[petsc-dev] Data types in petsc4py

[Samar Khatiwala]

Hello,

I’m writing a Python extension module to expose various classes, functions in my C PETSc-based application code. I’ve taken a lot of inspiration from petsc4py and borrowed bits of code from it. Everything works but one niggling issue I’m facing is how to automatically inherit the data-types used to build PETSc in my Cython code. Specifically, I want to make sure that if PETSc is built with 64-bit ints then I use the correct type when I call my PETSc C functions or create Numpy arrays with PyArray_SimpleNewFromData, etc.

I tried to understand how petsc4py does it, but looking at PETSc/PETSc.pyx it seems that PetscInt is hardwired:

cdef extern from * nogil:
    ctypedef long   PetscInt
    ctypedef double PetscReal
    ctypedef double PetscScalar

PetscInt is then used elsewhere Ωin petsc4py (e.g., in Mat.pyx) when calling PETSc functions. But wouldn’t this be incompatible with the PetscInt defined in petscsystypes.h if PETSc were built with --with-64-bit-indices?

On the other hand, I see (sort of; the code is very complicated!) that arraynpy.pxi correctly picks up (via numpy.h) the mappings between PETSc and Numpy arrays and these are used in the few places where an enumerated data type is required as an argument (e.g., in DMSwarm.pyx).

I don’t believe there’s any incompatibility with PetscInt in petsc4py so I’m likely missing something here. Could someone please explain how PetscInt, etc pick up their correct values and how I can do the same in my code?

Thanks very much!

Best,

Samar