xpath_node : 0;
+	}
+	
+	PUGI__FN bool xpath_node::operator!() const
+	{
+		return !(_node || _attribute);
+	}
+
+	PUGI__FN bool xpath_node::operator==(const xpath_node& n) const
+	{
+		return _node == n._node && _attribute == n._attribute;
+	}
+	
+	PUGI__FN bool xpath_node::operator!=(const xpath_node& n) const
+	{
+		return _node != n._node || _attribute != n._attribute;
+	}
+
+#ifdef __BORLANDC__
+	PUGI__FN bool operator&&(const xpath_node& lhs, bool rhs)
+	{
+		return (bool)lhs && rhs;
+	}
+
+	PUGI__FN bool operator||(const xpath_node& lhs, bool rhs)
+	{
+		return (bool)lhs || rhs;
+	}
+#endif
+
+	PUGI__FN void xpath_node_set::_assign(const_iterator begin_, const_iterator end_, type_t type_)
+	{
+		assert(begin_ <= end_);
+
+		size_t size_ = static_cast<size_t>(end_ - begin_);
+
+		if (size_ <= 1)
+		{
+			// deallocate old buffer
+			if (_begin != &_storage) impl::xml_memory::deallocate(_begin);
+
+			// use internal buffer
+			if (begin_ != end_) _storage = *begin_;
+
+			_begin = &_storage;
+			_end = &_storage + size_;
+			_type = type_;
+		}
+		else
+		{
+			// make heap copy
+			xpath_node* storage = static_cast<xpath_node*>(impl::xml_memory::allocate(size_ * sizeof(xpath_node)));
+
+			if (!storage)
+			{
+			#ifdef PUGIXML_NO_EXCEPTIONS
+				return;
+			#else
+				throw std::bad_alloc();
+			#endif
+			}
+
+			memcpy(storage, begin_, size_ * sizeof(xpath_node));
+			
+			// deallocate old buffer
+			if (_begin != &_storage) impl::xml_memory::deallocate(_begin);
+
+			// finalize
+			_begin = storage;
+			_end = storage + size_;
+			_type = type_;
+		}
+	}
+
+#if __cplusplus >= 201103
+	PUGI__FN void xpath_node_set::_move(xpath_node_set& rhs)
+	{
+		_type = rhs._type;
+		_storage = rhs._storage;
+		_begin = (rhs._begin == &rhs._storage) ? &_storage : rhs._begin;
+		_end = _begin + (rhs._end - rhs._begin);
+
+		rhs._type = type_unsorted;
+		rhs._begin = &rhs._storage;
+		rhs._end = rhs._begin;
+	}
+#endif
+
+	PUGI__FN xpath_node_set::xpath_node_set(): _type(type_unsorted), _begin(&_storage), _end(&_storage)
+	{
+	}
+
+	PUGI__FN xpath_node_set::xpath_node_set(const_iterator begin_, const_iterator end_, type_t type_): _type(type_unsorted), _begin(&_storage), _end(&_storage)
+	{
+		_assign(begin_, end_, type_);
+	}
+
+	PUGI__FN xpath_node_set::~xpath_node_set()
+	{
+		if (_begin != &_storage)
+			impl::xml_memory::deallocate(_begin);
+	}
+		
+	PUGI__FN xpath_node_set::xpath_node_set(const xpath_node_set& ns): _type(type_unsorted), _begin(&_storage), _end(&_storage)
+	{
+		_assign(ns._begin, ns._end, ns._type);
+	}
+	
+	PUGI__FN xpath_node_set& xpath_node_set::operator=(const xpath_node_set& ns)
+	{
+		if (this == &ns) return *this;
+
+		_assign(ns._begin, ns._end, ns._type);
+
+		return *this;
+	}
+
+#if __cplusplus >= 201103
+	PUGI__FN xpath_node_set::xpath_node_set(xpath_node_set&& rhs): _type(type_unsorted), _begin(&_storage), _end(&_storage)
+	{
+		_move(rhs);
+	}
+
+	PUGI__FN xpath_node_set& xpath_node_set::operator=(xpath_node_set&& rhs)
+	{
+		if (this == &rhs) return *this;
+
+		if (_begin != &_storage)
+			impl::xml_memory::deallocate(_begin);
+
+		_move(rhs);
+
+		return *this;
+	}
+#endif
+
+	PUGI__FN xpath_node_set::type_t xpath_node_set::type() const
+	{
+		return _type;
+	}
+		
+	PUGI__FN size_t xpath_node_set::size() const
+	{
+		return _end - _begin;
+	}
+		
+	PUGI__FN bool xpath_node_set::empty() const
+	{
+		return _begin == _end;
+	}
+		
+	PUGI__FN const xpath_node& xpath_node_set::operator[](size_t index) const
+	{
+		assert(index < size());
+		return _begin[index];
+	}
+
+	PUGI__FN xpath_node_set::const_iterator xpath_node_set::begin() const
+	{
+		return _begin;
+	}
+		
+	PUGI__FN xpath_node_set::const_iterator xpath_node_set::end() const
+	{
+		return _end;
+	}
+	
+	PUGI__FN void xpath_node_set::sort(bool reverse)
+	{
+		_type = impl::xpath_sort(_begin, _end, _type, reverse);
+	}
+
+	PUGI__FN xpath_node xpath_node_set::first() const
+	{
+		return impl::xpath_first(_begin, _end, _type);
+	}
+
+	PUGI__FN xpath_parse_result::xpath_parse_result(): error("Internal error"), offset(0)
+	{
+	}
+
+	PUGI__FN xpath_parse_result::operator bool() const
+	{
+		return error == 0;
+	}
+
+	PUGI__FN const char* xpath_parse_result::description() const
+	{
+		return error ? error : "No error";
+	}
+
+	PUGI__FN xpath_variable::xpath_variable(xpath_value_type type_): _type(type_), _next(0)
+	{
+	}
+
+	PUGI__FN const char_t* xpath_variable::name() const
+	{
+		switch (_type)
+		{
+		case xpath_type_node_set:
+			return static_cast<const impl::xpath_variable_node_set*>(this)->name;
+
+		case xpath_type_number:
+			return static_cast<const impl::xpath_variable_number*>(this)->name;
+
+		case xpath_type_string:
+			return static_cast<const impl::xpath_variable_string*>(this)->name;
+
+		case xpath_type_boolean:
+			return static_cast<const impl::xpath_variable_boolean*>(this)->name;
+
+		default:
+			assert(!"Invalid variable type");
+			return 0;
+		}
+	}
+
+	PUGI__FN xpath_value_type xpath_variable::type() const
+	{
+		return _type;
+	}
+
+	PUGI__FN bool xpath_variable::get_boolean() const
+	{
+		return (_type == xpath_type_boolean) ? static_cast<const impl::xpath_variable_boolean*>(this)->value : false;
+	}
+
+	PUGI__FN double xpath_variable::get_number() const
+	{
+		return (_type == xpath_type_number) ? static_cast<const impl::xpath_variable_number*>(this)->value : impl::gen_nan();
+	}
+
+	PUGI__FN const char_t* xpath_variable::get_string() const
+	{
+		const char_t* value = (_type == xpath_type_string) ? static_cast<const impl::xpath_variable_string*>(this)->value : 0;
+		return value ? value : PUGIXML_TEXT("");
+	}
+
+	PUGI__FN const xpath_node_set& xpath_variable::get_node_set() const
+	{
+		return (_type == xpath_type_node_set) ? static_cast<const impl::xpath_variable_node_set*>(this)->value : impl::dummy_node_set;
+	}
+
+	PUGI__FN bool xpath_variable::set(bool value)
+	{
+		if (_type != xpath_type_boolean) return false;
+
+		static_cast<impl::xpath_variable_boolean*>(this)->value = value;
+		return true;
+	}
+
+	PUGI__FN bool xpath_variable::set(double value)
+	{
+		if (_type != xpath_type_number) return false;
+
+		static_cast<impl::xpath_variable_number*>(this)->value = value;
+		return true;
+	}
+
+	PUGI__FN bool xpath_variable::set(const char_t* value)
+	{
+		if (_type != xpath_type_string) return false;
+
+		impl::xpath_variable_string* var = static_cast<impl::xpath_variable_string*>(this);
+
+		// duplicate string
+		size_t size = (impl::strlength(value) + 1) * sizeof(char_t);
+
+		char_t* copy = static_cast<char_t*>(impl::xml_memory::allocate(size));
+		if (!copy) return false;
+
+		memcpy(copy, value, size);
+
+		// replace old string
+		if (var->value) impl::xml_memory::deallocate(var->value);
+		var->value = copy;
+
+		return true;
+	}
+
+	PUGI__FN bool xpath_variable::set(const xpath_node_set& value)
+	{
+		if (_type != xpath_type_node_set) return false;
+
+		static_cast<impl::xpath_variable_node_set*>(this)->value = value;
+		return true;
+	}
+
+	PUGI__FN xpath_variable_set::xpath_variable_set()
+	{
+		for (size_t i = 0; i < sizeof(_data) / sizeof(_data[0]); ++i)
+			_data[i] = 0;
+	}
+
+	PUGI__FN xpath_variable_set::~xpath_variable_set()
+	{
+		for (size_t i = 0; i < sizeof(_data) / sizeof(_data[0]); ++i)
+			_destroy(_data[i]);
+	}
+
+	PUGI__FN xpath_variable_set::xpath_variable_set(const xpath_variable_set& rhs)
+	{
+		for (size_t i = 0; i < sizeof(_data) / sizeof(_data[0]); ++i)
+			_data[i] = 0;
+
+		_assign(rhs);
+	}
+
+	PUGI__FN xpath_variable_set& xpath_variable_set::operator=(const xpath_variable_set& rhs)
+	{
+		if (this == &rhs) return *this;
+
+		_assign(rhs);
+
+		return *this;
+	}
+
+#if __cplusplus >= 201103
+	PUGI__FN xpath_variable_set::xpath_variable_set(xpath_variable_set&& rhs)
+	{
+		for (size_t i = 0; i < sizeof(_data) / sizeof(_data[0]); ++i)
+		{
+			_data[i] = rhs._data[i];
+			rhs._data[i] = 0;
+		}
+	}
+
+	PUGI__FN xpath_variable_set& xpath_variable_set::operator=(xpath_variable_set&& rhs)
+	{
+		for (size_t i = 0; i < sizeof(_data) / sizeof(_data[0]); ++i)
+		{
+			_destroy(_data[i]);
+
+			_data[i] = rhs._data[i];
+			rhs._data[i] = 0;
+		}
+
+		return *this;
+	}
+#endif
+
+	PUGI__FN void xpath_variable_set::_assign(const xpath_variable_set& rhs)
+	{
+		xpath_variable_set temp;
+
+		for (size_t i = 0; i < sizeof(_data) / sizeof(_data[0]); ++i)
+			if (rhs._data[i] && !_clone(rhs._data[i], &temp._data[i]))
+				return;
+
+		_swap(temp);
+	}
+
+	PUGI__FN void xpath_variable_set::_swap(xpath_variable_set& rhs)
+	{
+		for (size_t i = 0; i < sizeof(_data) / sizeof(_data[0]); ++i)
+		{
+			xpath_variable* chain = _data[i];
+
+			_data[i] = rhs._data[i];
+			rhs._data[i] = chain;
+		}
+	}
+
+	PUGI__FN xpath_variable* xpath_variable_set::_find(const char_t* name) const
+	{
+		const size_t hash_size = sizeof(_data) / sizeof(_data[0]);
+		size_t hash = impl::hash_string(name) % hash_size;
+
+		// look for existing variable
+		for (xpath_variable* var = _data[hash]; var; var = var->_next)
+			if (impl::strequal(var->name(), name))
+				return var;
+
+		return 0;
+	}
+
+	PUGI__FN bool xpath_variable_set::_clone(xpath_variable* var, xpath_variable** out_result)
+	{
+		xpath_variable* last = 0;
+
+		while (var)
+		{
+			// allocate storage for new variable
+			xpath_variable* nvar = impl::new_xpath_variable(var->_type, var->name());
+			if (!nvar) return false;
+
+			// link the variable to the result immediately to handle failures gracefully
+			if (last)
+				last->_next = nvar;
+			else
+				*out_result = nvar;
+
+			last = nvar;
+
+			// copy the value; this can fail due to out-of-memory conditions
+			if (!impl::copy_xpath_variable(nvar, var)) return false;
+
+			var = var->_next;
+		}
+
+		return true;
+	}
+
+	PUGI__FN void xpath_variable_set::_destroy(xpath_variable* var)
+	{
+		while (var)
+		{
+			xpath_variable* next = var->_next;
+
+			impl::delete_xpath_variable(var->_type, var);
+
+			var = next;
+		}
+	}
+
+	PUGI__FN xpath_variable* xpath_variable_set::add(const char_t* name, xpath_value_type type)
+	{
+		const size_t hash_size = sizeof(_data) / sizeof(_data[0]);
+		size_t hash = impl::hash_string(name) % hash_size;
+
+		// look for existing variable
+		for (xpath_variable* var = _data[hash]; var; var = var->_next)
+			if (impl::strequal(var->name(), name))
+				return var->type() == type ? var : 0;
+
+		// add new variable
+		xpath_variable* result = impl::new_xpath_variable(type, name);
+
+		if (result)
+		{
+			result->_next = _data[hash];
+
+			_data[hash] = result;
+		}
+
+		return result;
+	}
+
+	PUGI__FN bool xpath_variable_set::set(const char_t* name, bool value)
+	{
+		xpath_variable* var = add(name, xpath_type_boolean);
+		return var ? var->set(value) : false;
+	}
+
+	PUGI__FN bool xpath_variable_set::set(const char_t* name, double value)
+	{
+		xpath_variable* var = add(name, xpath_type_number);
+		return var ? var->set(value) : false;
+	}
+
+	PUGI__FN bool xpath_variable_set::set(const char_t* name, const char_t* value)
+	{
+		xpath_variable* var = add(name, xpath_type_string);
+		return var ? var->set(value) : false;
+	}
+
+	PUGI__FN bool xpath_variable_set::set(const char_t* name, const xpath_node_set& value)
+	{
+		xpath_variable* var = add(name, xpath_type_node_set);
+		return var ? var->set(value) : false;
+	}
+
+	PUGI__FN xpath_variable* xpath_variable_set::get(const char_t* name)
+	{
+		return _find(name);
+	}
+
+	PUGI__FN const xpath_variable* xpath_variable_set::get(const char_t* name) const
+	{
+		return _find(name);
+	}
+
+	PUGI__FN xpath_query::xpath_query(const char_t* query, xpath_variable_set* variables): _impl(0)
+	{
+		impl::xpath_query_impl* qimpl = impl::xpath_query_impl::create();
+
+		if (!qimpl)
+		{
+		#ifdef PUGIXML_NO_EXCEPTIONS
+			_result.error = "Out of memory";
+		#else
+			throw std::bad_alloc();
+		#endif
+		}
+		else
+		{
+			using impl::auto_deleter; // MSVC7 workaround
+			auto_deleter<impl::xpath_query_impl> impl(qimpl, impl::xpath_query_impl::destroy);
+
+			qimpl->root = impl::xpath_parser::parse(query, variables, &qimpl->alloc, &_result);
+
+			if (qimpl->root)
+			{
+				qimpl->root->optimize(&qimpl->alloc);
+
+				_impl = impl.release();
+				_result.error = 0;
+			}
+		}
+	}
+
+	PUGI__FN xpath_query::xpath_query(): _impl(0)
+	{
+	}
+
+	PUGI__FN xpath_query::~xpath_query()
+	{
+		if (_impl)
+			impl::xpath_query_impl::destroy(static_cast<impl::xpath_query_impl*>(_impl));
+	}
+
+#if __cplusplus >= 201103
+	PUGI__FN xpath_query::xpath_query(xpath_query&& rhs)
+	{
+		_impl = rhs._impl;
+		rhs._impl = 0;
+	}
+
+	xpath_query& xpath_query::operator=(xpath_query&& rhs)
+	{
+		if (this == &rhs) return *this;
+
+		if (_impl)
+			impl::xpath_query_impl::destroy(static_cast<impl::xpath_query_impl*>(_impl));
+
+		_impl = rhs._impl;
+		rhs._impl = 0;
+
+		return *this;
+	}
+#endif
+
+	PUGI__FN xpath_value_type xpath_query::return_type() const
+	{
+		if (!_impl) return xpath_type_none;
+
+		return static_cast<impl::xpath_query_impl*>(_impl)->root->rettype();
+	}
+
+	PUGI__FN bool xpath_query::evaluate_boolean(const xpath_node& n) const
+	{
+		if (!_impl) return false;
+		
+		impl::xpath_context c(n, 1, 1);
+		impl::xpath_stack_data sd;
+
+	#ifdef PUGIXML_NO_EXCEPTIONS
+		if (setjmp(sd.error_handler)) return false;
+	#endif
+		
+		return static_cast<impl::xpath_query_impl*>(_impl)->root->eval_boolean(c, sd.stack);
+	}
+	
+	PUGI__FN double xpath_query::evaluate_number(const xpath_node& n) const
+	{
+		if (!_impl) return impl::gen_nan();
+		
+		impl::xpath_context c(n, 1, 1);
+		impl::xpath_stack_data sd;
+
+	#ifdef PUGIXML_NO_EXCEPTIONS
+		if (setjmp(sd.error_handler)) return impl::gen_nan();
+	#endif
+
+		return static_cast<impl::xpath_query_impl*>(_impl)->root->eval_number(c, sd.stack);
+	}
+
+#ifndef PUGIXML_NO_STL
+	PUGI__FN string_t xpath_query::evaluate_string(const xpath_node& n) const
+	{
+		impl::xpath_stack_data sd;
+
+		impl::xpath_string r = impl::evaluate_string_impl(static_cast<impl::xpath_query_impl*>(_impl), n, sd);
+
+		return string_t(r.c_str(), r.length());
+	}
+#endif
+
+	PUGI__FN size_t xpath_query::evaluate_string(char_t* buffer, size_t capacity, const xpath_node& n) const
+	{
+		impl::xpath_stack_data sd;
+
+		impl::xpath_string r = impl::evaluate_string_impl(static_cast<impl::xpath_query_impl*>(_impl), n, sd);
+
+		size_t full_size = r.length() + 1;
+		
+		if (capacity > 0)
+		{
+			size_t size = (full_size < capacity) ? full_size : capacity;
+			assert(size > 0);
+
+			memcpy(buffer, r.c_str(), (size - 1) * sizeof(char_t));
+			buffer[size - 1] = 0;
+		}
+		
+		return full_size;
+	}
+
+	PUGI__FN xpath_node_set xpath_query::evaluate_node_set(const xpath_node& n) const
+	{
+		impl::xpath_ast_node* root = impl::evaluate_node_set_prepare(static_cast<impl::xpath_query_impl*>(_impl));
+		if (!root) return xpath_node_set();
+
+		impl::xpath_context c(n, 1, 1);
+		impl::xpath_stack_data sd;
+
+	#ifdef PUGIXML_NO_EXCEPTIONS
+		if (setjmp(sd.error_handler)) return xpath_node_set();
+	#endif
+
+		impl::xpath_node_set_raw r = root->eval_node_set(c, sd.stack, impl::nodeset_eval_all);
+
+		return xpath_node_set(r.begin(), r.end(), r.type());
+	}
+
+	PUGI__FN xpath_node xpath_query::evaluate_node(const xpath_node& n) const
+	{
+		impl::xpath_ast_node* root = impl::evaluate_node_set_prepare(static_cast<impl::xpath_query_impl*>(_impl));
+		if (!root) return xpath_node();
+
+		impl::xpath_context c(n, 1, 1);
+		impl::xpath_stack_data sd;
+
+	#ifdef PUGIXML_NO_EXCEPTIONS
+		if (setjmp(sd.error_handler)) return xpath_node();
+	#endif
+
+		impl::xpath_node_set_raw r = root->eval_node_set(c, sd.stack, impl::nodeset_eval_first);
+
+		return r.first();
+	}
+
+	PUGI__FN const xpath_parse_result& xpath_query::result() const
+	{
+		return _result;
+	}
+
+	PUGI__FN static void unspecified_bool_xpath_query(xpath_query***)
+	{
+	}
+
+	PUGI__FN xpath_query::operator xpath_query::unspecified_bool_type() const
+	{
+		return _impl ? unspecified_bool_xpath_query : 0;
+	}
+
+	PUGI__FN bool xpath_query::operator!() const
+	{
+		return !_impl;
+	}
+
+	PUGI__FN xpath_node xml_node::select_node(const char_t* query, xpath_variable_set* variables) const
+	{
+		xpath_query q(query, variables);
+		return select_node(q);
+	}
+
+	PUGI__FN xpath_node xml_node::select_node(const xpath_query& query) const
+	{
+		return query.evaluate_node(*this);
+	}
+
+	PUGI__FN xpath_node_set xml_node::select_nodes(const char_t* query, xpath_variable_set* variables) const
+	{
+		xpath_query q(query, variables);
+		return select_nodes(q);
+	}
+
+	PUGI__FN xpath_node_set xml_node::select_nodes(const xpath_query& query) const
+	{
+		return query.evaluate_node_set(*this);
+	}
+
+	PUGI__FN xpath_node xml_node::select_single_node(const char_t* query, xpath_variable_set* variables) const
+	{
+		xpath_query q(query, variables);
+		return select_single_node(q);
+	}
+
+	PUGI__FN xpath_node xml_node::select_single_node(const xpath_query& query) const
+	{
+		return query.evaluate_node(*this);
+	}
+}
+
+#endif
+
+#ifdef __BORLANDC__
+#	pragma option pop
+#endif
+
+// Intel C++ does not properly keep warning state for function templates,
+// so popping warning state at the end of translation unit leads to warnings in the middle.
+#if defined(_MSC_VER) && !defined(__INTEL_COMPILER)
+#	pragma warning(pop)
+#endif
+
+// Undefine all local macros (makes sure we're not leaking macros in header-only mode)
+#undef PUGI__NO_INLINE
+#undef PUGI__UNLIKELY
+#undef PUGI__STATIC_ASSERT
+#undef PUGI__DMC_VOLATILE
+#undef PUGI__MSVC_CRT_VERSION
+#undef PUGI__NS_BEGIN
+#undef PUGI__NS_END
+#undef PUGI__FN
+#undef PUGI__FN_NO_INLINE
+#undef PUGI__GETPAGE_IMPL
+#undef PUGI__GETPAGE
+#undef PUGI__NODETYPE
+#undef PUGI__IS_CHARTYPE_IMPL
+#undef PUGI__IS_CHARTYPE
+#undef PUGI__IS_CHARTYPEX
+#undef PUGI__ENDSWITH
+#undef PUGI__SKIPWS
+#undef PUGI__OPTSET
+#undef PUGI__PUSHNODE
+#undef PUGI__POPNODE
+#undef PUGI__SCANFOR
+#undef PUGI__SCANWHILE
+#undef PUGI__SCANWHILE_UNROLL
+#undef PUGI__ENDSEG
+#undef PUGI__THROW_ERROR
+#undef PUGI__CHECK_ERROR
+
+#endif
+
+/**
+ * Copyright (c) 2006-2015 Arseny Kapoulkine
+ *
+ * Permission is hereby granted, free of charge, to any person
+ * obtaining a copy of this software and associated documentation
+ * files (the "Software"), to deal in the Software without
+ * restriction, including without limitation the rights to use,
+ * copy, modify, merge, publish, distribute, sublicense, and/or sell
+ * copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following
+ * conditions:
+ *
+ * The above copyright notice and this permission notice shall be
+ * included in all copies or substantial portions of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES
+ * OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT
+ * HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY,
+ * WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
+ * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ */
diff --git a/lib/pugixml/pugixml.hpp b/lib/pugixml/pugixml.hpp
new file mode 100644
index 00000000..cdd24b6d
--- /dev/null
+++ b/lib/pugixml/pugixml.hpp
@@ -0,0 +1,1400 @@
+/**
+ * pugixml parser - version 1.6
+ * --------------------------------------------------------
+ * Copyright (C) 2006-2015, by Arseny Kapoulkine (arseny.kapoulkine@gmail.com)
+ * Report bugs and download new versions at http://pugixml.org/
+ *
+ * This library is distributed under the MIT License. See notice at the end
+ * of this file.
+ *
+ * This work is based on the pugxml parser, which is:
+ * Copyright (C) 2003, by Kristen Wegner (kristen@tima.net)
+ */
+
+#ifndef PUGIXML_VERSION
+// Define version macro; evaluates to major * 100 + minor so that it's safe to use in less-than comparisons
+#	define PUGIXML_VERSION 160
+#endif
+
+// Include user configuration file (this can define various configuration macros)
+#include "pugiconfig.hpp"
+
+#ifndef HEADER_PUGIXML_HPP
+#define HEADER_PUGIXML_HPP
+
+// Include stddef.h for size_t and ptrdiff_t
+#include <stddef.h>
+
+// Include exception header for XPath
+#if !defined(PUGIXML_NO_XPATH) && !defined(PUGIXML_NO_EXCEPTIONS)
+#	include <exception>
+#endif
+
+// Include STL headers
+#ifndef PUGIXML_NO_STL
+#	include <iterator>
+#	include <iosfwd>
+#	include <string>
+#endif
+
+// Macro for deprecated features
+#ifndef PUGIXML_DEPRECATED
+#	if defined(__GNUC__)
+#		define PUGIXML_DEPRECATED __attribute__((deprecated))
+#	elif defined(_MSC_VER) && _MSC_VER >= 1300
+#		define PUGIXML_DEPRECATED __declspec(deprecated)
+#	else
+#		define PUGIXML_DEPRECATED
+#	endif
+#endif
+
+// If no API is defined, assume default
+#ifndef PUGIXML_API
+#	define PUGIXML_API
+#endif
+
+// If no API for classes is defined, assume default
+#ifndef PUGIXML_CLASS
+#	define PUGIXML_CLASS PUGIXML_API
+#endif
+
+// If no API for functions is defined, assume default
+#ifndef PUGIXML_FUNCTION
+#	define PUGIXML_FUNCTION PUGIXML_API
+#endif
+
+// If the platform is known to have long long support, enable long long functions
+#ifndef PUGIXML_HAS_LONG_LONG
+#	if __cplusplus >= 201103
+#		define PUGIXML_HAS_LONG_LONG
+#	elif defined(_MSC_VER) && _MSC_VER >= 1400
+#		define PUGIXML_HAS_LONG_LONG
+#	endif
+#endif
+
+// Character interface macros
+#ifdef PUGIXML_WCHAR_MODE
+#	define PUGIXML_TEXT(t) L ## t
+#	define PUGIXML_CHAR wchar_t
+#else
+#	define PUGIXML_TEXT(t) t
+#	define PUGIXML_CHAR char
+#endif
+
+namespace pugi
+{
+	// Character type used for all internal storage and operations; depends on PUGIXML_WCHAR_MODE
+	typedef PUGIXML_CHAR char_t;
+
+#ifndef PUGIXML_NO_STL
+	// String type used for operations that work with STL string; depends on PUGIXML_WCHAR_MODE
+	typedef std::basic_string<PUGIXML_CHAR, std::char_traits<PUGIXML_CHAR>, std::allocator<PUGIXML_CHAR> > string_t;
+#endif
+}
+
+// The PugiXML namespace
+namespace pugi
+{
+	// Tree node types
+	enum xml_node_type
+	{
+		node_null,			// Empty (null) node handle
+		node_document,		// A document tree's absolute root
+		node_element,		// Element tag, i.e. '<node/>'
+		node_pcdata,		// Plain character data, i.e. 'text'
+		node_cdata,			// Character data, i.e. '<![CDATA[text]]>'
+		node_comment,		// Comment tag, i.e. '<!-- text -->'
+		node_pi,			// Processing instruction, i.e. '<?name?>'
+		node_declaration,	// Document declaration, i.e. '<?xml version="1.0"?>'
+		node_doctype		// Document type declaration, i.e. '<!DOCTYPE doc>'
+	};
+
+	// Parsing options
+
+	// Minimal parsing mode (equivalent to turning all other flags off).
+	// Only elements and PCDATA sections are added to the DOM tree, no text conversions are performed.
+	const unsigned int parse_minimal = 0x0000;
+
+	// This flag determines if processing instructions (node_pi) are added to the DOM tree. This flag is off by default.
+	const unsigned int parse_pi = 0x0001;
+
+	// This flag determines if comments (node_comment) are added to the DOM tree. This flag is off by default.
+	const unsigned int parse_comments = 0x0002;
+
+	// This flag determines if CDATA sections (node_cdata) are added to the DOM tree. This flag is on by default.
+	const unsigned int parse_cdata = 0x0004;
+
+	// This flag determines if plain character data (node_pcdata) that consist only of whitespace are added to the DOM tree.
+	// This flag is off by default; turning it on usually results in slower parsing and more memory consumption.
+	const unsigned int parse_ws_pcdata = 0x0008;
+
+	// This flag determines if character and entity references are expanded during parsing. This flag is on by default.
+	const unsigned int parse_escapes = 0x0010;
+
+	// This flag determines if EOL characters are normalized (converted to #xA) during parsing. This flag is on by default.
+	const unsigned int parse_eol = 0x0020;
+	
+	// This flag determines if attribute values are normalized using CDATA normalization rules during parsing. This flag is on by default.
+	const unsigned int parse_wconv_attribute = 0x0040;
+
+	// This flag determines if attribute values are normalized using NMTOKENS normalization rules during parsing. This flag is off by default.
+	const unsigned int parse_wnorm_attribute = 0x0080;
+	
+	// This flag determines if document declaration (node_declaration) is added to the DOM tree. This flag is off by default.
+	const unsigned int parse_declaration = 0x0100;
+
+	// This flag determines if document type declaration (node_doctype) is added to the DOM tree. This flag is off by default.
+	const unsigned int parse_doctype = 0x0200;
+
+	// This flag determines if plain character data (node_pcdata) that is the only child of the parent node and that consists only
+	// of whitespace is added to the DOM tree.
+	// This flag is off by default; turning it on may result in slower parsing and more memory consumption.
+	const unsigned int parse_ws_pcdata_single = 0x0400;
+
+	// This flag determines if leading and trailing whitespace is to be removed from plain character data. This flag is off by default.
+	const unsigned int parse_trim_pcdata = 0x0800;
+
+	// This flag determines if plain character data that does not have a parent node is added to the DOM tree, and if an empty document
+	// is a valid document. This flag is off by default.
+	const unsigned int parse_fragment = 0x1000;
+
+	// The default parsing mode.
+	// Elements, PCDATA and CDATA sections are added to the DOM tree, character/reference entities are expanded,
+	// End-of-Line characters are normalized, attribute values are normalized using CDATA normalization rules.
+	const unsigned int parse_default = parse_cdata | parse_escapes | parse_wconv_attribute | parse_eol;
+
+	// The full parsing mode.
+	// Nodes of all types are added to the DOM tree, character/reference entities are expanded,
+	// End-of-Line characters are normalized, attribute values are normalized using CDATA normalization rules.
+	const unsigned int parse_full = parse_default | parse_pi | parse_comments | parse_declaration | parse_doctype;
+
+	// These flags determine the encoding of input data for XML document
+	enum xml_encoding
+	{
+		encoding_auto,		// Auto-detect input encoding using BOM or < / <? detection; use UTF8 if BOM is not found
+		encoding_utf8,		// UTF8 encoding
+		encoding_utf16_le,	// Little-endian UTF16
+		encoding_utf16_be,	// Big-endian UTF16
+		encoding_utf16,		// UTF16 with native endianness
+		encoding_utf32_le,	// Little-endian UTF32
+		encoding_utf32_be,	// Big-endian UTF32
+		encoding_utf32,		// UTF32 with native endianness
+		encoding_wchar,		// The same encoding wchar_t has (either UTF16 or UTF32)
+		encoding_latin1
+	};
+
+	// Formatting flags
+	
+	// Indent the nodes that are written to output stream with as many indentation strings as deep the node is in DOM tree. This flag is on by default.
+	const unsigned int format_indent = 0x01;
+	
+	// Write encoding-specific BOM to the output stream. This flag is off by default.
+	const unsigned int format_write_bom = 0x02;
+
+	// Use raw output mode (no indentation and no line breaks are written). This flag is off by default.
+	const unsigned int format_raw = 0x04;
+	
+	// Omit default XML declaration even if there is no declaration in the document. This flag is off by default.
+	const unsigned int format_no_declaration = 0x08;
+
+	// Don't escape attribute values and PCDATA contents. This flag is off by default.
+	const unsigned int format_no_escapes = 0x10;
+
+	// Open file using text mode in xml_document::save_file. This enables special character (i.e. new-line) conversions on some systems. This flag is off by default.
+	const unsigned int format_save_file_text = 0x20;
+
+	// Write every attribute on a new line with appropriate indentation. This flag is off by default.
+	const unsigned int format_indent_attributes = 0x40;
+
+	// The default set of formatting flags.
+	// Nodes are indented depending on their depth in DOM tree, a default declaration is output if document has none.
+	const unsigned int format_default = format_indent;
+
+	// Forward declarations
+	struct xml_attribute_struct;
+	struct xml_node_struct;
+
+	class xml_node_iterator;
+	class xml_attribute_iterator;
+	class xml_named_node_iterator;
+
+	class xml_tree_walker;
+
+	struct xml_parse_result;
+
+	class xml_node;
+
+	class xml_text;
+	
+	#ifndef PUGIXML_NO_XPATH
+	class xpath_node;
+	class xpath_node_set;
+	class xpath_query;
+	class xpath_variable_set;
+	#endif
+
+	// Range-based for loop support
+	template <typename It> class xml_object_range
+	{
+	public:
+		typedef It const_iterator;
+		typedef It iterator;
+
+		xml_object_range(It b, It e): _begin(b), _end(e)
+		{
+		}
+
+		It begin() const { return _begin; }
+		It end() const { return _end; }
+
+	private:
+		It _begin, _end;
+	};
+
+	// Writer interface for node printing (see xml_node::print)
+	class PUGIXML_CLASS xml_writer
+	{
+	public:
+		virtual ~xml_writer() {}
+
+		// Write memory chunk into stream/file/whatever
+		virtual void write(const void* data, size_t size) = 0;
+	};
+
+	// xml_writer implementation for FILE*
+	class PUGIXML_CLASS xml_writer_file: public xml_writer
+	{
+	public:
+		// Construct writer from a FILE* object; void* is used to avoid header dependencies on stdio
+		xml_writer_file(void* file);
+
+		virtual void write(const void* data, size_t size);
+
+	private:
+		void* file;
+	};
+
+	#ifndef PUGIXML_NO_STL
+	// xml_writer implementation for streams
+	class PUGIXML_CLASS xml_writer_stream: public xml_writer
+	{
+	public:
+		// Construct writer from an output stream object
+		xml_writer_stream(std::basic_ostream<char, std::char_traits<char> >& stream);
+		xml_writer_stream(std::basic_ostream<wchar_t, std::char_traits<wchar_t> >& stream);
+
+		virtual void write(const void* data, size_t size);
+
+	private:
+		std::basic_ostream<char, std::char_traits<char> >* narrow_stream;
+		std::basic_ostream<wchar_t, std::char_traits<wchar_t> >* wide_stream;
+	};
+	#endif
+
+	// A light-weight handle for manipulating attributes in DOM tree
+	class PUGIXML_CLASS xml_attribute
+	{
+		friend class xml_attribute_iterator;
+		friend class xml_node;
+
+	private:
+		xml_attribute_struct* _attr;
+	
+		typedef void (*unspecified_bool_type)(xml_attribute***);
+
+	public:
+		// Default constructor. Constructs an empty attribute.
+		xml_attribute();
+		
+		// Constructs attribute from internal pointer
+		explicit xml_attribute(xml_attribute_struct* attr);
+
+		// Safe bool conversion operator
+		operator unspecified_bool_type() const;
+
+		// Borland C++ workaround
+		bool operator!() const;
+
+		// Comparison operators (compares wrapped attribute pointers)
+		bool operator==(const xml_attribute& r) const;
+		bool operator!=(const xml_attribute& r) const;
+		bool operator<(const xml_attribute& r) const;
+		bool operator>(const xml_attribute& r) const;
+		bool operator<=(const xml_attribute& r) const;
+		bool operator>=(const xml_attribute& r) const;
+
+		// Check if attribute is empty
+		bool empty() const;
+
+		// Get attribute name/value, or "" if attribute is empty
+		const char_t* name() const;
+		const char_t* value() const;
+
+		// Get attribute value, or the default value if attribute is empty
+		const char_t* as_string(const char_t* def = PUGIXML_TEXT("")) const;
+
+		// Get attribute value as a number, or the default value if conversion did not succeed or attribute is empty
+		int as_int(int def = 0) const;
+		unsigned int as_uint(unsigned int def = 0) const;
+		double as_double(double def = 0) const;
+		float as_float(float def = 0) const;
+
+	#ifdef PUGIXML_HAS_LONG_LONG
+		long long as_llong(long long def = 0) const;
+		unsigned long long as_ullong(unsigned long long def = 0) const;
+	#endif
+
+		// Get attribute value as bool (returns true if first character is in '1tTyY' set), or the default value if attribute is empty
+		bool as_bool(bool def = false) const;
+
+		// Set attribute name/value (returns false if attribute is empty or there is not enough memory)
+		bool set_name(const char_t* rhs);
+		bool set_value(const char_t* rhs);
+
+		// Set attribute value with type conversion (numbers are converted to strings, boolean is converted to "true"/"false")
+		bool set_value(int rhs);
+		bool set_value(unsigned int rhs);
+		bool set_value(double rhs);
+		bool set_value(float rhs);
+		bool set_value(bool rhs);
+
+	#ifdef PUGIXML_HAS_LONG_LONG
+		bool set_value(long long rhs);
+		bool set_value(unsigned long long rhs);
+	#endif
+
+		// Set attribute value (equivalent to set_value without error checking)
+		xml_attribute& operator=(const char_t* rhs);
+		xml_attribute& operator=(int rhs);
+		xml_attribute& operator=(unsigned int rhs);
+		xml_attribute& operator=(double rhs);
+		xml_attribute& operator=(float rhs);
+		xml_attribute& operator=(bool rhs);
+
+	#ifdef PUGIXML_HAS_LONG_LONG
+		xml_attribute& operator=(long long rhs);
+		xml_attribute& operator=(unsigned long long rhs);
+	#endif
+
+		// Get next/previous attribute in the attribute list of the parent node
+		xml_attribute next_attribute() const;
+		xml_attribute previous_attribute() const;
+
+		// Get hash value (unique for handles to the same object)
+		size_t hash_value() const;
+
+		// Get internal pointer
+		xml_attribute_struct* internal_object() const;
+	};
+
+#ifdef __BORLANDC__
+	// Borland C++ workaround
+	bool PUGIXML_FUNCTION operator&&(const xml_attribute& lhs, bool rhs);
+	bool PUGIXML_FUNCTION operator||(const xml_attribute& lhs, bool rhs);
+#endif
+
+	// A light-weight handle for manipulating nodes in DOM tree
+	class PUGIXML_CLASS xml_node
+	{
+		friend class xml_attribute_iterator;
+		friend class xml_node_iterator;
+		friend class xml_named_node_iterator;
+
+	protected:
+		xml_node_struct* _root;
+
+		typedef void (*unspecified_bool_type)(xml_node***);
+
+	public:
+		// Default constructor. Constructs an empty node.
+		xml_node();
+
+		// Constructs node from internal pointer
+		explicit xml_node(xml_node_struct* p);
+
+		// Safe bool conversion operator
+		operator unspecified_bool_type() const;
+
+		// Borland C++ workaround
+		bool operator!() const;
+	
+		// Comparison operators (compares wrapped node pointers)
+		bool operator==(const xml_node& r) const;
+		bool operator!=(const xml_node& r) const;
+		bool operator<(const xml_node& r) const;
+		bool operator>(const xml_node& r) const;
+		bool operator<=(const xml_node& r) const;
+		bool operator>=(const xml_node& r) const;
+
+		// Check if node is empty.
+		bool empty() const;
+
+		// Get node type
+		xml_node_type type() const;
+
+		// Get node name, or "" if node is empty or it has no name
+		const char_t* name() const;
+
+		// Get node value, or "" if node is empty or it has no value
+		// Note: For <node>text</node> node.value() does not return "text"! Use child_value() or text() methods to access text inside nodes.
+		const char_t* value() const;
+	
+		// Get attribute list
+		xml_attribute first_attribute() const;
+		xml_attribute last_attribute() const;
+
+		// Get children list
+		xml_node first_child() const;
+		xml_node last_child() const;
+
+		// Get next/previous sibling in the children list of the parent node
+		xml_node next_sibling() const;
+		xml_node previous_sibling() const;
+		
+		// Get parent node
+		xml_node parent() const;
+
+		// Get root of DOM tree this node belongs to
+		xml_node root() const;
+
+		// Get text object for the current node
+		xml_text text() const;
+
+		// Get child, attribute or next/previous sibling with the specified name
+		xml_node child(const char_t* name) const;
+		xml_attribute attribute(const char_t* name) const;
+		xml_node next_sibling(const char_t* name) const;
+		xml_node previous_sibling(const char_t* name) const;
+
+		// Get attribute, starting the search from a hint (and updating hint so that searching for a sequence of attributes is fast)
+		xml_attribute attribute(const char_t* name, xml_attribute& hint) const;
+
+		// Get child value of current node; that is, value of the first child node of type PCDATA/CDATA
+		const char_t* child_value() const;
+
+		// Get child value of child with specified name. Equivalent to child(name).child_value().
+		const char_t* child_value(const char_t* name) const;
+
+		// Set node name/value (returns false if node is empty, there is not enough memory, or node can not have name/value)
+		bool set_name(const char_t* rhs);
+		bool set_value(const char_t* rhs);
