diff --git a/Fltk/FlGui.cpp b/Fltk/FlGui.cpp
index abd35b0f0b0dd31e6ed7ab72d133d4ff60828fc4..3f1dc378c2a6170539886dffb390e9f29d2eaf5a 100644
--- a/Fltk/FlGui.cpp
+++ b/Fltk/FlGui.cpp
@@ -44,6 +44,15 @@
 #include "Generator.h"
 #include "gl2ps.h"
 
+// check (now!) if there are any pending events, and process them
+void FlGui::check(){ Fl::check(); }
+
+// wait (possibly indefinitely) for any events, then process them
+void FlGui::wait(){ Fl::wait(); }
+
+// wait (at most time seconds) for any events, then process them
+void FlGui::wait(double time){ Fl::wait(time); }
+
 class drawContextFltk : public drawContextGlobal{
  public:
   void draw()
diff --git a/Fltk/FlGui.h b/Fltk/FlGui.h
index b79330756304d47e259158f3fabc4c589c9b292b..3388bb5a8b6876754468fd2aed94ef9eec977384 100644
--- a/Fltk/FlGui.h
+++ b/Fltk/FlGui.h
@@ -8,7 +8,6 @@
 
 #include <string>
 #include <vector>
-#include <FL/Fl.H>
 
 #define GMSH_WINDOW_BOX FL_FLAT_BOX
 #define NB_BUTT_SCROLL 25
@@ -33,6 +32,7 @@ class geometryContextWindow;
 class meshContextWindow;
 class aboutWindow;
 class onelabWindow;
+class Fl_Widget;
 
 class GVertex;
 class GEdge;
@@ -76,11 +76,11 @@ class FlGui{
   // run the GUI until there's no window left
   static int run();
   // check (now!) if there are any pending events, and process them
-  static void check(){ Fl::check(); }
+  static void check();
   // wait (possibly indefinitely) for any events, then process them
-  static void wait(){ Fl::wait(); }
+  static void wait();
   // wait (at most time seconds) for any events, then process them
-  static void wait(double time){ Fl::wait(time); }
+  static void wait(double time);
   // is a file opened through the Mac Finder?
   void setOpenedThroughMacFinder(bool val){ _openedThroughMacFinder = val; }
   bool getOpenedThroughMacFinder(){ return _openedThroughMacFinder; }
diff --git a/Geo/GRbf.cpp b/Geo/GRbf.cpp
index cd8af08eff2cce8b1a0728f08877f4c39eb37ef2..d6094e6dbcfa16b1b25aa109a746b3faacde1dee 100644
--- a/Geo/GRbf.cpp
+++ b/Geo/GRbf.cpp
@@ -153,10 +153,12 @@ GRbf::GRbf (double sizeBox, int variableEps, int rbfFun, std::map<MVertex*, SVec
 
 GRbf::~GRbf(){
 #if defined (HAVE_ANN)
+  ANNpointArray XYZNodes = XYZkdtree->thePoints();
+  ANNpointArray UVNodes = UVkdtree->thePoints();
+  annDeallocPts(XYZNodes);
+  annDeallocPts(UVNodes);
   delete XYZkdtree;
   delete UVkdtree;
-  annDeallocPts(XYZnodes);
-  annDeallocPts(UVnodes);
 #endif
 
