Fix/nb leaks

[bosilca]

A possible fix for #8763

[bosilca]

We need to converge on this and once we reached a consensus we need to apply the same logic to basically all the non-blocking Fortran functions. This includes:

• ineighbor_alltoallw_f.c,
• ineighbor_alltoallv_f.c,
• neighbor_allgatherv_init_f.c,
• neighbor_alltoallw_init_f.c,
• neighbor_alltoallv_init_f.c,
• ineighbor_allgatherv_f.c,
• igatherv_f.c,
• iallgatherv_f.c,
• iscatterv_f.c,
• ialltoallv_f.c,
• ialltoallw_f.c,
• ireduce_scatter_f.c,
• reduce_scatter_init_f.c,
• alltoallv_init_f.c,
• alltoallw_init_f.c,
• allgatherv_init_f.c,
• gatherv_init_f.c,
• scatterv_init_f.c.

[devreal]

LGTM, will approve once the other functions are in 👍

[devreal]

A small formatting nitpick, looks good otherwise.

[gpaulsen]

Looks like Mellanox build failed with two compile errors:

                        ^
coll_hcoll_rte.c: In function 'get_coll_handle':
coll_hcoll_rte.c:355:19: error: 'struct <anonymous>' has no member named 'objs'
     ompi_req->data.objs.objs[0]          = NULL;
                   ^
coll_hcoll_rte.c:356:19: error: 'struct <anonymous>' has no member named 'objs'
     ompi_req->data.objs.objs[1]          = NULL;
                   ^

[bosilca]

I don't have hcoll on mu clusters, but I think I pushed a fix for this compile error. The PR is ready for final review and them merge. @ggouaillardet and @devreal please one last review, maybe we can make it in the pending releases.

[gpaulsen]

/azp run

[azure-pipelines]

Azure Pipelines successfully started running 1 pipeline(s).

[gpaulsen]

@ggouaillardet @devreal can you please re-review?

[gpaulsen]

@devreal @ggouaillardet ?

[gpaulsen]

/azp run

[azure-pipelines]

Azure Pipelines successfully started running 1 pipeline(s).

[awlauria]

@bosilca I fixed the compile error. Does that look right?

@janjust ^^ this touches hcoll.

[awlauria]

If that's right can you squash and rebase on current main?

[awlauria]

Closing for #10288

It's a squash of the two commits here and a rebase on top of main.