GitLab

With this change to the python wrappers, special objects like
vtkVariant can be automatically constructed in order to resolve
method arguments.  So vtkVariantArray.InsertNextValue(vtkVariant v)
can take int, float, string, etc. and the vtkVariant will be
constructed automatically.  Only one level of conversion is allowed.

I've added enum constants for how well an arg matches the arg type
required by a particular method signature.  The enums are
EXACT_MATCH, GOOD_MATCH, NEEDS_CONVERSION, and INCOMPATIBLE.
For vtkObjectBase args, the code counts the number of generations
between the required arg class and the provided arg class.  A
difference of one generation is GOOD_MATCH, two generations is
GOOD_MATCH+1, etc. getting progressively worse but never as bad
as NEEDS_CONVERSION. Lower values signify better matches.

For overloaded method checks, if "self" was a PyVTKClass, then
the code was always assuming the first arg would be a PyVTKObject
instance of that PyVTKClass.  This assumption was wrong for static
methods.

The semantics for finding the right overload in the python wrappers
should match C++ semantics.  That means that array are matched based
only on the type, not on the size of the array.

The python wrappers used to call each overloaded method in turn until
one of them was able to resolve the arguments.  A new method has been
added to vtkPythonUtil.cxx called PyVTKCallOverloadedMethod() that will
carefully check the overloads against the provided arguments.  Overloads
that have a perfect type-match to the args get first priorty, followed
by trivial conversions like "int" to "short" or resolving a VTK subclass,
followed finally by more advanced conversions like "float" to "int".

of Pat's changes between Jun 22, 2009 and Sept. 8 2009.

BUG: When using the Py_NewInterpreter api there was no way to tell vtkPythonCommand which interpreter to use when executing the callback.  Added a SetThreadState method so that vtkPythonCommand may call PyThreadState_Swap in order to make the interpreter for the specified threadstate the active interpreter.  See vtk-developers email from June 18, 2009 with subject 'Py_NewInterpreter and vtkPythonCallback issue'

variable.
BUG: Fixing python pointer sscanf to use %p so that it works reliably on 64 bit
windows machines.

variable.
BUG: Fixing python pointer sscanf to use %p so that it works reliably on 64 bit
windows machines.

COMP:Fixed warnings with Python 2.6 about missing initializer for member PyBufferProcs::bf_getbuffer, PyBufferProcs::bf_releasebuffer and _typeobject::tp_version_tag. I have been told it was safe to apply this patch.

ENH: Merge changes from main tree into VTK-5-2 branch. (cvs -q up -j1.87 -j1.90 Common/vtkPythonUtil.cxx)

the Python interpreter would crash on exit.  This has been applied.
Thanks Chris!  More details here:
http://public.kitware.com/pipermail/vtk-developers/2008-August/005418.html

Kruszynski here:

http://public.kitware.com/pipermail/vtk-developers/2008-August/005409.html

The problem being that we are DECREFing a Python object after the Python
interpreter has been finalized leading to segfaults.

to be called after the interpreter has been finalized causing a
segfault.  This patch ensures that the interpreter is alive before
calling the callback preventing these problems.

ENH: Merge changes from main tree into VTK-5-2 branch. (cvs -q up -j1.83 -j1.87 Common/vtkPythonUtil.cxx)

COMP:Fixed warning in gcc release mode about dereferencing type-punned pointer will break strict-aliasing rules.

  - The vtkPythonUtil hash entry owns a reference
    for each object in it.
  - Therefore we do not need to Register/UnRegister
    a reference when the hash entry is created/destroyed.
  - This reduces all python-held reference counts by 1.
  - Also remove old delete observer code that is unused.
  - The observer is not needed because the python object
    has its own python reference count.

ENH: Added buffer interface to vtk objects (arrays). This allows passing VTK array pointers to python modules such as numpy

BUG: Fix bug 1953: replace strstream use with stringstream now that kwsys provides stringstream on all platforms. Many thanks to Sean McBride and Mathieu Coursolle from Rogue Research for all their effort developing the patch (and patience in waiting for its commit...!). COMP: Eliminate performance warning in generated Java wrappers regarding jboolean cast to bool. ENH: Exclude Java and Python wrappers (in addition to already excluded Tcl wrappers) from coverage analysis.

COMP: configure a value into vtkToolkits.h to determine whether to support
multithreading in the python wrappers or not

Alex

COMP: prepare VTK for crosscompiling
-fix some problems if no development packages for tcl or tk are installed
-prefer the own cmake modules over the ones provided by cmake
-fix build of vtkSocket for systems which don't have SO_REUSEADDR (e.g. BlueGene)
-add a RenderWindow which uses OSMesa only, i.e. no X at all
-support build on platforms without shared libs
-use ADD_SUBDIRECTORY() instead of SUBDIRS() in VTK/CMakeLists.txt,
otherwise GET_TARGET_PROPERTY(LOCATION) doesn't work, since the targets
haven't been created yet
-add FindPythonLibs, FindTCL and FindPackageHandleStandardArgs from current
cmake cvs

tested with cmake cvs on Linux and BlueGene and cmake 2.4.6 on Linux

Alex

ENH: Merge changes from main tree into VTK-5-0 branch. (cvs -q up -j1.76 -j1.77 Common/vtkPythonUtil.cxx)

COMP: Fix compile error when using Python 2.5 (thanks to Scott Harris and Charl Botha on the VTK mailing lists)

acquisition API for Python versions >= 2.3.  Without this fix,
observer callbacks from multiple threads into Python would break.  See
this thread for more details:
http://public.kitware.com/pipermail/vtk-developers/2005-July/003384.html

all types.

weakly reference VTK objects.  Also cleaned up the
VTK_PYTHON_UTIL_SUPRESS_UNINITIALIZED macro definitions a bit.

ENH: Removing need for VTK_ID_TYPE_IS_NOT_BASIC_TYPE.  Providing vtkPythonCheckArray overloads for long long and __int64 when they are available.  This will cover all cases of vtkIdType.  Also adding overload for signed char.

ENH: Removing inclusion of windows.h by top VTK headers.  This results in a substantial compile time reduction (~70%) on Windows compilers.  It also avoids forcing windows name mangling on user code.  Preprocessor work-arounds for this mangling has been added for VTK methods.  A new vtkWindows.h has been created for VTK code that really needs the windows API.  A vtkOpenGL.h has been created for VTK code that needs GL/gl.h because this header requires windows.h to be included first on some compilers.  This addresses bug #836.

     1) this makes help() work nicely on vtk classes, since __dict__ is used for introspection
     2) since __dict__ is a hash, VTK function calls no longer require a linear lookup and should be faster than before
     The __dict__ is only created for VTK classes that are actually used, so the memory overhead should be small.  Note that VTK classes and instances no longer have __members__ or __methods__ special attributes.