+		
+		// Add attribute with specified name. Returns added attribute, or empty attribute on errors.
+		xml_attribute append_attribute(const char_t* name);
+		xml_attribute prepend_attribute(const char_t* name);
+		xml_attribute insert_attribute_after(const char_t* name, const xml_attribute& attr);
+		xml_attribute insert_attribute_before(const char_t* name, const xml_attribute& attr);
+
+		// Add a copy of the specified attribute. Returns added attribute, or empty attribute on errors.
+		xml_attribute append_copy(const xml_attribute& proto);
+		xml_attribute prepend_copy(const xml_attribute& proto);
+		xml_attribute insert_copy_after(const xml_attribute& proto, const xml_attribute& attr);
+		xml_attribute insert_copy_before(const xml_attribute& proto, const xml_attribute& attr);
+
+		// Add child node with specified type. Returns added node, or empty node on errors.
+		xml_node append_child(xml_node_type type = node_element);
+		xml_node prepend_child(xml_node_type type = node_element);
+		xml_node insert_child_after(xml_node_type type, const xml_node& node);
+		xml_node insert_child_before(xml_node_type type, const xml_node& node);
+
+		// Add child element with specified name. Returns added node, or empty node on errors.
+		xml_node append_child(const char_t* name);
+		xml_node prepend_child(const char_t* name);
+		xml_node insert_child_after(const char_t* name, const xml_node& node);
+		xml_node insert_child_before(const char_t* name, const xml_node& node);
+
+		// Add a copy of the specified node as a child. Returns added node, or empty node on errors.
+		xml_node append_copy(const xml_node& proto);
+		xml_node prepend_copy(const xml_node& proto);
+		xml_node insert_copy_after(const xml_node& proto, const xml_node& node);
+		xml_node insert_copy_before(const xml_node& proto, const xml_node& node);
+
+		// Move the specified node to become a child of this node. Returns moved node, or empty node on errors.
+		xml_node append_move(const xml_node& moved);
+		xml_node prepend_move(const xml_node& moved);
+		xml_node insert_move_after(const xml_node& moved, const xml_node& node);
+		xml_node insert_move_before(const xml_node& moved, const xml_node& node);
+
+		// Remove specified attribute
+		bool remove_attribute(const xml_attribute& a);
+		bool remove_attribute(const char_t* name);
+
+		// Remove specified child
+		bool remove_child(const xml_node& n);
+		bool remove_child(const char_t* name);
+
+		// Parses buffer as an XML document fragment and appends all nodes as children of the current node.
+		// Copies/converts the buffer, so it may be deleted or changed after the function returns.
+		// Note: append_buffer allocates memory that has the lifetime of the owning document; removing the appended nodes does not immediately reclaim that memory.
+		xml_parse_result append_buffer(const void* contents, size_t size, unsigned int options = parse_default, xml_encoding encoding = encoding_auto);
+
+		// Find attribute using predicate. Returns first attribute for which predicate returned true.
+		template <typename Predicate> xml_attribute find_attribute(Predicate pred) const
+		{
+			if (!_root) return xml_attribute();
+			
+			for (xml_attribute attrib = first_attribute(); attrib; attrib = attrib.next_attribute())
+				if (pred(attrib))
+					return attrib;
+		
+			return xml_attribute();
+		}
+
+		// Find child node using predicate. Returns first child for which predicate returned true.
+		template <typename Predicate> xml_node find_child(Predicate pred) const
+		{
+			if (!_root) return xml_node();
+	
+			for (xml_node node = first_child(); node; node = node.next_sibling())
+				if (pred(node))
+					return node;
+		
+			return xml_node();
+		}
+
+		// Find node from subtree using predicate. Returns first node from subtree (depth-first), for which predicate returned true.
+		template <typename Predicate> xml_node find_node(Predicate pred) const
+		{
+			if (!_root) return xml_node();
+
+			xml_node cur = first_child();
+			
+			while (cur._root && cur._root != _root)
+			{
+				if (pred(cur)) return cur;
+
+				if (cur.first_child()) cur = cur.first_child();
+				else if (cur.next_sibling()) cur = cur.next_sibling();
+				else
+				{
+					while (!cur.next_sibling() && cur._root != _root) cur = cur.parent();
+
+					if (cur._root != _root) cur = cur.next_sibling();
+				}
+			}
+
+			return xml_node();
+		}
+
+		// Find child node by attribute name/value
+		xml_node find_child_by_attribute(const char_t* name, const char_t* attr_name, const char_t* attr_value) const;
+		xml_node find_child_by_attribute(const char_t* attr_name, const char_t* attr_value) const;
+
+	#ifndef PUGIXML_NO_STL
+		// Get the absolute node path from root as a text string.
+		string_t path(char_t delimiter = '/') const;
+	#endif
+
+		// Search for a node by path consisting of node names and . or .. elements.
+		xml_node first_element_by_path(const char_t* path, char_t delimiter = '/') const;
+
+		// Recursively traverse subtree with xml_tree_walker
+		bool traverse(xml_tree_walker& walker);
+	
+	#ifndef PUGIXML_NO_XPATH
+		// Select single node by evaluating XPath query. Returns first node from the resulting node set.
+		xpath_node select_node(const char_t* query, xpath_variable_set* variables = 0) const;
+		xpath_node select_node(const xpath_query& query) const;
+
+		// Select node set by evaluating XPath query
+		xpath_node_set select_nodes(const char_t* query, xpath_variable_set* variables = 0) const;
+		xpath_node_set select_nodes(const xpath_query& query) const;
+
+		// (deprecated: use select_node instead) Select single node by evaluating XPath query.
+		xpath_node select_single_node(const char_t* query, xpath_variable_set* variables = 0) const;
+		xpath_node select_single_node(const xpath_query& query) const;
+
+	#endif
+		
+		// Print subtree using a writer object
+		void print(xml_writer& writer, const char_t* indent = PUGIXML_TEXT("\t"), unsigned int flags = format_default, xml_encoding encoding = encoding_auto, unsigned int depth = 0) const;
+
+	#ifndef PUGIXML_NO_STL
+		// Print subtree to stream
+		void print(std::basic_ostream<char, std::char_traits<char> >& os, const char_t* indent = PUGIXML_TEXT("\t"), unsigned int flags = format_default, xml_encoding encoding = encoding_auto, unsigned int depth = 0) const;
+		void print(std::basic_ostream<wchar_t, std::char_traits<wchar_t> >& os, const char_t* indent = PUGIXML_TEXT("\t"), unsigned int flags = format_default, unsigned int depth = 0) const;
+	#endif
+
+		// Child nodes iterators
+		typedef xml_node_iterator iterator;
+
+		iterator begin() const;
+		iterator end() const;
+
+		// Attribute iterators
+		typedef xml_attribute_iterator attribute_iterator;
+
+		attribute_iterator attributes_begin() const;
+		attribute_iterator attributes_end() const;
+
+		// Range-based for support
+		xml_object_range<xml_node_iterator> children() const;
+		xml_object_range<xml_named_node_iterator> children(const char_t* name) const;
+		xml_object_range<xml_attribute_iterator> attributes() const;
+
+		// Get node offset in parsed file/string (in char_t units) for debugging purposes
+		ptrdiff_t offset_debug() const;
+
+		// Get hash value (unique for handles to the same object)
+		size_t hash_value() const;
+
+		// Get internal pointer
+		xml_node_struct* internal_object() const;
+	};
+
+#ifdef __BORLANDC__
+	// Borland C++ workaround
+	bool PUGIXML_FUNCTION operator&&(const xml_node& lhs, bool rhs);
+	bool PUGIXML_FUNCTION operator||(const xml_node& lhs, bool rhs);
+#endif
+
+	// A helper for working with text inside PCDATA nodes
+	class PUGIXML_CLASS xml_text
+	{
+		friend class xml_node;
+
+		xml_node_struct* _root;
+
+		typedef void (*unspecified_bool_type)(xml_text***);
+
+		explicit xml_text(xml_node_struct* root);
+
+		xml_node_struct* _data_new();
+		xml_node_struct* _data() const;
+
+	public:
+		// Default constructor. Constructs an empty object.
+		xml_text();
+
+		// Safe bool conversion operator
+		operator unspecified_bool_type() const;
+
+		// Borland C++ workaround
+		bool operator!() const;
+
+		// Check if text object is empty
+		bool empty() const;
+
+		// Get text, or "" if object is empty
+		const char_t* get() const;
+
+		// Get text, or the default value if object is empty
+		const char_t* as_string(const char_t* def = PUGIXML_TEXT("")) const;
+
+		// Get text as a number, or the default value if conversion did not succeed or object is empty
+		int as_int(int def = 0) const;
+		unsigned int as_uint(unsigned int def = 0) const;
+		double as_double(double def = 0) const;
+		float as_float(float def = 0) const;
+
+	#ifdef PUGIXML_HAS_LONG_LONG
+		long long as_llong(long long def = 0) const;
+		unsigned long long as_ullong(unsigned long long def = 0) const;
+	#endif
+
+		// Get text as bool (returns true if first character is in '1tTyY' set), or the default value if object is empty
+		bool as_bool(bool def = false) const;
+
+		// Set text (returns false if object is empty or there is not enough memory)
+		bool set(const char_t* rhs);
+
+		// Set text with type conversion (numbers are converted to strings, boolean is converted to "true"/"false")
+		bool set(int rhs);
+		bool set(unsigned int rhs);
+		bool set(double rhs);
+		bool set(float rhs);
+		bool set(bool rhs);
+
+	#ifdef PUGIXML_HAS_LONG_LONG
+		bool set(long long rhs);
+		bool set(unsigned long long rhs);
+	#endif
+
+		// Set text (equivalent to set without error checking)
+		xml_text& operator=(const char_t* rhs);
+		xml_text& operator=(int rhs);
+		xml_text& operator=(unsigned int rhs);
+		xml_text& operator=(double rhs);
+		xml_text& operator=(float rhs);
+		xml_text& operator=(bool rhs);
+
+	#ifdef PUGIXML_HAS_LONG_LONG
+		xml_text& operator=(long long rhs);
+		xml_text& operator=(unsigned long long rhs);
+	#endif
+
+		// Get the data node (node_pcdata or node_cdata) for this object
+		xml_node data() const;
+	};
+
+#ifdef __BORLANDC__
+	// Borland C++ workaround
+	bool PUGIXML_FUNCTION operator&&(const xml_text& lhs, bool rhs);
+	bool PUGIXML_FUNCTION operator||(const xml_text& lhs, bool rhs);
+#endif
+
+	// Child node iterator (a bidirectional iterator over a collection of xml_node)
+	class PUGIXML_CLASS xml_node_iterator
+	{
+		friend class xml_node;
+
+	private:
+		mutable xml_node _wrap;
+		xml_node _parent;
+
+		xml_node_iterator(xml_node_struct* ref, xml_node_struct* parent);
+
+	public:
+		// Iterator traits
+		typedef ptrdiff_t difference_type;
+		typedef xml_node value_type;
+		typedef xml_node* pointer;
+		typedef xml_node& reference;
+
+	#ifndef PUGIXML_NO_STL
+		typedef std::bidirectional_iterator_tag iterator_category;
+	#endif
+
+		// Default constructor
+		xml_node_iterator();
+
+		// Construct an iterator which points to the specified node
+		xml_node_iterator(const xml_node& node);
+
+		// Iterator operators
+		bool operator==(const xml_node_iterator& rhs) const;
+		bool operator!=(const xml_node_iterator& rhs) const;
+
+		xml_node& operator*() const;
+		xml_node* operator->() const;
+
+		const xml_node_iterator& operator++();
+		xml_node_iterator operator++(int);
+
+		const xml_node_iterator& operator--();
+		xml_node_iterator operator--(int);
+	};
+
+	// Attribute iterator (a bidirectional iterator over a collection of xml_attribute)
+	class PUGIXML_CLASS xml_attribute_iterator
+	{
+		friend class xml_node;
+
+	private:
+		mutable xml_attribute _wrap;
+		xml_node _parent;
+
+		xml_attribute_iterator(xml_attribute_struct* ref, xml_node_struct* parent);
+
+	public:
+		// Iterator traits
+		typedef ptrdiff_t difference_type;
+		typedef xml_attribute value_type;
+		typedef xml_attribute* pointer;
+		typedef xml_attribute& reference;
+
+	#ifndef PUGIXML_NO_STL
+		typedef std::bidirectional_iterator_tag iterator_category;
+	#endif
+
+		// Default constructor
+		xml_attribute_iterator();
+
+		// Construct an iterator which points to the specified attribute
+		xml_attribute_iterator(const xml_attribute& attr, const xml_node& parent);
+
+		// Iterator operators
+		bool operator==(const xml_attribute_iterator& rhs) const;
+		bool operator!=(const xml_attribute_iterator& rhs) const;
+
+		xml_attribute& operator*() const;
+		xml_attribute* operator->() const;
+
+		const xml_attribute_iterator& operator++();
+		xml_attribute_iterator operator++(int);
+
+		const xml_attribute_iterator& operator--();
+		xml_attribute_iterator operator--(int);
+	};
+
+	// Named node range helper
+	class PUGIXML_CLASS xml_named_node_iterator
+	{
+		friend class xml_node;
+
+	public:
+		// Iterator traits
+		typedef ptrdiff_t difference_type;
+		typedef xml_node value_type;
+		typedef xml_node* pointer;
+		typedef xml_node& reference;
+
+	#ifndef PUGIXML_NO_STL
+		typedef std::bidirectional_iterator_tag iterator_category;
+	#endif
+
+		// Default constructor
+		xml_named_node_iterator();
+
+		// Construct an iterator which points to the specified node
+		xml_named_node_iterator(const xml_node& node, const char_t* name);
+
+		// Iterator operators
+		bool operator==(const xml_named_node_iterator& rhs) const;
+		bool operator!=(const xml_named_node_iterator& rhs) const;
+
+		xml_node& operator*() const;
+		xml_node* operator->() const;
+
+		const xml_named_node_iterator& operator++();
+		xml_named_node_iterator operator++(int);
+
+		const xml_named_node_iterator& operator--();
+		xml_named_node_iterator operator--(int);
+
+	private:
+		mutable xml_node _wrap;
+		xml_node _parent;
+		const char_t* _name;
+
+		xml_named_node_iterator(xml_node_struct* ref, xml_node_struct* parent, const char_t* name);
+	};
+
+	// Abstract tree walker class (see xml_node::traverse)
+	class PUGIXML_CLASS xml_tree_walker
+	{
+		friend class xml_node;
+
+	private:
+		int _depth;
+	
+	protected:
+		// Get current traversal depth
+		int depth() const;
+	
+	public:
+		xml_tree_walker();
+		virtual ~xml_tree_walker();
+
+		// Callback that is called when traversal begins
+		virtual bool begin(xml_node& node);
+
+		// Callback that is called for each node traversed
+		virtual bool for_each(xml_node& node) = 0;
+
+		// Callback that is called when traversal ends
+		virtual bool end(xml_node& node);
+	};
+
+	// Parsing status, returned as part of xml_parse_result object
+	enum xml_parse_status
+	{
+		status_ok = 0,				// No error
+
+		status_file_not_found,		// File was not found during load_file()
+		status_io_error,			// Error reading from file/stream
+		status_out_of_memory,		// Could not allocate memory
+		status_internal_error,		// Internal error occurred
+
+		status_unrecognized_tag,	// Parser could not determine tag type
+
+		status_bad_pi,				// Parsing error occurred while parsing document declaration/processing instruction
+		status_bad_comment,			// Parsing error occurred while parsing comment
+		status_bad_cdata,			// Parsing error occurred while parsing CDATA section
+		status_bad_doctype,			// Parsing error occurred while parsing document type declaration
+		status_bad_pcdata,			// Parsing error occurred while parsing PCDATA section
+		status_bad_start_element,	// Parsing error occurred while parsing start element tag
+		status_bad_attribute,		// Parsing error occurred while parsing element attribute
+		status_bad_end_element,		// Parsing error occurred while parsing end element tag
+		status_end_element_mismatch,// There was a mismatch of start-end tags (closing tag had incorrect name, some tag was not closed or there was an excessive closing tag)
+
+		status_append_invalid_root,	// Unable to append nodes since root type is not node_element or node_document (exclusive to xml_node::append_buffer)
+
+		status_no_document_element	// Parsing resulted in a document without element nodes
+	};
+
+	// Parsing result
+	struct PUGIXML_CLASS xml_parse_result
+	{
+		// Parsing status (see xml_parse_status)
+		xml_parse_status status;
+
+		// Last parsed offset (in char_t units from start of input data)
+		ptrdiff_t offset;
+
+		// Source document encoding
+		xml_encoding encoding;
+
+		// Default constructor, initializes object to failed state
+		xml_parse_result();
+
+		// Cast to bool operator
+		operator bool() const;
+
+		// Get error description
+		const char* description() const;
+	};
+
+	// Document class (DOM tree root)
+	class PUGIXML_CLASS xml_document: public xml_node
+	{
+	private:
+		char_t* _buffer;
+
+		char _memory[192];
+		
+		// Non-copyable semantics
+		xml_document(const xml_document&);
+		const xml_document& operator=(const xml_document&);
+
+		void create();
+		void destroy();
+
+	public:
+		// Default constructor, makes empty document
+		xml_document();
+
+		// Destructor, invalidates all node/attribute handles to this document
+		~xml_document();
+
+		// Removes all nodes, leaving the empty document
+		void reset();
+
+		// Removes all nodes, then copies the entire contents of the specified document
+		void reset(const xml_document& proto);
+
+	#ifndef PUGIXML_NO_STL
+		// Load document from stream.
+		xml_parse_result load(std::basic_istream<char, std::char_traits<char> >& stream, unsigned int options = parse_default, xml_encoding encoding = encoding_auto);
+		xml_parse_result load(std::basic_istream<wchar_t, std::char_traits<wchar_t> >& stream, unsigned int options = parse_default);
+	#endif
+
+		// (deprecated: use load_string instead) Load document from zero-terminated string. No encoding conversions are applied.
+		xml_parse_result load(const char_t* contents, unsigned int options = parse_default);
+
+		// Load document from zero-terminated string. No encoding conversions are applied.
+		xml_parse_result load_string(const char_t* contents, unsigned int options = parse_default);
+
+		// Load document from file
+		xml_parse_result load_file(const char* path, unsigned int options = parse_default, xml_encoding encoding = encoding_auto);
+		xml_parse_result load_file(const wchar_t* path, unsigned int options = parse_default, xml_encoding encoding = encoding_auto);
+
+		// Load document from buffer. Copies/converts the buffer, so it may be deleted or changed after the function returns.
+		xml_parse_result load_buffer(const void* contents, size_t size, unsigned int options = parse_default, xml_encoding encoding = encoding_auto);
+
+		// Load document from buffer, using the buffer for in-place parsing (the buffer is modified and used for storage of document data).
+		// You should ensure that buffer data will persist throughout the document's lifetime, and free the buffer memory manually once document is destroyed.
+		xml_parse_result load_buffer_inplace(void* contents, size_t size, unsigned int options = parse_default, xml_encoding encoding = encoding_auto);
+
+		// Load document from buffer, using the buffer for in-place parsing (the buffer is modified and used for storage of document data).
+		// You should allocate the buffer with pugixml allocation function; document will free the buffer when it is no longer needed (you can't use it anymore).
+		xml_parse_result load_buffer_inplace_own(void* contents, size_t size, unsigned int options = parse_default, xml_encoding encoding = encoding_auto);
+
+		// Save XML document to writer (semantics is slightly different from xml_node::print, see documentation for details).
+		void save(xml_writer& writer, const char_t* indent = PUGIXML_TEXT("\t"), unsigned int flags = format_default, xml_encoding encoding = encoding_auto) const;
+
+	#ifndef PUGIXML_NO_STL
+		// Save XML document to stream (semantics is slightly different from xml_node::print, see documentation for details).
+		void save(std::basic_ostream<char, std::char_traits<char> >& stream, const char_t* indent = PUGIXML_TEXT("\t"), unsigned int flags = format_default, xml_encoding encoding = encoding_auto) const;
+		void save(std::basic_ostream<wchar_t, std::char_traits<wchar_t> >& stream, const char_t* indent = PUGIXML_TEXT("\t"), unsigned int flags = format_default) const;
+	#endif
+
+		// Save XML to file
+		bool save_file(const char* path, const char_t* indent = PUGIXML_TEXT("\t"), unsigned int flags = format_default, xml_encoding encoding = encoding_auto) const;
+		bool save_file(const wchar_t* path, const char_t* indent = PUGIXML_TEXT("\t"), unsigned int flags = format_default, xml_encoding encoding = encoding_auto) const;
+
+		// Get document element
+		xml_node document_element() const;
+	};
+
+#ifndef PUGIXML_NO_XPATH
+	// XPath query return type
+	enum xpath_value_type
+	{
+		xpath_type_none,	  // Unknown type (query failed to compile)
+		xpath_type_node_set,  // Node set (xpath_node_set)
+		xpath_type_number,	  // Number
+		xpath_type_string,	  // String
+		xpath_type_boolean	  // Boolean
+	};
+
+	// XPath parsing result
+	struct PUGIXML_CLASS xpath_parse_result
+	{
+		// Error message (0 if no error)
+		const char* error;
+
+		// Last parsed offset (in char_t units from string start)
+		ptrdiff_t offset;
+
+		// Default constructor, initializes object to failed state
+		xpath_parse_result();
+
+		// Cast to bool operator
+		operator bool() const;
+
+		// Get error description
+		const char* description() const;
+	};
+
+	// A single XPath variable
+	class PUGIXML_CLASS xpath_variable
+	{
+		friend class xpath_variable_set;
+
+	protected:
+		xpath_value_type _type;
+		xpath_variable* _next;
+
+		xpath_variable(xpath_value_type type);
+
+		// Non-copyable semantics
+		xpath_variable(const xpath_variable&);
+		xpath_variable& operator=(const xpath_variable&);
+		
+	public:
+		// Get variable name
+		const char_t* name() const;
+
+		// Get variable type
+		xpath_value_type type() const;
+
+		// Get variable value; no type conversion is performed, default value (false, NaN, empty string, empty node set) is returned on type mismatch error
+		bool get_boolean() const;
+		double get_number() const;
+		const char_t* get_string() const;
+		const xpath_node_set& get_node_set() const;
+
+		// Set variable value; no type conversion is performed, false is returned on type mismatch error
+		bool set(bool value);
+		bool set(double value);
+		bool set(const char_t* value);
+		bool set(const xpath_node_set& value);
+	};
+
+	// A set of XPath variables
+	class PUGIXML_CLASS xpath_variable_set
+	{
+	private:
+		xpath_variable* _data[64];
+
+		void _assign(const xpath_variable_set& rhs);
+		void _swap(xpath_variable_set& rhs);
+
+		xpath_variable* _find(const char_t* name) const;
+
+		static bool _clone(xpath_variable* var, xpath_variable** out_result);
+		static void _destroy(xpath_variable* var);
+
+	public:
+		// Default constructor/destructor
+		xpath_variable_set();
+		~xpath_variable_set();
+
+		// Copy constructor/assignment operator
+		xpath_variable_set(const xpath_variable_set& rhs);
+		xpath_variable_set& operator=(const xpath_variable_set& rhs);
+
+	#if __cplusplus >= 201103
+		// Move semantics support
+		xpath_variable_set(xpath_variable_set&& rhs);
+		xpath_variable_set& operator=(xpath_variable_set&& rhs);
+	#endif
+
+		// Add a new variable or get the existing one, if the types match
+		xpath_variable* add(const char_t* name, xpath_value_type type);
+
+		// Set value of an existing variable; no type conversion is performed, false is returned if there is no such variable or if types mismatch
+		bool set(const char_t* name, bool value);
+		bool set(const char_t* name, double value);
+		bool set(const char_t* name, const char_t* value);
+		bool set(const char_t* name, const xpath_node_set& value);
+
+		// Get existing variable by name
+		xpath_variable* get(const char_t* name);
+		const xpath_variable* get(const char_t* name) const;
+	};
+
+	// A compiled XPath query object
+	class PUGIXML_CLASS xpath_query
+	{
+	private:
+		void* _impl;
+		xpath_parse_result _result;
+
+		typedef void (*unspecified_bool_type)(xpath_query***);
+
+		// Non-copyable semantics
+		xpath_query(const xpath_query&);
+		xpath_query& operator=(const xpath_query&);
+
+	public:
+		// Construct a compiled object from XPath expression.
+		// If PUGIXML_NO_EXCEPTIONS is not defined, throws xpath_exception on compilation errors.
+		explicit xpath_query(const char_t* query, xpath_variable_set* variables = 0);
+
+		// Constructor
+		xpath_query();
+
+		// Destructor
+		~xpath_query();
+
+	#if __cplusplus >= 201103
+		// Move semantics support
+		xpath_query(xpath_query&& rhs);
+		xpath_query& operator=(xpath_query&& rhs);
+	#endif
+
+		// Get query expression return type
+		xpath_value_type return_type() const;
+		
+		// Evaluate expression as boolean value in the specified context; performs type conversion if necessary.
+		// If PUGIXML_NO_EXCEPTIONS is not defined, throws std::bad_alloc on out of memory errors.
+		bool evaluate_boolean(const xpath_node& n) const;
+		
+		// Evaluate expression as double value in the specified context; performs type conversion if necessary.
+		// If PUGIXML_NO_EXCEPTIONS is not defined, throws std::bad_alloc on out of memory errors.
+		double evaluate_number(const xpath_node& n) const;
+		
+	#ifndef PUGIXML_NO_STL
+		// Evaluate expression as string value in the specified context; performs type conversion if necessary.
+		// If PUGIXML_NO_EXCEPTIONS is not defined, throws std::bad_alloc on out of memory errors.
+		string_t evaluate_string(const xpath_node& n) const;
+	#endif
+		
+		// Evaluate expression as string value in the specified context; performs type conversion if necessary.
+		// At most capacity characters are written to the destination buffer, full result size is returned (includes terminating zero).
+		// If PUGIXML_NO_EXCEPTIONS is not defined, throws std::bad_alloc on out of memory errors.
+		// If PUGIXML_NO_EXCEPTIONS is defined, returns empty  set instead.
+		size_t evaluate_string(char_t* buffer, size_t capacity, const xpath_node& n) const;
+
+		// Evaluate expression as node set in the specified context.
+		// If PUGIXML_NO_EXCEPTIONS is not defined, throws xpath_exception on type mismatch and std::bad_alloc on out of memory errors.
+		// If PUGIXML_NO_EXCEPTIONS is defined, returns empty node set instead.
+		xpath_node_set evaluate_node_set(const xpath_node& n) const;
+
+		// Evaluate expression as node set in the specified context.
+		// Return first node in document order, or empty node if node set is empty.
+		// If PUGIXML_NO_EXCEPTIONS is not defined, throws xpath_exception on type mismatch and std::bad_alloc on out of memory errors.
+		// If PUGIXML_NO_EXCEPTIONS is defined, returns empty node instead.
+		xpath_node evaluate_node(const xpath_node& n) const;
+
+		// Get parsing result (used to get compilation errors in PUGIXML_NO_EXCEPTIONS mode)
+		const xpath_parse_result& result() const;
+
+		// Safe bool conversion operator
+		operator unspecified_bool_type() const;
+
+		// Borland C++ workaround
+		bool operator!() const;
+	};
+	
+	#ifndef PUGIXML_NO_EXCEPTIONS
+	// XPath exception class
+	class PUGIXML_CLASS xpath_exception: public std::exception
+	{
+	private:
+		xpath_parse_result _result;
+
+	public:
+		// Construct exception from parse result
+		explicit xpath_exception(const xpath_parse_result& result);
+
+		// Get error message
+		virtual const char* what() const throw();
+
+		// Get parse result
+		const xpath_parse_result& result() const;
+	};
+	#endif
+	
+	// XPath node class (either xml_node or xml_attribute)
+	class PUGIXML_CLASS xpath_node
+	{
+	private:
+		xml_node _node;
+		xml_attribute _attribute;
+	
+		typedef void (*unspecified_bool_type)(xpath_node***);
+
+	public:
+		// Default constructor; constructs empty XPath node
+		xpath_node();
+		
+		// Construct XPath node from XML node/attribute
+		xpath_node(const xml_node& node);
+		xpath_node(const xml_attribute& attribute, const xml_node& parent);
+
+		// Get node/attribute, if any
+		xml_node node() const;
+		xml_attribute attribute() const;
+		
+		// Get parent of contained node/attribute
+		xml_node parent() const;
+
+		// Safe bool conversion operator
+		operator unspecified_bool_type() const;
+		
+		// Borland C++ workaround
+		bool operator!() const;
+
+		// Comparison operators
+		bool operator==(const xpath_node& n) const;
+		bool operator!=(const xpath_node& n) const;
+	};
+
+#ifdef __BORLANDC__
+	// Borland C++ workaround
+	bool PUGIXML_FUNCTION operator&&(const xpath_node& lhs, bool rhs);
+	bool PUGIXML_FUNCTION operator||(const xpath_node& lhs, bool rhs);
+#endif
+
+	// A fixed-size collection of XPath nodes
+	class PUGIXML_CLASS xpath_node_set
+	{
+	public:
+		// Collection type
+		enum type_t
+		{
+			type_unsorted,			// Not ordered
+			type_sorted,			// Sorted by document order (ascending)
+			type_sorted_reverse		// Sorted by document order (descending)
+		};
+		
+		// Constant iterator type
+		typedef const xpath_node* const_iterator;
+
+		// We define non-constant iterator to be the same as constant iterator so that various generic algorithms (i.e. boost foreach) work
+		typedef const xpath_node* iterator;
+	
+		// Default constructor. Constructs empty set.
+		xpath_node_set();
+
+		// Constructs a set from iterator range; data is not checked for duplicates and is not sorted according to provided type, so be careful
+		xpath_node_set(const_iterator begin, const_iterator end, type_t type = type_unsorted);
+
+		// Destructor
+		~xpath_node_set();
+		
+		// Copy constructor/assignment operator
+		xpath_node_set(const xpath_node_set& ns);
+		xpath_node_set& operator=(const xpath_node_set& ns);
+
+	#if __cplusplus >= 201103
+		// Move semantics support
+		xpath_node_set(xpath_node_set&& rhs);
+		xpath_node_set& operator=(xpath_node_set&& rhs);
+	#endif
+
+		// Get collection type
+		type_t type() const;
+		
+		// Get collection size
+		size_t size() const;
+
+		// Indexing operator
+		const xpath_node& operator[](size_t index) const;
+		
+		// Collection iterators
+		const_iterator begin() const;
+		const_iterator end() const;
+
+		// Sort the collection in ascending/descending order by document order
+		void sort(bool reverse = false);
+		
+		// Get first node in the collection by document order
+		xpath_node first() const;
+		
+		// Check if collection is empty
+		bool empty() const;
+	
+	private:
+		type_t _type;
+		
+		xpath_node _storage;
+		
+		xpath_node* _begin;
+		xpath_node* _end;
+
+		void _assign(const_iterator begin, const_iterator end, type_t type);
+		void _move(xpath_node_set& rhs);
+	};
+#endif
+
+#ifndef PUGIXML_NO_STL
+	// Convert wide string to UTF8
+	std::basic_string<char, std::char_traits<char>, std::allocator<char> > PUGIXML_FUNCTION as_utf8(const wchar_t* str);
+	std::basic_string<char, std::char_traits<char>, std::allocator<char> > PUGIXML_FUNCTION as_utf8(const std::basic_string<wchar_t, std::char_traits<wchar_t>, std::allocator<wchar_t> >& str);
+	
+	// Convert UTF8 to wide string
+	std::basic_string<wchar_t, std::char_traits<wchar_t>, std::allocator<wchar_t> > PUGIXML_FUNCTION as_wide(const char* str);
+	std::basic_string<wchar_t, std::char_traits<wchar_t>, std::allocator<wchar_t> > PUGIXML_FUNCTION as_wide(const std::basic_string<char, std::char_traits<char>, std::allocator<char> >& str);
+#endif
+
+	// Memory allocation function interface; returns pointer to allocated memory or NULL on failure
+	typedef void* (*allocation_function)(size_t size);
+	
+	// Memory deallocation function interface
+	typedef void (*deallocation_function)(void* ptr);
+
+	// Override default memory management functions. All subsequent allocations/deallocations will be performed via supplied functions.
+	void PUGIXML_FUNCTION set_memory_management_functions(allocation_function allocate, deallocation_function deallocate);
+	
+	// Get current memory management functions
+	allocation_function PUGIXML_FUNCTION get_memory_allocation_function();
+	deallocation_function PUGIXML_FUNCTION get_memory_deallocation_function();
+}
+
+#if !defined(PUGIXML_NO_STL) && (defined(_MSC_VER) || defined(__ICC))
+namespace std
+{
+	// Workarounds for (non-standard) iterator category detection for older versions (MSVC7/IC8 and earlier)
+	std::bidirectional_iterator_tag PUGIXML_FUNCTION _Iter_cat(const pugi::xml_node_iterator&);
+	std::bidirectional_iterator_tag PUGIXML_FUNCTION _Iter_cat(const pugi::xml_attribute_iterator&);
+	std::bidirectional_iterator_tag PUGIXML_FUNCTION _Iter_cat(const pugi::xml_named_node_iterator&);
+}
+#endif
+
+#if !defined(PUGIXML_NO_STL) && defined(__SUNPRO_CC)
+namespace std
+{
+	// Workarounds for (non-standard) iterator category detection
+	std::bidirectional_iterator_tag PUGIXML_FUNCTION __iterator_category(const pugi::xml_node_iterator&);
+	std::bidirectional_iterator_tag PUGIXML_FUNCTION __iterator_category(const pugi::xml_attribute_iterator&);
+	std::bidirectional_iterator_tag PUGIXML_FUNCTION __iterator_category(const pugi::xml_named_node_iterator&);
+}
+#endif
+
+#endif
+
+// Make sure implementation is included in header-only mode
+// Use macro expansion in #include to work around QMake (QTBUG-11923)
+#if defined(PUGIXML_HEADER_ONLY) && !defined(PUGIXML_SOURCE)
+#	define PUGIXML_SOURCE "pugixml.cpp"
+#	include PUGIXML_SOURCE
+#endif
+
+/**
+ * Copyright (c) 2006-2015 Arseny Kapoulkine
+ *
+ * Permission is hereby granted, free of charge, to any person
+ * obtaining a copy of this software and associated documentation
+ * files (the "Software"), to deal in the Software without
+ * restriction, including without limitation the rights to use,
+ * copy, modify, merge, publish, distribute, sublicense, and/or sell
+ * copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following
+ * conditions:
+ *
+ * The above copyright notice and this permission notice shall be
+ * included in all copies or substantial portions of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES
+ * OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT
+ * HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY,
+ * WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
+ * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ */
diff --git a/lib/pugixml/readme.txt b/lib/pugixml/readme.txt
new file mode 100644
index 00000000..faa41d37
--- /dev/null
+++ b/lib/pugixml/readme.txt
@@ -0,0 +1,52 @@
+pugixml 1.6 - an XML processing library
+
+Copyright (C) 2006-2015, by Arseny Kapoulkine (arseny.kapoulkine@gmail.com)
+Report bugs and download new versions at http://pugixml.org/
+
+This is the distribution of pugixml, which is a C++ XML processing library,
+which consists of a DOM-like interface with rich traversal/modification
+capabilities, an extremely fast XML parser which constructs the DOM tree from
+an XML file/buffer, and an XPath 1.0 implementation for complex data-driven
+tree queries. Full Unicode support is also available, with Unicode interface
+variants and conversions between different Unicode encodings (which happen
+automatically during parsing/saving).
+
+The distribution contains the following folders:
+
+	contrib/ - various contributions to pugixml
+
+	docs/ - documentation
+		docs/samples - pugixml usage examples
+		docs/quickstart.html - quick start guide
+		docs/manual.html - complete manual
+
+	scripts/ - project files for IDE/build systems
+
+	src/ - header and source files
+
+	readme.txt - this file.
+
+This library is distributed under the MIT License:
+
+Copyright (c) 2006-2015 Arseny Kapoulkine
+
+Permission is hereby granted, free of charge, to any person
+obtaining a copy of this software and associated documentation
+files (the "Software"), to deal in the Software without
+restriction, including without limitation the rights to use,
+copy, modify, merge, publish, distribute, sublicense, and/or sell
+copies of the Software, and to permit persons to whom the
+Software is furnished to do so, subject to the following
+conditions:
+
+The above copyright notice and this permission notice shall be
+included in all copies or substantial portions of the Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES
+OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT
+HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY,
+WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
+FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+OTHER DEALINGS IN THE SOFTWARE.
diff --git a/lib/qcd/action/ActionBase.h b/lib/qcd/action/ActionBase.h
index 05edce86..df5ee50b 100644
--- a/lib/qcd/action/ActionBase.h
+++ b/lib/qcd/action/ActionBase.h
@@ -7,14 +7,30 @@ template<class GaugeField>
 class Action { 
 
