Error in @for_petsc function KSPSetComputeOperators ??

[priyanshu109]

I was playing with the branch bk/wrapping_tests.

I was trying to follow the examples. It seems as if the codes/functions have been updated since the examples were written.

But somehow, I could make things work. I tried laplacian.jl and then I was trying ex_50.jl.

But, when we call, PETSc.KSPSetComputeOperators!(ksp, jacobian!),

the higher level function calls a function in petsc_library.jl.

I kept on getting the same error , " Cannot convert a value to nothing for assignment".

When I looked deeper, this KSPSetComputeOperators behaves similarly to KSPSetComputeRHS and there is only one small difference in the structure of two functions.

 PetscErrorCode,
        (KSP, Ptr{KSPComputeRHSFn}, Ptr{Cvoid}),
        arg1,
        arg2,
        arg3, in KSPComputeRHS 

but

@for_petsc function KSPSetComputeOperators(::$UnionPetscLib, arg1, arg2, arg3)
    @chk ccall(
        (:KSPSetComputeOperators, $petsc_library),
        PetscErrorCode,
        (KSP, KSPComputeOperatorsFn, Ptr{Cvoid}),
        arg1,
        arg2,
        arg3,
    )

end. No Ptr in this.

Then, I defined my own function with Ptr, such that Julia higher level function will call this one, it worked fine.

I am just pointing it out since I am not 100 percent sure if I am doing this correct ( I have almost no experience with C) or was there another way to call the same function.

Thank you in advance.

[boriskaus]

Thanks for spotting this!
This is most likely a bug, This branch is very much work in progress, and I haven't been able to start fixing the tests yet.
In fact if you, or anyone else reading this, would like to help with this here some explanations of what I am trying to do:

1. Initially I tried using Clang.jl to automatically create wrappers, continuing what Jeremy Kozdon start a long time ago. In this way a good part of the petsc library is autowrapped into one large file, petsc_library.jl. As you'll see the function calls are very much in-transparent, so we started writing more julia-friendly codes for this along with tests. See for example dmstag_wrapped.jl. This also automatically extracts the documentation of PETSc functions and adds it to the julia functions. Yet much remains to be done.

2. Following a discussion with Matt Knepley (@knepley) a little while back, he pointed out that Barry Smith (@BarrySmith) had created a python script to automatically generate fortran wrappers for the whole PETSc library. My current idea is therefore to basically re-use that infrastructure to generate automatic wrappers for all Julia functions, structs, enums etc.
   Ofcourse, as always, some hiccups occur along the way. In this case, because it seems that not all structs are listed in the python routine (private structs are missing). Moreover, some enums define several variables that point to the same numbers, which julia dislikes. The script that does this is here

As always my time is limited, so perhaps someone can have a look at this as well and can help moving it forward. In general I am quite excited to re-use the PETSc infrastructure itself to autogenerate bindings, because that will make this easy to maintain.

[boriskaus]

To follow up from above, the new wrapping method (#2) produces the correct routine:

"""
	KSPSetComputeOperators(ksp::KSP, func::KSPComputeOperatorsFn, ctx::void) 
"""
@for_petsc function KSPSetComputeOperators(::$UnionPetscLib, ksp::KSP, func::KSPComputeOperatorsFn, ctx::void)
    @chk ccall(
               (:KSPSetComputeOperators, $petsc_library),
               PetscErrorCode,
               (KSP, Ptr{KSPComputeOperatorsFn}, Ptr{void}),
               ksp, func, ctx,
              )
end 

this also nicely illustrates that the variable names are more descriptive (although void should be changed to Cvoid).