 }
@@ -164,7 +166,7 @@ GRbf::~GRbf(){
 void GRbf::buildXYZkdtree(){
 
 #if defined (HAVE_ANN)
-  XYZnodes = annAllocPts(nbNodes, 3);
+  ANNpointArray XYZnodes = annAllocPts(nbNodes, 3);
   for(int i = 0; i < nbNodes; i++){
     XYZnodes[i][0] = centers(i,0); 
     XYZnodes[i][1] = centers(i,1); 
@@ -821,7 +823,7 @@ void GRbf::solveHarmonicMap(fullMatrix<double> Oper,
   //ANN UVtree
   double dist_min = 1.e6;
 #if defined (HAVE_ANN)
-  UVnodes = annAllocPts(nbNodes, 3);
+  ANNpointArray UVnodes = annAllocPts(nbNodes, 3);
   for(int i = 0; i < nbNodes; i++){
     UVnodes[i][0] = UV(i,0); 
     UVnodes[i][1] = UV(i,1); 
diff --git a/Geo/GRbf.h b/Geo/GRbf.h
index c3a565e77fe719080f9b697aff8f8b3ef4832dc3..6d2f89c23acabe6f606b72d9eb2cbfd7d9d567a9 100644
--- a/Geo/GRbf.h
+++ b/Geo/GRbf.h
@@ -9,7 +9,7 @@
 #include "MVertex.h"
 #include "Context.h"
 #if defined(HAVE_ANN)
-#include <ANN/ANN.h>
+class ANNkd_tree;
 #endif
 
 
@@ -57,8 +57,6 @@ class GRbf {
 #if defined (HAVE_ANN)
   ANNkd_tree *XYZkdtree; 
   ANNkd_tree *UVkdtree; 
-  ANNpointArray XYZnodes;
-  ANNpointArray UVnodes;
 #endif
 
  public:
diff --git a/Mesh/Levy3D.cpp b/Mesh/Levy3D.cpp
index e6d88c023a8ec88b90ba65d6d4752d450b4c0e74..61d146b059fccc37c213d5c1af3431cd9484992d 100755
--- a/Mesh/Levy3D.cpp
+++ b/Mesh/Levy3D.cpp
@@ -17,6 +17,13 @@
 #include "MElement.h"
 #include "MElementOctree.h"
 #include "meshGRegion.h"
+#include "ap.h"
+#include "alglibinternal.h"
+#include "alglibmisc.h" 
+#include "linalg.h"
+#include "optimization.h"
+
+void call_back(const alglib::real_1d_array&,double&,alglib::real_1d_array&,void*);
 
 /*********class VoronoiVertex*********/
 
diff --git a/Mesh/Levy3D.h b/Mesh/Levy3D.h
index 8118deca98e36c9e4509aa4778e4f8c1da81803e..42e164ea5d457f50576fbc7ec0f1450ec98854a6 100755
--- a/Mesh/Levy3D.h
+++ b/Mesh/Levy3D.h
@@ -10,12 +10,7 @@
 #include <list>
 #include "fullMatrix.h"
 #include "GRegion.h"
-#include "MElementOctree.h"
-#include "ap.h"
-#include "alglibinternal.h"
-#include "alglibmisc.h" 
-#include "linalg.h"
-#include "optimization.h"
+class MElementOctree;
 
 /*********class VoronoiVertex*********/
 
@@ -191,4 +186,3 @@ class Wrap{
 /*********functions*********/
 
 bool inside_domain(MElementOctree*,double,double,double);
-void call_back(const alglib::real_1d_array&,double&,alglib::real_1d_array&,void*);
diff --git a/Mesh/meshGFaceLloyd.cpp b/Mesh/meshGFaceLloyd.cpp
index 897dd7fbc426f9971cf73e8ff8eb7f8b68fb659a..2f81ae95ecc4b570c70eb47e6cb40c980c9411a4 100644
--- a/Mesh/meshGFaceLloyd.cpp
+++ b/Mesh/meshGFaceLloyd.cpp
@@ -28,6 +28,47 @@
 #include "polynomialBasis.h"
 #include "MElementOctree.h"
 