  public:
-  virtual void  init(const GaugeField &U, GridParallelRNG& pRNG) = 0;
-  virtual RealD S(const GaugeField &U)                           = 0;  // evaluate the action
-  virtual void  deriv(const GaugeField &U,GaugeField & dSdU )    = 0;  // evaluate the action derivative
-  //virtual void  refresh(const GaugeField & ) {}                ; 
+  virtual void  init (const GaugeField &U, GridParallelRNG& pRNG) = 0;  // 
+  virtual RealD S    (const GaugeField &U)                        = 0;  // evaluate the action
+  virtual void  deriv(const GaugeField &U,GaugeField & dSdU )     = 0;  // evaluate the action derivative
+  virtual void  refresh(const GaugeField & ) {};                        // Default to no-op for actions with no internal fields
   // Boundary conditions?
   // Heatbath?
   virtual ~Action() {};
 };
 
+// Could derive PseudoFermion action with a PF field, FermionField, and a Grid; implement refresh
+template<class GaugeField, class FermionField>
+class PseudoFermionAction : public Action<GaugeField> {
+ public:
+  FermionField Phi;
+  GridParallelRNG &pRNG;
+  GridBase &Grid;
+
+  PseudoFermionAction(GridBase &_Grid,GridParallelRNG &_pRNG) : Grid(_Grid), Phi(&_Grid), pRNG(_pRNG) {
+  };
+
+  virtual void refresh(const GaugeField &gauge) {
+    gaussian(Phi,pRNG);
+  };
+
+};
 }}
 #endif
diff --git a/lib/qcd/action/Actions.h b/lib/qcd/action/Actions.h
index 94809723..26f26b8f 100644
--- a/lib/qcd/action/Actions.h
+++ b/lib/qcd/action/Actions.h
@@ -79,4 +79,10 @@
 ///////////////////////////////////////////////////////////////////////////////
 #include <qcd/action/fermion/g5HermitianLinop.h>
 
+////////////////////////////////////////
+// Pseudo fermion combinations
+////////////////////////////////////////
+#include <qcd/action/pseudofermion/TwoFlavour.h>
+
+
 #endif
diff --git a/lib/qcd/action/fermion/ContinuedFractionFermion5D.cc b/lib/qcd/action/fermion/ContinuedFractionFermion5D.cc
index 0d54f272..2206bbc8 100644
--- a/lib/qcd/action/fermion/ContinuedFractionFermion5D.cc
+++ b/lib/qcd/action/fermion/ContinuedFractionFermion5D.cc
@@ -10,12 +10,12 @@ namespace Grid {
     void ContinuedFractionFermion5D::SetCoefficientsZolotarev(RealD zolo_hi,Approx::zolotarev_data *zdata)
     {
       // How to check Ls matches??
-      //      std::cout << Ls << " Ls"<<std::endl;
-      //      std::cout << zdata->n  << " - n"<<std::endl;
-      //      std::cout << zdata->da << " -da "<<std::endl;
-      //      std::cout << zdata->db << " -db"<<std::endl;
-      //      std::cout << zdata->dn << " -dn"<<std::endl;
-      //      std::cout << zdata->dd << " -dd"<<std::endl;
+      //      std::cout<<GridLogMessage << Ls << " Ls"<<std::endl;
+      //      std::cout<<GridLogMessage << zdata->n  << " - n"<<std::endl;
+      //      std::cout<<GridLogMessage << zdata->da << " -da "<<std::endl;
+      //      std::cout<<GridLogMessage << zdata->db << " -db"<<std::endl;
+      //      std::cout<<GridLogMessage << zdata->dn << " -dn"<<std::endl;
+      //      std::cout<<GridLogMessage << zdata->dd << " -dd"<<std::endl;
 
       assert(zdata->db==Ls);// Beta has Ls coeffs
 
@@ -55,7 +55,7 @@ namespace Grid {
 	See[s] = Aee[s] - 1.0/See[s-1];
       }
       for(int s=0;s<Ls;s++){
-	std::cout <<"s = "<<s<<" Beta "<<Beta[s]<<" Aee "<<Aee[s] <<" See "<<See[s] <<std::endl;
+	std::cout<<GridLogMessage <<"s = "<<s<<" Beta "<<Beta[s]<<" Aee "<<Aee[s] <<" See "<<See[s] <<std::endl;
       }
     }
 
diff --git a/lib/qcd/action/fermion/DomainWallFermion.h b/lib/qcd/action/fermion/DomainWallFermion.h
index 34f01be5..498d8b34 100644
--- a/lib/qcd/action/fermion/DomainWallFermion.h
+++ b/lib/qcd/action/fermion/DomainWallFermion.h
@@ -32,7 +32,7 @@ namespace Grid {
 	Approx::zolotarev_data *zdata = Approx::higham(eps,this->Ls);// eps is ignored for higham
 	assert(zdata->n==this->Ls);
 	
-	std::cout << "DomainWallFermion with Ls="<<Ls<<std::endl;
+	std::cout<<GridLogMessage << "DomainWallFermion with Ls="<<Ls<<std::endl;
 	// Call base setter
 	this->CayleyFermion5D::SetCoefficientsTanh(zdata,1.0,0.0);
 
diff --git a/lib/qcd/action/fermion/FermionOperator.h b/lib/qcd/action/fermion/FermionOperator.h
index 1aa2f8f0..b6ba2d08 100644
--- a/lib/qcd/action/fermion/FermionOperator.h
+++ b/lib/qcd/action/fermion/FermionOperator.h
@@ -39,10 +39,27 @@ namespace Grid {
       virtual void Dhop  (const FermionField &in, FermionField &out,int dag)=0;
       virtual void DhopOE(const FermionField &in, FermionField &out,int dag)=0;
       virtual void DhopEO(const FermionField &in, FermionField &out,int dag)=0;
+      virtual void DhopDir(const FermionField &in, FermionField &out,int dir,int disp)=0; // implemented by WilsonFermion and WilsonFermion5D
 
-      virtual void  Mdiag(const FermionField &in, FermionField &out) { Mooee(in,out);};   // Same as Mooee applied to both CB's
-      virtual void  Mdir (const FermionField &in, FermionField &out,int dir,int disp)=0;   // case by case Wilson, Clover, Cayley, ContFrac, PartFrac
-      virtual void  DhopDir(const FermionField &in, FermionField &out,int dir,int disp)=0; // implemented by WilsonFermion and WilsonFermion5D
+      // force terms; five routines; default to Dhop on diagonal
+      virtual void MDeriv  (LatticeGaugeField &mat,const FermionField &U,const FermionField &V,int dag){DhopDeriv(mat,U,V,dag);};
+      virtual void MoeDeriv(LatticeGaugeField &mat,const FermionField &U,const FermionField &V,int dag){DhopDerivOE(mat,U,V,dag);};
+      virtual void MeoDeriv(LatticeGaugeField &mat,const FermionField &U,const FermionField &V,int dag){DhopDerivEO(mat,U,V,dag);};
+      virtual void MooDeriv(LatticeGaugeField &mat,const FermionField &U,const FermionField &V,int dag){mat=zero;};
+      virtual void MeeDeriv(LatticeGaugeField &mat,const FermionField &U,const FermionField &V,int dag){mat=zero;};
+
+      virtual void DhopDeriv  (LatticeGaugeField &mat,const FermionField &U,const FermionField &V,int dag)=0;
+      virtual void DhopDerivEO(LatticeGaugeField &mat,const FermionField &U,const FermionField &V,int dag)=0;
+      virtual void DhopDerivOE(LatticeGaugeField &mat,const FermionField &U,const FermionField &V,int dag)=0;
+
+
+      virtual void  Mdiag  (const FermionField &in, FermionField &out) { Mooee(in,out);};   // Same as Mooee applied to both CB's
+      virtual void  Mdir   (const FermionField &in, FermionField &out,int dir,int disp)=0;   // case by case Wilson, Clover, Cayley, ContFrac, PartFrac
+
+      ///////////////////////////////////////////////
+      // Updates gauge field during HMC
+      ///////////////////////////////////////////////
+      virtual void ImportGauge(const GaugeField & _U);
 
     };
 
diff --git a/lib/qcd/action/fermion/MobiusFermion.h b/lib/qcd/action/fermion/MobiusFermion.h
index 6603e3f2..5aaaa188 100644
--- a/lib/qcd/action/fermion/MobiusFermion.h
+++ b/lib/qcd/action/fermion/MobiusFermion.h
@@ -30,7 +30,7 @@ namespace Grid {
       {
 	RealD eps = 1.0;
 
-	std::cout << "MobiusFermion (b="<<b<<",c="<<c<<") with Ls= "<<Ls<<" Tanh approx"<<std::endl;
+	std::cout<<GridLogMessage << "MobiusFermion (b="<<b<<",c="<<c<<") with Ls= "<<Ls<<" Tanh approx"<<std::endl;
 	Approx::zolotarev_data *zdata = Approx::higham(eps,this->Ls);// eps is ignored for higham
 	assert(zdata->n==this->Ls);
 	
diff --git a/lib/qcd/action/fermion/MobiusZolotarevFermion.h b/lib/qcd/action/fermion/MobiusZolotarevFermion.h
index d3c0b3a0..5f74702c 100644
--- a/lib/qcd/action/fermion/MobiusZolotarevFermion.h
+++ b/lib/qcd/action/fermion/MobiusZolotarevFermion.h
@@ -34,7 +34,7 @@ namespace Grid {
 	Approx::zolotarev_data *zdata = Approx::zolotarev(eps,this->Ls,0);
 	assert(zdata->n==this->Ls);
 
-	std::cout << "MobiusZolotarevFermion (b="<<b<<",c="<<c<<") with Ls= "<<Ls<<" Zolotarev range ["<<lo<<","<<hi<<"]"<<std::endl;
+	std::cout<<GridLogMessage << "MobiusZolotarevFermion (b="<<b<<",c="<<c<<") with Ls= "<<Ls<<" Zolotarev range ["<<lo<<","<<hi<<"]"<<std::endl;
 	
 	// Call base setter
 	this->CayleyFermion5D::SetCoefficientsZolotarev(hi,zdata,b,c);
diff --git a/lib/qcd/action/fermion/PartialFractionFermion5D.cc b/lib/qcd/action/fermion/PartialFractionFermion5D.cc
index a148fe40..958cea30 100644
--- a/lib/qcd/action/fermion/PartialFractionFermion5D.cc
+++ b/lib/qcd/action/fermion/PartialFractionFermion5D.cc
@@ -260,12 +260,12 @@ namespace Grid {
     void  PartialFractionFermion5D::SetCoefficientsZolotarev(RealD zolo_hi,Approx::zolotarev_data *zdata){
 
       // check on degree matching
-      //      std::cout << Ls << " Ls"<<std::endl;
-      //      std::cout << zdata->n  << " - n"<<std::endl;
-      //      std::cout << zdata->da << " -da "<<std::endl;
-      //      std::cout << zdata->db << " -db"<<std::endl;
-      //      std::cout << zdata->dn << " -dn"<<std::endl;
-      //      std::cout << zdata->dd << " -dd"<<std::endl;
+      //      std::cout<<GridLogMessage << Ls << " Ls"<<std::endl;
+      //      std::cout<<GridLogMessage << zdata->n  << " - n"<<std::endl;
+      //      std::cout<<GridLogMessage << zdata->da << " -da "<<std::endl;
+      //      std::cout<<GridLogMessage << zdata->db << " -db"<<std::endl;
+      //      std::cout<<GridLogMessage << zdata->dn << " -dn"<<std::endl;
+      //      std::cout<<GridLogMessage << zdata->dd << " -dd"<<std::endl;
       assert(Ls == (2*zdata->da -1) );
 
       // Part frac
diff --git a/lib/qcd/action/fermion/WilsonFermion.cc b/lib/qcd/action/fermion/WilsonFermion.cc
index f5ff7cb0..2c2132d1 100644
--- a/lib/qcd/action/fermion/WilsonFermion.cc
+++ b/lib/qcd/action/fermion/WilsonFermion.cc
@@ -24,6 +24,10 @@ WilsonFermion::WilsonFermion(LatticeGaugeField &_Umu,
 {
   // Allocate the required comms buffer
   comm_buf.resize(Stencil._unified_buffer_size); // this is always big enough to contain EO
+  ImportGauge(_Umu);
+}
+void WilsonFermion::ImportGauge(const LatticeGaugeField &_Umu)
+{
   DoubleStore(Umu,_Umu);
   pickCheckerboard(Even,UmuEven,Umu);
   pickCheckerboard(Odd ,UmuOdd,Umu);
@@ -98,7 +102,9 @@ void WilsonFermion::Mdir (const LatticeFermion &in, LatticeFermion &out,int dir,
   DhopDir(in,out,dir,disp);
 }
 void WilsonFermion::DhopDir(const LatticeFermion &in, LatticeFermion &out,int dir,int disp){
+
   WilsonCompressor compressor(DaggerNo);
+
   Stencil.HaloExchange<vSpinColourVector,vHalfSpinColourVector,WilsonCompressor>(in,comm_buf,compressor);
   
   assert( (disp==1)||(disp==-1) );
@@ -109,9 +115,22 @@ void WilsonFermion::DhopDir(const LatticeFermion &in, LatticeFermion &out,int di
 
 PARALLEL_FOR_LOOP
   for(int sss=0;sss<in._grid->oSites();sss++){
-    DiracOptDhopDir(Stencil,Umu,comm_buf,sss,sss,in,out,dirdisp);
+    DiracOptDhopDir(Stencil,Umu,comm_buf,sss,sss,in,out,dirdisp,dirdisp);
   }
+ 
+};
+void WilsonFermion::DhopDirDisp(const LatticeFermion &in, LatticeFermion &out,int dirdisp,int gamma,int dag)
+{
 
+  WilsonCompressor compressor(dag);
+
+  Stencil.HaloExchange<vSpinColourVector,vHalfSpinColourVector,WilsonCompressor>(in,comm_buf,compressor);
+
+PARALLEL_FOR_LOOP
+  for(int sss=0;sss<in._grid->oSites();sss++){
+    DiracOptDhopDir(Stencil,Umu,comm_buf,sss,sss,in,out,dirdisp,gamma);
+  }
+ 
 };
 
 void WilsonFermion::DhopInternal(CartesianStencil & st,LatticeDoubledGaugeField & U,
@@ -177,6 +196,77 @@ void WilsonFermion::Dhop(const LatticeFermion &in, LatticeFermion &out,int dag)
   DhopInternal(Stencil,Umu,in,out,dag);
 }
 
+void WilsonFermion::DerivInternal(CartesianStencil & st,LatticeDoubledGaugeField & U,
+				  LatticeGaugeField &mat,const LatticeFermion &A,const LatticeFermion &B,int dag)
+{
+  assert((dag==DaggerNo) ||(dag==DaggerYes));
+
+  WilsonCompressor compressor(dag);
+
+  LatticeColourMatrix tmp(B._grid);
+  LatticeFermion Btilde(B._grid);
+
+  st.HaloExchange<vSpinColourVector,vHalfSpinColourVector,WilsonCompressor>(B,comm_buf,compressor);
+  
+  for(int mu=0;mu<Nd;mu++){
+      
+    ////////////////////////////////////////////////////////////////////////
+    // Flip gamma (1+g)<->(1-g) if dag
+    ////////////////////////////////////////////////////////////////////////
+    int gamma = mu;
+    if ( dag ) gamma+= Nd;
+
+    ////////////////////////
+    // Call the single hop
+    ////////////////////////
+PARALLEL_FOR_LOOP
+    for(int sss=0;sss<B._grid->oSites();sss++){
+      DiracOptDhopDir(st,U,comm_buf,sss,sss,B,Btilde,mu,gamma);
+    }
+
+    //////////////////////////////////////////////////
+    // spin trace outer product
+    //////////////////////////////////////////////////
+    tmp = TraceIndex<SpinIndex>(outerProduct(Btilde,A)); 
+    PokeIndex<LorentzIndex>(mat,tmp,mu);
+
+  }
+}
+void WilsonFermion::DhopDeriv(LatticeGaugeField &mat,const LatticeFermion &U,const LatticeFermion &V,int dag)
+{
+  conformable(U._grid,_grid);  
+  conformable(U._grid,V._grid);
+  conformable(U._grid,mat._grid);
+
+  mat.checkerboard = U.checkerboard;
+
+  DerivInternal(Stencil,Umu,mat,U,V,dag);
+}
+void WilsonFermion::DhopDerivOE(LatticeGaugeField &mat,const LatticeFermion &U,const LatticeFermion &V,int dag)
+{
+  conformable(U._grid,_cbgrid);  
+  conformable(U._grid,V._grid);
+  conformable(U._grid,mat._grid);
+
+  assert(V.checkerboard==Even);
+  assert(U.checkerboard==Odd);
+  mat.checkerboard = Odd;
+
+  DerivInternal(StencilEven,UmuOdd,mat,U,V,dag);
+}
+void WilsonFermion::DhopDerivEO(LatticeGaugeField &mat,const LatticeFermion &U,const LatticeFermion &V,int dag)
+{
+  conformable(U._grid,_cbgrid);  
+  conformable(U._grid,V._grid);
+  conformable(U._grid,mat._grid);
+
+  assert(V.checkerboard==Odd);
+  assert(U.checkerboard==Even);
+  mat.checkerboard = Even;
+
+  DerivInternal(StencilOdd,UmuEven,mat,U,V,dag);
+}
+
 }}
 
 
diff --git a/lib/qcd/action/fermion/WilsonFermion.h b/lib/qcd/action/fermion/WilsonFermion.h
index 60726fdd..79d3e626 100644
--- a/lib/qcd/action/fermion/WilsonFermion.h
+++ b/lib/qcd/action/fermion/WilsonFermion.h
@@ -24,11 +24,95 @@ namespace Grid {
       // half checkerboard operaions
       void   Meooe       (const LatticeFermion &in, LatticeFermion &out);
       void   MeooeDag    (const LatticeFermion &in, LatticeFermion &out);
+
       virtual void   Mooee       (const LatticeFermion &in, LatticeFermion &out); // remain virtual so we 
       virtual void   MooeeDag    (const LatticeFermion &in, LatticeFermion &out); // can derive Clover
       virtual void   MooeeInv    (const LatticeFermion &in, LatticeFermion &out); // from Wilson base
       virtual void   MooeeInvDag (const LatticeFermion &in, LatticeFermion &out);
 
+      ////////////////////////
+      // 
+      // Force term: d/dtau S = 0
+      //
+      // It is simplest to consider the two flavour force term 
+      // 
+      //       S[U,phi] = phidag (MdagM)^-1 phi
+      //
+      // But simplify even this to
+      //
+      //       S[U,phi] = phidag MdagM phi
+      //
+      // (other options exist depending on nature of action fragment.)
+      // 
+      // Require momentum be traceless anti-hermitian to move within group manifold [ P = i P^a T^a ]
+      //
+      // Define the HMC hamiltonian
+      //
+      //       H = 1/2 Tr P^2 + S(U,phi)
+      // 
+      // .
+      // U =  P U    (lorentz & color indices multiplied)
+      //
+      // Hence
+      //
+      // .c    c  c       c
+      // U =  U  P   = - U  P        (c == dagger)
+      //
+      // So, taking some liberty with implicit indices
+      //                  .      .          .c      c
+      // dH/dt = 0 = Tr P P +Tr[ U  dS/dU + U  dS/dU  ]
+      //
+      //                .                          c      c
+      //           = Tr P P + i Tr[  P U dS/dU  - U   P dS/dU  ]
+      //
+      //                   .                        c  c
+      //           = Tr P (P + i ( U dS/dU - P dS/dU  U ]
+      //
+      //              .                        c  c
+      //           => P  = -i [ U dS/dU - dS/dU  U ]      generates HMC EoM
+      //
+      // Simple case work this out using S = phi^dag MdagM phi for wilson:
+      //                               c       c
+      // dSdt     = dU_xdt  dSdUx  + dUxdt dSdUx
+      //          
+      //         = Tr i P U_x [ (\phi^\dag)_x (1+g) (M \phi)_x+\mu   +(\phi^\dag M^\dag)_x (1-g) \phi_{x+\mu} ]
+      //                 c
+      //            - i U_x P [ (\phi^\dag)_x+mu (1-g) (M \phi)_x    +(\phi^\dag M^\dag)_(x+\mu) (1+g) \phi_{x} ]
+      //
+      //         = i [(\phi^\dag)_x      ]_j P_jk  [U_x(1+g) (M \phi)_x+\mu]_k        (1)
+      //         + i [(\phi^\dagM^\dag)_x]_j P_jk  [U_x(1-g) (\phi)_x+\mu]_k          (2)
+      //         - i [(\phi^\dag)_x+mu (1-g) U^dag_x]_j P_jk  [(M \phi)_xk            (3)
+      //         - i [(\phi^\dagM^\dag)_x+mu (1+g) U^dag_x]_j P_jk  [ \phi]_xk        (4)
+      //
+      // Observe that (1)* = (4)
+      //              (2)* = (3)
+      //
+      // Write as    .
+      //             P_{kj}  = - i (  [U_x(1+g) (M \phi)_x+\mu] (x) [(\phi^\dag)_x] + [U_x(1-g) (\phi)_x+\mu] (x) [(\phi^\dagM^\dag)_x] - h.c )
+      //
+      // where (x) denotes outer product in colour and spins are traced.
+      //
+      // Need only evaluate (1) and (2)  [Chroma] or (2) and (4) [IroIro] and take the 
+      // traceless anti hermitian part (of term in brackets w/o the "i")
+      // 
+      // Generalisation to S=phi^dag (MdagM)^{-1} phi is simple:
+      //
+      // For more complicated DWF etc... apply product rule in differentiation
+      //
+      ////////////////////////
+      void DhopDeriv  (LatticeGaugeField &mat,const LatticeFermion &U,const LatticeFermion &V,int dag);
+      void DhopDerivEO(LatticeGaugeField &mat,const LatticeFermion &U,const LatticeFermion &V,int dag);
+      void DhopDerivOE(LatticeGaugeField &mat,const LatticeFermion &U,const LatticeFermion &V,int dag);
+
+      // Extra support internal
+      void DerivInternal(CartesianStencil & st,
+			 LatticeDoubledGaugeField & U,
+			 LatticeGaugeField &mat,
+			 const LatticeFermion &A,
+			 const LatticeFermion &B,
+			 int dag);
+
+
       // non-hermitian hopping term; half cb or both
       void Dhop  (const LatticeFermion &in, LatticeFermion &out,int dag);
       void DhopOE(const LatticeFermion &in, LatticeFermion &out,int dag);
@@ -37,6 +121,7 @@ namespace Grid {
       // Multigrid assistance
       void   Mdir (const LatticeFermion &in, LatticeFermion &out,int dir,int disp);
       void DhopDir(const LatticeFermion &in, LatticeFermion &out,int dir,int disp);
+      void DhopDirDisp(const LatticeFermion &in, LatticeFermion &out,int dirdisp,int gamma,int dag);
 
       ///////////////////////////////////////////////////////////////
       // Extra methods added by derived
@@ -51,6 +136,7 @@ namespace Grid {
       WilsonFermion(LatticeGaugeField &_Umu,GridCartesian &Fgrid,GridRedBlackCartesian &Hgrid,RealD _mass);
 
       // DoubleStore
+      virtual void ImportGauge(const LatticeGaugeField &_Umu);
       void DoubleStore(LatticeDoubledGaugeField &Uds,const LatticeGaugeField &Umu);
 
       ///////////////////////////////////////////////////////////////
@@ -59,7 +145,8 @@ namespace Grid {
       static int HandOptDslash; // these are a temporary hack
       static int MortonOrder;
 
-    protected:
+      //    protected:
+    public:
 
       RealD                        mass;
 
diff --git a/lib/qcd/action/fermion/WilsonFermion5D.cc b/lib/qcd/action/fermion/WilsonFermion5D.cc
index 316b0676..c3329503 100644
--- a/lib/qcd/action/fermion/WilsonFermion5D.cc
+++ b/lib/qcd/action/fermion/WilsonFermion5D.cc
@@ -65,7 +65,10 @@ namespace QCD {
 
   // Allocate the required comms buffer
   comm_buf.resize(Stencil._unified_buffer_size); // this is always big enough to contain EO
-  
+  ImportGauge(_Umu);
+}  
+void WilsonFermion5D::ImportGauge(const LatticeGaugeField &_Umu)
+{
   DoubleStore(Umu,_Umu);
   pickCheckerboard(Even,UmuEven,Umu);
   pickCheckerboard(Odd ,UmuOdd,Umu);
@@ -100,19 +103,111 @@ void WilsonFermion5D::DhopDir(const LatticeFermion &in, LatticeFermion &out,int
   assert(dirdisp<=7);
   assert(dirdisp>=0);
 
-//PARALLEL_FOR_LOOP
+PARALLEL_FOR_LOOP
   for(int ss=0;ss<Umu._grid->oSites();ss++){
     for(int s=0;s<Ls;s++){
       int sU=ss;
       int sF = s+Ls*sU; 
-      DiracOptDhopDir(Stencil,Umu,comm_buf,sF,sU,in,out,dirdisp);
+      DiracOptDhopDir(Stencil,Umu,comm_buf,sF,sU,in,out,dirdisp,dirdisp);
     }
   }
 };
 
+void WilsonFermion5D::DerivInternal(CartesianStencil & st,
+				    LatticeDoubledGaugeField & U,
+				    LatticeGaugeField &mat,
+				    const LatticeFermion &A,
+				    const LatticeFermion &B,
+				    int dag)
+{
+  assert((dag==DaggerNo) ||(dag==DaggerYes));
+
+  WilsonCompressor compressor(dag);
+  
+  LatticeColourMatrix tmp(B._grid);
+  LatticeFermion Btilde(B._grid);
+
+  st.HaloExchange<vSpinColourVector,vHalfSpinColourVector,WilsonCompressor>(B,comm_buf,compressor);
+  
+  for(int mu=0;mu<Nd;mu++){
+      
+    ////////////////////////////////////////////////////////////////////////
+    // Flip gamma if dag
+    ////////////////////////////////////////////////////////////////////////
+    int gamma = mu;
+    if ( dag ) gamma+= Nd;
+
+    ////////////////////////
+    // Call the single hop
+    ////////////////////////
+PARALLEL_FOR_LOOP
+    for(int sss=0;sss<B._grid->oSites();sss++){
+      for(int s=0;s<Ls;s++){
+	int sU=sss;
+	int sF = s+Ls*sU;
+	DiracOptDhopDir(st,U,comm_buf,sF,sU,B,Btilde,mu,gamma);
+      }
+    }
+
+    ////////////////////////////
+    // spin trace outer product
+    ////////////////////////////
+// FIXME : need to sum over fifth direction.
+    tmp = TraceIndex<SpinIndex>(outerProduct(Btilde,A)); // ordering here
+    PokeIndex<LorentzIndex>(mat,tmp,mu);
+
+  }
+}
+
+void WilsonFermion5D::DhopDeriv(      LatticeGaugeField &mat,
+				const LatticeFermion &A,
+				const LatticeFermion &B,
+				int dag)
+{
+  conformable(A._grid,FermionGrid());  
+  conformable(A._grid,B._grid);
+  conformable(GaugeGrid(),mat._grid);
+
+  mat.checkerboard = A.checkerboard;
+
+  DerivInternal(Stencil,Umu,mat,A,B,dag);
+}
+
+void WilsonFermion5D::DhopDerivEO(LatticeGaugeField &mat,
+				  const LatticeFermion &A,
+				  const LatticeFermion &B,
+				  int dag)
+{
+  conformable(A._grid,FermionRedBlackGrid());
+  conformable(GaugeRedBlackGrid(),mat._grid);
+  conformable(A._grid,B._grid);
+
+  assert(B.checkerboard==Odd);
+  assert(A.checkerboard==Even);
+  mat.checkerboard = Even;
+
+  DerivInternal(StencilOdd,UmuEven,mat,A,B,dag);
+}
+
+void WilsonFermion5D::DhopDerivOE(LatticeGaugeField &mat,
+				  const LatticeFermion &A,
+				  const LatticeFermion &B,
+				  int dag)
+{
+  conformable(A._grid,FermionRedBlackGrid());
+  conformable(GaugeRedBlackGrid(),mat._grid);
+  conformable(A._grid,B._grid);
+
+  assert(B.checkerboard==Even);
+  assert(A.checkerboard==Odd);
+  mat.checkerboard = Odd;
+
+  DerivInternal(StencilEven,UmuOdd,mat,A,B,dag);
+}
+
 void WilsonFermion5D::DhopInternal(CartesianStencil & st, LebesgueOrder &lo,
 				   LatticeDoubledGaugeField & U,
-			   const LatticeFermion &in, LatticeFermion &out,int dag)
+				   const LatticeFermion &in, LatticeFermion &out,int dag)
 {
   //  assert((dag==DaggerNo) ||(dag==DaggerYes));
 
diff --git a/lib/qcd/action/fermion/WilsonFermion5D.h b/lib/qcd/action/fermion/WilsonFermion5D.h
index e345d2a4..d5cff3e6 100644
--- a/lib/qcd/action/fermion/WilsonFermion5D.h
+++ b/lib/qcd/action/fermion/WilsonFermion5D.h
@@ -44,12 +44,18 @@ namespace Grid {
 
       // half checkerboard operations; leave unimplemented as abstract for now
       virtual void   Meooe       (const LatticeFermion &in, LatticeFermion &out){assert(0);};
-      virtual void   MeooeDag    (const LatticeFermion &in, LatticeFermion &out){assert(0);};
       virtual void   Mooee       (const LatticeFermion &in, LatticeFermion &out){assert(0);};
-      virtual void   MooeeDag    (const LatticeFermion &in, LatticeFermion &out){assert(0);};
       virtual void   MooeeInv    (const LatticeFermion &in, LatticeFermion &out){assert(0);};
+
+      virtual void   MeooeDag    (const LatticeFermion &in, LatticeFermion &out){assert(0);};
+      virtual void   MooeeDag    (const LatticeFermion &in, LatticeFermion &out){assert(0);};
       virtual void   MooeeInvDag (const LatticeFermion &in, LatticeFermion &out){assert(0);};
 
+      // These can be overridden by fancy 5d chiral actions
+      virtual void DhopDeriv  (LatticeGaugeField &mat,const LatticeFermion &U,const LatticeFermion &V,int dag);
+      virtual void DhopDerivEO(LatticeGaugeField &mat,const LatticeFermion &U,const LatticeFermion &V,int dag);
+      virtual void DhopDerivOE(LatticeGaugeField &mat,const LatticeFermion &U,const LatticeFermion &V,int dag);
+
       // Implement hopping term non-hermitian hopping term; half cb or both
       // Implement s-diagonal DW
       void DW    (const LatticeFermion &in, LatticeFermion &out,int dag);
@@ -64,6 +70,14 @@ namespace Grid {
       ///////////////////////////////////////////////////////////////
       // New methods added 
       ///////////////////////////////////////////////////////////////
+
+      void DerivInternal(CartesianStencil & st,
+			 LatticeDoubledGaugeField & U,
+			 LatticeGaugeField &mat,
+			 const LatticeFermion &A,
+			 const LatticeFermion &B,
+			 int dag);
+
       void DhopInternal(CartesianStencil & st,
 			LebesgueOrder &lo,
 			LatticeDoubledGaugeField &U,
@@ -80,6 +94,7 @@ namespace Grid {
 			  double _M5);
 
       // DoubleStore
+      virtual void ImportGauge(const LatticeGaugeField &_Umu);
       void DoubleStore(LatticeDoubledGaugeField &Uds,const LatticeGaugeField &Umu);
 
       ///////////////////////////////////////////////////////////////
diff --git a/lib/qcd/action/fermion/WilsonKernels.cc b/lib/qcd/action/fermion/WilsonKernels.cc
index bdc18eee..43ccff63 100644
--- a/lib/qcd/action/fermion/WilsonKernels.cc
+++ b/lib/qcd/action/fermion/WilsonKernels.cc
@@ -294,8 +294,8 @@ void DiracOptDhopSiteDag(CartesianStencil &st,LatticeDoubledGaugeField &U,
 }
 
 void DiracOptDhopDir(CartesianStencil &st,LatticeDoubledGaugeField &U,
-			std::vector<vHalfSpinColourVector,alignedAllocator<vHalfSpinColourVector> >  &buf,
-		       int sF,int sU,const LatticeFermion &in, LatticeFermion &out,int dirdisp)
+		     std::vector<vHalfSpinColourVector,alignedAllocator<vHalfSpinColourVector> >  &buf,
+		     int sF,int sU,const LatticeFermion &in, LatticeFermion &out,int dir,int gamma)
 {
     vHalfSpinColourVector  tmp;    
     vHalfSpinColourVector  chi;    
@@ -304,13 +304,13 @@ void DiracOptDhopDir(CartesianStencil &st,LatticeDoubledGaugeField &U,
     int offset,local,perm, ptype;
     int ss=sF;
     
-    offset = st._offsets [dirdisp][ss];
-    local  = st._is_local[dirdisp][ss];
-    perm   = st._permute[dirdisp][ss];
-    ptype  = st._permute_type[dirdisp];
+    offset = st._offsets [dir][ss];
+    local  = st._is_local[dir][ss];
+    perm   = st._permute[dir][ss];
+    ptype  = st._permute_type[dir];
 
     // Xp
-    if(dirdisp==Xp){
+    if(gamma==Xp){
       if ( local && perm ) {
 	spProjXp(tmp,in._odata[offset]);
 	permute(chi,tmp,ptype);
@@ -319,12 +319,12 @@ void DiracOptDhopDir(CartesianStencil &st,LatticeDoubledGaugeField &U,
       } else { 
 	chi=buf[offset];
       }
-      mult(&Uchi(),&U._odata[sU](Xp),&chi());
+      mult(&Uchi(),&U._odata[sU](dir),&chi());
       spReconXp(result,Uchi);
     }
 
     // Yp
-    if ( dirdisp==Yp ){
+    if ( gamma==Yp ){
       if ( local && perm ) {
 	spProjYp(tmp,in._odata[offset]);
 	permute(chi,tmp,ptype);
@@ -333,12 +333,12 @@ void DiracOptDhopDir(CartesianStencil &st,LatticeDoubledGaugeField &U,
       } else { 
 	chi=buf[offset];
       }
-      mult(&Uchi(),&U._odata[sU](Yp),&chi());
+      mult(&Uchi(),&U._odata[sU](dir),&chi());
       spReconYp(result,Uchi);
     }
 
     // Zp
-    if ( dirdisp ==Zp ){
+    if ( gamma ==Zp ){
       if ( local && perm ) {
 	spProjZp(tmp,in._odata[offset]);
 	permute(chi,tmp,ptype);
@@ -347,12 +347,12 @@ void DiracOptDhopDir(CartesianStencil &st,LatticeDoubledGaugeField &U,
       } else { 
 	chi=buf[offset];
       }
-      mult(&Uchi(),&U._odata[sU](Zp),&chi());
+      mult(&Uchi(),&U._odata[sU](dir),&chi());
       spReconZp(result,Uchi);
     }
 
     // Tp
-    if ( dirdisp ==Tp ){
+    if ( gamma ==Tp ){
       if ( local && perm ) {
 	spProjTp(tmp,in._odata[offset]);
 	permute(chi,tmp,ptype);
@@ -361,12 +361,12 @@ void DiracOptDhopDir(CartesianStencil &st,LatticeDoubledGaugeField &U,
       } else { 
 	chi=buf[offset];
       }
-      mult(&Uchi(),&U._odata[sU](Tp),&chi());
+      mult(&Uchi(),&U._odata[sU](dir),&chi());
       spReconTp(result,Uchi);
     }
 
     // Xm
-    if ( dirdisp==Xm ){
+    if ( gamma==Xm ){
       if ( local && perm ) {
 	spProjXm(tmp,in._odata[offset]);
 	permute(chi,tmp,ptype);
@@ -375,12 +375,12 @@ void DiracOptDhopDir(CartesianStencil &st,LatticeDoubledGaugeField &U,
       } else { 
 	chi=buf[offset];
       }
-      mult(&Uchi(),&U._odata[sU](Xm),&chi());
+      mult(&Uchi(),&U._odata[sU](dir),&chi());
       spReconXm(result,Uchi);
     }
 
     // Ym
-    if ( dirdisp == Ym ){
+    if ( gamma == Ym ){
       if ( local && perm ) {
 	spProjYm(tmp,in._odata[offset]);
 	permute(chi,tmp,ptype);
@@ -389,12 +389,12 @@ void DiracOptDhopDir(CartesianStencil &st,LatticeDoubledGaugeField &U,
       } else { 
 	chi=buf[offset];
       }
-      mult(&Uchi(),&U._odata[sU](Ym),&chi());
+      mult(&Uchi(),&U._odata[sU](dir),&chi());
       spReconYm(result,Uchi);
     }
 
     // Zm
-    if ( dirdisp == Zm ){
+    if ( gamma == Zm ){
       if ( local && perm ) {
 	spProjZm(tmp,in._odata[offset]);
 	permute(chi,tmp,ptype);
@@ -403,12 +403,12 @@ void DiracOptDhopDir(CartesianStencil &st,LatticeDoubledGaugeField &U,
       } else { 
 	chi=buf[offset];
       }
-      mult(&Uchi(),&U._odata[sU](Zm),&chi());
+      mult(&Uchi(),&U._odata[sU](dir),&chi());
       spReconZm(result,Uchi);
     }
 
     // Tm
-    if ( dirdisp==Tm ) {
+    if ( gamma==Tm ) {
       if ( local && perm ) {
 	spProjTm(tmp,in._odata[offset]);
 	permute(chi,tmp,ptype);
@@ -417,7 +417,7 @@ void DiracOptDhopDir(CartesianStencil &st,LatticeDoubledGaugeField &U,
       } else { 
 	chi=buf[offset];
       }
-      mult(&Uchi(),&U._odata[sU](Tm),&chi());
+      mult(&Uchi(),&U._odata[sU](dir),&chi());
       spReconTm(result,Uchi);
     }
 
diff --git a/lib/qcd/action/fermion/WilsonKernels.h b/lib/qcd/action/fermion/WilsonKernels.h
index ec807af5..cb0e07bb 100644
--- a/lib/qcd/action/fermion/WilsonKernels.h
+++ b/lib/qcd/action/fermion/WilsonKernels.h
@@ -22,7 +22,7 @@ namespace Grid {
 			       int sF,int sU,const LatticeFermion &in, LatticeFermion &out);
       void DiracOptDhopDir(CartesianStencil &st,LatticeDoubledGaugeField &U,
 			   std::vector<vHalfSpinColourVector,alignedAllocator<vHalfSpinColourVector> >  &buf,
-			   int sF,int sU,const LatticeFermion &in, LatticeFermion &out,int dirdisp);
+			   int sF,int sU,const LatticeFermion &in, LatticeFermion &out,int dirdisp,int gamma);
       
       //  };
 
diff --git a/lib/qcd/action/fermion/WilsonKernelsHand.cc b/lib/qcd/action/fermion/WilsonKernelsHand.cc
index 8c31d65b..cff700fb 100644
--- a/lib/qcd/action/fermion/WilsonKernelsHand.cc
+++ b/lib/qcd/action/fermion/WilsonKernelsHand.cc
@@ -360,11 +360,11 @@ void DiracOptHandDhopSite(CartesianStencil &st,LatticeDoubledGaugeField &U,
     MULT_2SPIN(Xp);
   }
   XP_RECON;
-  //  std::cout << "XP_RECON"<<std::endl;
-  //  std::cout << result_00 <<" "<<result_01 <<" "<<result_02 <<std::endl;
-  //  std::cout << result_10 <<" "<<result_11 <<" "<<result_12 <<std::endl;
-  //  std::cout << result_20 <<" "<<result_21 <<" "<<result_22 <<std::endl;
-  //  std::cout << result_30 <<" "<<result_31 <<" "<<result_32 <<std::endl;
+  //  std::cout<<GridLogMessage << "XP_RECON"<<std::endl;
+  //  std::cout<<GridLogMessage << result_00 <<" "<<result_01 <<" "<<result_02 <<std::endl;
+  //  std::cout<<GridLogMessage << result_10 <<" "<<result_11 <<" "<<result_12 <<std::endl;
+  //  std::cout<<GridLogMessage << result_20 <<" "<<result_21 <<" "<<result_22 <<std::endl;
+  //  std::cout<<GridLogMessage << result_30 <<" "<<result_31 <<" "<<result_32 <<std::endl;
 
