Bind(C) prif_coarray_handle

[bonachea]

Implements https://github.com/BerkeleyLab/prif/issues/157 , proposals 2, 2.1. 2.2, 2.3

[bonachea]

@certik This branch contains some new development in Caffeine for a forthcoming feature.

The latest commit (028cfd7) currently appears to be ICE'ing LFortran 0.62 and LFortran:recent on Linux, despite passing on all supported versions of gfortran and flang.

How can we narrow this down further?

[certik]

@bonachea Let me have a look to isolate an MRE, then we can fix it. I think this is unrelated to lfortran/lfortran#11175 ?

I can reproduce the failure. I am creating an MRE.

Here is the MRE:

program mre
  implicit none
  type :: t
    integer :: x
  end type
  type(t), pointer :: src, dest
  allocate(src)
  src%x = 1
  allocate(dest, source=src)
  if (dest%x /= 1) error stop "FAIL"
  print *, "PASS"
  deallocate(src)
  deallocate(dest)
end program

Which gives:

~/repos/caffeine(build4-mre)$ lfortran mre_allocate_source_ptr.f90
LCOMPILERS_ASSERT failed: /Users/ondrej/repos/lfortran/src/libasr/codegen/llvm_utils.cpp
function create_gep2(), line number 1774 at
ds_pointee_type == t
ERROR MESSAGE:
Type mismatch in create_gep2: GEP target type does not match pointer's pointee type. This would cause crashes in LLVM <= 8 constant folder. Target type: %mre.t = type { i32 }, Pointer pointee type: %mre.t*
Traceback (most recent call last):
...

I am fixing it.

[certik]

There were a few more bugs, so I fixed them all. With this PR lfortran/lfortran#11179 the bind-c-handle branch builds and passes all tests:

_____ 78 of 80 tests passed. 2 tests were skipped _____

I also tested the old main (b8dd1f9) and that also works with the PR:

_____ 78 of 80 tests passed. 2 tests were skipped _____

I also tested the latest main with your main, and that also failed, so I think we didn't have enough integration tests to keep it running. I'll see if I can set it up at our CI to test Caffeine with every commit.

[certik]

@bonachea ok, so lfortran/lfortran#11181 is in, the latest main in LFortran now works.

I have some more fixes / cleanup on top, but that shouldn't be a blocker for you.

You can also remove 028cfd7 which now works.

[bonachea]

@bonachea ok, so lfortran/lfortran#11181 is in, the latest main in LFortran now works.

I have some more fixes / cleanup on top, but that shouldn't be a blocker for you.

Thanks for the amazingly quick turnaround on these fixes @certik !!

I've re-run the CI for this PR against lfortran:latest and confirm that it's working!

This PR will ideally merge to caffeine:main in the next few weeks, and since LFortran 0.62 is no longer sufficient I'd prefer to simultaneously update our CI from LFortran 0.62 to a new stable release tag. What's your timeframe for the next LFortran release?

[certik]

I can release it any time you want. I am trying to get caffeine tested at our CI: lfortran/lfortran#11190, to ensure it is reproducible (but since it already works for you, I think we have it) and to prevent future breakages. Then we can make a release.

[bonachea]

Superseded by BerkeleyLab#320