TSGetConvergedReason is unusable: out-parameter wrapped as an input #247

TSGetConvergedReason can't be called usefully. The only method takes reason::TSConvergedReason by value and ccalls Ptr{TSConvergedReason} without allocating a Ref, so there's no way to get the value back:

julia> methods(PETSc.LibPETSc.TSGetConvergedReason)
# (petsclib, ts::TS, reason::TSConvergedReason)   <- the only one

The same function is wrapped correctly for the sibling solver classes. KSP_wrappers.jl:4066:

@for_petsc function KSPGetConvergedReason(petsclib::$UnionPetscLib, ksp::PetscKSP)
    reason = Ref{KSPConvergedReason}()
    @chk ccall((:KSPGetConvergedReason, $petsc_library), PetscErrorCode,
               (CKSP, Ptr{KSPConvergedReason}), ksp, reason)
    return reason[]
end

TS_wrappers.jl:3664:

@for_petsc function TSGetConvergedReason(petsclib::$UnionPetscLib, ts::TS, reason::TSConvergedReason)
    @chk ccall((:TSGetConvergedReason, $petsc_library), PetscErrorCode,
               (CTS, Ptr{TSConvergedReason}), ts, reason)
    return nothing
end

SNESGetConvergedReason has both forms, the 2-arg one that allocates the Ref and the 3-arg one that doesn't. TS only got the second. TSGetSolveTime, further down the same file, allocates its Ref correctly, so the generator seems to classify the out-parameter for some signatures and not others.

I hit this writing a TS wrapper and worked around it by deriving convergence from TSGetSolveTime and TSGetStepNumber, which can't tell a diverged step from a step-limit stop.

A crude scan for the same shape (a *Get* function whose last argument is passed as Ptr{T} with no Ref allocated, returning nothing) turns up around 247 candidates, DMGetGlobalSection and DMGetLocalToGlobalMapping among them. I haven't checked those individually, so that's a hint rather than a count.

Is wrapping/ the right place to fix this, or are the generated files edited directly?