   // Yp
   offset = st._offsets [Yp][ss];
@@ -446,11 +446,11 @@ void DiracOptHandDhopSite(CartesianStencil &st,LatticeDoubledGaugeField &U,
     MULT_2SPIN(Xm);
   }
   XM_RECON_ACCUM;
-  //  std::cout << "XM_RECON_ACCUM"<<std::endl;
-  //  std::cout << result_00 <<" "<<result_01 <<" "<<result_02 <<std::endl;
-  //  std::cout << result_10 <<" "<<result_11 <<" "<<result_12 <<std::endl;
-  //  std::cout << result_20 <<" "<<result_21 <<" "<<result_22 <<std::endl;
-  //  std::cout << result_30 <<" "<<result_31 <<" "<<result_32 <<std::endl;
+  //  std::cout<<GridLogMessage << "XM_RECON_ACCUM"<<std::endl;
+  //  std::cout<<GridLogMessage << result_00 <<" "<<result_01 <<" "<<result_02 <<std::endl;
+  //  std::cout<<GridLogMessage << result_10 <<" "<<result_11 <<" "<<result_12 <<std::endl;
+  //  std::cout<<GridLogMessage << result_20 <<" "<<result_21 <<" "<<result_22 <<std::endl;
+  //  std::cout<<GridLogMessage << result_30 <<" "<<result_31 <<" "<<result_32 <<std::endl;
   
   
   // Ym
diff --git a/lib/qcd/action/gauge/WilsonGaugeAction.h b/lib/qcd/action/gauge/WilsonGaugeAction.h
index ad1aecab..9e0b79d9 100644
--- a/lib/qcd/action/gauge/WilsonGaugeAction.h
+++ b/lib/qcd/action/gauge/WilsonGaugeAction.h
@@ -18,9 +18,11 @@ namespace Grid{
       
       virtual RealD S(const GaugeField &U) {
 	RealD plaq = WilsonLoops<MatrixField,GaugeField>::avgPlaquette(U);
-	std::cout << "Plaq : "<<plaq << "\n";
-	double vol = U._grid->gSites();
-	return beta*(1.0 -plaq)*(Nd*(Nd-1.0))*vol*0.5;
+	std::cout<<GridLogMessage << "Plaq : "<<plaq << "\n";
+	RealD vol = U._grid->gSites();
+	RealD action=beta*(1.0 -plaq)*(Nd*(Nd-1.0))*vol*0.5;
+	std::cout << GridLogMessage << "WilsonGauge action "<<action<<std::endl;
+	return action;
       };
       virtual void deriv(const GaugeField &U,GaugeField & dSdU) {
 	//not optimal implementation FIXME
diff --git a/lib/qcd/action/pseudofermion/TwoFlavour.h b/lib/qcd/action/pseudofermion/TwoFlavour.h
new file mode 100644
index 00000000..67f3101c
--- /dev/null
+++ b/lib/qcd/action/pseudofermion/TwoFlavour.h
@@ -0,0 +1,209 @@
+#ifndef QCD_PSEUDOFERMION_TWO_FLAVOUR_H
+#define QCD_PSEUDOFERMION_TWO_FLAVOUR_H
+
+namespace Grid{
+  namespace QCD{
+
+    // Placeholder comments:
+
+    ///////////////////////////////////////
+    // Two flavour ratio
+    ///////////////////////////////////////
+    // S = phi^dag V (Mdag M)^-1 V^dag  phi
+    // dS/du = phi^dag dV (Mdag M)^-1 V^dag  phi
+    //       - phi^dag V (Mdag M)^-1 [ Mdag dM + dMdag M ]  (Mdag M)^-1 V^dag  phi
+    //       + phi^dag V (Mdag M)^-1 dV^dag  phi
+
+    ///////////////////////////////////////
+    // One flavour rational
+    ///////////////////////////////////////
+
+    // S_f = chi^dag *  N(M^dag*M)/D(M^dag*M) * chi
+    //
+    // Here, M is some operator 
+    // N and D makeup the rat. poly 
+    //
+    // Need
+    // dS_f/dU = chi^dag   P/Q d[N/D]  P/Q  chi
+    //
+    // Here N/D \sim R_{-1/2} ~ (M^dagM)^{-1/2}
+    //
+    // N/D is expressed as partial fraction expansion:
+    //
+    //           a0 + \sum_k ak/(M^dagM + bk)
+    //
+    // d[N/D] is then
+    //
+    //          \sum_k -ak [M^dagM+bk]^{-1}  [ dM^dag M + M^dag dM ] [M^dag M + bk]^{-1}
+    //
+    // Need
+    //
+    //       Mf Phi_k = [MdagM+bk]^{-1} Phi
+    //       Mf Phi   = \sum_k ak [MdagM+bk]^{-1} Phi
+    //
+    // With these building blocks
+    //
+    //       dS/dU =  \sum_k -ak Mf Phi_k^dag      [ dM^dag M + M^dag dM ] Mf Phi_k
+    //        S    = innerprodReal(Phi,Mf Phi);
+    
+    ///////////////////////////////////////
+    // One flavour rational ratio
+    ///////////////////////////////////////
+
+    // S_f = chi^dag* P(V^dag*V)/Q(V^dag*V)* N(M^dag*M)/D(M^dag*M)* P(V^dag*V)/Q(V^dag*V)* chi
+    //
+    // Here, M is some 5D operator and V is the Pauli-Villars field
+    // N and D makeup the rat. poly of the M term and P and & makeup the rat.poly of the denom term
+    //
+    // Need
+    // dS_f/dU =  chi^dag d[P/Q]  N/D   P/Q  chi
+    //         +  chi^dag   P/Q d[N/D]  P/Q  chi
+    //         +  chi^dag   P/Q   N/D d[P/Q] chi
+    //
+    // Here P/Q \sim R_{1/4}  ~ (V^dagV)^{1/4}
+    // Here N/D \sim R_{-1/2} ~ (M^dagM)^{-1/2}
+    //
+    // P/Q is expressed as partial fraction expansion:
+    //
+    //           a0 + \sum_k ak/(V^dagV + bk)
+    //
+    // d[P/Q] is then
+    //
+    //          \sum_k -ak [V^dagV+bk]^{-1}  [ dV^dag V + V^dag dV ] [V^dag V + bk]^{-1}
+    //
+    // and similar for N/D.
+    // 
+    // Need
+    //       MpvPhi_k   = [Vdag V + bk]^{-1} chi
+    //
+    //       MpvPhi     = {a0 +  \sum_k ak [Vdag V + bk]^{-1} }chi
+    //
+    //       MfMpvPhi_k = [MdagM+bk]^{-1} MpvPhi
+    //      
+    //       MfMpvPhi   = {a0 +  \sum_k ak [Mdag M + bk]^{-1} } MpvPhi
+    //
+    //       MpvMfMpvPhi_k = [Vdag V + bk]^{-1} MfMpvchi
+    //
+    // With these building blocks
+    //
+    //       dS/dU =  
+    //                 \sum_k -ak MpvPhi_k^dag        [ dV^dag V + V^dag dV ] MpvMfMpvPhi_k           <- deriv on P left
+    //             +   \sum_k -ak MpvMfMpvPhi_k^\dag  [ dV^dag V + V^dag dV ] MpvPhi_k
+    //             +   \sum_k -ak MfMpvPhi_k^dag      [ dM^dag M + M^dag dM ] MfMpvPhi_k
+
+    
+    ////////////////////////////////////////////////////////////////////////
+    // Two flavour pseudofermion action for any dop
+    ////////////////////////////////////////////////////////////////////////
+    template<class GaugeField,class MatrixField,class FermionField>
+      class TwoFlavourPseudoFermionAction : public Action<GaugeField> {
+
+    private:
+      
+      FermionOperator<FermionField,GaugeField> & FermOp;// the basic operator
+
+      OperatorFunction<FermionField> &DerivativeSolver;
+
+      OperatorFunction<FermionField> &ActionSolver;
+
+      GridBase &Grid;
+
+      FermionField Phi; // the pseudo fermion field for this trajectory
+
+    public:
+      /////////////////////////////////////////////////
+      // Pass in required objects.
+      /////////////////////////////////////////////////
+    TwoFlavourPseudoFermionAction(FermionOperator<FermionField,GaugeField>  &Op, 
+				  OperatorFunction<FermionField> & DS,
+				  OperatorFunction<FermionField> & AS,
+				  GridBase &_Grid
+				  ) : FermOp(Op), DerivativeSolver(DS), ActionSolver(AS), Phi(&_Grid), Grid(_Grid) {
+      };
+      
+      //////////////////////////////////////////////////////////////////////////////////////
+      // Push the gauge field in to the dops. Assume any BC's and smearing already applied
+      //////////////////////////////////////////////////////////////////////////////////////
+      virtual void init(const GaugeField &U, GridParallelRNG& pRNG) {
+
+	// P(phi) = e^{- phi^dag (MdagM)^-1 phi}
+	// Phi = Mdag eta 
+	// P(eta) = e^{- eta^dag eta}
+	//
+	// e^{x^2/2 sig^2} => sig^2 = 0.5.
+	// 
+	// So eta should be of width sig = 1/sqrt(2).
+	// and must multiply by 0.707....
+	//
+	// Chroma has this scale factor: two_flavor_monomial_w.h
+	// IroIro: does not use this scale. It is absorbed by a change of vars
+	//         in the Phi integral, and thus is only an irrelevant prefactor for the partition function.
+	//
+	RealD scale = std::sqrt(0.5);
+	FermionField eta(&Grid);
+
+	gaussian(pRNG,eta);
+
+	FermOp.Mdag(eta,Phi);
+
+	Phi=Phi*scale;
+	
+      };
+
+      //////////////////////////////////////////////////////
+      // S = phi^dag (Mdag M)^-1 phi
+      //////////////////////////////////////////////////////
+      virtual RealD S(const GaugeField &U) {
+
+	FermOp.ImportGauge(U);
+
+	FermionField X(&Grid);
+	FermionField Y(&Grid);
+	
+	MdagMLinearOperator<FermionOperator<FermionField,GaugeField> ,FermionField> MdagMOp(FermOp);
+	X=zero;
+	ActionSolver(MdagMOp,Phi,X);
+	MdagMOp.Op(X,Y);
+
+	RealD action = norm2(Y);
+	std::cout << GridLogMessage << "Pseudofermion action "<<action<<std::endl;
+	return action;
+      };
+
+      //////////////////////////////////////////////////////
+      // dS/du = - phi^dag  (Mdag M)^-1 [ Mdag dM + dMdag M ]  (Mdag M)^-1 phi
+      //       = - phi^dag M^-1 dM (MdagM)^-1 phi -  phi^dag (MdagM)^-1 dMdag dM (Mdag)^-1 phi 
+      //
+      //       = - Ydag dM X  - Xdag dMdag Y
+      //
+      //////////////////////////////////////////////////////
+      virtual void deriv(const GaugeField &U,GaugeField & dSdU) {
+
+	FermOp.ImportGauge(U);
+
+	FermionField X(&Grid);
+	FermionField Y(&Grid);
+	GaugeField   tmp(&Grid);
+
+	MdagMLinearOperator<FermionOperator<FermionField,GaugeField> ,FermionField> MdagMOp(FermOp);
+
+	X=zero;
+	DerivativeSolver(MdagMOp,Phi,X);
+	MdagMOp.Op(X,Y);
+
+	// Our conventions really make this UdSdU; We do not differentiate wrt Udag here.
+	// So must take dSdU - adj(dSdU) and left multiply by mom to get dS/dt.
+
+	FermOp.MDeriv(tmp , Y, X,DaggerNo );  dSdU=tmp;
+	FermOp.MDeriv(tmp , X, Y,DaggerYes);  dSdU=dSdU+tmp;
+	
+	dSdU = Ta(dSdU);
+
+      };
+
+    };
+    
+  }
+}
+
+#endif
diff --git a/lib/qcd/hmc/HMC.cc b/lib/qcd/hmc/HMC.cc
index 9db2c604..f7a5d0b8 100644
--- a/lib/qcd/hmc/HMC.cc
+++ b/lib/qcd/hmc/HMC.cc
@@ -7,8 +7,8 @@ namespace Grid{
 	// FIXME fill this constructor  now just default values
 	  
 	////////////////////////////// Default values
-	Nsweeps             = 100;
-	TotalSweeps         = 20;
+	Nsweeps             = 200;
+	TotalSweeps         = 220;
 	ThermalizationSteps = 20;
 	StartingConfig      = 0;
 	SaveInterval        = 1;
@@ -17,8 +17,5 @@ namespace Grid{
 	  
       }
 
-
-   
-
   }
 }
diff --git a/lib/qcd/hmc/HMC.h b/lib/qcd/hmc/HMC.h
index 2da8dd0f..2d410917 100644
--- a/lib/qcd/hmc/HMC.h
+++ b/lib/qcd/hmc/HMC.h
@@ -3,7 +3,7 @@
  * @brief Classes for Hybrid Monte Carlo update
  *
  * @author Guido Cossu
- * Time-stamp: <2015-07-07 14:58:13 neo>
+ * Time-stamp: <2015-07-30 16:58:26 neo>
  */
 //--------------------------------------------------------------------
 #ifndef HMC_INCLUDED
@@ -28,75 +28,89 @@ namespace Grid{
     
     template <class Algorithm> 
     class HybridMonteCarlo{
+
       const HMCparameters Params;
-      GridSerialRNG sRNG;
+
+      GridSerialRNG sRNG; // Fixme: need a RNG management strategy.
+
       Integrator<Algorithm>& MD;
-      
+
+      /////////////////////////////////////////////////////////
+      // Metropolis step
+      /////////////////////////////////////////////////////////
       bool metropolis_test(const RealD DeltaH){
+
 	RealD rn_test;
+
 	RealD prob = std::exp(-DeltaH);
+
 	random(sRNG,rn_test);
       
-	std::cout<< "--------------------------------------------\n";
-	std::cout<< "dH = "<<DeltaH << "  Random = "<< rn_test 
-		 << "\nAcc. Probability = " << ((prob<1.0)? prob: 1.0)<< "   ";
+	std::cout<<GridLogMessage<< "--------------------------------------------\n";
+	std::cout<<GridLogMessage<< "dH = "<<DeltaH << "  Random = "<< rn_test <<"\n";
+	std::cout<<GridLogMessage<< "Acc. Probability = " << ((prob<1.0)? prob: 1.0)<< "   ";
       
 	if((prob >1.0) || (rn_test <= prob)){       // accepted
-	  std::cout <<"-- ACCEPTED\n";
+	  std::cout<<GridLogMessage <<"-- ACCEPTED\n";
 	  return true;
 	} else {                               // rejected
-	  std::cout <<"-- REJECTED\n";
+	  std::cout<<GridLogMessage <<"-- REJECTED\n";
 	  return false;
 	}
+
       }
 
+      /////////////////////////////////////////////////////////
+      // Evolution
+      /////////////////////////////////////////////////////////
       RealD evolve_step(LatticeGaugeField& U){
 	MD.init(U); // set U and initialize P and phi's 
+
 	RealD H0 = MD.S(U); // initial state action  
-	std::cout<<"Total H before = "<< H0 << "\n";
-      
+	std::cout<<GridLogMessage<<"Total H before = "<< H0 << "\n";
+
 	MD.integrate(U);
       
 	RealD H1 = MD.S(U); // updated state action            
-	std::cout<<"Total H after = "<< H1 << "\n";
-      
+	std::cout<<GridLogMessage<<"Total H after = "<< H1 << "\n";
 	return (H1-H0);
       }
       
     public:
-    HybridMonteCarlo(HMCparameters Pms, 
-		     Integrator<Algorithm>& MolDyn):
-      Params(Pms),MD(MolDyn){
-	//FIXME
 
-	// initialize RNGs also with seed
+      /////////////////////////////////////////
+      // Constructor
+      /////////////////////////////////////////
+      HybridMonteCarlo(HMCparameters Pms,  Integrator<Algorithm>& MolDyn): Params(Pms),MD(MolDyn) {
+
+	//FIXME...  initialize RNGs also with seed ; RNG management strategy
 	sRNG.SeedRandomDevice();
+
       }
       ~HybridMonteCarlo(){};
 
+
       void evolve(LatticeGaugeField& Uin){
 	Real DeltaH;
 	
 	// Thermalizations
 	for(int iter=1; iter <= Params.ThermalizationSteps; ++iter){
-	  std::cout << "-- # Thermalization step = "<< iter <<  "\n";
+	  std::cout<<GridLogMessage << "-- # Thermalization step = "<< iter <<  "\n";
 	
 	  DeltaH = evolve_step(Uin);
-	  std::cout<< " dH = "<< DeltaH << "\n";
+	  std::cout<<GridLogMessage<< "dH = "<< DeltaH << "\n";
 	}
 
 	// Actual updates (evolve a copy Ucopy then copy back eventually)
 	LatticeGaugeField Ucopy(Uin._grid);
 	for(int iter=Params.StartingConfig; 
 	    iter < Params.Nsweeps+Params.StartingConfig; ++iter){
-	  std::cout << "-- # Sweep = "<< iter <<  "\n";
+	  std::cout<<GridLogMessage << "-- # Sweep = "<< iter <<  "\n";
 	  
 	  Ucopy = Uin;
 	  DeltaH = evolve_step(Ucopy);
 		
 	  if(metropolis_test(DeltaH)) Uin = Ucopy;
-	  
-	  // here save config and RNG seed
 	}
       }
     };
diff --git a/lib/qcd/hmc/integrators/Integrator.h b/lib/qcd/hmc/integrators/Integrator.h
index f1643339..79143cd6 100644
--- a/lib/qcd/hmc/integrators/Integrator.h
+++ b/lib/qcd/hmc/integrators/Integrator.h
@@ -3,7 +3,7 @@
  * @brief Classes for the Molecular Dynamics integrator
  *
  * @author Guido Cossu
- * Time-stamp: <2015-07-07 14:58:40 neo>
+ * Time-stamp: <2015-07-30 16:21:29 neo>
  */
 //--------------------------------------------------------------------
 
@@ -71,7 +71,6 @@ namespace Grid{
 	}
       }
 
-
       void update_U(LatticeGaugeField&U, double ep){
 	//rewrite exponential to deal automatically  with the lorentz index?
 	LatticeColourMatrix Umu(U._grid);
@@ -85,7 +84,6 @@ namespace Grid{
 
       }
       
-
       
       friend void IntegratorAlgorithm::step (LatticeGaugeField& U, 
 					     int level, std::vector<int>& clock,
@@ -99,11 +97,9 @@ namespace Grid{
       
       ~Integrator(){}
 
-
       //Initialization of momenta and actions
       void init(LatticeGaugeField& U){
-	std::cout<< "Integrator init\n";
-
+	std::cout<<GridLogMessage<< "Integrator init\n";
 	MDutils::generate_momenta(*P,pRNG);
 	for(int level=0; level< as.size(); ++level){
 	  for(int actionID=0; actionID<as[level].actions.size(); ++actionID){
@@ -112,7 +108,6 @@ namespace Grid{
 	}
       }
 
-      
       // Calculate action
       RealD S(LatticeGaugeField& U){
 	LatticeComplex Hloc(U._grid);
@@ -126,12 +121,14 @@ namespace Grid{
 	
 	RealD H = Hsum.real();
 
-	std::cout << "H_p = "<< H << "\n";
+	std::cout<<GridLogMessage << "Momentum action H_p = "<< H << "\n";
 
 	// Actions
 	for(int level=0; level<as.size(); ++level)
 	  for(int actionID=0; actionID<as[level].actions.size(); ++actionID)
 	    H += as[level].actions.at(actionID)->S(U);
+
+	std::cout<<GridLogMessage << "Total action H = "<< H << "\n";
 	
 	return H;
       }
diff --git a/lib/qcd/hmc/integrators/Integrator_algorithm.h b/lib/qcd/hmc/integrators/Integrator_algorithm.h
index c192b404..48974c90 100644
--- a/lib/qcd/hmc/integrators/Integrator_algorithm.h
+++ b/lib/qcd/hmc/integrators/Integrator_algorithm.h
@@ -33,35 +33,32 @@ namespace Grid{
 	for(int l=1; l<=level; ++l) fin*= 2.0*Integ->as[l].multiplier;
 	fin = 3*Integ->Params.MDsteps*fin -1;
 	
-	
 	for(int e=0; e<Integ->as[level].multiplier; ++e){
-	  
 	  if(clock[level] == 0){    // initial half step 
 	    Integ->update_P(U,level,lambda*eps);
 	    ++clock[level];
-	    for(int l=0; l<level;++l) std::cout<<"   ";
-	    std::cout<<"P "<< clock[level] <<std::endl;
+	    for(int l=0; l<level;++l) std::cout<<GridLogMessage<<"   ";
+	    std::cout<<GridLogMessage<<"P "<< clock[level] <<std::endl;
 	  }
 	  
 	  if(level == fl){          // lowest level 
 	    Integ->update_U(U,0.5*eps);
-	    
-	    for(int l=0; l<level;++l) std::cout<<"   ";
-	    std::cout<<"U "<< (clock[level]+1) <<std::endl;
+	    for(int l=0; l<level;++l) std::cout<<GridLogMessage<<"   ";
+	    std::cout<<GridLogMessage<<"U "<< (clock[level]+1) <<std::endl;
 	  }else{                 // recursive function call 
 	    step(U,level+1,clock, Integ);
 	  }
 	  
 	  Integ->update_P(U,level,(1.0-2.0*lambda)*eps);
 	  ++clock[level];
-	  for(int l=0; l<level;++l) std::cout<<"   ";
-	  std::cout<<"P "<< (clock[level]) <<std::endl;
+	  for(int l=0; l<level;++l) std::cout<<GridLogMessage<<"   ";
+	  std::cout<<GridLogMessage<<"P "<< (clock[level]) <<std::endl;
 	  
 	  if(level == fl){          // lowest level 
 	    Integ->update_U(U,0.5*eps);
 	    
-	    for(int l=0; l<level;++l) std::cout<<"   ";
-	    std::cout<<"U "<< (clock[level]+1) <<std::endl;
+	    for(int l=0; l<level;++l) std::cout<<GridLogMessage<<"   ";
+	    std::cout<<GridLogMessage<<"U "<< (clock[level]+1) <<std::endl;
 	  }else{                 // recursive function call 
 	    step(U,level+1,clock, Integ);
 	  }    
@@ -71,19 +68,17 @@ namespace Grid{
 	    Integ->update_P(U,level,lambda*eps);
 	    
 	    ++clock[level];
-	    for(int l=0; l<level;++l) std::cout<<"   ";
-	    std::cout<<"P "<< clock[level] <<std::endl;
+	    for(int l=0; l<level;++l) std::cout<<GridLogMessage<<"   ";
+	    std::cout<<GridLogMessage<<"P "<< clock[level] <<std::endl;
 	  }else{                  // bulk step
 	    Integ->update_P(U,level,lambda*2.0*eps);
 	    
 	    clock[level]+=2;
-	    for(int l=0; l<level;++l) std::cout<<"   ";
-	    std::cout<<"P "<< clock[level] <<std::endl;
+	    for(int l=0; l<level;++l) std::cout<<GridLogMessage<<"   ";
+	    std::cout<<GridLogMessage<<"P "<< clock[level] <<std::endl;
 	  }
 	}
-	
-	
-	
+		
       }
       
     };
@@ -93,6 +88,7 @@ namespace Grid{
       void step (LatticeLorentzColourMatrix& U, 
 		 int level, std::vector<int>& clock,
 		 Integrator<LeapFrog>* Integ){
+
 	// level  : current level
 	// fl     : final level
 	// eps    : current step size
@@ -112,34 +108,32 @@ namespace Grid{
 	  if(clock[level] == 0){    // initial half step
 	    Integ->update_P(U, level,eps/2.0);
 	    ++clock[level];
-	    for(int l=0; l<level;++l) std::cout<<"   ";
-	    std::cout<<"P "<< 0.5*clock[level] <<std::endl;
+	    for(int l=0; l<level;++l) std::cout<<GridLogMessage<<"   ";
+	    std::cout<<GridLogMessage<<"P "<< 0.5*clock[level] <<std::endl;
 	  }
+
 	  if(level == fl){          // lowest level
 	    Integ->update_U(U, eps);
-	    for(int l=0; l<level;++l) std::cout<<"   ";
-	    std::cout<<"U "<< 0.5*(clock[level]+1) <<std::endl;
+	    for(int l=0; l<level;++l) std::cout<<GridLogMessage<<"   ";
+	    std::cout<<GridLogMessage<<"U "<< 0.5*(clock[level]+1) <<std::endl;
 	  }else{                 // recursive function call
 	    step(U, level+1,clock, Integ);
 	  }
+
 	  if(clock[level] == fin){  // final half step
 	    Integ->update_P(U, level,eps/2.0);
-	    
 	    ++clock[level];
-	    for(int l=0; l<level;++l) std::cout<<"   ";
-	    std::cout<<"P "<< 0.5*clock[level] <<std::endl;
+	    for(int l=0; l<level;++l) std::cout<<GridLogMessage<<"   ";
+	    std::cout<<GridLogMessage<<"P "<< 0.5*clock[level] <<std::endl;
 	  }else{                  // bulk step
 	    Integ->update_P(U, level,eps);
-	    
 	    clock[level]+=2;
-	    for(int l=0; l<level;++l) std::cout<<"   ";
-	    std::cout<<"P "<< 0.5*clock[level] <<std::endl;
+	    for(int l=0; l<level;++l) std::cout<<GridLogMessage<<"   ";
+	    std::cout<<GridLogMessage<<"P "<< 0.5*clock[level] <<std::endl;
 	  }
+
 	}
 
-
-
-
       }
     };
 
diff --git a/lib/qcd/hmc/integrators/Integrator_base.h b/lib/qcd/hmc/integrators/Integrator_base.h
index 38798d67..1cc32808 100644
--- a/lib/qcd/hmc/integrators/Integrator_base.h
+++ b/lib/qcd/hmc/integrators/Integrator_base.h
@@ -65,6 +65,18 @@ namespace Grid{
 	for(int a=0; a<as[level].size(); ++a){
 	  LatticeLorentzColourMatrix force(U._grid);
 	  as[level].at(a)->deriv(U,force);
+
+	  Complex dSdt=0.0;
+	  for(int mu=0;mu<Nd;mu++){
+	    LatticeColourMatrix forcemu(U._grid);
+	    LatticeColourMatrix mommu(U._grid);
+	    forcemu=PeekIndex<LorentzIndex>(force,mu);
+	    mommu=PeekIndex<LorentzIndex>(*P,mu);
+
+	    dSdt += sum(trace(forcemu*(*P)));
+
+	  }	  
+	  std::cout << GridLogMessage << " action "<<level<<","<<a<<" dSdt "<< dSdt << " dt "<<ep  <<std::endl;
 	  *P -= force*ep;
 	}
       }
@@ -101,7 +113,7 @@ namespace Grid{
       //Initialization of momenta and actions
       void init(LatticeLorentzColourMatrix& U,
 		GridParallelRNG& pRNG){
-	std::cout<< "Integrator init\n";
+	std::cout<<GridLogMessage<< "Integrator init\n";
 	if (!P)
 	  P = new LatticeLorentzColourMatrix(U._grid);
 	MDutils::generate_momenta(*P,pRNG);
@@ -172,13 +184,13 @@ namespace Grid{
 	  if(clock[level] == 0){    // initial half step
 	    Integ->update_P(U, level,eps/2);
 	    ++clock[level];
-	    for(int l=0; l<level;++l) std::cout<<"   ";
-	    std::cout<<"P "<< 0.5*clock[level] <<std::endl;
+	    for(int l=0; l<level;++l) std::cout<<GridLogMessage<<"   ";
+	    std::cout<<GridLogMessage<<"P "<< 0.5*clock[level] <<std::endl;
 	  }
 	  if(level == fl){          // lowest level
 	    Integ->update_U(U, eps);
-	    for(int l=0; l<level;++l) std::cout<<"   ";
-	    std::cout<<"U "<< 0.5*(clock[level]+1) <<std::endl;
+	    for(int l=0; l<level;++l) std::cout<<GridLogMessage<<"   ";
+	    std::cout<<GridLogMessage<<"U "<< 0.5*(clock[level]+1) <<std::endl;
 	  }else{                 // recursive function call
 	    step(U, level+1,clock, Integ);
 	  }
@@ -186,14 +198,14 @@ namespace Grid{
 	    Integ->update_P(U, level,eps/2);
 	    
 	    ++clock[level];
-	    for(int l=0; l<level;++l) std::cout<<"   ";
-	    std::cout<<"P "<< 0.5*clock[level] <<std::endl;
+	    for(int l=0; l<level;++l) std::cout<<GridLogMessage<<"   ";
+	    std::cout<<GridLogMessage<<"P "<< 0.5*clock[level] <<std::endl;
 	  }else{                  // bulk step
 	    Integ->update_P(U, level,eps);
 	    
 	    clock[level]+=2;
-	    for(int l=0; l<level;++l) std::cout<<"   ";
-	    std::cout<<"P "<< 0.5*clock[level] <<std::endl;
+	    for(int l=0; l<level;++l) std::cout<<GridLogMessage<<"   ";
+	    std::cout<<GridLogMessage<<"P "<< 0.5*clock[level] <<std::endl;
 	  }
 	}
 
diff --git a/lib/qcd/utils/SUn.h b/lib/qcd/utils/SUn.h
index 10313f5c..66e5f8ab 100644
--- a/lib/qcd/utils/SUn.h
+++ b/lib/qcd/utils/SUn.h
@@ -372,7 +372,7 @@ Note that in step D setting B ~ X - A and using B in place of A in step E will g
     LatticeReal d(grid); d=zero;
     LatticeReal alpha(grid);
 
-    //    std::cout<<"xi "<<xi <<std::endl;
+    //    std::cout<<GridLogMessage<<"xi "<<xi <<std::endl;
     alpha = toReal(2.0*xi);
 
     do { 
@@ -468,11 +468,11 @@ Note that in step D setting B ~ X - A and using B in place of A in step E will g
     LatticeMatrix Vcheck(grid);
     Vcheck = zero;
     Vcheck = where(Accepted,V*adj(V) - 1.0,Vcheck);
-    //    std::cout << "SU3 check " <<norm2(Vcheck)<<std::endl;
+    //    std::cout<<GridLogMessage << "SU3 check " <<norm2(Vcheck)<<std::endl;
     assert(norm2(Vcheck)<1.0e-4);
     
     // Verify the link stays in SU(3)
-    //    std::cout <<"Checking the modified link"<<std::endl;
+    //    std::cout<<GridLogMessage <<"Checking the modified link"<<std::endl;
     Vcheck = link*adj(link) - 1.0;
     assert(norm2(Vcheck)<1.0e-4);
     /////////////////////////////////
@@ -483,42 +483,42 @@ Note that in step D setting B ~ X - A and using B in place of A in step E will g
     for(int gen=0;gen<generators();gen++){
       Matrix ta;
       generator(gen,ta);
-      std::cout<< "Nc = "<<ncolour<<" t_"<<gen<<std::endl;
-      std::cout<<ta<<std::endl;
+      std::cout<<GridLogMessage<< "Nc = "<<ncolour<<" t_"<<gen<<std::endl;
+      std::cout<<GridLogMessage<<ta<<std::endl;
     }
   }
 
   static void testGenerators(void){
     Matrix ta;
     Matrix tb;
-    std::cout<<"Checking trace ta tb is 0.5 delta_ab"<<std::endl;
+    std::cout<<GridLogMessage<<"Checking trace ta tb is 0.5 delta_ab"<<std::endl;
     for(int a=0;a<generators();a++){
       for(int b=0;b<generators();b++){
 	generator(a,ta);
 	generator(b,tb);
 	Complex tr =TensorRemove(trace(ta*tb)); 
-	std::cout<<tr<<" ";
+	std::cout<<GridLogMessage<<tr<<" ";
 	if(a==b) assert(abs(tr-Complex(0.5))<1.0e-6);
 	if(a!=b) assert(abs(tr)<1.0e-6);
       }
-      std::cout<<std::endl;
+      std::cout<<GridLogMessage<<std::endl;
     }
-    std::cout<<"Checking hermitian"<<std::endl;
+    std::cout<<GridLogMessage<<"Checking hermitian"<<std::endl;
     for(int a=0;a<generators();a++){
       generator(a,ta);
-      std::cout<<a<<" ";
+      std::cout<<GridLogMessage<<a<<" ";
       assert(norm2(ta-adj(ta))<1.0e-6);
     }    
-    std::cout<<std::endl;
+    std::cout<<GridLogMessage<<std::endl;
 
-    std::cout<<"Checking traceless"<<std::endl;
+    std::cout<<GridLogMessage<<"Checking traceless"<<std::endl;
     for(int a=0;a<generators();a++){
       generator(a,ta);
       Complex tr =TensorRemove(trace(ta)); 
-      std::cout<<a<<" ";
+      std::cout<<GridLogMessage<<a<<" ";
       assert(abs(tr)<1.0e-6);
     }    
-    std::cout<<std::endl;
+    std::cout<<GridLogMessage<<std::endl;
   }
 
   // reunitarise??
@@ -554,9 +554,7 @@ Note that in step D setting B ~ X - A and using B in place of A in step E will g
     for(int a=0;a<generators();a++){
       gaussian(pRNG,ca); 
       generator(a,ta);
-      
       la=toComplex(ca)*ci*ta;
-   
       out += la; 
     }
 
diff --git a/lib/simd/Grid_vector_types.h b/lib/simd/Grid_vector_types.h
index 4a0ea33b..fe3b6cba 100644
--- a/lib/simd/Grid_vector_types.h
+++ b/lib/simd/Grid_vector_types.h
@@ -414,7 +414,7 @@ namespace Grid {
   template<class S, class V >
   inline Grid_simd< S, V> outerProduct(const Grid_simd< S, V> &l, const Grid_simd< S, V> & r)
   {
-    return l*r;
+    return l*conjugate(r);
   }
 
   template<class S, class V >
diff --git a/lib/stamp-h1 b/lib/stamp-h1
index 2a8e9ffe..1b3b5f4c 100644
--- a/lib/stamp-h1
+++ b/lib/stamp-h1
@@ -1 +1 @@
-timestamp for lib/GridConfig.h
+timestamp for lib/Config.h
diff --git a/lib/tensors/Tensor_Ta.h b/lib/tensors/Tensor_Ta.h
index 724cbd6e..e71e54fd 100644
--- a/lib/tensors/Tensor_Ta.h
+++ b/lib/tensors/Tensor_Ta.h
@@ -7,11 +7,12 @@ namespace Grid {
   /////////////////////////////////////////////// 
   // Ta function for scalar, vector, matrix
   /////////////////////////////////////////////// 
+  /*
   inline ComplexF Ta( const ComplexF &arg){    return arg;}
   inline ComplexD Ta( const ComplexD &arg){    return arg;}
   inline RealF Ta( const RealF &arg){    return arg;}
   inline RealD Ta( const RealD &arg){    return arg;}
-
+  */
 
   template<class vtype> inline iScalar<vtype> Ta(const iScalar<vtype>&r)
     {
@@ -29,10 +30,11 @@ namespace Grid {
     }
   template<class vtype,int N> inline iMatrix<vtype,N> Ta(const iMatrix<vtype,N> &arg)
     {
-      iMatrix<vtype,N> ret(arg);
-      double factor = (1/(double)N);
-      ret = (ret - adj(arg))*0.5;
-      ret -= trace(ret)*factor;
+      iMatrix<vtype,N> ret;
+
+      double factor = (1.0/(double)N);
+      ret= (arg - adj(arg))*0.5;
+      ret=ret - (trace(ret)*factor);
       return ret;
     }
 
diff --git a/lib/tensors/Tensor_extract_merge.h b/lib/tensors/Tensor_extract_merge.h
index 21ffd778..6209209f 100644
--- a/lib/tensors/Tensor_extract_merge.h
+++ b/lib/tensors/Tensor_extract_merge.h
@@ -57,11 +57,11 @@ inline void extract(typename std::enable_if<!isGridTensor<vsimd>::value, const v
     extracted[i]=buf[i*s];
     for(int ii=1;ii<s;ii++){
       if ( buf[i*s]!=buf[i*s+ii] ){
-	std::cout << " SIMD extract failure splat = "<<s<<" ii "<<ii<<" " <<Nextr<<" "<< Nsimd<<" "<<std::endl;
+	std::cout<<GridLogMessage << " SIMD extract failure splat = "<<s<<" ii "<<ii<<" " <<Nextr<<" "<< Nsimd<<" "<<std::endl;
 	for(int vv=0;vv<Nsimd;vv++) {
-	  std::cout<< buf[vv]<<" ";
+	  std::cout<<GridLogMessage<< buf[vv]<<" ";
 	}
-	std::cout<<std::endl;
+	std::cout<<GridLogMessage<<std::endl;
 	assert(0);
       }
       assert(buf[i*s]==buf[i*s+ii]);
diff --git a/lib/tensors/Tensor_outer.h b/lib/tensors/Tensor_outer.h
index 816f6ad8..d0150709 100644
--- a/lib/tensors/Tensor_outer.h
+++ b/lib/tensors/Tensor_outer.h
@@ -28,11 +28,13 @@ auto outerProduct (const iScalar<l>& lhs,const iScalar<r>& rhs) -> iScalar<declt
 
 inline ComplexF outerProduct(const ComplexF &l, const ComplexF& r)
 {
-  return l*r;
+  std::cout << "outer product taking conj "<<r<<" "<<conj(r)<<std::endl;
+  return l*conj(r);
 }
 inline ComplexD outerProduct(const ComplexD &l, const ComplexD& r)
 {
-  return l*r;
+  std::cout << "outer product taking conj "<<r<<" "<<conj(r)<<std::endl;
+  return l*conj(r);
 }
 inline RealF outerProduct(const RealF &l, const RealF& r)
 {
diff --git a/lib/tensors/Tensor_trace.h b/lib/tensors/Tensor_trace.h
index 4cf5edc2..bd75bb91 100644
--- a/lib/tensors/Tensor_trace.h
+++ b/lib/tensors/Tensor_trace.h
@@ -31,6 +31,16 @@ inline auto trace(const iScalar<vtype> &arg) -> iScalar<decltype(trace(arg._inte
     return ret;
 }
 
+template<class vtype,int N>
+  inline auto trace(const iVector<vtype,N> &arg) -> iVector<decltype(trace(arg._internal[0])),N>
+{
+  iVector<decltype(trace(arg._internal[0])),N> ret;
+  for(int i=0;i<N;i++){
+    ret._internal[i]=trace(arg._internal[i]);
+  }
+  return ret;
+}
+
 
 }
 #endif
diff --git a/tests/Make.inc b/tests/Make.inc
index 31c39cd4..00d3be99 100644
--- a/tests/Make.inc
+++ b/tests/Make.inc
@@ -1,5 +1,5 @@
 
