I want to ask all you for adding better support for petsc4py in core
PETSc.
All what I need to to add to opaque pointers in the base PetscObject
structure, like this
void *pycontext
PetscErrorcode (*pydestroy)(void*)
As you see, this will not make core PETSc depend on Python in any way.
Additionally, the extra memory overhead is small, just to pointers..
My intention is to put a Python dictionary in the pycontext pointer,
and call pydestroy function pointer (I'll set if when needed) at the
time PetscHeaderDestroy is called.
Such a feature is currently being emulated with PetscContainer and
PetscObjectCompose, but I would really like to cleanup this hackery
and implement things the right way.
I've started to migrate my SWIG-based wrappers to use Cython. This is
preliminar and not yet ready for public consumption, but at firts I
anticipate this approach if far easier to understand, extend, and
maintain than the previous one. In the near future, making petsc4py
interoperate with C/C++/Fortran user code will be possible and easy.
For example, I saw the new Picard implementation in SNES. Such a
feature could be implemented with a few lines of pure Python code, no
point to have to write a full SNES type implementation (unless it is
really needed). Just write in Python a custom nonlinear solver, and
now run your C/C++/Fortran code with an special option in the command
line, and PETSc picks-up and used the Python code implementing your
solver.
--
Lisandro Dalcín
---------------
Centro Internacional de Métodos Computacionales en Ingeniería (CIMEC)
Instituto de Desarrollo Tecnológico para la Industria Química (INTEC)
Consejo Nacional de Investigaciones Científicas y Técnicas (CONICET)
PTLC - Güemes 3450, (3000) Santa Fe, Argentina
Tel/Fax: +54-(0)342-451.1594