+bool domain_search(MElementOctree* octree,double x,double y){
+  MElement* element;
+	
+  element = (MElement*)octree->find(x,y,0.0,2,true);
+  if(element!=NULL) return 1;
+  else return 0;
+}
+
+
+
+class wrapper{
+ private :
+  int p;
+  int dimension;
+  GFace* gf;
+  int iteration;
+  int max;
+  double start;
+  DocRecord* triangulator;
+  MElementOctree* octree;
+ public :
+  wrapper();
+  ~wrapper();
+  int get_p();
+  void set_p(int);
+  int get_dimension();
+  void set_dimension(int);
+  GFace* get_face();
+  void set_face(GFace*);
+  int get_iteration();
+  void set_iteration(int);
+  int get_max();
+  void set_max(int);
+  double get_start();
+  void set_start(double);
+  DocRecord* get_triangulator();
+  void set_triangulator(DocRecord*);
+  MElementOctree* get_octree();
+  void set_octree(MElementOctree*);
+};
+
 
 
 /****************fonction callback****************/
@@ -147,15 +188,6 @@ void callback(const alglib::real_1d_array& x,double& func,alglib::real_1d_array&
   }	
 }
 
-bool domain_search(MElementOctree* octree,double x,double y){
-  MElement* element;
-	
-  element = (MElement*)octree->find(x,y,0.0,2,true);
-  if(element!=NULL) return 1;
-  else return 0;
-}
-
-
 
 /****************class smoothing****************/
 
diff --git a/Mesh/meshGFaceLloyd.h b/Mesh/meshGFaceLloyd.h
index 59f62124b623589bbd112753c3d7d5eb28b2944f..8e80fc7103aed9195cb517fa06747b24376fc102 100644
--- a/Mesh/meshGFaceLloyd.h
+++ b/Mesh/meshGFaceLloyd.h
@@ -13,12 +13,6 @@
 #include "fullMatrix.h"
 #include "DivideAndConquer.h"
 #include <queue>
-#include "ap.h"
-#include "alglibinternal.h"
-#include "alglibmisc.h" 
-#include "linalg.h"
-#include "optimization.h"
-#include "MElementOctree.h"
 
 class GFace;
 class voronoi_vertex;
@@ -28,8 +22,6 @@ class segment;
 class segment_list;
 class metric;
 
-void callback(const alglib::real_1d_array&,double&,alglib::real_1d_array&,void*);
-bool domain_search(MElementOctree*,double,double);
 
 class smoothing{
   int ITER_MAX;
@@ -216,37 +208,6 @@ class segment_list{
   bool add_segment(segment);
 };
 
-class wrapper{
- private :
-  int p;
-  int dimension;
-  GFace* gf;
-  int iteration;
-  int max;
-  double start;
-  DocRecord* triangulator;
-  MElementOctree* octree;
- public :
-  wrapper();
-  ~wrapper();
-  int get_p();
-  void set_p(int);
-  int get_dimension();
-  void set_dimension(int);
-  GFace* get_face();
-  void set_face(GFace*);
-  int get_iteration();
-  void set_iteration(int);
-  int get_max();
-  void set_max(int);
-  double get_start();
-  void set_start(double);
-  DocRecord* get_triangulator();
-  void set_triangulator(DocRecord*);
-  MElementOctree* get_octree();
-  void set_octree(MElementOctree*);
-};
-
 #endif
 