-bin_PROGRAMS = Test_cayley_cg Test_cayley_coarsen_support Test_cayley_even_odd Test_cayley_ldop_cr Test_cf_coarsen_support Test_cf_cr_unprec Test_contfrac_cg Test_contfrac_even_odd Test_cshift Test_cshift_red_black Test_dwf_cg_prec Test_dwf_cg_schur Test_dwf_cg_unprec Test_dwf_cr_unprec Test_dwf_even_odd Test_dwf_fpgcr Test_dwf_hdcr Test_gamma Test_GaugeAction Test_hmc_WilsonGauge Test_lie_generators Test_main Test_multishift_sqrt Test_nersc_io Test_quenched_update Test_remez Test_rng Test_rng_fixed Test_simd Test_stencil Test_wilson_cg_prec Test_wilson_cg_schur Test_wilson_cg_unprec Test_wilson_cr_unprec Test_wilson_even_odd
+bin_PROGRAMS = Test_cayley_cg Test_cayley_coarsen_support Test_cayley_even_odd Test_cayley_ldop_cr Test_cf_coarsen_support Test_cf_cr_unprec Test_contfrac_cg Test_contfrac_even_odd Test_cshift Test_cshift_red_black Test_dwf_cg_prec Test_dwf_cg_schur Test_dwf_cg_unprec Test_dwf_cr_unprec Test_dwf_even_odd Test_dwf_fpgcr Test_dwf_hdcr Test_gamma Test_GaugeAction Test_hmc_WilsonFermionGauge Test_hmc_WilsonGauge Test_lie_generators Test_main Test_multishift_sqrt Test_nersc_io Test_quenched_update Test_remez Test_rng Test_rng_fixed Test_simd Test_stencil Test_wilson_cg_prec Test_wilson_cg_schur Test_wilson_cg_unprec Test_wilson_cr_unprec Test_wilson_even_odd Test_wilson_force Test_wilson_force_phiMdagMphi Test_wilson_force_phiMphi
 
 
 Test_cayley_cg_SOURCES=Test_cayley_cg.cc
@@ -78,6 +78,10 @@ Test_GaugeAction_SOURCES=Test_GaugeAction.cc
 Test_GaugeAction_LDADD=-lGrid
 
 
+Test_hmc_WilsonFermionGauge_SOURCES=Test_hmc_WilsonFermionGauge.cc
+Test_hmc_WilsonFermionGauge_LDADD=-lGrid
+
+
 Test_hmc_WilsonGauge_SOURCES=Test_hmc_WilsonGauge.cc
 Test_hmc_WilsonGauge_LDADD=-lGrid
 
@@ -141,3 +145,15 @@ Test_wilson_cr_unprec_LDADD=-lGrid
 Test_wilson_even_odd_SOURCES=Test_wilson_even_odd.cc
 Test_wilson_even_odd_LDADD=-lGrid
 
+
+Test_wilson_force_SOURCES=Test_wilson_force.cc
+Test_wilson_force_LDADD=-lGrid
+
+
+Test_wilson_force_phiMdagMphi_SOURCES=Test_wilson_force_phiMdagMphi.cc
+Test_wilson_force_phiMdagMphi_LDADD=-lGrid
+
+
+Test_wilson_force_phiMphi_SOURCES=Test_wilson_force_phiMphi.cc
+Test_wilson_force_phiMphi_LDADD=-lGrid
+
diff --git a/tests/Test_GaugeAction.cc b/tests/Test_GaugeAction.cc
index 612eb693..d779cd20 100644
--- a/tests/Test_GaugeAction.cc
+++ b/tests/Test_GaugeAction.cc
@@ -94,7 +94,7 @@ int main (int argc, char ** argv)
   double vol = Fine.gSites();
   Complex PlaqScale(1.0/vol/6.0/3.0);
   RealD   StapScale(1.0/vol/6.0/3.0);
-  std::cout <<"PlaqScale" << PlaqScale<<std::endl;
+  std::cout<<GridLogMessage <<"PlaqScale" << PlaqScale<<std::endl;
   std::vector<TComplex> Plaq_T(orthosz);
   sliceSum(Plaq,Plaq_T,Nd-1);
   int Nt = Plaq_T.size();
@@ -105,20 +105,20 @@ int main (int argc, char ** argv)
   for(int t=0;t<Nt;t++){
     Plaq_T_sum = Plaq_T_sum+Plaq_T[t];
     Complex Pt=TensorRemove(Plaq_T[t]);
-    std::cout << "sliced ["<<t<<"]" <<Pt*PlaqScale*Real(Nt) << std::endl;
+    std::cout<<GridLogMessage << "sliced ["<<t<<"]" <<Pt*PlaqScale*Real(Nt) << std::endl;
   }
 
   {
     Complex Pt = TensorRemove(Plaq_T_sum);
-    std::cout << "total " <<Pt*PlaqScale<<std::endl;
+    std::cout<<GridLogMessage << "total " <<Pt*PlaqScale<<std::endl;
   }  
 
   TComplex Tp = sum(Plaq);
   Complex p  = TensorRemove(Tp);
-  std::cout << "calculated plaquettes " <<p*PlaqScale<<std::endl;
+  std::cout<<GridLogMessage << "calculated plaquettes " <<p*PlaqScale<<std::endl;
 
   RealD avg_plaq = ColourWilsonLoops::avgPlaquette(Umu);
-  std::cout << "NEW : calculated real plaquettes " <<avg_plaq<<std::endl;
+  std::cout<<GridLogMessage << "NEW : calculated real plaquettes " <<avg_plaq<<std::endl;
 
   RealD stap_plaq=0.0;
   LatticeColourMatrix stap(&Fine);
@@ -130,16 +130,16 @@ int main (int argc, char ** argv)
     Complex s  = TensorRemove(Ts);
     stap_plaq+=real(s);
   }
-  std::cout << "NEW : plaquette via staples"<< stap_plaq*StapScale*0.25<< std::endl;
+  std::cout<<GridLogMessage << "NEW : plaquette via staples"<< stap_plaq*StapScale*0.25<< std::endl;
   Complex LinkTraceScale(1.0/vol/4.0/3.0);
   TComplex Tl = sum(LinkTrace);
   Complex l  = TensorRemove(Tl);
-  std::cout << "calculated link trace " <<l*LinkTraceScale<<std::endl;
+  std::cout<<GridLogMessage << "calculated link trace " <<l*LinkTraceScale<<std::endl;
 
   blockSum(cPlaq,Plaq);
   TComplex TcP = sum(cPlaq);
   Complex ll= TensorRemove(TcP);
-  std::cout << "coarsened plaquettes sum to " <<ll*PlaqScale<<std::endl;
+  std::cout<<GridLogMessage << "coarsened plaquettes sum to " <<ll*PlaqScale<<std::endl;
 
 
   Grid_finalize();
diff --git a/tests/Test_cayley_cg.cc b/tests/Test_cayley_cg.cc
index 23004be9..f8d1187c 100644
--- a/tests/Test_cayley_cg.cc
+++ b/tests/Test_cayley_cg.cc
@@ -52,7 +52,7 @@ int main (int argc, char ** argv)
   Grid_init(&argc,&argv);
 
   int threads = GridThread::GetThreads();
-  std::cout << "Grid is setup to use "<<threads<<" threads"<<std::endl;
+  std::cout<<GridLogMessage << "Grid is setup to use "<<threads<<" threads"<<std::endl;
 
   const int Ls=8;
   GridCartesian         * UGrid   = SpaceTimeGrid::makeFourDimGrid(GridDefaultLatt(), GridDefaultSimd(Nd,vComplex::Nsimd()),GridDefaultMpi());
@@ -71,33 +71,33 @@ int main (int argc, char ** argv)
 
   RealD mass=0.1;
   RealD M5  =1.8;
-  std::cout <<"DomainWallFermion test"<<std::endl;
+  std::cout<<GridLogMessage <<"DomainWallFermion test"<<std::endl;
   DomainWallFermion Ddwf(Umu,*FGrid,*FrbGrid,*UGrid,*UrbGrid,mass,M5);
   TestCGinversions<DomainWallFermion>(Ddwf,FGrid,FrbGrid,UGrid,UrbGrid,mass,M5,&RNG4,&RNG5);
 
   RealD b=1.5;// Scale factor b+c=2, b-c=1
   RealD c=0.5;
-  std::cout <<"MobiusFermion test"<<std::endl;
+  std::cout<<GridLogMessage <<"MobiusFermion test"<<std::endl;
   MobiusFermion Dmob(Umu,*FGrid,*FrbGrid,*UGrid,*UrbGrid,mass,M5,b,c);
   TestCGinversions<MobiusFermion>(Dmob,FGrid,FrbGrid,UGrid,UrbGrid,mass,M5,&RNG4,&RNG5);
 
-  std::cout <<"MobiusZolotarevFermion test"<<std::endl;
+  std::cout<<GridLogMessage <<"MobiusZolotarevFermion test"<<std::endl;
   MobiusZolotarevFermion Dzolo(Umu,*FGrid,*FrbGrid,*UGrid,*UrbGrid,mass,M5,b,c,0.1,2.0);
   TestCGinversions<MobiusZolotarevFermion>(Dzolo,FGrid,FrbGrid,UGrid,UrbGrid,mass,M5,&RNG4,&RNG5);
 
-  std::cout <<"ScaledShamirFermion test"<<std::endl;
+  std::cout<<GridLogMessage <<"ScaledShamirFermion test"<<std::endl;
   ScaledShamirFermion Dsham(Umu,*FGrid,*FrbGrid,*UGrid,*UrbGrid,mass,M5,2.0);
   TestCGinversions<ScaledShamirFermion>(Dsham,FGrid,FrbGrid,UGrid,UrbGrid,mass,M5,&RNG4,&RNG5);
 
-  std::cout <<"ShamirZolotarevFermion test"<<std::endl;
+  std::cout<<GridLogMessage <<"ShamirZolotarevFermion test"<<std::endl;
   ShamirZolotarevFermion Dshamz(Umu,*FGrid,*FrbGrid,*UGrid,*UrbGrid,mass,M5,0.1,2.0);
   TestCGinversions<ShamirZolotarevFermion>(Dshamz,FGrid,FrbGrid,UGrid,UrbGrid,mass,M5,&RNG4,&RNG5);
 
-  std::cout <<"OverlapWilsonCayleyTanhFermion test"<<std::endl;
+  std::cout<<GridLogMessage <<"OverlapWilsonCayleyTanhFermion test"<<std::endl;
   OverlapWilsonCayleyTanhFermion Dov(Umu,*FGrid,*FrbGrid,*UGrid,*UrbGrid,mass,M5,1.0);
   TestCGinversions<OverlapWilsonCayleyTanhFermion>(Dov,FGrid,FrbGrid,UGrid,UrbGrid,mass,M5,&RNG4,&RNG5);
 
-  std::cout <<"OverlapWilsonCayleyZolotarevFermion test"<<std::endl;
+  std::cout<<GridLogMessage <<"OverlapWilsonCayleyZolotarevFermion test"<<std::endl;
   OverlapWilsonCayleyZolotarevFermion Dovz(Umu,*FGrid,*FrbGrid,*UGrid,*UrbGrid,mass,M5,0.1,2.0);
   TestCGinversions<OverlapWilsonCayleyZolotarevFermion>(Dovz,FGrid,FrbGrid,UGrid,UrbGrid,mass,M5,&RNG4,&RNG5);
 
@@ -111,11 +111,11 @@ void  TestCGinversions(What & Ddwf,
 		       GridParallelRNG *RNG4,
 		       GridParallelRNG *RNG5)
 {
-  std::cout << "Testing unpreconditioned inverter"<<std::endl;
+  std::cout<<GridLogMessage << "Testing unpreconditioned inverter"<<std::endl;
   TestCGunprec<What>(Ddwf,FGrid,FrbGrid,UGrid,UrbGrid,mass,M5,RNG4,RNG5);
-  std::cout << "Testing red black preconditioned inverter"<<std::endl;
+  std::cout<<GridLogMessage << "Testing red black preconditioned inverter"<<std::endl;
   TestCGprec<What>(Ddwf,FGrid,FrbGrid,UGrid,UrbGrid,mass,M5,RNG4,RNG5);
-  std::cout << "Testing red black Schur inverter"<<std::endl;
+  std::cout<<GridLogMessage << "Testing red black Schur inverter"<<std::endl;
   TestCGschur<What>(Ddwf,FGrid,FrbGrid,UGrid,UrbGrid,mass,M5,RNG4,RNG5);
 }
 
diff --git a/tests/Test_cayley_coarsen_support.cc b/tests/Test_cayley_coarsen_support.cc
index 473fa10a..3ddf6205 100644
--- a/tests/Test_cayley_coarsen_support.cc
+++ b/tests/Test_cayley_coarsen_support.cc
@@ -57,8 +57,8 @@ int main (int argc, char ** argv)
   Complex cone(1.0,0.0);
   for(int nn=0;nn<Nd;nn++){
     if(1) {
-      if (nn>2) { U[nn]=zero; std::cout << "zeroing gauge field in dir "<<nn<<std::endl; }
-      else      { U[nn]=cone; std::cout << "unit gauge field in dir "<<nn<<std::endl; }
+      if (nn>2) { U[nn]=zero; std::cout<<GridLogMessage << "zeroing gauge field in dir "<<nn<<std::endl; }
+      else      { U[nn]=cone; std::cout<<GridLogMessage << "unit gauge field in dir "<<nn<<std::endl; }
     }
     pokeIndex<LorentzIndex>(Umu,U[nn],nn);
   }
@@ -75,19 +75,19 @@ int main (int argc, char ** argv)
   
   for(int d=0;d<4;d++){
     HermIndefOp.OpDir(src,tmp,d+1,+1); result=result+tmp; 
-    std::cout<<"dir "<<d<<" tmp "<<norm2(tmp)<<std::endl;
+    std::cout<<GridLogMessage<<"dir "<<d<<" tmp "<<norm2(tmp)<<std::endl;
     HermIndefOp.OpDir(src,tmp,d+1,-1); result=result+tmp;
-    std::cout<<"dir "<<d<<" tmp "<<norm2(tmp)<<std::endl;
+    std::cout<<GridLogMessage<<"dir "<<d<<" tmp "<<norm2(tmp)<<std::endl;
   }
   err = result-ref;
-  std::cout<<"Error "<<norm2(err)<<std::endl;
+  std::cout<<GridLogMessage<<"Error "<<norm2(err)<<std::endl;
 
   const int nbasis = 2;
   LatticeFermion prom(FGrid);
 
   std::vector<LatticeFermion> subspace(nbasis,FGrid);
 
-  std::cout<<"Calling Aggregation class" <<std::endl;
+  std::cout<<GridLogMessage<<"Calling Aggregation class" <<std::endl;
 
   MdagMLinearOperator<DomainWallFermion,LatticeFermion> HermDefOp(Ddwf);
   typedef Aggregation<vSpinColourVector,vTComplex,nbasis> Subspace;
@@ -96,7 +96,7 @@ int main (int argc, char ** argv)
 
   subspace=Aggregates.subspace;
 
-  std::cout << "Called aggregation class"<< std::endl;
+  std::cout<<GridLogMessage << "Called aggregation class"<< std::endl;
 
   typedef CoarsenedMatrix<vSpinColourVector,vTComplex,nbasis> LittleDiracOperator;
   typedef LittleDiracOperator::CoarseVector CoarseVector;
@@ -119,7 +119,7 @@ int main (int argc, char ** argv)
     prom=prom+subspace[b];
   }
   err=err-prom; 
-  std::cout<<"Promoted back from subspace err "<<norm2(err)<<std::endl;
+  std::cout<<GridLogMessage<<"Promoted back from subspace err "<<norm2(err)<<std::endl;
 
   HermIndefOp.HermOp(prom,tmp);
   blockProject(c_proj,tmp,subspace);
@@ -127,15 +127,15 @@ int main (int argc, char ** argv)
   LittleDiracOp.M(c_src,c_res);
 
   c_proj = c_proj - c_res;
-  std::cout<<"Representation of ldop within subspace "<<norm2(c_proj)<<std::endl;
+  std::cout<<GridLogMessage<<"Representation of ldop within subspace "<<norm2(c_proj)<<std::endl;
 
-  std::cout << "Multiplying by LittleDiracOp "<< std::endl;
+  std::cout<<GridLogMessage << "Multiplying by LittleDiracOp "<< std::endl;
   LittleDiracOp.M(c_src,c_res);
 
-  std::cout<<"Testing hermiticity explicitly by inspecting matrix elements"<<std::endl;
+  std::cout<<GridLogMessage<<"Testing hermiticity explicitly by inspecting matrix elements"<<std::endl;
   LittleDiracOp.AssertHermitian();
 
-  std::cout << "Testing Hermiticity stochastically "<< std::endl;
+  std::cout<<GridLogMessage << "Testing Hermiticity stochastically "<< std::endl;
   CoarseVector phi(Coarse5d);
   CoarseVector chi(Coarse5d);
   CoarseVector Aphi(Coarse5d);
@@ -145,7 +145,7 @@ int main (int argc, char ** argv)
   random(CRNG,chi);
 
 
-  std::cout<<"Made randoms"<<std::endl;
+  std::cout<<GridLogMessage<<"Made randoms"<<std::endl;
 
   LittleDiracOp.M(phi,Aphi);
   LittleDiracOp.Mdag(chi,Achi);
@@ -155,11 +155,11 @@ int main (int argc, char ** argv)
   ComplexD cAc = innerProduct(chi,Achi);
   ComplexD pAp = innerProduct(phi,Aphi);
 
-  std::cout<< "pAc "<<pAc<<" cAp "<< cAp<< " diff "<<pAc-adj(cAp)<<std::endl;
+  std::cout<<GridLogMessage<< "pAc "<<pAc<<" cAp "<< cAp<< " diff "<<pAc-adj(cAp)<<std::endl;
 
-  std::cout<< "pAp "<<pAp<<" cAc "<< cAc<<"Should be real"<< std::endl;
+  std::cout<<GridLogMessage<< "pAp "<<pAp<<" cAc "<< cAc<<"Should be real"<< std::endl;
 
-  std::cout<<"Testing linearity"<<std::endl;
+  std::cout<<GridLogMessage<<"Testing linearity"<<std::endl;
   CoarseVector PhiPlusChi(Coarse5d);
   CoarseVector APhiPlusChi(Coarse5d);
   CoarseVector linerr(Coarse5d);
@@ -168,9 +168,9 @@ int main (int argc, char ** argv)
 
   linerr= APhiPlusChi-Aphi;
   linerr= linerr-Achi;
-  std::cout<<"**Diff "<<norm2(linerr)<<std::endl;
+  std::cout<<GridLogMessage<<"**Diff "<<norm2(linerr)<<std::endl;
 
 
-  std::cout << "Done "<< std::endl;
+  std::cout<<GridLogMessage << "Done "<< std::endl;
   Grid_finalize();
 }
diff --git a/tests/Test_cayley_even_odd.cc b/tests/Test_cayley_even_odd.cc
index 21c828db..d0c7b74a 100644
--- a/tests/Test_cayley_even_odd.cc
+++ b/tests/Test_cayley_even_odd.cc
@@ -29,7 +29,7 @@ int main (int argc, char ** argv)
   Grid_init(&argc,&argv);
 
   int threads = GridThread::GetThreads();
-  std::cout << "Grid is setup to use "<<threads<<" threads"<<std::endl;
+  std::cout<<GridLogMessage << "Grid is setup to use "<<threads<<" threads"<<std::endl;
 
   const int Ls=8;
   GridCartesian         * UGrid   = SpaceTimeGrid::makeFourDimGrid(GridDefaultLatt(), GridDefaultSimd(Nd,vComplex::Nsimd()),GridDefaultMpi());
@@ -48,34 +48,34 @@ int main (int argc, char ** argv)
 
   RealD mass=0.1;
   RealD M5  =1.8;
-  std::cout <<"DomainWallFermion test"<<std::endl;
+  std::cout<<GridLogMessage <<"DomainWallFermion test"<<std::endl;
   DomainWallFermion Ddwf(Umu,*FGrid,*FrbGrid,*UGrid,*UrbGrid,mass,M5);
   TestWhat<DomainWallFermion>(Ddwf,FGrid,FrbGrid,UGrid,UrbGrid,mass,M5,&RNG4,&RNG5);
 
   RealD b=1.5;// Scale factor b+c=2, b-c=1
   RealD c=0.5;
-  std::cout <<"MobiusFermion test"<<std::endl;
+  std::cout<<GridLogMessage <<"MobiusFermion test"<<std::endl;
   MobiusFermion Dmob(Umu,*FGrid,*FrbGrid,*UGrid,*UrbGrid,mass,M5,b,c);
   TestWhat<MobiusFermion>(Dmob,FGrid,FrbGrid,UGrid,UrbGrid,mass,M5,&RNG4,&RNG5);
 
-  std::cout <<"MobiusZolotarevFermion test"<<std::endl;
+  std::cout<<GridLogMessage <<"MobiusZolotarevFermion test"<<std::endl;
   MobiusZolotarevFermion Dzolo(Umu,*FGrid,*FrbGrid,*UGrid,*UrbGrid,mass,M5,b,c,0.1,2.0);
   TestWhat<MobiusZolotarevFermion>(Dzolo,FGrid,FrbGrid,UGrid,UrbGrid,mass,M5,&RNG4,&RNG5);
 
-  std::cout <<"ScaledShamirFermion test"<<std::endl;
+  std::cout<<GridLogMessage <<"ScaledShamirFermion test"<<std::endl;
   ScaledShamirFermion Dsham(Umu,*FGrid,*FrbGrid,*UGrid,*UrbGrid,mass,M5,2.0);
   TestWhat<ScaledShamirFermion>(Dsham,FGrid,FrbGrid,UGrid,UrbGrid,mass,M5,&RNG4,&RNG5);
 
 
-  std::cout <<"ShamirZolotarevFermion test"<<std::endl;
+  std::cout<<GridLogMessage <<"ShamirZolotarevFermion test"<<std::endl;
   ShamirZolotarevFermion Dshamz(Umu,*FGrid,*FrbGrid,*UGrid,*UrbGrid,mass,M5,0.1,2.0);
   TestWhat<ShamirZolotarevFermion>(Dshamz,FGrid,FrbGrid,UGrid,UrbGrid,mass,M5,&RNG4,&RNG5);
 
-  std::cout <<"OverlapWilsonCayleyTanhFermion test"<<std::endl;
+  std::cout<<GridLogMessage <<"OverlapWilsonCayleyTanhFermion test"<<std::endl;
   OverlapWilsonCayleyTanhFermion Dov(Umu,*FGrid,*FrbGrid,*UGrid,*UrbGrid,mass,M5,1.0);
   TestWhat<OverlapWilsonCayleyTanhFermion>(Dov,FGrid,FrbGrid,UGrid,UrbGrid,mass,M5,&RNG4,&RNG5);
 
-  std::cout <<"OverlapWilsonCayleyZolotarevFermion test"<<std::endl;
+  std::cout<<GridLogMessage <<"OverlapWilsonCayleyZolotarevFermion test"<<std::endl;
   OverlapWilsonCayleyZolotarevFermion Dovz(Umu,*FGrid,*FrbGrid,*UGrid,*UrbGrid,mass,M5,0.1,2.0);
   TestWhat<OverlapWilsonCayleyZolotarevFermion>(Dovz,FGrid,FrbGrid,UGrid,UrbGrid,mass,M5,&RNG4,&RNG5);
 
@@ -106,40 +106,40 @@ void  TestWhat(What & Ddwf,
   LatticeFermion r_eo  (FGrid);
   LatticeFermion r_eeoo(FGrid);
 
-  std::cout<<"=========================================================="<<std::endl;
-  std::cout<<"= Testing that Meo + Moe + Moo + Mee = Munprec "<<std::endl;
-  std::cout<<"=========================================================="<<std::endl;
+  std::cout<<GridLogMessage<<"=========================================================="<<std::endl;
+  std::cout<<GridLogMessage<<"= Testing that Meo + Moe + Moo + Mee = Munprec "<<std::endl;
+  std::cout<<GridLogMessage<<"=========================================================="<<std::endl;
 
   pickCheckerboard(Even,src_e,src);
   pickCheckerboard(Odd,src_o,src);
 
-  Ddwf.Meooe(src_e,r_o);  std::cout<<"Applied Meo"<<std::endl;
-  Ddwf.Meooe(src_o,r_e);  std::cout<<"Applied Moe"<<std::endl;
+  Ddwf.Meooe(src_e,r_o);  std::cout<<GridLogMessage<<"Applied Meo"<<std::endl;
+  Ddwf.Meooe(src_o,r_e);  std::cout<<GridLogMessage<<"Applied Moe"<<std::endl;
   setCheckerboard(r_eo,r_o);
   setCheckerboard(r_eo,r_e);
 
-  Ddwf.Mooee(src_e,r_e);  std::cout<<"Applied Mee"<<std::endl;
-  Ddwf.Mooee(src_o,r_o);  std::cout<<"Applied Moo"<<std::endl;
+  Ddwf.Mooee(src_e,r_e);  std::cout<<GridLogMessage<<"Applied Mee"<<std::endl;
+  Ddwf.Mooee(src_o,r_o);  std::cout<<GridLogMessage<<"Applied Moo"<<std::endl;
   setCheckerboard(r_eeoo,r_e);
   setCheckerboard(r_eeoo,r_o);
 
   r_eo=r_eo+r_eeoo;
   Ddwf.M(src,ref);  
 
-  //  std::cout << r_eo<<std::endl;
-  //  std::cout << ref <<std::endl;
+  //  std::cout<<GridLogMessage << r_eo<<std::endl;
+  //  std::cout<<GridLogMessage << ref <<std::endl;
 
   err= ref - r_eo;
-  std::cout << "EO norm diff   "<< norm2(err)<< " "<<norm2(ref)<< " " << norm2(r_eo) <<std::endl;
+  std::cout<<GridLogMessage << "EO norm diff   "<< norm2(err)<< " "<<norm2(ref)<< " " << norm2(r_eo) <<std::endl;
     
   LatticeComplex cerr(FGrid);
   cerr = localInnerProduct(err,err);
-  //  std::cout << cerr<<std::endl;
+  //  std::cout<<GridLogMessage << cerr<<std::endl;
 
-  std::cout<<"=============================================================="<<std::endl;
-  std::cout<<"= Test Ddagger is the dagger of D by requiring                "<<std::endl;
-  std::cout<<"=  < phi | Deo | chi > * = < chi | Deo^dag| phi>  "<<std::endl;
-  std::cout<<"=============================================================="<<std::endl;
+  std::cout<<GridLogMessage<<"=============================================================="<<std::endl;
+  std::cout<<GridLogMessage<<"= Test Ddagger is the dagger of D by requiring                "<<std::endl;
+  std::cout<<GridLogMessage<<"=  < phi | Deo | chi > * = < chi | Deo^dag| phi>  "<<std::endl;
+  std::cout<<GridLogMessage<<"=============================================================="<<std::endl;
   
   LatticeFermion chi_e   (FrbGrid);
   LatticeFermion chi_o   (FrbGrid);
@@ -169,15 +169,15 @@ void  TestWhat(What & Ddwf,
   ComplexD cDpe = innerProduct(chi_e,dphi_e);
   ComplexD cDpo = innerProduct(chi_o,dphi_o);
 
-  std::cout <<"e "<<pDce<<" "<<cDpe <<std::endl;
-  std::cout <<"o "<<pDco<<" "<<cDpo <<std::endl;
+  std::cout<<GridLogMessage <<"e "<<pDce<<" "<<cDpe <<std::endl;
+  std::cout<<GridLogMessage <<"o "<<pDco<<" "<<cDpo <<std::endl;
 
-  std::cout <<"pDce - conj(cDpo) "<< pDce-conj(cDpo) <<std::endl;
-  std::cout <<"pDco - conj(cDpe) "<< pDco-conj(cDpe) <<std::endl;
+  std::cout<<GridLogMessage <<"pDce - conj(cDpo) "<< pDce-conj(cDpo) <<std::endl;
+  std::cout<<GridLogMessage <<"pDco - conj(cDpe) "<< pDco-conj(cDpe) <<std::endl;
 
-  std::cout<<"=============================================================="<<std::endl;
-  std::cout<<"= Test MeeInv Mee = 1                                         "<<std::endl;
-  std::cout<<"=============================================================="<<std::endl;
+  std::cout<<GridLogMessage<<"=============================================================="<<std::endl;
+  std::cout<<GridLogMessage<<"= Test MeeInv Mee = 1                                         "<<std::endl;
+  std::cout<<GridLogMessage<<"=============================================================="<<std::endl;
 
   pickCheckerboard(Even,chi_e,chi);
   pickCheckerboard(Odd ,chi_o,chi);
@@ -192,11 +192,11 @@ void  TestWhat(What & Ddwf,
   setCheckerboard(phi,phi_o);
 
   err = phi-chi;
-  std::cout << "norm diff   "<< norm2(err)<< std::endl;
+  std::cout<<GridLogMessage << "norm diff   "<< norm2(err)<< std::endl;
 
-  std::cout<<"=============================================================="<<std::endl;
-  std::cout<<"= Test MeeInvDag MeeDag = 1                                   "<<std::endl;
-  std::cout<<"=============================================================="<<std::endl;
+  std::cout<<GridLogMessage<<"=============================================================="<<std::endl;
+  std::cout<<GridLogMessage<<"= Test MeeInvDag MeeDag = 1                                   "<<std::endl;
+  std::cout<<GridLogMessage<<"=============================================================="<<std::endl;
 
   pickCheckerboard(Even,chi_e,chi);
   pickCheckerboard(Odd ,chi_o,chi);
@@ -211,11 +211,11 @@ void  TestWhat(What & Ddwf,
   setCheckerboard(phi,phi_o);
 
   err = phi-chi;
-  std::cout << "norm diff   "<< norm2(err)<< std::endl;
+  std::cout<<GridLogMessage << "norm diff   "<< norm2(err)<< std::endl;
 
-  std::cout<<"=============================================================="<<std::endl;
-  std::cout<<"= Test DiagMoo MpcDagMpc is Hermitian              "<<std::endl;
-  std::cout<<"=============================================================="<<std::endl;
+  std::cout<<GridLogMessage<<"=============================================================="<<std::endl;
+  std::cout<<GridLogMessage<<"= Test DiagMoo MpcDagMpc is Hermitian              "<<std::endl;
+  std::cout<<GridLogMessage<<"=============================================================="<<std::endl;
   
   random(*RNG5,phi);
   random(*RNG5,chi);
@@ -237,10 +237,10 @@ void  TestWhat(What & Ddwf,
   cDpe = innerProduct(chi_e,dphi_e);
   cDpo = innerProduct(chi_o,dphi_o);
 
-  std::cout <<"e "<<pDce<<" "<<cDpe <<std::endl;
-  std::cout <<"o "<<pDco<<" "<<cDpo <<std::endl;
+  std::cout<<GridLogMessage <<"e "<<pDce<<" "<<cDpe <<std::endl;
+  std::cout<<GridLogMessage <<"o "<<pDco<<" "<<cDpo <<std::endl;
 
-  std::cout <<"pDce - conj(cDpo) "<< pDco-conj(cDpo) <<std::endl;
-  std::cout <<"pDco - conj(cDpe) "<< pDce-conj(cDpe) <<std::endl;
+  std::cout<<GridLogMessage <<"pDce - conj(cDpo) "<< pDco-conj(cDpo) <<std::endl;
+  std::cout<<GridLogMessage <<"pDco - conj(cDpe) "<< pDce-conj(cDpe) <<std::endl;
   
 }
diff --git a/tests/Test_cayley_ldop_cr.cc b/tests/Test_cayley_ldop_cr.cc
index 2e67df26..48d7c9dc 100644
--- a/tests/Test_cayley_ldop_cr.cc
+++ b/tests/Test_cayley_ldop_cr.cc
@@ -52,9 +52,9 @@ int main (int argc, char ** argv)
   RealD mass=0.1;
   RealD M5=1.5;
 
-  std::cout << "**************************************************"<< std::endl;
-  std::cout << "Building g5R5 hermitian DWF operator" <<std::endl;
-  std::cout << "**************************************************"<< std::endl;
+  std::cout<<GridLogMessage << "**************************************************"<< std::endl;
+  std::cout<<GridLogMessage << "Building g5R5 hermitian DWF operator" <<std::endl;
+  std::cout<<GridLogMessage << "**************************************************"<< std::endl;
   DomainWallFermion Ddwf(Umu,*FGrid,*FrbGrid,*UGrid,*UrbGrid,mass,M5);
   Gamma5R5HermitianLinearOperator<DomainWallFermion,LatticeFermion> HermIndefOp(Ddwf);
 
@@ -64,9 +64,9 @@ int main (int argc, char ** argv)
   typedef CoarsenedMatrix<vSpinColourVector,vTComplex,nbasis> LittleDiracOperator;
   typedef LittleDiracOperator::CoarseVector                   CoarseVector;
 
-  std::cout << "**************************************************"<< std::endl;
-  std::cout << "Calling Aggregation class to build subspace" <<std::endl;
-  std::cout << "**************************************************"<< std::endl;
+  std::cout<<GridLogMessage << "**************************************************"<< std::endl;
+  std::cout<<GridLogMessage << "Calling Aggregation class to build subspace" <<std::endl;
+  std::cout<<GridLogMessage << "**************************************************"<< std::endl;
   MdagMLinearOperator<DomainWallFermion,LatticeFermion> HermDefOp(Ddwf);
   Subspace Aggregates(Coarse5d,FGrid);
   Aggregates.CreateSubspace(RNG5,HermDefOp);
@@ -80,22 +80,22 @@ int main (int argc, char ** argv)
   gaussian(CRNG,c_src);
   c_res=zero;
 
-  std::cout << "**************************************************"<< std::endl;
-  std::cout << "Solving mdagm-CG on coarse space "<< std::endl;
-  std::cout << "**************************************************"<< std::endl;
+  std::cout<<GridLogMessage << "**************************************************"<< std::endl;
+  std::cout<<GridLogMessage << "Solving mdagm-CG on coarse space "<< std::endl;
+  std::cout<<GridLogMessage << "**************************************************"<< std::endl;
   MdagMLinearOperator<LittleDiracOperator,CoarseVector> PosdefLdop(LittleDiracOp);
   ConjugateGradient<CoarseVector> CG(1.0e-6,10000);
   CG(PosdefLdop,c_src,c_res);
 
-  std::cout << "**************************************************"<< std::endl;
-  std::cout << "Solving indef-MCR on coarse space "<< std::endl;
-  std::cout << "**************************************************"<< std::endl;
+  std::cout<<GridLogMessage << "**************************************************"<< std::endl;
+  std::cout<<GridLogMessage << "Solving indef-MCR on coarse space "<< std::endl;
+  std::cout<<GridLogMessage << "**************************************************"<< std::endl;
   HermitianLinearOperator<LittleDiracOperator,CoarseVector> HermIndefLdop(LittleDiracOp);
   ConjugateResidual<CoarseVector> MCR(1.0e-6,10000);
   MCR(HermIndefLdop,c_src,c_res);
 
-  std::cout << "**************************************************"<< std::endl;
-  std::cout << "Done "<< std::endl;
-  std::cout << "**************************************************"<< std::endl;
+  std::cout<<GridLogMessage << "**************************************************"<< std::endl;
+  std::cout<<GridLogMessage << "Done "<< std::endl;
+  std::cout<<GridLogMessage << "**************************************************"<< std::endl;
   Grid_finalize();
 }
diff --git a/tests/Test_cf_coarsen_support.cc b/tests/Test_cf_coarsen_support.cc
index 959f5a9b..caa862be 100644
--- a/tests/Test_cf_coarsen_support.cc
+++ b/tests/Test_cf_coarsen_support.cc
@@ -56,12 +56,12 @@ int main (int argc, char ** argv)
     
     for(int d=0;d<4;d++){
       HermIndefOp.OpDir(src,tmp,d,+1); result=result+tmp; 
-      std::cout<<"dir "<<d<<" tmp "<<norm2(tmp)<<std::endl;
+      std::cout<<GridLogMessage<<"dir "<<d<<" tmp "<<norm2(tmp)<<std::endl;
       HermIndefOp.OpDir(src,tmp,d,-1); result=result+tmp;
-      std::cout<<"dir "<<d<<" tmp "<<norm2(tmp)<<std::endl;
+      std::cout<<GridLogMessage<<"dir "<<d<<" tmp "<<norm2(tmp)<<std::endl;
     }
     err = result-ref;
-    std::cout<<"Error "<<norm2(err)<<std::endl;
+    std::cout<<GridLogMessage<<"Error "<<norm2(err)<<std::endl;
   }
 
   {
@@ -73,13 +73,13 @@ int main (int argc, char ** argv)
     
     for(int d=0;d<4;d++){
       HermIndefOp.OpDir(src,tmp,d,+1); result=result+tmp; 
-      std::cout<<"dir "<<d<<" tmp "<<norm2(tmp)<<std::endl;
+      std::cout<<GridLogMessage<<"dir "<<d<<" tmp "<<norm2(tmp)<<std::endl;
       HermIndefOp.OpDir(src,tmp,d,-1); result=result+tmp;
-      std::cout<<"dir "<<d<<" tmp "<<norm2(tmp)<<std::endl;
+      std::cout<<GridLogMessage<<"dir "<<d<<" tmp "<<norm2(tmp)<<std::endl;
     }
 
     err = result-ref;
-    std::cout<<"Error "<<norm2(err)<<std::endl;
+    std::cout<<GridLogMessage<<"Error "<<norm2(err)<<std::endl;
   }
 
 
diff --git a/tests/Test_contfrac_cg.cc b/tests/Test_contfrac_cg.cc
index 1b937cef..da6a81b9 100644
--- a/tests/Test_contfrac_cg.cc
+++ b/tests/Test_contfrac_cg.cc
@@ -53,7 +53,7 @@ int main (int argc, char ** argv)
   Grid_init(&argc,&argv);
 
   int threads = GridThread::GetThreads();
-  std::cout << "Grid is setup to use "<<threads<<" threads"<<std::endl;
+  std::cout<<GridLogMessage << "Grid is setup to use "<<threads<<" threads"<<std::endl;
 
   const int Ls=9;
   GridCartesian         * UGrid   = SpaceTimeGrid::makeFourDimGrid(GridDefaultLatt(), GridDefaultSimd(Nd,vComplex::Nsimd()),GridDefaultMpi());
@@ -74,20 +74,20 @@ int main (int argc, char ** argv)
   RealD M5  =1.8;
 
 
-  std::cout <<"OverlapWilsonContFracTanhFermion  test"<<std::endl;
+  std::cout<<GridLogMessage <<"OverlapWilsonContFracTanhFermion  test"<<std::endl;
   OverlapWilsonContFracTanhFermion Dcf(Umu,*FGrid,*FrbGrid,*UGrid,*UrbGrid,mass,M5,1.0);
   TestCGinversions<OverlapWilsonContFracTanhFermion>(Dcf,FGrid,FrbGrid,UGrid,UrbGrid,mass,M5,&RNG4,&RNG5);
 
-  std::cout <<"OverlapWilsonContFracZolotarevFermion  test"<<std::endl;
+  std::cout<<GridLogMessage <<"OverlapWilsonContFracZolotarevFermion  test"<<std::endl;
   OverlapWilsonContFracZolotarevFermion Dcfz(Umu,*FGrid,*FrbGrid,*UGrid,*UrbGrid,mass,M5,0.1,6.0);
   TestCGinversions<OverlapWilsonContFracZolotarevFermion>(Dcfz,FGrid,FrbGrid,UGrid,UrbGrid,mass,M5,&RNG4,&RNG5);
 
 
-  std::cout <<"OverlapWilsonPartialFractionTanhFermion  test"<<std::endl;
+  std::cout<<GridLogMessage <<"OverlapWilsonPartialFractionTanhFermion  test"<<std::endl;
   OverlapWilsonPartialFractionTanhFermion Dpf(Umu,*FGrid,*FrbGrid,*UGrid,*UrbGrid,mass,M5,1.0);
   TestCGinversions<OverlapWilsonPartialFractionTanhFermion>(Dpf,FGrid,FrbGrid,UGrid,UrbGrid,mass,M5,&RNG4,&RNG5);
 
-  std::cout <<"OverlapWilsonPartialFractionZolotarevFermion  test"<<std::endl;
+  std::cout<<GridLogMessage <<"OverlapWilsonPartialFractionZolotarevFermion  test"<<std::endl;
   OverlapWilsonPartialFractionZolotarevFermion Dpfz(Umu,*FGrid,*FrbGrid,*UGrid,*UrbGrid,mass,M5,0.1,6.0);
   TestCGinversions<OverlapWilsonPartialFractionZolotarevFermion>(Dpfz,FGrid,FrbGrid,UGrid,UrbGrid,mass,M5,&RNG4,&RNG5);
 
