Fix calls in caffeine.c

[ktras]

Closes #74.

[ktras]

The following two warnings still exist because when attempting to remove the warnings by dereferencing the user_op args to match what the functions seems to expect, then there is a runtime crash during the tests. These function calls involves macro expansions for both the functions and for the argument types, so I may have misread what was needed to fix the call and remove the warning. Or perhaps the call doesn't need fixing.

././src/caffeine/caffeine.c:142:103: warning: passing argument 9 of 'gasnete_tm_reduce_nb' from incompatible pointer type [-Wincompatible-pointer-types]
  142 |       team, result_image-1, a_address, a_address, GEX_DT_USER, c_sizeof_a, num_elements, GEX_OP_USER, user_op, &c_sizeof_a, 0
      |                                                                                                       ^~~~~~~
      |                                                                                                       |
      |                                                                                                       void (**)(const void *, void *, size_t,  const void *) {aka void (**)(const void *, void *, long unsigned int,  const void *)}
/Users/kateras/.local/include/gasnet_coll.h:400:60: note: in definition of macro 'gex_Coll_ReduceToOneNB'
  400 |         gasnete_tm_reduce_nb(tm,root,dst,src,dt,dts,dtc,op,fn,cdata,flags,0 GASNETI_THREAD_GET)
      |                                                            ^~
/Users/kateras/.local/include/gasnet_coll.h:396:58: note: expected 'gex_Coll_ReduceFn_t' {aka 'void (*)(const void *, void *, long unsigned int,  const void *)'} but argument is of type 'void (**)(const void *, void *, size_t,  const void *)' {aka 'void (**)(const void *, void *, long unsigned int,  const void *)'}
  396 |                        gex_OP_t _op, gex_Coll_ReduceFn_t _user_op, void * _user_cdata,
      |                                      ~~~~~~~~~~~~~~~~~~~~^~~~~~~~
././src/caffeine/caffeine.c:146:103: warning: passing argument 8 of 'gasnete_tm_reduce_all_nb' from incompatible pointer type [-Wincompatible-pointer-types]
  146 |       team,                 a_address, a_address, GEX_DT_USER, c_sizeof_a, num_elements, GEX_OP_USER, user_op, &c_sizeof_a, 0
      |                                                                                                       ^~~~~~~
      |                                                                                                       |
      |                                                                                                       void (**)(const void *, void *, size_t,  const void *) {aka void (**)(const void *, void *, long unsigned int,  const void *)}
/Users/kateras/.local/include/gasnet_coll.h:413:59: note: in definition of macro 'gex_Coll_ReduceToAllNB'
  413 |         gasnete_tm_reduce_all_nb(tm,dst,src,dt,dts,dtc,op,fn,cdata,flags,0 GASNETI_THREAD_GET)
      |                                                           ^~
/Users/kateras/.local/include/gasnet_coll.h:409:59: note: expected 'gex_Coll_ReduceFn_t' {aka 'void (*)(const void *, void *, long unsigned int,  const void *)'} but argument is of type 'void (**)(const void *, void *, size_t,  const void *)' {aka 'void (**)(const void *, void *, long unsigned int,  const void *)'}
  409 |                         gex_OP_t _op, gex_Coll_ReduceFn_t _user_op, void * _user_cdata,
      |                                       ~~~~~~~~~~~~~~~~~~~~^~~~~~~~

[bonachea]

The following two warnings still exist because when attempting to remove the warnings by dereferencing the user_op args to match what the functions seems to expect, then there is a runtime crash during the tests.

I suspect you deployed the wrong fix. This is a warning about mismatched types in code that is already believed to operate correctly. In such a situation adding or removing a deference operator to assuage the typechecker is almost never the right fix, because it changes the semantics of the code that already runs correctly (to something that usually does not).

I believe the actual problem here is with the type declarations, specifically the C declaration of the caf_co_reduce function. gex_Coll_ReduceFn_t is already a pointer-to-function type, and thus compatible with the c_funptr provided from the Fortran side of the call. The problem is the extraneous * in the C declaration of the user_op formal argument to caf_co_reduce.

[bonachea]

Code changes so far look right.

I agree there are several problems revealed in issue #74 that remain unaddressed so far. I'd say it's your choice whether to merge this incremental improvement as-is, or attempt to fix the rest of the warnings here.

On a related note, I'd like to suggest/request a convention where commits which fix or address issues include the issue number in the commit message. Right now the issue number appears only in the PR description, but years from now it will be much easier to understand the commit in isolation if the issue number also appears in the commit message.

[everythingfunctional]

Looks fine to me, but I'll defer to Dan on the intricacies of C type-casting.