 #endif
diff --git a/Solver/dofManager.cpp b/Solver/dofManager.cpp
index 2ce4b596a4e1fc7acaf1626333d71967fb386be7..a2a6c14253bfd4f3a088bcce07a1f29c41f6116a 100644
--- a/Solver/dofManager.cpp
+++ b/Solver/dofManager.cpp
@@ -1,4 +1,8 @@
 #include <dofManager.h>
+#include "GmshConfig.h"
+#ifdef HAVE_MPI
+#include "mpi.h"
+#endif
 template<>
 void dofManager<double>::scatterSolution()
 {
@@ -31,3 +35,112 @@ void dofManager<double>::scatterSolution()
   MPI_Waitall (Msg::GetCommSize(), &reqSend[0], MPI_STATUS_IGNORE);
 #endif
 }
+
+void dofManagerBase::_parallelFinalize()
+{
+  _localSize = unknown.size();
+#ifdef HAVE_MPI
+  int numStart;
+  int numTotal;
+  MPI_Status status;
+  parentByProc.resize(Msg::GetCommSize());
+  ghostByProc.resize(Msg::GetCommSize());
+  if (Msg::GetCommRank() == 0){
+    numStart = 0;
+  }
+  else
+    MPI_Recv (&numStart, 1, MPI_INT, Msg::GetCommRank()-1, 0, MPI_COMM_WORLD, &status);
+  numTotal = numStart + _localSize;
+  if (Msg::GetCommRank() != Msg::GetCommSize()-1)
+    MPI_Send (&numTotal, 1, MPI_INT, Msg::GetCommRank()+1, 0, MPI_COMM_WORLD);
+  MPI_Bcast(&numTotal, 1, MPI_INT, Msg::GetCommSize()-1, MPI_COMM_WORLD);
+  for (std::map <Dof, int> ::iterator it = unknown.begin(); it!= unknown.end(); it++)
+    it->second += numStart;
+  std::vector<std::list<Dof> >  ghostedByProc;
+  int *nRequest = new int[Msg::GetCommSize()];
+  int *nRequested = new int[Msg::GetCommSize()];
+  for (int i = 0; i<Msg::GetCommSize(); i++)
+    nRequest[i] = 0;
+  for (std::map <Dof, std::pair<int, int> >::iterator it = ghostByDof.begin(); it != ghostByDof.end(); it++) {
+    int procId = it->second.first;
+    it->second.second = nRequest[procId]++;
+  }
+  MPI_Alltoall(nRequest, 1, MPI_INT, nRequested, 1, MPI_INT, MPI_COMM_WORLD);
+  long int **recv0 = new long int*[Msg::GetCommSize()];
+  int **recv1 = new int*[Msg::GetCommSize()];
+  long int **send0 = new long int*[Msg::GetCommSize()];
+  int **send1 = new int*[Msg::GetCommSize()];
+  MPI_Request *reqRecv0 = new MPI_Request[2*Msg::GetCommSize()];
+  MPI_Request *reqRecv1 = reqRecv0 + Msg::GetCommSize();
+  MPI_Request *reqSend0 = new MPI_Request[Msg::GetCommSize()];
+  MPI_Request *reqSend1 = new MPI_Request[Msg::GetCommSize()];
+  for (int i = 0; i < Msg::GetCommSize(); i++) {
+    send0[i] = new long int[nRequest[i]*2];
+    recv0[i] = new long int[nRequested[i]*2];
+    send1[i] = new int[nRequested[i]];
+    recv1[i] = new int[nRequest[i]];
+    reqSend0[i] = reqSend1[i] = reqRecv0[i] = reqRecv1[i] = MPI_REQUEST_NULL;
+    parentByProc[i].resize(nRequested[i], Dof(0,0));
+    ghostByProc[i].resize(nRequest[i], Dof(0,0));
+  }
+  for (int i = 0; i<Msg::GetCommSize(); i++)
+    nRequest [i] = 0;
+  for (std::map <Dof, std::pair<int, int> >::iterator it = ghostByDof.begin(); it != ghostByDof.end(); it++) {
+    int proc = it->second.first;
+    send0 [proc] [nRequest[proc]*2] = it->first.getEntity();
+    send0 [proc] [nRequest[proc]*2+1] = it->first.getType();
+    ghostByProc[proc][nRequest[proc]] = it->first;
+    nRequest [proc] ++;
+  }
+  for (int i = 0; i<Msg::GetCommSize(); i++) {
+    if (nRequested[i] > 0) {
+      MPI_Irecv (recv0[i], 2*nRequested[i], MPI_LONG, i, 0, MPI_COMM_WORLD, &reqRecv0[i]);
+    }
+    if (nRequest[i] > 0) {
+      MPI_Irecv (recv1[i], 2*nRequest[i], MPI_INT, i, 1, MPI_COMM_WORLD, &reqRecv1[i]);
+      MPI_Isend (send0[i], 2*nRequest[i], MPI_LONG, i, 0, MPI_COMM_WORLD, &reqSend0[i]);
+    }
+  }
+  int index;
+  while (MPI_Waitany (2*Msg::GetCommSize(), reqRecv0, &index, &status) == 0 &&
+         index != MPI_UNDEFINED) {
+    if (status.MPI_TAG == 0) {
+      for (int j = 0; j < nRequested[index]; j++) {
+        Dof d(recv0[index][j*2], recv0[index][j*2+1]);
+        std::map<Dof, int>::iterator it = unknown.find(d);
+        if (it == unknown.end ())
+          Msg::Error ("ghost Dof does not exist on parent process");
+        send1[index][j] = it->second;
+        parentByProc[index][j] = d;
+      }
+      MPI_Isend(send1[index], nRequested[index], MPI_INT, index, 1,
+                MPI_COMM_WORLD, &reqSend1[index]);
+    }
+  }
+  for (int i = 0; i<Msg::GetCommSize(); i++)
+  for (int i = 0; i<Msg::GetCommSize(); i++)
+    nRequest[i] = 0;
+  for (std::map <Dof, std::pair<int, int> >::iterator it = ghostByDof.begin(); it != ghostByDof.end(); it++) {
+    int proc = it->second.first;
+    unknown[it->first] = recv1 [proc][nRequest[proc] ++];
+  }
+  MPI_Waitall (Msg::GetCommSize(), reqSend0, MPI_STATUS_IGNORE);
+  MPI_Waitall (Msg::GetCommSize(), reqSend1, MPI_STATUS_IGNORE);
+  for (int i = 0; i < Msg::GetCommSize(); i++) {
+    delete [] send0[i];
+    delete [] send1[i];
+    delete [] recv0[i];
+    delete [] recv1[i];
+  }
+  delete [] send0;
+  delete [] send1;
+  delete [] recv0;
+  delete [] recv1;
+  delete [] reqSend0;
+  delete [] reqSend1;
+  delete [] reqRecv0;
+#endif
+  _parallelFinalized = true;
+}
+
+
diff --git a/Solver/dofManager.h b/Solver/dofManager.h
index 4783b3eee3b259df5a19110e13e1c56e2452ba7d..87ebbedc1a9a9e7f614d87c41cc20cb1877b5db4 100644
--- a/Solver/dofManager.h
+++ b/Solver/dofManager.h
@@ -12,15 +12,10 @@
 #include <map>
 #include <list>
 #include <iostream>
-#include "GmshConfig.h"
 #include "MVertex.h"
 #include "linearSystem.h"
 #include "fullMatrix.h"
 
