[petsc-dev] C++ and hypre long long int vs PETSc int64_t for 64 bit integer builds

[Jed Brown]

C/C++ integer hierarchy was a mistake, with the language-specified types overlapping and the size-specific (int64_t) being typedefs. So long is "different" from long long even if they're both 64-bit signed integers.

I think they should test that sizeof(HYPRE_Int) == sizeof(PetscInt), but not that they're "the same". They might still need some casts depending on usage.

Barry Smith <bsmith at petsc.dev> writes:

> For 64 bit integer builds  AMReX has this test in their code
>
> CMakeFiles/amrex.dir/F_Interfaces/Base/AMReX_multifab_mod.F90.o
> /lcrc/project/PhloughToolz/petsc/petsc-3.17.0_oneAPI_2021.4.0.3422/externalpackages/git.amrex/Src/Extern/PETSc/AMReX_PETSc.cpp(44): error: static assertion failed with "HYPRE_Int != PetscInt"
>       static_assert(std::is_same<HYPRE_Int, PetscInt>::value, "HYPRE_Int != PetscInt");
>
>  Currently, hypre uses long lont int for 64 bit integers, PETSc depending on what MPI and integer types are available uses int64_t but may fall back to long long int.
>
>   This means depending on configuration information the PETSc built hypre and PETSc may or may not be compatible by the AMReX tests for C++ code.
>
>  What is the correct general fix for this three-body problem?
>
>    Will the C++ code in AMReX, which presumably is template-and-polymorphic-method-city, fail if the long long int and int64_t are mixed and treated as the same (that is one just removes the is_same() check)? Clearly, the easiest fix 
>
>    Do HYPRE and PETSc configure need to have command-line options to allow controlling exactly the type used to ensure matching? 
>
>    Something else?
>
>   Barry