@@ -102,11 +102,11 @@ void  TestCGinversions(What & Ddwf,
 		       GridParallelRNG *RNG4,
 		       GridParallelRNG *RNG5)
 {
-  std::cout << "Testing unpreconditioned inverter"<<std::endl;
+  std::cout<<GridLogMessage << "Testing unpreconditioned inverter"<<std::endl;
   TestCGunprec<What>(Ddwf,FGrid,FrbGrid,UGrid,UrbGrid,mass,M5,RNG4,RNG5);
-  std::cout << "Testing red black preconditioned inverter"<<std::endl;
+  std::cout<<GridLogMessage << "Testing red black preconditioned inverter"<<std::endl;
   TestCGprec<What>(Ddwf,FGrid,FrbGrid,UGrid,UrbGrid,mass,M5,RNG4,RNG5);
-  std::cout << "Testing red black Schur inverter"<<std::endl;
+  std::cout<<GridLogMessage << "Testing red black Schur inverter"<<std::endl;
   TestCGschur<What>(Ddwf,FGrid,FrbGrid,UGrid,UrbGrid,mass,M5,RNG4,RNG5);
 }
 
diff --git a/tests/Test_contfrac_even_odd.cc b/tests/Test_contfrac_even_odd.cc
index 5d39b7df..c8e5b093 100644
--- a/tests/Test_contfrac_even_odd.cc
+++ b/tests/Test_contfrac_even_odd.cc
@@ -29,7 +29,7 @@ int main (int argc, char ** argv)
   Grid_init(&argc,&argv);
 
   int threads = GridThread::GetThreads();
-  std::cout << "Grid is setup to use "<<threads<<" threads"<<std::endl;
+  std::cout<<GridLogMessage << "Grid is setup to use "<<threads<<" threads"<<std::endl;
 
   const int Ls=9;
   GridCartesian         * UGrid   = SpaceTimeGrid::makeFourDimGrid(GridDefaultLatt(), GridDefaultSimd(Nd,vComplex::Nsimd()),GridDefaultMpi());
@@ -49,19 +49,19 @@ int main (int argc, char ** argv)
   RealD mass=0.1;
   RealD M5  =1.8;
 
-  std::cout <<"OverlapWilsonContFracTanhFermion  test"<<std::endl;
+  std::cout<<GridLogMessage <<"OverlapWilsonContFracTanhFermion  test"<<std::endl;
   OverlapWilsonContFracTanhFermion Dcf(Umu,*FGrid,*FrbGrid,*UGrid,*UrbGrid,mass,M5,1.0);
   TestWhat<OverlapWilsonContFracTanhFermion>(Dcf,FGrid,FrbGrid,UGrid,UrbGrid,mass,M5,&RNG4,&RNG5);
 
-  std::cout <<"OverlapWilsonContFracZolotarevFermion  test"<<std::endl;
+  std::cout<<GridLogMessage <<"OverlapWilsonContFracZolotarevFermion  test"<<std::endl;
   OverlapWilsonContFracZolotarevFermion Dcfz(Umu,*FGrid,*FrbGrid,*UGrid,*UrbGrid,mass,M5,0.1,6.0);
   TestWhat<OverlapWilsonContFracZolotarevFermion>(Dcfz,FGrid,FrbGrid,UGrid,UrbGrid,mass,M5,&RNG4,&RNG5);
 
-  std::cout <<"OverlapWilsonPartialFractionTanhFermion  test"<<std::endl;
+  std::cout<<GridLogMessage <<"OverlapWilsonPartialFractionTanhFermion  test"<<std::endl;
   OverlapWilsonPartialFractionTanhFermion Dpf(Umu,*FGrid,*FrbGrid,*UGrid,*UrbGrid,mass,M5,1.0);
   TestWhat<OverlapWilsonPartialFractionTanhFermion>(Dpf,FGrid,FrbGrid,UGrid,UrbGrid,mass,M5,&RNG4,&RNG5);
 
-  std::cout <<"OverlapWilsonPartialFractionZolotarevFermion  test"<<std::endl;
+  std::cout<<GridLogMessage <<"OverlapWilsonPartialFractionZolotarevFermion  test"<<std::endl;
   OverlapWilsonPartialFractionZolotarevFermion Dpfz(Umu,*FGrid,*FrbGrid,*UGrid,*UrbGrid,mass,M5,0.1,6.0);
   TestWhat<OverlapWilsonPartialFractionZolotarevFermion>(Dpfz,FGrid,FrbGrid,UGrid,UrbGrid,mass,M5,&RNG4,&RNG5);
 
@@ -92,40 +92,40 @@ void  TestWhat(What & Ddwf,
   LatticeFermion r_eo  (FGrid);
   LatticeFermion r_eeoo(FGrid);
 
-  std::cout<<"=========================================================="<<std::endl;
-  std::cout<<"= Testing that Meo + Moe + Moo + Mee = Munprec "<<std::endl;
-  std::cout<<"=========================================================="<<std::endl;
+  std::cout<<GridLogMessage<<"=========================================================="<<std::endl;
+  std::cout<<GridLogMessage<<"= Testing that Meo + Moe + Moo + Mee = Munprec "<<std::endl;
+  std::cout<<GridLogMessage<<"=========================================================="<<std::endl;
 
   pickCheckerboard(Even,src_e,src);
   pickCheckerboard( Odd,src_o,src);
 
-  Ddwf.Meooe(src_e,r_o);  std::cout<<"Applied Meo "<<norm2(r_o)<<std::endl;
-  Ddwf.Meooe(src_o,r_e);  std::cout<<"Applied Moe "<<norm2(r_e)<<std::endl;
+  Ddwf.Meooe(src_e,r_o);  std::cout<<GridLogMessage<<"Applied Meo "<<norm2(r_o)<<std::endl;
+  Ddwf.Meooe(src_o,r_e);  std::cout<<GridLogMessage<<"Applied Moe "<<norm2(r_e)<<std::endl;
   setCheckerboard(r_eo,r_o);
   setCheckerboard(r_eo,r_e);
 
-  Ddwf.Mooee(src_e,r_e);  std::cout<<"Applied Mee"<<norm2(r_e)<<std::endl;
-  Ddwf.Mooee(src_o,r_o);  std::cout<<"Applied Moo"<<norm2(r_o)<<std::endl;
+  Ddwf.Mooee(src_e,r_e);  std::cout<<GridLogMessage<<"Applied Mee"<<norm2(r_e)<<std::endl;
+  Ddwf.Mooee(src_o,r_o);  std::cout<<GridLogMessage<<"Applied Moo"<<norm2(r_o)<<std::endl;
   setCheckerboard(r_eeoo,r_e);
   setCheckerboard(r_eeoo,r_o);
 
   r_eo=r_eo+r_eeoo;
   Ddwf.M(src,ref);  
 
-  //  std::cout << r_eo<<std::endl;
-  //  std::cout << ref <<std::endl;
+  //  std::cout<<GridLogMessage << r_eo<<std::endl;
+  //  std::cout<<GridLogMessage << ref <<std::endl;
 
   err= ref - r_eo;
-  std::cout << "EO norm diff   "<< norm2(err)<< " "<<norm2(ref)<< " " << norm2(r_eo) <<std::endl;
+  std::cout<<GridLogMessage << "EO norm diff   "<< norm2(err)<< " "<<norm2(ref)<< " " << norm2(r_eo) <<std::endl;
     
   LatticeComplex cerr(FGrid);
   cerr = localInnerProduct(err,err);
-  //  std::cout << cerr<<std::endl;
+  //  std::cout<<GridLogMessage << cerr<<std::endl;
 
-  std::cout<<"=============================================================="<<std::endl;
-  std::cout<<"= Test Ddagger is the dagger of D by requiring                "<<std::endl;
-  std::cout<<"=  < phi | Deo | chi > * = < chi | Deo^dag| phi>  "<<std::endl;
-  std::cout<<"=============================================================="<<std::endl;
+  std::cout<<GridLogMessage<<"=============================================================="<<std::endl;
+  std::cout<<GridLogMessage<<"= Test Ddagger is the dagger of D by requiring                "<<std::endl;
+  std::cout<<GridLogMessage<<"=  < phi | Deo | chi > * = < chi | Deo^dag| phi>  "<<std::endl;
+  std::cout<<GridLogMessage<<"=============================================================="<<std::endl;
   
   LatticeFermion chi_e   (FrbGrid);
   LatticeFermion chi_o   (FrbGrid);
@@ -155,15 +155,15 @@ void  TestWhat(What & Ddwf,
   ComplexD cDpe = innerProduct(chi_e,dphi_e);
   ComplexD cDpo = innerProduct(chi_o,dphi_o);
 
-  std::cout <<"e "<<pDce<<" "<<cDpe <<std::endl;
-  std::cout <<"o "<<pDco<<" "<<cDpo <<std::endl;
+  std::cout<<GridLogMessage <<"e "<<pDce<<" "<<cDpe <<std::endl;
+  std::cout<<GridLogMessage <<"o "<<pDco<<" "<<cDpo <<std::endl;
 
-  std::cout <<"pDce - conj(cDpo) "<< pDce-conj(cDpo) <<std::endl;
-  std::cout <<"pDco - conj(cDpe) "<< pDco-conj(cDpe) <<std::endl;
+  std::cout<<GridLogMessage <<"pDce - conj(cDpo) "<< pDce-conj(cDpo) <<std::endl;
+  std::cout<<GridLogMessage <<"pDco - conj(cDpe) "<< pDco-conj(cDpe) <<std::endl;
 
-  std::cout<<"=============================================================="<<std::endl;
-  std::cout<<"= Test MeeInv Mee = 1                                         "<<std::endl;
-  std::cout<<"=============================================================="<<std::endl;
+  std::cout<<GridLogMessage<<"=============================================================="<<std::endl;
+  std::cout<<GridLogMessage<<"= Test MeeInv Mee = 1                                         "<<std::endl;
+  std::cout<<GridLogMessage<<"=============================================================="<<std::endl;
 
   pickCheckerboard(Even,chi_e,chi);
   pickCheckerboard(Odd ,chi_o,chi);
@@ -178,11 +178,11 @@ void  TestWhat(What & Ddwf,
   setCheckerboard(phi,phi_o);
 
   err = phi-chi;
-  std::cout << "norm diff   "<< norm2(err)<< std::endl;
+  std::cout<<GridLogMessage << "norm diff   "<< norm2(err)<< std::endl;
 
-  std::cout<<"=============================================================="<<std::endl;
-  std::cout<<"= Test MeeInvDag MeeDag = 1                                   "<<std::endl;
-  std::cout<<"=============================================================="<<std::endl;
+  std::cout<<GridLogMessage<<"=============================================================="<<std::endl;
+  std::cout<<GridLogMessage<<"= Test MeeInvDag MeeDag = 1                                   "<<std::endl;
+  std::cout<<GridLogMessage<<"=============================================================="<<std::endl;
 
   pickCheckerboard(Even,chi_e,chi);
   pickCheckerboard(Odd ,chi_o,chi);
@@ -197,11 +197,11 @@ void  TestWhat(What & Ddwf,
   setCheckerboard(phi,phi_o);
 
   err = phi-chi;
-  std::cout << "norm diff   "<< norm2(err)<< std::endl;
+  std::cout<<GridLogMessage << "norm diff   "<< norm2(err)<< std::endl;
 
-  std::cout<<"=============================================================="<<std::endl;
-  std::cout<<"= Test MpcDagMpc is Hermitian              "<<std::endl;
-  std::cout<<"=============================================================="<<std::endl;
+  std::cout<<GridLogMessage<<"=============================================================="<<std::endl;
+  std::cout<<GridLogMessage<<"= Test MpcDagMpc is Hermitian              "<<std::endl;
+  std::cout<<GridLogMessage<<"=============================================================="<<std::endl;
   
   random(*RNG5,phi);
   random(*RNG5,chi);
@@ -223,10 +223,10 @@ void  TestWhat(What & Ddwf,
   cDpe = innerProduct(chi_e,dphi_e);
   cDpo = innerProduct(chi_o,dphi_o);
 
-  std::cout <<"e "<<pDce<<" "<<cDpe <<std::endl;
-  std::cout <<"o "<<pDco<<" "<<cDpo <<std::endl;
+  std::cout<<GridLogMessage <<"e "<<pDce<<" "<<cDpe <<std::endl;
+  std::cout<<GridLogMessage <<"o "<<pDco<<" "<<cDpo <<std::endl;
 
-  std::cout <<"pDce - conj(cDpo) "<< pDco-conj(cDpo) <<std::endl;
-  std::cout <<"pDco - conj(cDpe) "<< pDce-conj(cDpe) <<std::endl;
+  std::cout<<GridLogMessage <<"pDce - conj(cDpo) "<< pDco-conj(cDpo) <<std::endl;
+  std::cout<<GridLogMessage <<"pDco - conj(cDpe) "<< pDce-conj(cDpe) <<std::endl;
   
 }
diff --git a/tests/Test_cshift.cc b/tests/Test_cshift.cc
index cd85439c..febf1837 100644
--- a/tests/Test_cshift.cc
+++ b/tests/Test_cshift.cc
@@ -39,7 +39,7 @@ int main (int argc, char ** argv)
   for(int dir=0;dir<4;dir++){
     for(int shift=0;shift<latt_size[dir];shift++){
       if ( Fine.IsBoss() ) 
-	std::cout<<"Shifting by "<<shift<<" in direction"<<dir<<std::endl;
+	std::cout<<GridLogMessage<<"Shifting by "<<shift<<" in direction"<<dir<<std::endl;
 
 	ShiftU  = Cshift(U,dir,shift);    // Shift everything
 
diff --git a/tests/Test_cshift_red_black.cc b/tests/Test_cshift_red_black.cc
index 79ebcfc5..65af9857 100644
--- a/tests/Test_cshift_red_black.cc
+++ b/tests/Test_cshift_red_black.cc
@@ -47,9 +47,9 @@ int main (int argc, char ** argv)
   pickCheckerboard(Even,Ue,U);
   pickCheckerboard(Odd,Uo,U);
 
-  //  std::cout << U<<std::endl;
-  std::cout << "Ue " <<norm2(Ue)<<std::endl;
-  std::cout << "Uo " <<norm2(Uo)<<std::endl;
+  //  std::cout<<GridLogMessage << U<<std::endl;
+  std::cout<<GridLogMessage << "Ue " <<norm2(Ue)<<std::endl;
+  std::cout<<GridLogMessage << "Uo " <<norm2(Uo)<<std::endl;
 
 
   TComplex cm;
@@ -57,28 +57,28 @@ int main (int argc, char ** argv)
     if ( dir!=1 ) continue;
     for(int shift=0;shift<latt_size[dir];shift++){
 
-	std::cout<<"Shifting by "<<shift<<" in direction"<<dir<<std::endl;
+	std::cout<<GridLogMessage<<"Shifting by "<<shift<<" in direction"<<dir<<std::endl;
 
-	//	std::cout<<"Even grid"<<std::endl;
+	//	std::cout<<GridLogMessage<<"Even grid"<<std::endl;
 	ShiftUe = Cshift(Ue,dir,shift);    // Shift everything cb by cb
-	//	std::cout << "\tShiftUe " <<norm2(ShiftUe)<<std::endl;
+	//	std::cout<<GridLogMessage << "\tShiftUe " <<norm2(ShiftUe)<<std::endl;
 
-	//	std::cout<<"Odd grid"<<std::endl;
+	//	std::cout<<GridLogMessage<<"Odd grid"<<std::endl;
 	ShiftUo = Cshift(Uo,dir,shift);    
-	//	std::cout << "\tShiftUo " <<norm2(ShiftUo)<<std::endl;
+	//	std::cout<<GridLogMessage << "\tShiftUo " <<norm2(ShiftUo)<<std::endl;
 
-	//	std::cout<<"Recombined Even/Odd grids"<<std::endl;
+	//	std::cout<<GridLogMessage<<"Recombined Even/Odd grids"<<std::endl;
 	setCheckerboard(rbShiftU,ShiftUe);
 	setCheckerboard(rbShiftU,ShiftUo);
-	//	std::cout << "\trbShiftU " <<norm2(rbShiftU)<<std::endl;
+	//	std::cout<<GridLogMessage << "\trbShiftU " <<norm2(rbShiftU)<<std::endl;
 
-	//	std::cout<<"Full grid shift"<<std::endl;
+	//	std::cout<<GridLogMessage<<"Full grid shift"<<std::endl;
 	ShiftU  = Cshift(U,dir,shift);    // Shift everything
-	//	std::cout << "\tShiftU " <<norm2(rbShiftU)<<std::endl;
+	//	std::cout<<GridLogMessage << "\tShiftU " <<norm2(rbShiftU)<<std::endl;
 
 	std::vector<int> coor(4);
 
-	std::cout << "Checking the non-checkerboard shift"<<std::endl;
+	std::cout<<GridLogMessage << "Checking the non-checkerboard shift"<<std::endl;
 	for(coor[3]=0;coor[3]<latt_size[3];coor[3]++){
 	for(coor[2]=0;coor[2]<latt_size[2];coor[2]++){
 	for(coor[1]=0;coor[1]<latt_size[1];coor[1]++){
@@ -117,7 +117,7 @@ int main (int argc, char ** argv)
 	}}}}
 
 
-	std::cout << "Checking the checkerboard shift"<<std::endl;
+	std::cout<<GridLogMessage << "Checking the checkerboard shift"<<std::endl;
 	for(coor[3]=0;coor[3]<latt_size[3];coor[3]++){
 	for(coor[2]=0;coor[2]<latt_size[2];coor[2]++){
 	for(coor[1]=0;coor[1]<latt_size[1];coor[1]++){
@@ -153,7 +153,7 @@ int main (int argc, char ** argv)
 	    std::cerr<<"Expect "<<index<<" " << peer[0]<<","<<peer[1]<<","<<peer[2]<<","<<peer[3]<<std::endl;
 	    exit(-1);
 	  } else if (0) { 
-	    std::cout<<"PASS shift "<< shift<<" in dir "<< dir
+	    std::cout<<GridLogMessage<<"PASS shift "<< shift<<" in dir "<< dir
 		     <<" ["<<coor[0]<<","<<coor[1]<<","<<coor[2]<<","<<coor[3]<<"] = "
 		     << cm()()()<<" expect "<<scm<<"  "<<nrm<<std::endl;
 	  }
diff --git a/tests/Test_dwf_cg_prec.cc b/tests/Test_dwf_cg_prec.cc
index d429ee92..633d280f 100644
--- a/tests/Test_dwf_cg_prec.cc
+++ b/tests/Test_dwf_cg_prec.cc
@@ -34,7 +34,9 @@ int main (int argc, char ** argv)
 
   LatticeFermion    src(FGrid); random(RNG5,src);
   LatticeFermion result(FGrid); result=zero;
-  LatticeGaugeField Umu(UGrid); random(RNG4,Umu);
+  LatticeGaugeField Umu(UGrid); 
+
+  SU3::HotConfiguration(RNG4,Umu);
 
   std::vector<LatticeColourMatrix> U(4,UGrid);
   for(int mu=0;mu<Nd;mu++){
diff --git a/tests/Test_dwf_even_odd.cc b/tests/Test_dwf_even_odd.cc
index 3ad01076..4e3d9799 100644
--- a/tests/Test_dwf_even_odd.cc
+++ b/tests/Test_dwf_even_odd.cc
@@ -22,7 +22,7 @@ int main (int argc, char ** argv)
   Grid_init(&argc,&argv);
 
   int threads = GridThread::GetThreads();
-  std::cout << "Grid is setup to use "<<threads<<" threads"<<std::endl;
+  std::cout<<GridLogMessage << "Grid is setup to use "<<threads<<" threads"<<std::endl;
 
 
   const int Ls=8;
@@ -67,40 +67,40 @@ int main (int argc, char ** argv)
   LatticeFermion r_eo  (FGrid);
   LatticeFermion r_eeoo(FGrid);
 
-  std::cout<<"=========================================================="<<std::endl;
-  std::cout<<"= Testing that Meo + Moe + Moo + Mee = Munprec "<<std::endl;
-  std::cout<<"=========================================================="<<std::endl;
+  std::cout<<GridLogMessage<<"=========================================================="<<std::endl;
+  std::cout<<GridLogMessage<<"= Testing that Meo + Moe + Moo + Mee = Munprec "<<std::endl;
+  std::cout<<GridLogMessage<<"=========================================================="<<std::endl;
 
   pickCheckerboard(Even,src_e,src);
   pickCheckerboard(Odd,src_o,src);
 
-  Ddwf.Meooe(src_e,r_o);  std::cout<<"Applied Meo"<<std::endl;
-  Ddwf.Meooe(src_o,r_e);  std::cout<<"Applied Moe"<<std::endl;
+  Ddwf.Meooe(src_e,r_o);  std::cout<<GridLogMessage<<"Applied Meo"<<std::endl;
+  Ddwf.Meooe(src_o,r_e);  std::cout<<GridLogMessage<<"Applied Moe"<<std::endl;
   setCheckerboard(r_eo,r_o);
   setCheckerboard(r_eo,r_e);
 
-  Ddwf.Mooee(src_e,r_e);  std::cout<<"Applied Mee"<<std::endl;
-  Ddwf.Mooee(src_o,r_o);  std::cout<<"Applied Moo"<<std::endl;
+  Ddwf.Mooee(src_e,r_e);  std::cout<<GridLogMessage<<"Applied Mee"<<std::endl;
+  Ddwf.Mooee(src_o,r_o);  std::cout<<GridLogMessage<<"Applied Moo"<<std::endl;
   setCheckerboard(r_eeoo,r_e);
   setCheckerboard(r_eeoo,r_o);
 
   r_eo=r_eo+r_eeoo;
   Ddwf.M(src,ref);  
 
-  //  std::cout << r_eo<<std::endl;
-  //  std::cout << ref <<std::endl;
+  //  std::cout<<GridLogMessage << r_eo<<std::endl;
+  //  std::cout<<GridLogMessage << ref <<std::endl;
 
   err= ref - r_eo;
-  std::cout << "EO norm diff   "<< norm2(err)<< " "<<norm2(ref)<< " " << norm2(r_eo) <<std::endl;
+  std::cout<<GridLogMessage << "EO norm diff   "<< norm2(err)<< " "<<norm2(ref)<< " " << norm2(r_eo) <<std::endl;
     
   LatticeComplex cerr(FGrid);
   cerr = localInnerProduct(err,err);
-  //  std::cout << cerr<<std::endl;
+  //  std::cout<<GridLogMessage << cerr<<std::endl;
 
-  std::cout<<"=============================================================="<<std::endl;
-  std::cout<<"= Test Ddagger is the dagger of D by requiring                "<<std::endl;
-  std::cout<<"=  < phi | Deo | chi > * = < chi | Deo^dag| phi>  "<<std::endl;
-  std::cout<<"=============================================================="<<std::endl;
+  std::cout<<GridLogMessage<<"=============================================================="<<std::endl;
+  std::cout<<GridLogMessage<<"= Test Ddagger is the dagger of D by requiring                "<<std::endl;
+  std::cout<<GridLogMessage<<"=  < phi | Deo | chi > * = < chi | Deo^dag| phi>  "<<std::endl;
+  std::cout<<GridLogMessage<<"=============================================================="<<std::endl;
   
   LatticeFermion chi_e   (FrbGrid);
   LatticeFermion chi_o   (FrbGrid);
@@ -130,15 +130,15 @@ int main (int argc, char ** argv)
   ComplexD cDpe = innerProduct(chi_e,dphi_e);
   ComplexD cDpo = innerProduct(chi_o,dphi_o);
 
-  std::cout <<"e "<<pDce<<" "<<cDpe <<std::endl;
-  std::cout <<"o "<<pDco<<" "<<cDpo <<std::endl;
+  std::cout<<GridLogMessage <<"e "<<pDce<<" "<<cDpe <<std::endl;
+  std::cout<<GridLogMessage <<"o "<<pDco<<" "<<cDpo <<std::endl;
 
-  std::cout <<"pDce - conj(cDpo) "<< pDce-conj(cDpo) <<std::endl;
-  std::cout <<"pDco - conj(cDpe) "<< pDco-conj(cDpe) <<std::endl;
+  std::cout<<GridLogMessage <<"pDce - conj(cDpo) "<< pDce-conj(cDpo) <<std::endl;
+  std::cout<<GridLogMessage <<"pDco - conj(cDpe) "<< pDco-conj(cDpe) <<std::endl;
 
-  std::cout<<"=============================================================="<<std::endl;
-  std::cout<<"= Test MeeInv Mee = 1                                         "<<std::endl;
-  std::cout<<"=============================================================="<<std::endl;
+  std::cout<<GridLogMessage<<"=============================================================="<<std::endl;
+  std::cout<<GridLogMessage<<"= Test MeeInv Mee = 1                                         "<<std::endl;
+  std::cout<<GridLogMessage<<"=============================================================="<<std::endl;
 
   pickCheckerboard(Even,chi_e,chi);
   pickCheckerboard(Odd ,chi_o,chi);
@@ -153,11 +153,11 @@ int main (int argc, char ** argv)
   setCheckerboard(phi,phi_o);
 
   err = phi-chi;
-  std::cout << "norm diff   "<< norm2(err)<< std::endl;
+  std::cout<<GridLogMessage << "norm diff   "<< norm2(err)<< std::endl;
 
-  std::cout<<"=============================================================="<<std::endl;
-  std::cout<<"= Test MeeInvDag MeeDag = 1                                   "<<std::endl;
-  std::cout<<"=============================================================="<<std::endl;
+  std::cout<<GridLogMessage<<"=============================================================="<<std::endl;
+  std::cout<<GridLogMessage<<"= Test MeeInvDag MeeDag = 1                                   "<<std::endl;
+  std::cout<<GridLogMessage<<"=============================================================="<<std::endl;
 
   pickCheckerboard(Even,chi_e,chi);
   pickCheckerboard(Odd ,chi_o,chi);
@@ -172,11 +172,11 @@ int main (int argc, char ** argv)
   setCheckerboard(phi,phi_o);
 
   err = phi-chi;
-  std::cout << "norm diff   "<< norm2(err)<< std::endl;
+  std::cout<<GridLogMessage << "norm diff   "<< norm2(err)<< std::endl;
 
-  std::cout<<"=============================================================="<<std::endl;
-  std::cout<<"= Test MpcDagMpc is Hermitian              "<<std::endl;
-  std::cout<<"=============================================================="<<std::endl;
+  std::cout<<GridLogMessage<<"=============================================================="<<std::endl;
+  std::cout<<GridLogMessage<<"= Test MpcDagMpc is Hermitian              "<<std::endl;
+  std::cout<<GridLogMessage<<"=============================================================="<<std::endl;
   
   random(RNG5,phi);
   random(RNG5,chi);
@@ -199,11 +199,11 @@ int main (int argc, char ** argv)
   cDpe = innerProduct(chi_e,dphi_e);
   cDpo = innerProduct(chi_o,dphi_o);
 
-  std::cout <<"e "<<pDce<<" "<<cDpe <<std::endl;
-  std::cout <<"o "<<pDco<<" "<<cDpo <<std::endl;
+  std::cout<<GridLogMessage <<"e "<<pDce<<" "<<cDpe <<std::endl;
+  std::cout<<GridLogMessage <<"o "<<pDco<<" "<<cDpo <<std::endl;
 
-  std::cout <<"pDce - conj(cDpo) "<< pDco-conj(cDpo) <<std::endl;
-  std::cout <<"pDco - conj(cDpe) "<< pDce-conj(cDpe) <<std::endl;
+  std::cout<<GridLogMessage <<"pDce - conj(cDpo) "<< pDco-conj(cDpo) <<std::endl;
+  std::cout<<GridLogMessage <<"pDco - conj(cDpe) "<< pDce-conj(cDpe) <<std::endl;
   
   Grid_finalize();
 }
diff --git a/tests/Test_dwf_fpgcr.cc b/tests/Test_dwf_fpgcr.cc
index 969377a0..6c589c6c 100644
--- a/tests/Test_dwf_fpgcr.cc
+++ b/tests/Test_dwf_fpgcr.cc
@@ -54,35 +54,35 @@ int main (int argc, char ** argv)
   RealD M5=1.8;
   DomainWallFermion Ddwf(Umu,*FGrid,*FrbGrid,*UGrid,*UrbGrid,mass,M5);
 
-  std::cout<<"*********************************************************"<<std::endl;
-  std::cout<<"* Solving with MdagM VPGCR "<<std::endl;
-  std::cout<<"*********************************************************"<<std::endl;
+  std::cout<<GridLogMessage<<"*********************************************************"<<std::endl;
+  std::cout<<GridLogMessage<<"* Solving with MdagM VPGCR "<<std::endl;
+  std::cout<<GridLogMessage<<"*********************************************************"<<std::endl;
   MdagMLinearOperator<DomainWallFermion,LatticeFermion> HermOp(Ddwf);
   result=zero;
   PGCR(HermOp,src,result);
 
-  std::cout<<"*********************************************************"<<std::endl;
-  std::cout<<"* Solving with g5-VPGCR "<<std::endl;
-  std::cout<<"*********************************************************"<<std::endl;
+  std::cout<<GridLogMessage<<"*********************************************************"<<std::endl;
+  std::cout<<GridLogMessage<<"* Solving with g5-VPGCR "<<std::endl;
+  std::cout<<GridLogMessage<<"*********************************************************"<<std::endl;
   Gamma5R5HermitianLinearOperator<DomainWallFermion,LatticeFermion> g5HermOp(Ddwf);
   result=zero;
   PGCR(g5HermOp,src,result);
 
-  std::cout<<"*********************************************************"<<std::endl;
-  std::cout<<"* Solving with MdagM-CR "<<std::endl;
-  std::cout<<"*********************************************************"<<std::endl;
+  std::cout<<GridLogMessage<<"*********************************************************"<<std::endl;
+  std::cout<<GridLogMessage<<"* Solving with MdagM-CR "<<std::endl;
+  std::cout<<GridLogMessage<<"*********************************************************"<<std::endl;
   result=zero;
   CR(HermOp,src,result);
 
-  std::cout<<"*********************************************************"<<std::endl;
-  std::cout<<"* Solving with g5-CR "<<std::endl;
-  std::cout<<"*********************************************************"<<std::endl;
+  std::cout<<GridLogMessage<<"*********************************************************"<<std::endl;
+  std::cout<<GridLogMessage<<"* Solving with g5-CR "<<std::endl;
+  std::cout<<GridLogMessage<<"*********************************************************"<<std::endl;
   result=zero;
   CR(g5HermOp,src,result);
 
-  std::cout<<"*********************************************************"<<std::endl;
-  std::cout<<"* Solving with MdagM-CG "<<std::endl;
-  std::cout<<"*********************************************************"<<std::endl;
+  std::cout<<GridLogMessage<<"*********************************************************"<<std::endl;
+  std::cout<<GridLogMessage<<"* Solving with MdagM-CG "<<std::endl;
+  std::cout<<GridLogMessage<<"*********************************************************"<<std::endl;
   result=zero;
   CG(HermOp,src,result);
 
diff --git a/tests/Test_dwf_hdcr.cc b/tests/Test_dwf_hdcr.cc
index 488f7277..29f6e998 100644
--- a/tests/Test_dwf_hdcr.cc
+++ b/tests/Test_dwf_hdcr.cc
@@ -53,7 +53,7 @@ public:
       //      _FineOperator.Op(p1,p2);// this is the G5 herm bit      
       RealD absp2=std::sqrt(norm2(p2));
       if(i%10==9)
-	std::cout << "Power method on mdagm "<<i<<" " << absp2/absp1<<std::endl;
+	std::cout<<GridLogMessage << "Power method on mdagm "<<i<<" " << absp2/absp1<<std::endl;
       p1=p2*(1.0/std::sqrt(absp2));
     }
   }
@@ -71,7 +71,7 @@ public:
     // Monitor completeness of low mode space
     _Aggregates.ProjectToSubspace  (Csrc,in);
     _Aggregates.PromoteFromSubspace(Csrc,out);
-    std::cout<<"Completeness: "<<std::sqrt(norm2(out)/norm2(in))<<std::endl;
+    std::cout<<GridLogMessage<<"Completeness: "<<std::sqrt(norm2(out)/norm2(in))<<std::endl;
 
     // Build some solvers
     ConjugateGradient<FineField>    fCG(1.0e-3,1000);
@@ -85,21 +85,21 @@ public:
     MdagMLinearOperator<Matrix,FineField> MdagMOp(_Matrix);
 
     Min=in;
-    std::cout<< " Preconditioner in  " << norm2(in)<<std::endl; 
+    std::cout<<GridLogMessage<< " Preconditioner in  " << norm2(in)<<std::endl; 
     _FineOperator.AdjOp(Min,tmp);
-    std::cout<< " Preconditioner tmp  " << norm2(in)<<std::endl; 
+    std::cout<<GridLogMessage<< " Preconditioner tmp  " << norm2(in)<<std::endl; 
 
     fCG(MdagMOp,tmp,out);
 
     _FineOperator.Op(out,tmp);
 
-    std::cout<< " Preconditioner in  " << norm2(in)<<std::endl; 
-    std::cout<< " Preconditioner out " << norm2(out)<<std::endl; 
-    std::cout<< " Preconditioner Aout" << norm2(tmp)<<std::endl; 
+    std::cout<<GridLogMessage<< " Preconditioner in  " << norm2(in)<<std::endl; 
+    std::cout<<GridLogMessage<< " Preconditioner out " << norm2(out)<<std::endl; 
+    std::cout<<GridLogMessage<< " Preconditioner Aout" << norm2(tmp)<<std::endl; 
 
     tmp = tmp - in;
     
-    std::cout<<"preconditioner thinks residual is "<<std::sqrt(norm2(tmp)/norm2(in))<<std::endl;
+    std::cout<<GridLogMessage<<"preconditioner thinks residual is "<<std::sqrt(norm2(tmp)/norm2(in))<<std::endl;
 
     /*
     //    _FineOperator.Op(Min,out);
@@ -145,7 +145,7 @@ public:
     // Monitor completeness of low mode space
     _Aggregates.ProjectToSubspace  (Csrc,in);
     _Aggregates.PromoteFromSubspace(Csrc,out);
-    std::cout<<"Coarse Grid Preconditioner\nCompleteness in: "<<std::sqrt(norm2(out)/norm2(in))<<std::endl;
+    std::cout<<GridLogMessage<<"Coarse Grid Preconditioner\nCompleteness in: "<<std::sqrt(norm2(out)/norm2(in))<<std::endl;
 
     // [PTM+Q] in = [1 - Q A] M in + Q in = Min + Q [ in -A Min]
     _FineOperator.Op(in,tmp);// this is the G5 herm bit
@@ -154,7 +154,7 @@ public:
     // Monitor completeness of low mode space
     _Aggregates.ProjectToSubspace  (Csrc,Min);
     _Aggregates.PromoteFromSubspace(Csrc,out);
-    std::cout<<"Completeness Min: "<<std::sqrt(norm2(out)/norm2(Min))<<std::endl;
+    std::cout<<GridLogMessage<<"Completeness Min: "<<std::sqrt(norm2(out)/norm2(Min))<<std::endl;
 
     _FineOperator.Op(Min,tmp);
     tmp = in - tmp;   // in - A Min
@@ -166,14 +166,14 @@ public:
 
     HermOp.Op(Csol,Ctmp);
     Ctmp=Ctmp-Csrc;
-    std::cout<<"coarse space true residual "<<std::sqrt(norm2(Ctmp)/norm2(Csrc))<<std::endl;
+    std::cout<<GridLogMessage<<"coarse space true residual "<<std::sqrt(norm2(Ctmp)/norm2(Csrc))<<std::endl;
     _Aggregates.PromoteFromSubspace(Csol,out);
 
     _FineOperator.Op(out,res);
     res=res-tmp;
-    std::cout<<"promoted sol residual "<<std::sqrt(norm2(res)/norm2(tmp))<<std::endl;
+    std::cout<<GridLogMessage<<"promoted sol residual "<<std::sqrt(norm2(res)/norm2(tmp))<<std::endl;
     _Aggregates.ProjectToSubspace  (Csrc,res);
-    std::cout<<"coarse space proj of residual "<<norm2(Csrc)<<std::endl;
+    std::cout<<GridLogMessage<<"coarse space proj of residual "<<norm2(Csrc)<<std::endl;
 
     
     out = out+Min; // additive coarse space correction
@@ -182,9 +182,9 @@ public:
     _FineOperator.Op(out,tmp);
     tmp=tmp-in;         // tmp is new residual
 
-    std::cout<< " Preconditioner in  " << norm2(in)<<std::endl; 
-    std::cout<< " Preconditioner out " << norm2(out)<<std::endl; 
-    std::cout<<"preconditioner thinks residual is "<<std::sqrt(norm2(tmp)/norm2(in))<<std::endl;
+    std::cout<<GridLogMessage<< " Preconditioner in  " << norm2(in)<<std::endl; 
+    std::cout<<GridLogMessage<< " Preconditioner out " << norm2(out)<<std::endl; 
+    std::cout<<GridLogMessage<<"preconditioner thinks residual is "<<std::sqrt(norm2(tmp)/norm2(in))<<std::endl;
 
   }
 #endif
@@ -210,7 +210,7 @@ public:
     // Monitor completeness of low mode space
     //    _Aggregates.ProjectToSubspace  (Csrc,in);
     //    _Aggregates.PromoteFromSubspace(Csrc,out);
-    //    std::cout<<"Coarse Grid Preconditioner\nCompleteness in: "<<std::sqrt(norm2(out)/norm2(in))<<std::endl;
+    //    std::cout<<GridLogMessage<<"Coarse Grid Preconditioner\nCompleteness in: "<<std::sqrt(norm2(out)/norm2(in))<<std::endl;
     
     _Aggregates.ProjectToSubspace  (Csrc,in);
     HermOp.AdjOp(Csrc,Ctmp);// Normal equations
@@ -229,7 +229,7 @@ public:
     _FineOperator.Op(out,tmp);
     tmp=tmp-in;         // tmp is new residual
 
-    std::cout<<"preconditioner thinks residual is "<<std::sqrt(norm2(tmp)/norm2(in))<<std::endl;
+    std::cout<<GridLogMessage<<"preconditioner thinks residual is "<<std::sqrt(norm2(tmp)/norm2(in))<<std::endl;
 
   }
 #endif
@@ -247,8 +247,8 @@ public:
 
     Ni = norm2(in);
 
-    for(int ilo=0;ilo<4;ilo++){
-      for(int ord=10;ord<60;ord+=10){
+    for(int ilo=0;ilo<3;ilo++){
+      for(int ord=5;ord<50;ord*=2){
 
 	_FineOperator.AdjOp(in,vec1);
 
@@ -258,7 +258,7 @@ public:
 	_FineOperator.Op(vec2,vec1);// this is the G5 herm bit
 	vec1  = in - vec1;   // tmp  = in - A Min
 	r=norm2(vec1);
-	std::cout << "Smoother resid "<<std::sqrt(r/Ni)<<std::endl;
+	std::cout<<GridLogMessage << "Smoother resid "<<std::sqrt(r/Ni)<<std::endl;
 
       }
     }
@@ -270,25 +270,25 @@ public:
     CoarseVector Ctmp(_CoarseOperator.Grid());
     CoarseVector Csol(_CoarseOperator.Grid()); Csol=zero;
 
-    ConjugateGradient<CoarseVector>  CG(1.0e-5,100000);
-    ConjugateGradient<FineField>    fCG(3.0e-2,1000);
+    ConjugateGradient<CoarseVector>  CG(1.0e-3,100000);
+    //    ConjugateGradient<FineField>    fCG(3.0e-2,1000);
 
