These are better tracked directly by Git rather than as ExternalData.

Data is now downloaded by each testing module as needed.

David Gobbi authored 7 years ago and Ben Boeckel committed 6 years ago

When a module is initialized, the 'depends' list for that module
is now allowed to include the absolute path to each dependency,
e.g. 'vtkmodule.vtkCommonCorePython'.

David Gobbi authored 7 years ago and Ben Boeckel committed 6 years ago

When the VTK Python extension modules are loaded more than once
(e.g. by multiple Python interpreters in the same process), we
must ignore any PyTypeObjects except for the ones from the first
load.  This fixes crashes that started occurring when we moved
the PyTypeObjects from the shared PythonD libraries into the
Python modules themselves.

David Gobbi authored 7 years ago and Ben Boeckel committed 6 years ago

This way, if one module depends on another, then it can check if
that module has already been imported before calling ImportModule.
In this way, people can avoid the ImportModule call completely
by calling "import" themselves inside their Python code, which
allows them to import the modules from a specific location.

David Gobbi authored 7 years ago and Ben Boeckel committed 6 years ago

If one vtk-python module depends on others, then it should load
those dependencies before doing anything else.

David Gobbi authored 7 years ago and Ben Boeckel committed 6 years ago

This removes all of the PythonD libraries, and moves their contents
into the python modules themselves. The result is a cleaner build.

David Gobbi authored 7 years ago and Ben Boeckel committed 6 years ago

If superclasses are linked via import, then it is no longer necessary
for the dynamic linker to be able to find the superclass objects.
This adds flexibility to the way the wrappers can be packaged.

David Gobbi authored 7 years ago and Ben Boeckel committed 6 years ago

For wrapped template specializations, the fully qualified name was
added to the module dictionary, e.g. vtkCommonDataModelPython.name,
instead of simply "name" by itself.

This patch also removes some obsolete code for handling old-style
classes, which were removed from the wrappers for Python3 and no
longer need to be supported.