-#if defined(HAVE_MPI)
-#include "mpi.h"
-#endif
-
 class Dof{
  protected:
   // v(x) = \sum_f \sum_i v_{fi} s^f_i(x)
@@ -88,18 +83,41 @@ class DofAffineConstraint{
   typename dofTraits<T>::VecType shift;
 };
 
+//non template part that can be implemented in the cxx file (and so avoid to include mpi.h in the .h file)
+class dofManagerBase{
+  protected:
+  // numbering of unknown dof blocks
+  std::map<Dof, int> unknown;
+
+  // associatations (not used ?)
+  std::map<Dof, Dof> associatedWith;
+
+  // parallel section
+  // those dof are images of ghost located on another proc (id givent by the map).
+  // this is a first try, maybe not the final implementation
+  std::map<Dof, std::pair<int, int> > ghostByDof;  // dof => procId, globalId
+  std::vector<std::vector<Dof> > ghostByProc, parentByProc;
+  int _localSize;
+  bool _parallelFinalized;
+  bool _isParallel;
+  void _parallelFinalize();
+  dofManagerBase(bool isParallel) {
+    _isParallel = isParallel;
+    _parallelFinalized = false;
+  }
+};
+
 // A manager for degrees of freedoms, templated on the value of a dof
 // (what the functional returns): float, double, complex<double>,
 // fullVecor<double>, ...
 template <class T>
-class dofManager{
+class dofManager : public dofManagerBase{
  public:
   typedef typename dofTraits<T>::VecType dataVec;
   typedef typename dofTraits<T>::MatType dataMat;
  protected:
   // general affine constraint on sub-blocks, treated by adding
   // equations:
-
   //   Dof = \sum_i dataMat_i x Dof_i + dataVec
   std::map<Dof, DofAffineConstraint< dataVec > > constraints;
 