     HermitianLinearOperator<CoarseOperator,CoarseVector>  HermOp(_CoarseOperator);
     MdagMLinearOperator<CoarseOperator,CoarseVector>     MdagMOp(_CoarseOperator);
     //    MdagMLinearOperator<Matrix,FineField>        fMdagMOp(_Matrix);
-    ShiftedMdagMLinearOperator<Matrix,FineField> fMdagMOp(_Matrix,0.5);
+    ShiftedMdagMLinearOperator<Matrix,FineField> fMdagMOp(_Matrix,0.0);
 
     FineField vec1(in._grid);
     FineField vec2(in._grid);
 
     //    Chebyshev<FineField> Cheby    (0.5,70.0,30,InverseApproximation);
     //    Chebyshev<FineField> ChebyAccu(0.5,70.0,30,InverseApproximation);
-    Chebyshev<FineField> Cheby    (1.0,70.0,20,InverseApproximation);
-    Chebyshev<FineField> ChebyAccu(1.0,70.0,20,InverseApproximation);
+    Chebyshev<FineField> Cheby    (2.0,70.0,10,InverseApproximation);
+    Chebyshev<FineField> ChebyAccu(2.0,70.0,10,InverseApproximation);
 
     _Aggregates.ProjectToSubspace  (Csrc,in);
     _Aggregates.PromoteFromSubspace(Csrc,out);
-    std::cout<<"Completeness: "<<std::sqrt(norm2(out)/norm2(in))<<std::endl;
+    std::cout<<GridLogMessage<<"Completeness: "<<std::sqrt(norm2(out)/norm2(in))<<std::endl;
     
     //    ofstream fout("smoother");
     //    Cheby.csv(fout);
@@ -306,7 +306,7 @@ public:
     _FineOperator.AdjOp(in,vec1);// this is the G5 herm bit
     ChebyAccu(fMdagMOp,vec1,out);    // solves  MdagM = g5 M g5M
 
-    std::cout << "Smoother norm "<<norm2(out)<<std::endl;
+    std::cout<<GridLogMessage << "Smoother norm "<<norm2(out)<<std::endl;
 
     // Update with residual for out
     _FineOperator.Op(out,vec1);// this is the G5 herm bit
@@ -314,7 +314,7 @@ public:
 
     RealD r = norm2(vec1);
 
-    std::cout << "Smoother resid "<<std::sqrt(r/Ni)<< " " << r << " " << Ni <<std::endl;
+    std::cout<<GridLogMessage << "Smoother resid "<<std::sqrt(r/Ni)<< " " << r << " " << Ni <<std::endl;
     
     _Aggregates.ProjectToSubspace  (Csrc,vec1);
     HermOp.AdjOp(Csrc,Ctmp);// Normal equations
@@ -329,7 +329,7 @@ public:
     vec1  = in - vec1;   // tmp  = in - A Min
     r=norm2(vec1);
 
-    std::cout << "Coarse resid "<<std::sqrt(r/Ni)<<std::endl;
+    std::cout<<GridLogMessage << "Coarse resid "<<std::sqrt(r/Ni)<<std::endl;
 
     // Reapply smoother
     _FineOperator.Op(vec1,vec2);  // this is the G5 herm bit
@@ -339,7 +339,7 @@ public:
     _FineOperator.Op(out,vec1);// this is the G5 herm bit
     vec1  = in - vec1;   // tmp  = in - A Min
     r=norm2(vec1);
-    std::cout << "Smoother resid "<<std::sqrt(r/Ni)<<std::endl;
+    std::cout<<GridLogMessage << "Smoother resid "<<std::sqrt(r/Ni)<<std::endl;
 
   }
 
@@ -385,7 +385,7 @@ int main (int argc, char ** argv)
   LatticeGaugeField Umu(UGrid); 
 
   NerscField header;
-  std::string file("./ckpoint_lat.4000");
+  std::string file("./ckpoint_lat.400");
   readNerscConfiguration(Umu,header,file);
 
   //  SU3::ColdConfiguration(RNG4,Umu);
@@ -396,9 +396,9 @@ int main (int argc, char ** argv)
   RealD mass=0.01;
   RealD M5=1.8;
 
-  std::cout << "**************************************************"<< std::endl;
-  std::cout << "Building g5R5 hermitian DWF operator" <<std::endl;
-  std::cout << "**************************************************"<< std::endl;
+  std::cout<<GridLogMessage << "**************************************************"<< std::endl;
+  std::cout<<GridLogMessage << "Building g5R5 hermitian DWF operator" <<std::endl;
+  std::cout<<GridLogMessage << "**************************************************"<< std::endl;
   DomainWallFermion Ddwf(Umu,*FGrid,*FrbGrid,*UGrid,*UrbGrid,mass,M5);
 
   const int nbasis = 32;
@@ -408,22 +408,22 @@ int main (int argc, char ** argv)
   typedef CoarsenedMatrix<vSpinColourVector,vTComplex,nbasis>          CoarseOperator;
   typedef CoarseOperator::CoarseVector                                 CoarseVector;
 
-  std::cout << "**************************************************"<< std::endl;
-  std::cout << "Calling Aggregation class to build subspace" <<std::endl;
-  std::cout << "**************************************************"<< std::endl;
+  std::cout<<GridLogMessage << "**************************************************"<< std::endl;
+  std::cout<<GridLogMessage << "Calling Aggregation class to build subspace" <<std::endl;
+  std::cout<<GridLogMessage << "**************************************************"<< std::endl;
   MdagMLinearOperator<DomainWallFermion,LatticeFermion> HermDefOp(Ddwf);
   Subspace Aggregates(Coarse5d,FGrid);
   //  Aggregates.CreateSubspace(RNG5,HermDefOp,nbasis);
   assert ( (nbasis & 0x1)==0);
   int nb=nbasis/2;
-  std::cout << " nbasis/2 = "<<nb<<std::endl;
+  std::cout<<GridLogMessage << " nbasis/2 = "<<nb<<std::endl;
   Aggregates.CreateSubspace(RNG5,HermDefOp,nb);
   for(int n=0;n<nb;n++){
     G5R5(Aggregates.subspace[n+nb],Aggregates.subspace[n]);
-    std::cout<<n<<" subspace "<<norm2(Aggregates.subspace[n+nb])<<" "<<norm2(Aggregates.subspace[n]) <<std::endl;
+    std::cout<<GridLogMessage<<n<<" subspace "<<norm2(Aggregates.subspace[n+nb])<<" "<<norm2(Aggregates.subspace[n]) <<std::endl;
   }
   for(int n=0;n<nbasis;n++){
-    std::cout << "vec["<<n<<"] = "<<norm2(Aggregates.subspace[n])  <<std::endl;
+    std::cout<<GridLogMessage << "vec["<<n<<"] = "<<norm2(Aggregates.subspace[n])  <<std::endl;
   }
 
 //  for(int i=0;i<nbasis;i++){
@@ -432,77 +432,90 @@ int main (int argc, char ** argv)
 //  }
   result=zero;
   
-  std::cout << "**************************************************"<< std::endl;
-  std::cout << "Building coarse representation of Indef operator" <<std::endl;
-  std::cout << "**************************************************"<< std::endl;
+  std::cout<<GridLogMessage << "**************************************************"<< std::endl;
+  std::cout<<GridLogMessage << "Building coarse representation of Indef operator" <<std::endl;
+  std::cout<<GridLogMessage << "**************************************************"<< std::endl;
   Gamma5R5HermitianLinearOperator<DomainWallFermion,LatticeFermion> HermIndefOp(Ddwf);
   CoarsenedMatrix<vSpinColourVector,vTComplex,nbasis> LDOp(*Coarse5d);
   LDOp.CoarsenOperator(FGrid,HermIndefOp,Aggregates);
 
-  std::cout << "**************************************************"<< std::endl;
-  std::cout << "Testing some coarse space solvers  " <<std::endl;
-  std::cout << "**************************************************"<< std::endl;
+  std::cout<<GridLogMessage << "**************************************************"<< std::endl;
+  std::cout<<GridLogMessage << "Testing some coarse space solvers  " <<std::endl;
+  std::cout<<GridLogMessage << "**************************************************"<< std::endl;
   CoarseVector c_src (Coarse5d);
   CoarseVector c_res (Coarse5d);
   gaussian(CRNG,c_src);
   c_res=zero;
 
-  std::cout << "**************************************************"<< std::endl;
-  std::cout << "Solving posdef-CG on coarse space "<< std::endl;
-  std::cout << "**************************************************"<< std::endl;
+  std::cout<<GridLogMessage << "**************************************************"<< std::endl;
+  std::cout<<GridLogMessage << "Solving posdef-CG on coarse space "<< std::endl;
+  std::cout<<GridLogMessage << "**************************************************"<< std::endl;
   MdagMLinearOperator<CoarseOperator,CoarseVector> PosdefLdop(LDOp);
   ConjugateGradient<CoarseVector> CG(1.0e-6,100000);
   CG(PosdefLdop,c_src,c_res);
 
-  //  std::cout << "**************************************************"<< std::endl;
-  //  std::cout << "Solving indef-MCR on coarse space "<< std::endl;
-  //  std::cout << "**************************************************"<< std::endl;
+  //  std::cout<<GridLogMessage << "**************************************************"<< std::endl;
+  //  std::cout<<GridLogMessage << "Solving indef-MCR on coarse space "<< std::endl;
+  //  std::cout<<GridLogMessage << "**************************************************"<< std::endl;
   //  HermitianLinearOperator<CoarseOperator,CoarseVector> HermIndefLdop(LDOp);
   //  ConjugateResidual<CoarseVector> MCR(1.0e-6,100000);
   //MCR(HermIndefLdop,c_src,c_res);
 
-  std::cout << "**************************************************"<< std::endl;
-  std::cout << "Building deflation preconditioner "<< std::endl;
-  std::cout << "**************************************************"<< std::endl;
+  std::cout<<GridLogMessage << "**************************************************"<< std::endl;
+  std::cout<<GridLogMessage << "Building deflation preconditioner "<< std::endl;
+  std::cout<<GridLogMessage << "**************************************************"<< std::endl;
 
   MultiGridPreconditioner <vSpinColourVector,vTComplex,nbasis,DomainWallFermion> Precon(Aggregates, LDOp,HermIndefOp,Ddwf);
   TrivialPrecon<LatticeFermion> simple;
 
-  std::cout << "**************************************************"<< std::endl;
-  std::cout << "Testing smoother efficacy"<< std::endl;
-  std::cout << "**************************************************"<< std::endl;
+  std::cout<<GridLogMessage << "**************************************************"<< std::endl;
+  std::cout<<GridLogMessage << "Testing smoother efficacy"<< std::endl;
+  std::cout<<GridLogMessage << "**************************************************"<< std::endl;
   Precon.SmootherTest(src);
 
-  std::cout << "**************************************************"<< std::endl;
-  std::cout << "Unprec CG "<< std::endl;
-  std::cout << "**************************************************"<< std::endl;
+  std::cout<<GridLogMessage << "**************************************************"<< std::endl;
+  std::cout<<GridLogMessage << "Unprec CG "<< std::endl;
+  std::cout<<GridLogMessage << "**************************************************"<< std::endl;
   //  TrivialPrecon<LatticeFermion> simple;
-  ConjugateGradient<LatticeFermion> fCG(1.0e-8,100000);
-  fCG(HermDefOp,src,result);
+  //  ConjugateGradient<LatticeFermion> fCG(1.0e-8,100000);
+  //  fCG(HermDefOp,src,result);
   //  exit(0);
 
-  std::cout << "**************************************************"<< std::endl;
-  std::cout << "Testing GCR on indef matrix "<< std::endl;
-  std::cout << "**************************************************"<< std::endl;
+  std::cout<<GridLogMessage << "**************************************************"<< std::endl;
+  std::cout<<GridLogMessage << "Testing GCR on indef matrix "<< std::endl;
+  std::cout<<GridLogMessage << "**************************************************"<< std::endl;
   //  PrecGeneralisedConjugateResidual<LatticeFermion> UPGCR(1.0e-8,100000,simple,8,128);
   //  UPGCR(HermIndefOp,src,result);
 
   
   /// Get themax eval
-  std::cout << "**************************************************"<< std::endl;
-  std::cout <<" Applying power method to find spectral range      "<<std::endl;
-  std::cout << "**************************************************"<< std::endl;
+  std::cout<<GridLogMessage << "**************************************************"<< std::endl;
+  std::cout<<GridLogMessage <<" Applying power method to find spectral range      "<<std::endl;
+  std::cout<<GridLogMessage << "**************************************************"<< std::endl;
   Precon.PowerMethod(src);
 
-  std::cout << "**************************************************"<< std::endl;
-  std::cout << "Building a two level PGCR "<< std::endl;
-  std::cout << "**************************************************"<< std::endl;
+  std::cout<<GridLogMessage << "**************************************************"<< std::endl;
+  std::cout<<GridLogMessage << "Building a two level PGCR "<< std::endl;
+  std::cout<<GridLogMessage << "**************************************************"<< std::endl;
   PrecGeneralisedConjugateResidual<LatticeFermion> PGCR(1.0e-8,100000,Precon,8,128);
-  std::cout<<"checking norm src "<<norm2(src)<<std::endl;
+  std::cout<<GridLogMessage<<"checking norm src "<<norm2(src)<<std::endl;
   PGCR(HermIndefOp,src,result);
 
-  std::cout << "**************************************************"<< std::endl;
-  std::cout << "Done "<< std::endl;
-  std::cout << "**************************************************"<< std::endl;
+  std::cout<<GridLogMessage << "**************************************************"<< std::endl;
+  std::cout<<GridLogMessage << "Red Black Prec CG "<< std::endl;
+  std::cout<<GridLogMessage << "**************************************************"<< std::endl;
+  SchurDiagMooeeOperator<DomainWallFermion,LatticeFermion> HermOpEO(Ddwf);
+  ConjugateGradient<LatticeFermion> pCG(1.0e-8,10000);
+
+  LatticeFermion    src_o(FrbGrid);
+  LatticeFermion result_o(FrbGrid);
+  pickCheckerboard(Odd,src_o,src);
+  result_o=zero;
+
+  pCG(HermOpEO,src_o,result_o);
+
+  std::cout<<GridLogMessage << "**************************************************"<< std::endl;
+  std::cout<<GridLogMessage << "Done "<< std::endl;
+  std::cout<<GridLogMessage << "**************************************************"<< std::endl;
   Grid_finalize();
 }
diff --git a/tests/Test_gamma.cc b/tests/Test_gamma.cc
index 1f9cb491..748409b1 100644
--- a/tests/Test_gamma.cc
+++ b/tests/Test_gamma.cc
@@ -35,19 +35,19 @@ int main (int argc, char ** argv)
   SpinVector lv; random(sRNG,lv);
   SpinVector rv; random(sRNG,rv);
 
-  //  std::cout << " Is pod " << std::is_pod<SpinVector>::value  << std::endl;
-  //  std::cout << " Is pod double   " << std::is_pod<double>::value  << std::endl;
-  //  std::cout << " Is pod ComplexF " << std::is_pod<ComplexF>::value  << std::endl;
-  //  std::cout << " Is triv double " << std::has_trivial_default_constructor<double>::value  << std::endl;
-  //  std::cout << " Is triv ComplexF " << std::has_trivial_default_constructor<ComplexF>::value  << std::endl;
-  //  std::cout << " Is pod Scalar<double> " << std::is_pod<iScalar<double> >::value  << std::endl;
-  //  std::cout << " Is pod Scalar<ComplexF> " << std::is_pod<iScalar<ComplexF> >::value  << std::endl;
-  //  std::cout << " Is pod Scalar<vComplexF> " << std::is_pod<iScalar<vComplexF> >::value  << std::endl;
-  //  std::cout << " Is pod Scalar<vComplexD> " << std::is_pod<iScalar<vComplexD> >::value  << std::endl;
-  //  std::cout << " Is pod Scalar<vRealF> " << std::is_pod<iScalar<vRealF> >::value  << std::endl;
-  //  std::cout << " Is pod Scalar<vRealD> " << std::is_pod<iScalar<vRealD> >::value  << std::endl;
-  //  std::cout << " Is triv Scalar<double> " <<std::has_trivial_default_constructor<iScalar<double> >::value << std::endl;
-  //  std::cout << " Is triv Scalar<vComplexD> "<<std::has_trivial_default_constructor<iScalar<vComplexD> >::value  << std::endl;
+  //  std::cout<<GridLogMessage << " Is pod " << std::is_pod<SpinVector>::value  << std::endl;
+  //  std::cout<<GridLogMessage << " Is pod double   " << std::is_pod<double>::value  << std::endl;
+  //  std::cout<<GridLogMessage << " Is pod ComplexF " << std::is_pod<ComplexF>::value  << std::endl;
+  //  std::cout<<GridLogMessage << " Is triv double " << std::has_trivial_default_constructor<double>::value  << std::endl;
+  //  std::cout<<GridLogMessage << " Is triv ComplexF " << std::has_trivial_default_constructor<ComplexF>::value  << std::endl;
+  //  std::cout<<GridLogMessage << " Is pod Scalar<double> " << std::is_pod<iScalar<double> >::value  << std::endl;
+  //  std::cout<<GridLogMessage << " Is pod Scalar<ComplexF> " << std::is_pod<iScalar<ComplexF> >::value  << std::endl;
+  //  std::cout<<GridLogMessage << " Is pod Scalar<vComplexF> " << std::is_pod<iScalar<vComplexF> >::value  << std::endl;
+  //  std::cout<<GridLogMessage << " Is pod Scalar<vComplexD> " << std::is_pod<iScalar<vComplexD> >::value  << std::endl;
+  //  std::cout<<GridLogMessage << " Is pod Scalar<vRealF> " << std::is_pod<iScalar<vRealF> >::value  << std::endl;
+  //  std::cout<<GridLogMessage << " Is pod Scalar<vRealD> " << std::is_pod<iScalar<vRealD> >::value  << std::endl;
+  //  std::cout<<GridLogMessage << " Is triv Scalar<double> " <<std::has_trivial_default_constructor<iScalar<double> >::value << std::endl;
+  //  std::cout<<GridLogMessage << " Is triv Scalar<vComplexD> "<<std::has_trivial_default_constructor<iScalar<vComplexD> >::value  << std::endl;
 
   for(int a=0;a<Ns;a++){
     ident()(a,a) = ComplexF(1.0);
@@ -66,8 +66,8 @@ int main (int argc, char ** argv)
 
     for(int i=0;i<Ns;i++){
 
-      if(i==0) std::cout << list[mu];
-      else     std::cout << list[12];
+      if(i==0) std::cout<<GridLogMessage << list[mu];
+      else     std::cout<<GridLogMessage << list[12];
 
       std::cout<<"(";
       for(int j=0;j<Ns;j++){
@@ -82,7 +82,7 @@ int main (int argc, char ** argv)
 	} else if ( abs(result()(i,j)() + Complex(1,0))==0){
 	  std::cout<< "-1";
 	}
-	std::cout<< ((j==Ns-1) ? ")" : "," );
+	std::cout<<((j==Ns-1) ? ")" : "," );
       }
       std::cout << std::endl;
     }
diff --git a/tests/Test_hmc_WilsonFermionGauge.cc b/tests/Test_hmc_WilsonFermionGauge.cc
new file mode 100644
index 00000000..70e68a44
--- /dev/null
+++ b/tests/Test_hmc_WilsonFermionGauge.cc
@@ -0,0 +1,56 @@
+#include "Grid.h"
+
+
+using namespace std;
+using namespace Grid;
+using namespace Grid::QCD;
+
+int main (int argc, char ** argv)
+{
+  Grid_init(&argc,&argv);
+
+  std::vector<int> latt_size   = GridDefaultLatt();
+  std::vector<int> simd_layout = GridDefaultSimd(4,vComplex::Nsimd());
+  std::vector<int> mpi_layout  = GridDefaultMpi();
+  
+  GridCartesian            Fine(latt_size,simd_layout,mpi_layout);
+  GridRedBlackCartesian  RBFine(latt_size,simd_layout,mpi_layout);
+  GridParallelRNG  pRNG(&Fine);
+  pRNG.SeedRandomDevice();
+  LatticeLorentzColourMatrix     U(&Fine);
+
+  SU3::HotConfiguration(pRNG, U);
+
+  // simplify template declaration? Strip the lorentz from the second template
+  WilsonGaugeAction<LatticeLorentzColourMatrix, LatticeColourMatrix> Waction(5.6);
+
+  Real mass=0.01;
+  WilsonFermion FermOp(U,Fine,RBFine,mass);
+  
+  ConjugateGradient<LatticeFermion>  CG(1.0e-8,10000);
+  
+  TwoFlavourPseudoFermionAction<LatticeLorentzColourMatrix, LatticeColourMatrix,LatticeFermion> 
+    WilsonNf2(FermOp,CG,CG,Fine);
+
+  
+  //Collect actions
+  ActionLevel Level1;
+  Level1.push_back(&WilsonNf2);
+  ActionLevel Level2(3);
+  Level2.push_back(&Waction);
+  ActionSet FullSet;
+  FullSet.push_back(Level1);
+  FullSet.push_back(Level2);
+
+  // Create integrator
+  typedef MinimumNorm2  IntegratorAlgorithm;// change here to modify the algorithm
+  IntegratorParameters MDpar(12,5,1.0);
+  Integrator<IntegratorAlgorithm> MDynamics(&Fine,MDpar, FullSet);
+
+  // Create HMC
+  HMCparameters HMCpar;
+  HybridMonteCarlo<IntegratorAlgorithm>  HMC(HMCpar, MDynamics);
+
+  HMC.evolve(U);
+
+}
diff --git a/tests/Test_lie_generators.cc b/tests/Test_lie_generators.cc
index 588baeda..25d11759 100644
--- a/tests/Test_lie_generators.cc
+++ b/tests/Test_lie_generators.cc
@@ -20,27 +20,27 @@ int main (int argc, char ** argv)
   
   GridRedBlackCartesian * rbGrid = SpaceTimeGrid::makeFourDimRedBlackGrid(grid);
 
-  std::cout<<"*********************************************"<<std::endl;
-  std::cout<<"* Generators for SU(2)"<<std::endl;
-  std::cout<<"*********************************************"<<std::endl;
+  std::cout<<GridLogMessage<<"*********************************************"<<std::endl;
+  std::cout<<GridLogMessage<<"* Generators for SU(2)"<<std::endl;
+  std::cout<<GridLogMessage<<"*********************************************"<<std::endl;
   SU2::printGenerators();
   SU2::testGenerators();
 
-  std::cout<<"*********************************************"<<std::endl;
-  std::cout<<"* Generators for SU(3)"<<std::endl;
-  std::cout<<"*********************************************"<<std::endl;
+  std::cout<<GridLogMessage<<"*********************************************"<<std::endl;
+  std::cout<<GridLogMessage<<"* Generators for SU(3)"<<std::endl;
+  std::cout<<GridLogMessage<<"*********************************************"<<std::endl;
   SU3::printGenerators();
   SU3::testGenerators();
 
-  //  std::cout<<"*********************************************"<<std::endl;
-  //  std::cout<<"* Generators for SU(4)"<<std::endl;
-  //  std::cout<<"*********************************************"<<std::endl;
+  //  std::cout<<GridLogMessage<<"*********************************************"<<std::endl;
+  //  std::cout<<GridLogMessage<<"* Generators for SU(4)"<<std::endl;
+  //  std::cout<<GridLogMessage<<"*********************************************"<<std::endl;
   //  SU4::printGenerators();
   //  SU4::testGenerators();
 
-  //  std::cout<<"*********************************************"<<std::endl;
-  //  std::cout<<"* Generators for SU(5)"<<std::endl;
-  //  std::cout<<"*********************************************"<<std::endl;
+  //  std::cout<<GridLogMessage<<"*********************************************"<<std::endl;
+  //  std::cout<<GridLogMessage<<"* Generators for SU(5)"<<std::endl;
+  //  std::cout<<GridLogMessage<<"*********************************************"<<std::endl;
   //  SU5::printGenerators();
   //  SU5::testGenerators();
 
diff --git a/tests/Test_main.cc b/tests/Test_main.cc
index a98fbeb7..b2b2ab31 100644
--- a/tests/Test_main.cc
+++ b/tests/Test_main.cc
@@ -213,29 +213,29 @@ int main (int argc, char ** argv)
     
    
     random(SerialRNG, cm);
-    std::cout << cm << std::endl;
+    std::cout<<GridLogMessage << cm << std::endl;
 
     cm = Ta(cm);
     TComplex tracecm= trace(cm);      
-    std::cout << cm << std::endl;
+    std::cout<<GridLogMessage << cm << std::endl;
 
 
     cm = Exponentiate(cm, 2.0, 12);
-    std::cout << cm << "  " << std::endl;
+    std::cout<<GridLogMessage << cm << "  " << std::endl;
     Complex det = Determinant(cm);
-    std::cout << "determinant: " << det <<  std::endl;
-    std::cout << "norm: " << norm2(cm) <<  std::endl;
+    std::cout<<GridLogMessage << "determinant: " << det <<  std::endl;
+    std::cout<<GridLogMessage << "norm: " << norm2(cm) <<  std::endl;
 
     cm = ProjectOnGroup(cm);
-    std::cout << cm << "  " << std::endl;
-    std::cout << "norm: " << norm2(cm) <<  std::endl;
+    std::cout<<GridLogMessage << cm << "  " << std::endl;
+    std::cout<<GridLogMessage << "norm: " << norm2(cm) <<  std::endl;
     cm = ProjectOnGroup(cm);
-    std::cout << cm << "  " << std::endl;
-    std::cout << "norm: " << norm2(cm) <<  std::endl;
+    std::cout<<GridLogMessage << cm << "  " << std::endl;
+    std::cout<<GridLogMessage << "norm: " << norm2(cm) <<  std::endl;
 
 
     //    det = Determinant(cm);
-    //    std::cout << "determinant: " << det <<  std::endl;
+    //    std::cout<<GridLogMessage << "determinant: " << det <<  std::endl;
 
 
 //    Foo = Foo+scalar; // LatticeColourMatrix+Scalar
@@ -253,19 +253,19 @@ int main (int argc, char ** argv)
     random(FineRNG,cMat);
     cMat = Ta(cMat);
     peekSite(cm, cMat, mysite);
-    std::cout << cm << "  " << std::endl;
+    std::cout<<GridLogMessage << cm << "  " << std::endl;
     cm = Exponentiate(cm, 1.0, 12);
-    std::cout << cm << "  " << std::endl;
-    std::cout << "norm: " << norm2(cm) <<  std::endl;
+    std::cout<<GridLogMessage << cm << "  " << std::endl;
+    std::cout<<GridLogMessage << "norm: " << norm2(cm) <<  std::endl;
 
 
-    std::cout << "norm cMmat : " << norm2(cMat) <<  std::endl;
+    std::cout<<GridLogMessage << "norm cMmat : " << norm2(cMat) <<  std::endl;
     cMat = expMat(cMat, ComplexD(1.0, 0.0));
-    std::cout << "norm expMat: " << norm2(cMat) <<  std::endl;
+    std::cout<<GridLogMessage << "norm expMat: " << norm2(cMat) <<  std::endl;
     peekSite(cm, cMat, mysite);
-    std::cout << cm << "  " << std::endl;
-    std::cout << "determinant: " << Determinant(cm) <<  std::endl;
-    std::cout << "norm: " << norm2(cm) <<  std::endl;
+    std::cout<<GridLogMessage << cm << "  " << std::endl;
+    std::cout<<GridLogMessage << "determinant: " << Determinant(cm) <<  std::endl;
+    std::cout<<GridLogMessage << "norm: " << norm2(cm) <<  std::endl;
 
 
     // LatticeComplex trlcMat(&Fine);
@@ -434,7 +434,7 @@ int main (int argc, char ** argv)
 	pickCheckerboard(1,bFoo,Foo);
     
 	if ( Fine.IsBoss() ) {
-	  std::cout << "Shifting both parities by "<< shift <<" direction "<< dir <<std::endl;
+	  std::cout<<GridLogMessage << "Shifting both parities by "<< shift <<" direction "<< dir <<std::endl;
 	}
 	Shifted  = Cshift(Foo,dir,shift);    // Shift everything
 
@@ -545,7 +545,7 @@ int main (int argc, char ** argv)
         }}
     }}}}
 	if( Fine.IsBoss() ){
-	  std::cout << "LatticeColorMatrix * LatticeColorMatrix nrm diff = "<<nrm<<std::endl;
+	  std::cout<<GridLogMessage << "LatticeColorMatrix * LatticeColorMatrix nrm diff = "<<nrm<<std::endl;
 	}
       }}
 
@@ -553,7 +553,7 @@ int main (int argc, char ** argv)
  } // loop for omp
 
 
- std::cout << sizeof(vComplexF) << std::endl;
+ std::cout<<GridLogMessage << sizeof(vComplexF) << std::endl;
  
  Grid_finalize();
 }
diff --git a/tests/Test_multishift_sqrt.cc b/tests/Test_multishift_sqrt.cc
index 4fff263f..111a521f 100644
--- a/tests/Test_multishift_sqrt.cc
+++ b/tests/Test_multishift_sqrt.cc
@@ -82,7 +82,7 @@ int main (int argc, char ** argv)
   ////////////////////////////////////////
   // sqrt and inverse sqrt
   ////////////////////////////////////////
-  std::cout << "Generating degree "<<degree<<" for x^(1/2)"<<std::endl;
+  std::cout<<GridLogMessage << "Generating degree "<<degree<<" for x^(1/2)"<<std::endl;
   remez.generateApprox(degree,1,2);
 
   MultiShiftFunction Sqrt(remez,1.0e-6,false);
@@ -120,19 +120,19 @@ int main (int argc, char ** argv)
 
   error = reference - combined;
 
-  std::cout << " Reference "<<norm2(reference)<<std::endl;
-  std::cout << " combined  "<<norm2(combined) <<std::endl;
-  std::cout << " error     "<<norm2(error)    <<std::endl;
+  std::cout<<GridLogMessage << " Reference "<<norm2(reference)<<std::endl;
+  std::cout<<GridLogMessage << " combined  "<<norm2(combined) <<std::endl;
+  std::cout<<GridLogMessage << " error     "<<norm2(error)    <<std::endl;
 
   MSCG(Diagonal,src,summed);
   error = summed - combined;
-  std::cout << " summed-combined "<<norm2(error)    <<std::endl;
+  std::cout<<GridLogMessage << " summed-combined "<<norm2(error)    <<std::endl;
 
 
   src=1.0;
   Chebyshev<LatticeFermion> Cheby(0.1,40.0,200,InverseApproximation);
 
-  std::cout<<"Chebuy approx vector "<<std::endl;
+  std::cout<<GridLogMessage<<"Chebuy approx vector "<<std::endl;
   Cheby(Diagonal,src,combined);
   std::ofstream of("cheby");
   Cheby.csv(of);
@@ -140,10 +140,10 @@ int main (int argc, char ** argv)
   Diagonal.ApplyInverse(src,reference);
   error = reference - combined;
 
-  std::cout << "Chebyshev inverse test "<<std::endl;
-  std::cout << " Reference "<<norm2(reference)<<std::endl;
-  std::cout << " combined  "<<norm2(combined) <<std::endl;
-  std::cout << " error     "<<norm2(error)    <<std::endl;
+  std::cout<<GridLogMessage << "Chebyshev inverse test "<<std::endl;
+  std::cout<<GridLogMessage << " Reference "<<norm2(reference)<<std::endl;
+  std::cout<<GridLogMessage << " combined  "<<norm2(combined) <<std::endl;
+  std::cout<<GridLogMessage << " error     "<<norm2(error)    <<std::endl;
 
   Grid_finalize();
 }
diff --git a/tests/Test_nersc_io.cc b/tests/Test_nersc_io.cc
index f4afb6d0..34821e97 100644
--- a/tests/Test_nersc_io.cc
+++ b/tests/Test_nersc_io.cc
@@ -55,7 +55,7 @@ int main (int argc, char ** argv)
 #endif
   double vol = Fine.gSites();
   Complex PlaqScale(1.0/vol/6.0/3.0);
-  std::cout <<"PlaqScale" << PlaqScale<<std::endl;
+  std::cout<<GridLogMessage <<"PlaqScale" << PlaqScale<<std::endl;
 
   std::vector<TComplex> Plaq_T(orthosz);
   sliceSum(Plaq,Plaq_T,Nd-1);
@@ -66,29 +66,29 @@ int main (int argc, char ** argv)
   for(int t=0;t<Nt;t++){
     Plaq_T_sum = Plaq_T_sum+Plaq_T[t];
     Complex Pt=TensorRemove(Plaq_T[t]);
-    std::cout << "sliced ["<<t<<"]" <<Pt*PlaqScale*Real(Nt)<<std::endl;
+    std::cout<<GridLogMessage << "sliced ["<<t<<"]" <<Pt*PlaqScale*Real(Nt)<<std::endl;
   }
 
   {
     Complex Pt = TensorRemove(Plaq_T_sum);
-    std::cout << "total " <<Pt*PlaqScale<<std::endl;
+    std::cout<<GridLogMessage << "total " <<Pt*PlaqScale<<std::endl;
   }  
 
 
   TComplex Tp = sum(Plaq);
   Complex p  = TensorRemove(Tp);
-  std::cout << "calculated plaquettes " <<p*PlaqScale<<std::endl;
+  std::cout<<GridLogMessage << "calculated plaquettes " <<p*PlaqScale<<std::endl;
 
 
   Complex LinkTraceScale(1.0/vol/4.0/3.0);
   TComplex Tl = sum(LinkTrace);
   Complex l  = TensorRemove(Tl);
-  std::cout << "calculated link trace " <<l*LinkTraceScale<<std::endl;
+  std::cout<<GridLogMessage << "calculated link trace " <<l*LinkTraceScale<<std::endl;
 
   blockSum(cPlaq,Plaq);
   TComplex TcP = sum(cPlaq);
   Complex ll= TensorRemove(TcP);
-  std::cout << "coarsened plaquettes sum to " <<ll*PlaqScale<<std::endl;
+  std::cout<<GridLogMessage << "coarsened plaquettes sum to " <<ll*PlaqScale<<std::endl;
   
   Grid_finalize();
 }
diff --git a/tests/Test_quenched_update.cc b/tests/Test_quenched_update.cc
index e419d002..b084c9c8 100644
--- a/tests/Test_quenched_update.cc
+++ b/tests/Test_quenched_update.cc
@@ -47,7 +47,7 @@ int main (int argc, char ** argv)
 
     RealD plaq = ColourWilsonLoops::avgPlaquette(Umu);
 
-    std::cout<<"sweep "<<sweep<<" PLAQUETTE "<<plaq<<std::endl;
+    std::cout<<GridLogMessage<<"sweep "<<sweep<<" PLAQUETTE "<<plaq<<std::endl;
 
     for( int cb=0;cb<2;cb++ ) {
 
@@ -55,7 +55,7 @@ int main (int argc, char ** argv)
       mask= zero;
       setCheckerboard(mask,one);
 
-      //      std::cout<<mask<<std::endl;
+      //      std::cout<<GridLogMessage<<mask<<std::endl;
       for(int mu=0;mu<Nd;mu++){
 	
 	// Get Link and Staple term in action; must contain Beta and 
diff --git a/tests/Test_remez.cc b/tests/Test_remez.cc
index 5a6300e7..2401d254 100644
--- a/tests/Test_remez.cc
+++ b/tests/Test_remez.cc
@@ -8,7 +8,7 @@ int main (int argc, char ** argv)
 {
   Grid_init(&argc,&argv);
 
-  std::cout << "Testing Remez"<<std::endl;
+  std::cout<<GridLogMessage << "Testing Remez"<<std::endl;
 
   double     lo=0.01;
   double     hi=1.0;
@@ -20,13 +20,13 @@ int main (int argc, char ** argv)
   // sqrt and inverse sqrt
   ////////////////////////////////////////
 
-  std::cout << "Generating degree "<<degree<<" for x^(1/2)"<<std::endl;
+  std::cout<<GridLogMessage << "Generating degree "<<degree<<" for x^(1/2)"<<std::endl;
   remez.generateApprox(degree,1,2);
   MultiShiftFunction Sqrt(remez,1.0,false);
   MultiShiftFunction InvSqrt(remez,1.0,true);
 
 
-  std::cout << "Generating degree "<<degree<<" for x^(1/4)"<<std::endl;
+  std::cout<<GridLogMessage << "Generating degree "<<degree<<" for x^(1/4)"<<std::endl;
   remez.generateApprox(degree,1,4);
   MultiShiftFunction SqrtSqrt(remez,1.0,false);
   MultiShiftFunction InvSqrtSqrt(remez,1.0,true);
@@ -49,10 +49,10 @@ int main (int argc, char ** argv)
   double aisx =InvSqrt.approx(x);
   double aissx=InvSqrtSqrt.approx(x);
 
-  std::cout << "x^(1/2) : "<<sx<<" "<<asx<<std::endl;
-  std::cout << "x^(1/4) : "<<ssx<<" "<<assx<<std::endl;
-  std::cout << "x^(-1/2): "<<isx<<" "<<aisx<<std::endl;
-  std::cout << "x^(-1/4): "<<issx<<" "<<aissx<<std::endl;
+  std::cout<<GridLogMessage << "x^(1/2) : "<<sx<<" "<<asx<<std::endl;
+  std::cout<<GridLogMessage << "x^(1/4) : "<<ssx<<" "<<assx<<std::endl;
+  std::cout<<GridLogMessage << "x^(-1/2): "<<isx<<" "<<aisx<<std::endl;
+  std::cout<<GridLogMessage << "x^(-1/4): "<<issx<<" "<<aissx<<std::endl;
   assert(fabs(sx-asx)<1.0e-6);
   assert(fabs(ssx-assx)<1.0e-6);
   assert(fabs(isx-aisx)<1.0e-6);
diff --git a/tests/Test_rng.cc b/tests/Test_rng.cc
index 9cd05a22..da8da734 100644
--- a/tests/Test_rng.cc
+++ b/tests/Test_rng.cc
@@ -24,27 +24,27 @@ int main (int argc, char ** argv)
 
   SpinMatrix rnd  ; 
   random(sRNG,rnd);
-  std::cout<<"Random Spin Matrix (random_device)\n"<< rnd<<std::endl;
+  std::cout<<GridLogMessage<<"Random Spin Matrix (random_device)\n"<< rnd<<std::endl;
 
   random(fsRNG,rnd);
-  std::cout<<"Random Spin Matrix (fixed seed)\n"<< rnd<<std::endl;
+  std::cout<<GridLogMessage<<"Random Spin Matrix (fixed seed)\n"<< rnd<<std::endl;
 
   SpinVector rv; 
   random(sRNG,rv);
-  std::cout<<"Random Spin Vector (random device)\n"<< rv<<std::endl;
+  std::cout<<GridLogMessage<<"Random Spin Vector (random device)\n"<< rv<<std::endl;
 
   random(fsRNG,rv);
-  std::cout<<"Random Spin Vector (fixed seed)\n"<< rv<<std::endl;
+  std::cout<<GridLogMessage<<"Random Spin Vector (fixed seed)\n"<< rv<<std::endl;
 
   gaussian(fsRNG,rv);
-  std::cout<<"Gaussian Spin Vector (fixed seed)\n"<< rv<<std::endl;
+  std::cout<<GridLogMessage<<"Gaussian Spin Vector (fixed seed)\n"<< rv<<std::endl;
 
