make instance name dynamic array

[Dupratj]

make the i_name element of the instance structure a dynamic
element. This change was suggested by hppritcha here
This allows us to keep the size of PREDEFINED_INSTANCE_PAD to 512 despite adding to the instance structure. Keeping it to 512 make it so we don't introduce breaking change.

[github-actions]

Hello! The Git Commit Checker CI bot found a few problems with this PR:

ed287bf: Dynamic instance name

• check_signed_off: does not contain a valid Signed-off-by line

Please fix these problems and, if necessary, force-push new commits back up to the PR branch. Thanks!

[bosilca]

Why are you mentioning PREDEFINED_COMMUNICATOR_PAD while talking about an instance ?

[Dupratj]

Why are you mentioning PREDEFINED_COMMUNICATOR_PAD while talking about an instance ?

My bad wrong variable, I meant PREDEFINED_INSTANCE_PAD but switched it up

[bosilca]

Digging a little further into this, I wonder if it is really necessary for the instance. The padding is necessary for every other case because we do have valid predefined MPI objects of that type that can be used. But for the
MPI instances, the only predefined MPI object is ompi_mpi_instance_null and this object is special and cannot be used as a valid MPI object.

[Dupratj]

I've dug into it a bit more and it seems you're right about it not being a problem during runtime.
The problem identified in the other pull request was for backward compatibility. Every program compiled on a version previous to the 13667 PR would need to recompiled on a version posterior to the 13667 PR.
This PR would allow the fix to be in 5.x, we have a similar fix working for the 4.x.

If we really think this is unnecessary I can push this PR only in 5.x and increase the PREDEFINED_INSTANCE_PAD in the PR 13667 when targeting the 6.x only. Such change would be less of a problem if we jump to another major version.

[Dupratj]

@bosilca what do you think ?

[Dupratj]

I've dug into it a bit more and it seems you're right about it not being a problem during runtime. The problem identified in the other pull request was for backward compatibility. Every program compiled on a version previous to the 13667 PR would need to recompiled on a version posterior to the 13667 PR. This PR would allow the fix to be in 5.x, we have a similar fix working for the 4.x.

If we really think this is unnecessary I can push this PR only in 5.x and increase the PREDEFINED_INSTANCE_PAD in the PR 13667 when targeting the 6.x only. Such change would be less of a problem if we jump to another major version.

What do you think @bosilca ? Maybe @hppritcha can also weigh in as he suggested the change in this PR.
I cannot advance this further for now.