@@ -110,38 +128,22 @@ class dofManager{
   // initial conditions (not used ?)
   std::map<Dof, std::vector<dataVec> > initial;
 
-  // numbering of unknown dof blocks
-  std::map<Dof, int> unknown;
-
-  // associatations (not used ?)
-  std::map<Dof, Dof> associatedWith;
-
   // linearSystems
-  std::map<const std::string, linearSystem<dataMat>*> _linearSystems;
   linearSystem<dataMat> *_current;
+  std::map<const std::string, linearSystem<dataMat>*> _linearSystems;
 
-  // parallel section
- private :
-  // those dof are images of ghost located on another proc (id givent by the map).
-  // this is a first try, maybe not the final implementation
-  std::map<Dof, std::pair<int, int> > ghostByDof;  // dof => procId, globalId
   std::map<Dof, T> ghostValue;
-  std::vector<std::vector<Dof> > ghostByProc, parentByProc;
-  int _localSize;
-  bool _parallelFinalized;
-  bool _isParallel;
-  void _parallelFinalize();
   public:
   void scatterSolution();
 
  public:
   dofManager(linearSystem<dataMat> *l, bool isParallel=false)
-    : _current(l), _isParallel(isParallel), _parallelFinalized(false)
+    :dofManagerBase(isParallel), _current(l)
   {
     _linearSystems["A"] = l;
   }
   dofManager(linearSystem<dataMat> *l1, linearSystem<dataMat> *l2)
-    : _current(l1), _isParallel(false), _parallelFinalized(false)
+    :dofManagerBase(false), _current(l1)
   {
     _linearSystems.insert(std::make_pair("A", l1));
     _linearSystems.insert(std::make_pair("B", l2));
@@ -557,112 +559,4 @@ class dofManager{
 	};
 };
 