   LatticeColourVector lcv(&Grid);
   random(pRNG,lcv);
-  std::cout<<"Random Lattice Colour Vector (random device)\n"<< lcv<<std::endl;
+  std::cout<<GridLogMessage<<"Random Lattice Colour Vector (random device)\n"<< lcv<<std::endl;
 
   random(fpRNG,lcv);
-  std::cout<<"Random Lattice Colour Vector (fixed seed)\n"<< lcv<<std::endl;
+  std::cout<<GridLogMessage<<"Random Lattice Colour Vector (fixed seed)\n"<< lcv<<std::endl;
 
   Grid_finalize();
 }
diff --git a/tests/Test_rng_fixed.cc b/tests/Test_rng_fixed.cc
index a4171401..f2271ca1 100644
--- a/tests/Test_rng_fixed.cc
+++ b/tests/Test_rng_fixed.cc
@@ -20,31 +20,31 @@ int main (int argc, char ** argv)
   GridParallelRNG          fpRNG(&Grid);  fpRNG.SeedFixedIntegers(seeds);
 
   vComplexF tmp; random(fsRNG,tmp);
-  std::cout<<"Random vComplexF (fixed seed)\n"<< tmp<<std::endl;
-  std::cout<<"conjugate(tmp)\n"<< conjugate(tmp)<<std::endl;
-  std::cout<<"conjugate(tmp)*tmp\n"<< conjugate(tmp)*tmp<<std::endl;
-  std::cout<<"innerProduct"<< innerProduct(tmp,tmp)<<std::endl;
-  std::cout<<"Reduce(innerProduct)"<< Reduce(innerProduct(tmp,tmp))<<std::endl;
+  std::cout<<GridLogMessage<<"Random vComplexF (fixed seed)\n"<< tmp<<std::endl;
+  std::cout<<GridLogMessage<<"conjugate(tmp)\n"<< conjugate(tmp)<<std::endl;
+  std::cout<<GridLogMessage<<"conjugate(tmp)*tmp\n"<< conjugate(tmp)*tmp<<std::endl;
+  std::cout<<GridLogMessage<<"innerProduct"<< innerProduct(tmp,tmp)<<std::endl;
+  std::cout<<GridLogMessage<<"Reduce(innerProduct)"<< Reduce(innerProduct(tmp,tmp))<<std::endl;
 
   SpinMatrix rnd  ; 
   random(fsRNG,rnd);
-  std::cout<<"Random Spin Matrix (fixed seed)\n"<< rnd<<std::endl;
+  std::cout<<GridLogMessage<<"Random Spin Matrix (fixed seed)\n"<< rnd<<std::endl;
 
   SpinVector rv; 
   random(fsRNG,rv);
-  std::cout<<"Random Spin Vector (fixed seed)\n"<< rv<<std::endl;
+  std::cout<<GridLogMessage<<"Random Spin Vector (fixed seed)\n"<< rv<<std::endl;
 
   gaussian(fsRNG,rv);
-  std::cout<<"Gaussian Spin Vector (fixed seed)\n"<< rv<<std::endl;
+  std::cout<<GridLogMessage<<"Gaussian Spin Vector (fixed seed)\n"<< rv<<std::endl;
 
   LatticeColourVector lcv(&Grid);
 
   LatticeFermion src(&Grid); random(fpRNG,src);
-  std::cout << "src norm : " << norm2(src)<<std::endl;
-  std::cout << "src " << src<<std::endl;
+  std::cout<<GridLogMessage << "src norm : " << norm2(src)<<std::endl;
+  std::cout<<GridLogMessage << "src " << src<<std::endl;
 
   random(fpRNG,lcv);
-  std::cout<<"Random Lattice Colour Vector (fixed seed)\n"<< lcv<<std::endl;
+  std::cout<<GridLogMessage<<"Random Lattice Colour Vector (fixed seed)\n"<< lcv<<std::endl;
 
 
   Grid_finalize();
diff --git a/tests/Test_simd.cc b/tests/Test_simd.cc
index ee50a312..004f2f52 100644
--- a/tests/Test_simd.cc
+++ b/tests/Test_simd.cc
@@ -113,18 +113,18 @@ void Tester(const functor &func)
   }
 
   extract<vec,scal>(v_result,result);
-  std::cout << " " << func.name()<<std::endl;
+  std::cout<<GridLogMessage << " " << func.name()<<std::endl;
 
   int ok=0;
   for(int i=0;i<Nsimd;i++){
     if ( abs(reference[i]-result[i])>0){
-      std::cout<< "*****" << std::endl;
-      std::cout<< "["<<i<<"] "<< abs(reference[i]-result[i]) << " " <<reference[i]<< " " << result[i]<<std::endl;
+      std::cout<<GridLogMessage<< "*****" << std::endl;
+      std::cout<<GridLogMessage<< "["<<i<<"] "<< abs(reference[i]-result[i]) << " " <<reference[i]<< " " << result[i]<<std::endl;
       ok++;
     }
   }
   if ( ok==0 ) {
-    std::cout << " OK!" <<std::endl;
+    std::cout<<GridLogMessage << " OK!" <<std::endl;
   }
   assert(ok==0);
 }
@@ -164,16 +164,16 @@ void ReductionTester(const functor &func)
     reference+=tmp;
   }
 
-  std::cout << " " << func.name()<<std::endl;
+  std::cout<<GridLogMessage << " " << func.name()<<std::endl;
 
   int ok=0;
   if ( abs(reference-result)/abs(reference) > 1.0e-6 ){ // rounding is possible for reduce order
-    std::cout<< "*****" << std::endl;
-    std::cout<< abs(reference-result) << " " <<reference<< " " << result<<std::endl;
+    std::cout<<GridLogMessage<< "*****" << std::endl;
+    std::cout<<GridLogMessage<< abs(reference-result) << " " <<reference<< " " << result<<std::endl;
     ok++;
   }
   if ( ok==0 ) {
-    std::cout << " OK!" <<std::endl;
+    std::cout<<GridLogMessage << " OK!" <<std::endl;
   }
   assert(ok==0);
 }
@@ -233,9 +233,9 @@ int main (int argc, char ** argv)
   Tester<ComplexF,vComplexF>(funcInnerProduct());
   ReductionTester<ComplexF,ComplexF,vComplexF>(funcReduce());
 
-  std::cout << "==================================="<<  std::endl;
-  std::cout << "Testing vComplexD "<<std::endl;
-  std::cout << "==================================="<<  std::endl;
+  std::cout<<GridLogMessage << "==================================="<<  std::endl;
+  std::cout<<GridLogMessage << "Testing vComplexD "<<std::endl;
+  std::cout<<GridLogMessage << "==================================="<<  std::endl;
 
 
   Tester<ComplexD,vComplexD>(funcTimesI());
@@ -248,6 +248,5 @@ int main (int argc, char ** argv)
   Tester<ComplexD,vComplexD>(funcInnerProduct());
   ReductionTester<ComplexD,ComplexD,vComplexD>(funcReduce());
 
-
   Grid_finalize();
 }
diff --git a/tests/Test_stencil.cc b/tests/Test_stencil.cc
index 3f6bb4fe..d455c1c1 100644
--- a/tests/Test_stencil.cc
+++ b/tests/Test_stencil.cc
@@ -34,7 +34,7 @@ int main (int argc, char ** argv)
     for(int dir=0;dir<4;dir++){
       for(int disp=0;disp<Fine._fdimensions[dir];disp++){
 
-	std::cout << "Using stencil to shift dim "<<dir<< " by "<<disp<<std::endl;
+	std::cout<<GridLogMessage << "Using stencil to shift dim "<<dir<< " by "<<disp<<std::endl;
 	// start to test the Cartesian npoint stencil infrastructure
 	int npoint=1;
 	std::vector<int> directions(npoint,dir);
@@ -73,7 +73,7 @@ int main (int argc, char ** argv)
 	Real nrmB = norm2(Bar);
 	Diff = Check-Bar;
 	Real nrm  = norm2(Diff);
-	std::cout<<"N2diff ="<<nrm<<" "<<nrmC<<" " <<nrmB<<std::endl;
+	std::cout<<GridLogMessage<<"N2diff ="<<nrm<<" "<<nrmC<<" " <<nrmB<<std::endl;
 
 	Real snrmC =0;
 	Real snrmB =0;
@@ -110,7 +110,7 @@ int main (int argc, char ** argv)
 	 
 	}}}}
 
-	std::cout<<"scalar N2diff = "<<snrm<<" " <<snrmC<<" "<<snrmB<<std::endl;
+	std::cout<<GridLogMessage<<"scalar N2diff = "<<snrm<<" " <<snrmC<<" "<<snrmB<<std::endl;
 
 
       }
diff --git a/tests/Test_wilson_even_odd.cc b/tests/Test_wilson_even_odd.cc
index 90d15e99..66ac89dc 100644
--- a/tests/Test_wilson_even_odd.cc
+++ b/tests/Test_wilson_even_odd.cc
@@ -27,7 +27,7 @@ int main (int argc, char ** argv)
   GridRedBlackCartesian     RBGrid(latt_size,simd_layout,mpi_layout);
 
   int threads = GridThread::GetThreads();
-  std::cout << "Grid is setup to use "<<threads<<" threads"<<std::endl;
+  std::cout<<GridLogMessage << "Grid is setup to use "<<threads<<" threads"<<std::endl;
 
   std::vector<int> seeds({1,2,3,4});
 
@@ -55,9 +55,9 @@ int main (int argc, char ** argv)
   Umu=zero;
   for(int nn=0;nn<Nd;nn++){
     random(pRNG,U[nn]);
-    std::cout<<"U[nn]"<<norm2(U[nn])<<std::endl;
+    std::cout<<GridLogMessage<<"U[nn]"<<norm2(U[nn])<<std::endl;
     PokeIndex<LorentzIndex>(Umu,U[nn],nn);
-    std::cout<<"Umu"<<norm2(Umu)<<std::endl;
+    std::cout<<GridLogMessage<<"Umu"<<norm2(Umu)<<std::endl;
   }
 
   RealD mass=0.1;
@@ -70,30 +70,30 @@ int main (int argc, char ** argv)
   LatticeFermion r_o   (&RBGrid);
   LatticeFermion r_eo  (&Grid);
 
-  std::cout<<"=========================================================="<<std::endl;
-  std::cout<<"= Testing that Deo + Doe = Dunprec "<<std::endl;
-  std::cout<<"=========================================================="<<std::endl;
+  std::cout<<GridLogMessage<<"=========================================================="<<std::endl;
+  std::cout<<GridLogMessage<<"= Testing that Deo + Doe = Dunprec "<<std::endl;
+  std::cout<<GridLogMessage<<"=========================================================="<<std::endl;
 
   pickCheckerboard(Even,src_e,src);
   pickCheckerboard(Odd,src_o,src);
 
-  Dw.Meooe(src_e,r_o);  std::cout<<"Applied Meo"<<std::endl;
-  Dw.Meooe(src_o,r_e);  std::cout<<"Applied Moe"<<std::endl;
+  Dw.Meooe(src_e,r_o);  std::cout<<GridLogMessage<<"Applied Meo"<<std::endl;
+  Dw.Meooe(src_o,r_e);  std::cout<<GridLogMessage<<"Applied Moe"<<std::endl;
   Dw.Dhop (src,ref,DaggerNo);
 
   setCheckerboard(r_eo,r_o);
   setCheckerboard(r_eo,r_e);
 
   err= ref - r_eo;
-  std::cout << "EO norm diff   "<< norm2(err)<< " "<<norm2(ref)<< " " << norm2(r_eo) <<std::endl;
+  std::cout<<GridLogMessage << "EO norm diff   "<< norm2(err)<< " "<<norm2(ref)<< " " << norm2(r_eo) <<std::endl;
 
   LatticeComplex cerr(&Grid);
   cerr = localInnerProduct(err,err);
 
-  std::cout<<"=============================================================="<<std::endl;
-  std::cout<<"= Test Ddagger is the dagger of D by requiring                "<<std::endl;
-  std::cout<<"=  < phi | Deo | chi > * = < chi | Deo^dag| phi>  "<<std::endl;
-  std::cout<<"=============================================================="<<std::endl;
+  std::cout<<GridLogMessage<<"=============================================================="<<std::endl;
+  std::cout<<GridLogMessage<<"= Test Ddagger is the dagger of D by requiring                "<<std::endl;
+  std::cout<<GridLogMessage<<"=  < phi | Deo | chi > * = < chi | Deo^dag| phi>  "<<std::endl;
+  std::cout<<GridLogMessage<<"=============================================================="<<std::endl;
   
   LatticeFermion chi_e   (&RBGrid);
   LatticeFermion chi_o   (&RBGrid);
@@ -123,15 +123,15 @@ int main (int argc, char ** argv)
   ComplexD cDpe = innerProduct(chi_e,dphi_e);
   ComplexD cDpo = innerProduct(chi_o,dphi_o);
 
-  std::cout <<"e "<<pDce<<" "<<cDpe <<std::endl;
-  std::cout <<"o "<<pDco<<" "<<cDpo <<std::endl;
+  std::cout<<GridLogMessage <<"e "<<pDce<<" "<<cDpe <<std::endl;
+  std::cout<<GridLogMessage <<"o "<<pDco<<" "<<cDpo <<std::endl;
 
-  std::cout <<"pDce - conj(cDpo) "<< pDce-conj(cDpo) <<std::endl;
-  std::cout <<"pDco - conj(cDpe) "<< pDco-conj(cDpe) <<std::endl;
+  std::cout<<GridLogMessage <<"pDce - conj(cDpo) "<< pDce-conj(cDpo) <<std::endl;
+  std::cout<<GridLogMessage <<"pDco - conj(cDpe) "<< pDco-conj(cDpe) <<std::endl;
 
-  std::cout<<"=============================================================="<<std::endl;
-  std::cout<<"= Test MeeInv Mee = 1                                         "<<std::endl;
-  std::cout<<"=============================================================="<<std::endl;
+  std::cout<<GridLogMessage<<"=============================================================="<<std::endl;
+  std::cout<<GridLogMessage<<"= Test MeeInv Mee = 1                                         "<<std::endl;
+  std::cout<<GridLogMessage<<"=============================================================="<<std::endl;
 
   pickCheckerboard(Even,chi_e,chi);
   pickCheckerboard(Odd ,chi_o,chi);
@@ -146,11 +146,11 @@ int main (int argc, char ** argv)
   setCheckerboard(phi,phi_o);
 
   err = phi-chi;
-  std::cout << "norm diff   "<< norm2(err)<< std::endl;
+  std::cout<<GridLogMessage << "norm diff   "<< norm2(err)<< std::endl;
 
-  std::cout<<"=============================================================="<<std::endl;
-  std::cout<<"= Test MeeInvDag MeeDag = 1                                   "<<std::endl;
-  std::cout<<"=============================================================="<<std::endl;
+  std::cout<<GridLogMessage<<"=============================================================="<<std::endl;
+  std::cout<<GridLogMessage<<"= Test MeeInvDag MeeDag = 1                                   "<<std::endl;
+  std::cout<<GridLogMessage<<"=============================================================="<<std::endl;
 
   pickCheckerboard(Even,chi_e,chi);
   pickCheckerboard(Odd ,chi_o,chi);
@@ -165,11 +165,11 @@ int main (int argc, char ** argv)
   setCheckerboard(phi,phi_o);
 
   err = phi-chi;
-  std::cout << "norm diff   "<< norm2(err)<< std::endl;
+  std::cout<<GridLogMessage << "norm diff   "<< norm2(err)<< std::endl;
 
-  std::cout<<"=============================================================="<<std::endl;
-  std::cout<<"= Test MpcDagMpc is Hermitian              "<<std::endl;
-  std::cout<<"=============================================================="<<std::endl;
+  std::cout<<GridLogMessage<<"=============================================================="<<std::endl;
+  std::cout<<GridLogMessage<<"= Test MpcDagMpc is Hermitian              "<<std::endl;
+  std::cout<<GridLogMessage<<"=============================================================="<<std::endl;
   
   random(pRNG,phi);
   random(pRNG,chi);
@@ -191,11 +191,11 @@ int main (int argc, char ** argv)
   cDpe = innerProduct(chi_e,dphi_e);
   cDpo = innerProduct(chi_o,dphi_o);
 
-  std::cout <<"e "<<pDce<<" "<<cDpe <<std::endl;
-  std::cout <<"o "<<pDco<<" "<<cDpo <<std::endl;
+  std::cout<<GridLogMessage <<"e "<<pDce<<" "<<cDpe <<std::endl;
+  std::cout<<GridLogMessage <<"o "<<pDco<<" "<<cDpo <<std::endl;
 
-  std::cout <<"pDce - conj(cDpo) "<< pDco-conj(cDpo) <<std::endl;
-  std::cout <<"pDco - conj(cDpe) "<< pDce-conj(cDpe) <<std::endl;
+  std::cout<<GridLogMessage <<"pDce - conj(cDpo) "<< pDco-conj(cDpo) <<std::endl;
+  std::cout<<GridLogMessage <<"pDco - conj(cDpe) "<< pDce-conj(cDpe) <<std::endl;
   
   Grid_finalize();
 }
diff --git a/tests/Test_wilson_force.cc b/tests/Test_wilson_force.cc
new file mode 100644
index 00000000..b08ccd45
--- /dev/null
+++ b/tests/Test_wilson_force.cc
@@ -0,0 +1,171 @@
+#include <Grid.h>
+
+using namespace std;
+using namespace Grid;
+using namespace Grid::QCD;
+
+#define parallel_for PARALLEL_FOR_LOOP for
+
+int main (int argc, char ** argv)
+{
+  Grid_init(&argc,&argv);
+
+  std::vector<int> latt_size   = GridDefaultLatt();
+  std::vector<int> simd_layout = GridDefaultSimd(Nd,vComplex::Nsimd());
+  std::vector<int> mpi_layout  = GridDefaultMpi();
+
+  GridCartesian               Grid(latt_size,simd_layout,mpi_layout);
+  GridRedBlackCartesian     RBGrid(latt_size,simd_layout,mpi_layout);
+
+  int threads = GridThread::GetThreads();
+  std::cout<<GridLogMessage << "Grid is setup to use "<<threads<<" threads"<<std::endl;
+
+  std::vector<int> seeds({1,2,3,4});
+
+  GridParallelRNG          pRNG(&Grid);
+  pRNG.SeedRandomDevice();
+
+  LatticeFermion phi        (&Grid); gaussian(pRNG,phi);
+  LatticeFermion Mphi       (&Grid); 
+  LatticeFermion MphiPrime  (&Grid); 
+
+  LatticeGaugeField U(&Grid);
+
+  SU3::HotConfiguration(pRNG,U);
+  //  SU3::ColdConfiguration(pRNG,U);
+  
+  ////////////////////////////////////
+  // Unmodified matrix element
+  ////////////////////////////////////
+  RealD mass=-4.0; //kills the diagonal term
+  WilsonFermion Dw     (U,     Grid,RBGrid,mass);
+  Dw.M   (phi,Mphi);
+
+  ComplexD S    = innerProduct(Mphi,Mphi); // pdag MdagM p
+
+  // get the deriv of phidag MdagM phi with respect to "U"
+  LatticeGaugeField UdSdU(&Grid);
+  LatticeGaugeField tmp(&Grid);
+
+  Dw.MDeriv(tmp , Mphi,  phi,DaggerNo );  UdSdU=tmp;
+  Dw.MDeriv(tmp , phi,  Mphi,DaggerYes ); UdSdU=(UdSdU+tmp);
+  
+  LatticeFermion Ftmp      (&Grid);
+
+  ////////////////////////////////////
+  // Modify the gauge field a little 
+  ////////////////////////////////////
+  RealD dt = 0.0001;
+  RealD Hmom = 0.0;
+  RealD Hmomprime = 0.0;
+  RealD Hmompp    = 0.0;
+  LatticeColourMatrix mommu(&Grid); 
+  LatticeColourMatrix forcemu(&Grid); 
+  LatticeGaugeField mom(&Grid); 
+  LatticeGaugeField Uprime(&Grid); 
+
+  for(int mu=0;mu<Nd;mu++){
+
+    SU3::GaussianLieAlgebraMatrix(pRNG, mommu); // Traceless antihermitian momentum; gaussian in lie alg
+
+    Hmom -= real(sum(trace(mommu*mommu)));
+
+    PokeIndex<LorentzIndex>(mom,mommu,mu);
+
+    // fourth order exponential approx
+    parallel_for(auto i=mom.begin();i<mom.end();i++){
+      Uprime[i](mu) =
+	  U[i](mu)
+	+ mom[i](mu)*U[i](mu)*dt 
+	+ mom[i](mu) *mom[i](mu) *U[i](mu)*(dt*dt/2.0)
+	+ mom[i](mu) *mom[i](mu) *mom[i](mu) *U[i](mu)*(dt*dt*dt/6.0)
+	+ mom[i](mu) *mom[i](mu) *mom[i](mu) *mom[i](mu) *U[i](mu)*(dt*dt*dt*dt/24.0)
+	+ mom[i](mu) *mom[i](mu) *mom[i](mu) *mom[i](mu) *mom[i](mu) *U[i](mu)*(dt*dt*dt*dt*dt/120.0)
+	+ mom[i](mu) *mom[i](mu) *mom[i](mu) *mom[i](mu) *mom[i](mu) *mom[i](mu) *U[i](mu)*(dt*dt*dt*dt*dt*dt/720.0)
+	;
+    }
+
+  }
+
+  std::cout << GridLogMessage <<"Initial mom hamiltonian is "<< Hmom <<std::endl;
+  Dw.DoubleStore(Dw.Umu,Uprime);
+  Dw.M          (phi,MphiPrime);
+
+  ComplexD Sprime    = innerProduct(MphiPrime   ,MphiPrime);
+
+  //////////////////////////////////////////////
+  // Use derivative to estimate dS
+  //////////////////////////////////////////////
+
+
+  for(int mu=0;mu<Nd;mu++){
+    std::cout << "" <<std::endl;
+    mommu   = PeekIndex<LorentzIndex>(mom,mu);
+    std::cout << GridLogMessage<< " Mommu  " << norm2(mommu)<<std::endl;
+    mommu   = mommu+adj(mommu);
+    std::cout << GridLogMessage<< " Mommu + Mommudag " << norm2(mommu)<<std::endl;
+    mommu   = PeekIndex<LorentzIndex>(UdSdU,mu);
+    std::cout << GridLogMessage<< " dsdumu  " << norm2(mommu)<<std::endl;
+    mommu   = mommu+adj(mommu);
+    std::cout << GridLogMessage<< " dsdumu + dag  " << norm2(mommu)<<std::endl;
+  }
+
+  LatticeComplex dS(&Grid); dS = zero;
+  LatticeComplex dSmom(&Grid); dSmom = zero;
+  LatticeComplex dSmom2(&Grid); dSmom2 = zero;
+  for(int mu=0;mu<Nd;mu++){
+    mommu   = PeekIndex<LorentzIndex>(UdSdU,mu);
+    mommu=Ta(mommu)*2.0;
+    PokeIndex<LorentzIndex>(UdSdU,mommu,mu);
+  }
+
+  for(int mu=0;mu<Nd;mu++){
+    mommu   = PeekIndex<LorentzIndex>(mom,mu);
+    std::cout << GridLogMessage<< " Mommu  " << norm2(mommu)<<std::endl;
+    mommu   = mommu+adj(mommu);
+    std::cout << GridLogMessage<< " Mommu + Mommudag " << norm2(mommu)<<std::endl;
+    mommu   = PeekIndex<LorentzIndex>(UdSdU,mu);
+    std::cout << GridLogMessage<< " dsdumu  " << norm2(mommu)<<std::endl;
+    mommu   = mommu+adj(mommu);
+    std::cout << GridLogMessage<< " dsdumu + dag  " << norm2(mommu)<<std::endl;
+  }
+
+  for(int mu=0;mu<Nd;mu++){
+    forcemu = PeekIndex<LorentzIndex>(UdSdU,mu);
+    mommu   = PeekIndex<LorentzIndex>(mom,mu);
+
+    // Update PF action density
+    dS = dS+trace(mommu*forcemu)*dt;
+
+    dSmom  = dSmom  - trace(mommu*forcemu) * dt;
+    dSmom2 = dSmom2 - trace(forcemu*forcemu) *(0.25* dt*dt);
+
+    // Update mom action density
+    mommu = mommu + forcemu*(dt*0.5);
+
+    Hmomprime -= real(sum(trace(mommu*mommu)));
+
+  }
+
+  Complex dSpred    = sum(dS);
+  Complex dSm       = sum(dSmom);
+  Complex dSm2      = sum(dSmom2);
+
+
+  std::cout << GridLogMessage <<"Initial mom hamiltonian is "<< Hmom <<std::endl;
+  std::cout << GridLogMessage <<"Final   mom hamiltonian is "<< Hmomprime <<std::endl;
+  std::cout << GridLogMessage <<"Delta   mom hamiltonian is "<< Hmomprime-Hmom <<std::endl;
+
+  std::cout << GridLogMessage << " S      "<<S<<std::endl;
+  std::cout << GridLogMessage << " Sprime "<<Sprime<<std::endl;
+  std::cout << GridLogMessage << "dS      "<<Sprime-S<<std::endl;
+  std::cout << GridLogMessage << "predict dS    "<< dSpred <<std::endl;
+  std::cout << GridLogMessage <<"dSm "<< dSm<<std::endl;
+  std::cout << GridLogMessage <<"dSm2"<< dSm2<<std::endl;
+
+  std::cout << GridLogMessage << "Total dS    "<< Hmomprime - Hmom + Sprime - S <<std::endl;
+
+
+  std::cout<< GridLogMessage << "Done" <<std::endl;
+  Grid_finalize();
+}
diff --git a/tests/Test_wilson_force_phiMdagMphi.cc b/tests/Test_wilson_force_phiMdagMphi.cc
new file mode 100644
index 00000000..33c5f6e1
--- /dev/null
+++ b/tests/Test_wilson_force_phiMdagMphi.cc
@@ -0,0 +1,140 @@
+#include <Grid.h>
+
+using namespace std;
+using namespace Grid;
+using namespace Grid::QCD;
+
+#define parallel_for PARALLEL_FOR_LOOP for
+
+int main (int argc, char ** argv)
+{
+  Grid_init(&argc,&argv);
+
+  std::vector<int> latt_size   = GridDefaultLatt();
+  std::vector<int> simd_layout = GridDefaultSimd(Nd,vComplex::Nsimd());
+  std::vector<int> mpi_layout  = GridDefaultMpi();
+
+  GridCartesian               Grid(latt_size,simd_layout,mpi_layout);
+  GridRedBlackCartesian     RBGrid(latt_size,simd_layout,mpi_layout);
+
+  int threads = GridThread::GetThreads();
+  std::cout<<GridLogMessage << "Grid is setup to use "<<threads<<" threads"<<std::endl;
+
+  std::vector<int> seeds({1,2,3,4});
+
+  GridParallelRNG          pRNG(&Grid);
+  pRNG.SeedRandomDevice();
+
+  LatticeFermion phi        (&Grid); gaussian(pRNG,phi);
+  LatticeFermion Mphi       (&Grid); 
+  LatticeFermion Mdagphi       (&Grid); 
+  LatticeFermion MphiPrime  (&Grid); 
+  LatticeFermion MdagphiPrime  (&Grid); 
+  LatticeFermion dMphi      (&Grid); 
+
+  LatticeGaugeField U(&Grid);
+
+
+  SU3::HotConfiguration(pRNG,U);
+  //  SU3::ColdConfiguration(pRNG,U);
+  
+  ////////////////////////////////////
+  // Unmodified matrix element
+  ////////////////////////////////////
+  RealD mass=-4.0; //kills the diagonal term
+  WilsonFermion Dw     (U,     Grid,RBGrid,mass);
+  Dw.M   (phi,Mphi);
+  Dw.Mdag(phi,Mdagphi);
+
+  ComplexD S    = innerProduct(Mphi,Mphi); // pdag MdagM p
+  ComplexD Sdag = innerProduct(Mdagphi,Mdagphi); // pdag MMdag p
+
+  // get the deriv of phidag MdagM phi with respect to "U"
+  LatticeGaugeField UdSdU(&Grid);
+  LatticeGaugeField UdSdUdag(&Grid);
+  LatticeGaugeField tmp(&Grid);
+
+  Dw.MDeriv(tmp , Mphi,  phi,DaggerNo );  UdSdU=tmp;
+
+  Dw.MDeriv(tmp , Mdagphi,  phi,DaggerYes );  UdSdUdag=tmp;
+
+
+  LatticeFermion dMdagphi      (&Grid);  dMdagphi=zero;
+  LatticeFermion Ftmp      (&Grid);
+
+
+  //  Dw.MDeriv(UdSdU,Mdagphi,  phi,DaggerYes );// UdSdU =UdSdU +tmp;
+
+  ////////////////////////////////////
+  // Modify the gauge field a little in one dir
+  ////////////////////////////////////
+  RealD dt = 1.0e-3;
+
+  LatticeColourMatrix mommu(&Grid); 
+  LatticeGaugeField mom(&Grid); 
+  LatticeGaugeField Uprime(&Grid); 
+
+  for(int mu=0;mu<Nd;mu++){
+
+    SU3::GaussianLieAlgebraMatrix(pRNG, mommu); // Traceless antihermitian momentum; gaussian in lie alg
+
+    //    Dw.DoubleStore(Dw.Umu,Uprime); // update U _and_ Udag
+    Dw.DhopDirDisp(phi,Ftmp,mu,mu+4,DaggerYes); 
+    dMdagphi=dMdagphi+mommu*Ftmp*dt;
+    
+    PokeIndex<LorentzIndex>(mom,mommu,mu);
+
+    parallel_for(auto i=mom.begin();i<mom.end();i++){
+      Uprime[i](mu) =U[i](mu)+ mom[i](mu)*U[i](mu)*dt;
+      Dw.Umu[i](mu) =Uprime[i](mu); // update U but _not_ Udag
+    }
+
+  }
+
+  Dw.Mdag(phi,MdagphiPrime);
+  Dw.M   (phi,MphiPrime);
+
+  std::cout << GridLogMessage << "deltaMdag phi    "<< norm2(dMdagphi) <<std::endl;
+  Ftmp=MdagphiPrime - Mdagphi;
+  std::cout << GridLogMessage << "diff Mdag phi    "<< norm2(Ftmp) <<std::endl;
+  Ftmp = Ftmp - dMdagphi;
+  std::cout << GridLogMessage << "err  Mdag phi    "<< norm2(Ftmp) <<std::endl;
+  std::cout << dMdagphi<<std::endl;
+  Ftmp=MdagphiPrime - Mdagphi;
+  std::cout << Ftmp<<std::endl;
+
+
+  ComplexD Sprime    = innerProduct(Mphi   ,MphiPrime);
+  ComplexD Sprimedag = innerProduct(Mdagphi,MdagphiPrime);
+
+  ComplexD deltaSdag = innerProduct(Mdagphi,dMdagphi);
+  std::cout << GridLogMessage << "deltaSdag from inner prod of mom* M[u]     "<<deltaSdag<<std::endl;
+
+  //////////////////////////////////////////////
+  // Use derivative to estimate dS
+  //////////////////////////////////////////////
+  LatticeComplex dS(&Grid); dS = zero;
+  LatticeComplex dSdag(&Grid); dSdag = zero;
+  parallel_for(auto i=mom.begin();i<mom.end();i++){
+    for(int mu=0;mu<Nd;mu++){
+      //      dS[i]() = dS[i]()+trace(mom[i](mu) * UdSdU[i](mu) - mom[i](mu)* adj( UdSdU[i](mu)) )*dt;
+      dS[i]()    =    dS[i]()+trace(mom[i](mu) * UdSdU[i](mu) )*dt;
+      dSdag[i]() = dSdag[i]()+trace(mom[i](mu) * UdSdUdag[i](mu) )*dt;
+    }
+  }
+  Complex dSpred    = sum(dS);
+  Complex dSdagpred = sum(dSdag);
+
+  std::cout << GridLogMessage << " S      "<<S<<std::endl;
+  std::cout << GridLogMessage << " Sprime "<<Sprime<<std::endl;
+  std::cout << GridLogMessage << "dS      "<<Sprime-S<<std::endl;
+  std::cout << GridLogMessage << "predict dS    "<< dSpred <<std::endl;
+  std::cout << "\n\n"<<std::endl;
+  std::cout << GridLogMessage << " Sdag      "<<Sdag<<std::endl;
+  std::cout << GridLogMessage << " Sprimedag "<<Sprimedag<<std::endl;
+  std::cout << GridLogMessage << "dSdag      "<<Sprimedag-Sdag<<std::endl;
+  std::cout << GridLogMessage << "predict dSdag    "<< dSdagpred <<std::endl;
+
+  std::cout<< GridLogMessage << "Done" <<std::endl;
+  Grid_finalize();
+}
diff --git a/tests/Test_wilson_force_phiMphi.cc b/tests/Test_wilson_force_phiMphi.cc
new file mode 100644
index 00000000..f1e42195
--- /dev/null
+++ b/tests/Test_wilson_force_phiMphi.cc
@@ -0,0 +1,162 @@
+#include <Grid.h>
+
+using namespace std;
+using namespace Grid;
+using namespace Grid::QCD;
+
+#define parallel_for PARALLEL_FOR_LOOP for
+
+int main (int argc, char ** argv)
+{
+  Grid_init(&argc,&argv);
+
+  std::vector<int> latt_size   = GridDefaultLatt();
+  std::vector<int> simd_layout = GridDefaultSimd(Nd,vComplex::Nsimd());
+  std::vector<int> mpi_layout  = GridDefaultMpi();
+
+  GridCartesian               Grid(latt_size,simd_layout,mpi_layout);
+  GridRedBlackCartesian     RBGrid(latt_size,simd_layout,mpi_layout);
+
+  int threads = GridThread::GetThreads();
+  std::cout<<GridLogMessage << "Grid is setup to use "<<threads<<" threads"<<std::endl;
+
+  std::vector<int> seeds({1,2,3,4});
+
+  GridParallelRNG          pRNG(&Grid);
+  pRNG.SeedRandomDevice();
+
+  LatticeFermion phi        (&Grid); gaussian(pRNG,phi);
+  LatticeFermion Mphi       (&Grid); 
+  LatticeFermion MphiPrime  (&Grid); 
+  LatticeFermion dMphi      (&Grid); 
+
+  LatticeGaugeField U(&Grid);
+
+
+  SU3::HotConfiguration(pRNG,U);
+
+  ////////////////////////////////////
+  // Unmodified matrix element
+  ////////////////////////////////////
+  RealD mass=-4.0; //kills the diagonal term
+  WilsonFermion Dw     (U,     Grid,RBGrid,mass);
+  Dw.M(phi,Mphi);
+
+  ComplexD S = innerProduct(phi,Mphi);
+
+  // get the deriv
+  LatticeGaugeField UdSdU(&Grid);
+  Dw.MDeriv(UdSdU,phi, phi,DaggerNo ); 
+
+
+  ////////////////////////////////////
+  // Modify the gauge field a little in one dir
+  ////////////////////////////////////
+  RealD dt = 1.0e-3;
+  Complex Complex_i(0,1);
+
+  LatticeColourMatrix Umu(&Grid);
+  LatticeColourMatrix Umu_save(&Grid);
+  LatticeColourMatrix dU (&Grid);
+  LatticeColourMatrix mom(&Grid); 
+  SU3::GaussianLieAlgebraMatrix(pRNG, mom); // Traceless antihermitian momentum; gaussian in lie alg
+
+
+  // check mom is as i expect
+  LatticeColourMatrix tmpmom(&Grid); 
+  tmpmom = mom+adj(mom);
+  std::cout << GridLogMessage << "mom anti-herm check "<< norm2(tmpmom)<<std::endl;
+  std::cout << GridLogMessage << "mom tr check "<< norm2(trace(mom))<<std::endl;
+  
+  const int mu=0;
+  Umu = PeekIndex<LorentzIndex>(U,mu);
+  Umu_save=Umu;
+  dU = mom * Umu * dt;
+  Umu= Umu+dU;
+  PokeIndex<LorentzIndex>(Dw.Umu,Umu,mu);
+
+  Dw.M(phi,MphiPrime);
+
+  ComplexD Sprime = innerProduct(phi,MphiPrime);
+
+  std::cout << GridLogMessage << " S      "<<S<<std::endl;
+  std::cout << GridLogMessage << " Sprime "<<Sprime<<std::endl;
+  std::cout << GridLogMessage << "dS      "<<Sprime-S<<std::endl;
+
+  Dw.Umu=zero;
+  PokeIndex<LorentzIndex>(Dw.Umu,dU,mu);
+  Dw.M(phi,dMphi);
+
+
+  ComplexD deltaS = innerProduct(phi,dMphi);
+  std::cout << GridLogMessage << "deltaS      "<<deltaS<<std::endl;
+
+  Dw.Umu=zero;
+  PokeIndex<LorentzIndex>(Dw.Umu,Umu_save,mu);
+  Dw.Mdir(phi,dMphi,mu,1);
+  dMphi = dt*mom*dMphi;
+
+  deltaS = innerProduct(phi,dMphi);
+  std::cout << GridLogMessage << "deltaS from inner prod of mom* M[u]     "<<deltaS<<std::endl;
+
+  deltaS = sum(trace(outerProduct(dMphi,phi)));
+  std::cout << GridLogMessage << "deltaS from trace outer prod of deltaM      "<<deltaS<<std::endl;
+
+/*
+  LatticeComplex lip(&Grid);
+  lip  = localInnerProduct(phi,dMphi);
+  
+  LatticeComplex trop(&Grid);
+  trop = trace(outerProduct(dMphi,phi));
+
+  LatticeSpinColourMatrix op(&Grid);
+  op = outerProduct(dMphi,phi);
+
+  LatticeSpinColourMatrix hop(&Grid);
+  LatticeComplex op_cpt(&Grid);
+  for(int s1=0;s1<Ns;s1++){
+  for(int s2=0;s2<Ns;s2++){
+  for(int c1=0;c1<Nc;c1++){
+  for(int c2=0;c2<Nc;c2++){
+
+    op_cpt = peekColour(peekSpin(dMphi,s1),c1) * adj(peekColour(peekSpin(phi,s2),c2));
+
+    parallel_for(auto i=hop.begin();i<hop.end();i++){
+      hop[i]()(s1,s2)(c1,c2) = op_cpt[i]()()();
+    }
+    
+  }}}}
+
+  LatticeSpinColourMatrix diffop(&Grid);
+
+  diffop = hop - op;
+  std::cout << GridLogMessage << "hand outer prod diff   "<<norm2(diffop)<<std::endl;
+
+  deltaS = sum(trace(hop));
+  std::cout << GridLogMessage << "deltaS hop   "<<deltaS<<std::endl;
+
+  std::cout << GridLogMessage<< "  phi[0] : "<<  phi._odata[0]<<std::endl;
+  std::cout << GridLogMessage<< "dMphi[0] : "<<dMphi._odata[0]<<std::endl;
+  std::cout << GridLogMessage<< "hop[0]   : "<<  hop._odata[0]<<std::endl;
+  std::cout << GridLogMessage<< " op[0]   : "<<   op._odata[0]<<std::endl;
+
+
+  std::cout << GridLogMessage << "lip      "<<lip<<std::endl;
+  std::cout << GridLogMessage << "trop     "<<trop<<std::endl;
+
+*/  
+  
+  //  std::cout << GridLogMessage << " UdSdU " << UdSdU << std::endl;
+
+  LatticeComplex dS(&Grid); dS = zero;
+  parallel_for(auto i=mom.begin();i<mom.end();i++){
+    dS[i]() = trace(mom[i]() * UdSdU[i](mu) )*dt;
+  }
+  Complex dSpred = sum(dS);
+
+  std::cout << GridLogMessage << "predict dS    "<< dSpred <<std::endl;
+
+
+  cout<< GridLogMessage << "Done" <<std::endl;
+  Grid_finalize();
+}