-template<class T>
-void dofManager<T>::_parallelFinalize()
-{
-  _localSize = unknown.size();
-#ifdef HAVE_MPI
-  int _numStart;
-  int _numTotal;
-  MPI_Status status;
-  parentByProc.resize(Msg::GetCommSize());
-  ghostByProc.resize(Msg::GetCommSize());
-  if (Msg::GetCommRank() == 0){
-    _numStart = 0;
-  }
-  else
-    MPI_Recv (&_numStart, 1, MPI_INT, Msg::GetCommRank()-1, 0, MPI_COMM_WORLD, &status);
-  _numTotal = _numStart + _localSize;
-  if (Msg::GetCommRank() != Msg::GetCommSize()-1)
-    MPI_Send (&_numTotal, 1, MPI_INT, Msg::GetCommRank()+1, 0, MPI_COMM_WORLD);
-  MPI_Bcast(&_numTotal, 1, MPI_INT, Msg::GetCommSize()-1, MPI_COMM_WORLD);
-  for (std::map <Dof, int> ::iterator it = unknown.begin(); it!= unknown.end(); it++)
-    it->second += _numStart;
-  std::vector<std::list<Dof> >  ghostedByProc;
-  int *nRequest = new int[Msg::GetCommSize()];
-  int *nRequested = new int[Msg::GetCommSize()];
-  for (int i = 0; i<Msg::GetCommSize(); i++)
-    nRequest[i] = 0;
-  for (std::map <Dof, std::pair<int, int> >::iterator it = ghostByDof.begin(); it != ghostByDof.end(); it++) {
-    int procId = it->second.first;
-    it->second.second = nRequest[procId]++;
-  }
-  MPI_Alltoall(nRequest, 1, MPI_INT, nRequested, 1, MPI_INT, MPI_COMM_WORLD);
-  long int **recv0 = new long int*[Msg::GetCommSize()];
-  int **recv1 = new int*[Msg::GetCommSize()];
-  long int **send0 = new long int*[Msg::GetCommSize()];
-  int **send1 = new int*[Msg::GetCommSize()];
-  MPI_Request *reqRecv0 = new MPI_Request[2*Msg::GetCommSize()];
-  MPI_Request *reqRecv1 = reqRecv0 + Msg::GetCommSize();
-  MPI_Request *reqSend0 = new MPI_Request[Msg::GetCommSize()];
-  MPI_Request *reqSend1 = new MPI_Request[Msg::GetCommSize()];
-  for (int i = 0; i < Msg::GetCommSize(); i++) {
-    send0[i] = new long int[nRequest[i]*2];
-    recv0[i] = new long int[nRequested[i]*2];
-    send1[i] = new int[nRequested[i]];
-    recv1[i] = new int[nRequest[i]];
-    reqSend0[i] = reqSend1[i] = reqRecv0[i] = reqRecv1[i] = MPI_REQUEST_NULL;
-    parentByProc[i].resize(nRequested[i], Dof(0,0));
-    ghostByProc[i].resize(nRequest[i], Dof(0,0));
-  }
-  for (int i = 0; i<Msg::GetCommSize(); i++)
-    nRequest [i] = 0;
-  for (std::map <Dof, std::pair<int, int> >::iterator it = ghostByDof.begin(); it != ghostByDof.end(); it++) {
-    int proc = it->second.first;
-    send0 [proc] [nRequest[proc]*2] = it->first.getEntity();
-    send0 [proc] [nRequest[proc]*2+1] = it->first.getType();
-    ghostByProc[proc][nRequest[proc]] = it->first;
-    nRequest [proc] ++;
-  }
-  for (int i = 0; i<Msg::GetCommSize(); i++) {
-    if (nRequested[i] > 0) {
-      MPI_Irecv (recv0[i], 2*nRequested[i], MPI_LONG, i, 0, MPI_COMM_WORLD, &reqRecv0[i]);
-    }
-    if (nRequest[i] > 0) {
-      MPI_Irecv (recv1[i], 2*nRequest[i], MPI_INT, i, 1, MPI_COMM_WORLD, &reqRecv1[i]);
-      MPI_Isend (send0[i], 2*nRequest[i], MPI_LONG, i, 0, MPI_COMM_WORLD, &reqSend0[i]);
-    }
-  }
-  int index;
-  while (MPI_Waitany (2*Msg::GetCommSize(), reqRecv0, &index, &status) == 0 &&
-         index != MPI_UNDEFINED) {
-    if (status.MPI_TAG == 0) {
-      for (int j = 0; j < nRequested[index]; j++) {
-        Dof d(recv0[index][j*2], recv0[index][j*2+1]);
-        std::map<Dof, int>::iterator it = unknown.find(d);
-        if (it == unknown.end ())
-          Msg::Error ("ghost Dof does not exist on parent process");
-        send1[index][j] = it->second;
-        parentByProc[index][j] = d;
-      }
-      MPI_Isend(send1[index], nRequested[index], MPI_INT, index, 1,
-                MPI_COMM_WORLD, &reqSend1[index]);
-    }
-  }
-  for (int i = 0; i<Msg::GetCommSize(); i++)
-  for (int i = 0; i<Msg::GetCommSize(); i++)
-    nRequest[i] = 0;
-  for (std::map <Dof, std::pair<int, int> >::iterator it = ghostByDof.begin(); it != ghostByDof.end(); it++) {
-    int proc = it->second.first;
-    unknown[it->first] = recv1 [proc][nRequest[proc] ++];
-  }
-  MPI_Waitall (Msg::GetCommSize(), reqSend0, MPI_STATUS_IGNORE);
-  MPI_Waitall (Msg::GetCommSize(), reqSend1, MPI_STATUS_IGNORE);
-  for (int i = 0; i < Msg::GetCommSize(); i++) {
-    delete [] send0[i];
-    delete [] send1[i];
-    delete [] recv0[i];
-    delete [] recv1[i];
-  }
-  delete [] send0;
-  delete [] send1;
-  delete [] recv0;
-  delete [] recv1;
-  delete [] reqSend0;
-  delete [] reqSend1;
-  delete [] reqRecv0;
-#endif
-  _parallelFinalized = true;
-}
-
 #endif
diff --git a/gmshpy/CMakeLists.txt b/gmshpy/CMakeLists.txt
index 4933e7db41169735efc143d123f106ff5c1b71bf..436155784f8c805e9a42c97fca5c59c6cee0250f 100644
--- a/gmshpy/CMakeLists.txt
+++ b/gmshpy/CMakeLists.txt
@@ -56,6 +56,11 @@ MACRO(SWIG_GET_WRAPPER_DEPENDENCIES swigFile genWrapper language DEST_VARIABLE)
   ENDIF(NOT ${swig_getdeps_error} EQUAL 0)
 ENDMACRO(SWIG_GET_WRAPPER_DEPENDENCIES)
 
+option(ENABLE_PYTHON_LIB_API "Export all C header files needed to build the python library" OFF)
+if(ENABLE_PYTHON_LIB_API)
+  set(GMSH_API ${GMSH_API} Geo/Curvature.h Mesh/Generator.h Mesh/highOrderTools.h Mesh/meshGFaceLloyd.h Numeric/DivideAndConquer.h Post/PViewFactory.h Solver/function.h Solver/functionDerivator.h Solver/functionPython.h Solver/functionNumpy.h Solver/linearSystemPETSc.h Fltk/FlGui.h Solver/functionSpace.h Solver/STensor43.h Solver/sparsityPattern.h Solver/SElement.h Solver/groupOfElements.h PARENT_SCOPE)
+endif(ENABLE_PYTHON_LIB_API)
+
 if(HAVE_SWIG)
   include(${SWIG_USE_FILE})
   include_directories(${PYTHON_INCLUDE_DIR})
diff --git a/gmshpy/gmshCommon.i b/gmshpy/gmshCommon.i
index 6d4dc71764d429dc1656c322819a707dac8f33f7..24f4ebd197393e2a2f4cc71bb63f73c8ec8edeb6 100644
--- a/gmshpy/gmshCommon.i
+++ b/gmshpy/gmshCommon.i
@@ -9,7 +9,6 @@
   #include "GmshConfig.h"
 
   #include "Context.h"
-  #include "DefaultOptions.h"
   #include "Gmsh.h"
   #include "GmshDefines.h"
   #include "GmshMessage.h"
@@ -46,7 +45,6 @@ namespace std {
 %include "GmshConfig.h"
 
 %include "Context.h"
-#include "DefaultOptions.h"
 #if defined(HAVE_FLTK)
 %include "FlGui.h"
 #endif
diff --git a/gmshpy/gmshSolver.i b/gmshpy/gmshSolver.i
index 97832aa66a0c13a42f954b253acd2f401ca53efd..e0aaa6982e780ab08a2ce1de907b9ba828476628 100644
--- a/gmshpy/gmshSolver.i
+++ b/gmshpy/gmshSolver.i
@@ -41,5 +41,7 @@ namespace std {
 %template(linearSystemTAUCSDouble) linearSystemCSRTaucs<double>;
 %include "linearSystemFull.h"
 %template(linearSystemFullDouble) linearSystemFull<double> ;
+#ifdef HAVE_PETSC
 %include "linearSystemPETSc.h"
 %template(linearSystemPETScDouble) linearSystemPETSc<double>;
+#endif