usnic: use fi_getname in newer libfabric

[goodell]

When using an external libfabric (or really any libfabric newer than
libfabric commit 607e863), we must use fi_getname to determine the local
port of our endpoint.

@jsquyres please review (and propagate to other OMPI branches/instances as needed)

[mellanox-github]

• Coverity found 882 errors for all: http://bgate.mellanox.com:8888/jenkins/job/gh-ompi-master-pr//ws/cov_build/all_537/output/errors/index.html
• Coverity found 5 errors for oshmem: http://bgate.mellanox.com:8888/jenkins/job/gh-ompi-master-pr//ws/cov_build/oshmem_537/output/errors/index.html
• Coverity found 2 errors for fca: http://bgate.mellanox.com:8888/jenkins/job/gh-ompi-master-pr//ws/cov_build/fca_537/output/errors/index.html
  
  Sent from GH.

[mellanox-github]

Refer to this link for build results (access rights to CI server needed):
http://bgate.mellanox.com/jenkins/job/gh-ompi-master-pr/537/
Test PASSed.

[jsquyres]

Doh -- sadness. When I run this with libfabric master, I get an assert fail in your new code:

ring_c: btl_usnic_module.c:1514: create_ep: Assertion `FI_SOCKADDR_IN == channel->info->addr_format' failed.

Works fine with libfabric 1.0.0, though. Sorry -- I don't have enough brain power to track this down ATM...

[goodell]

strange... I thought I tested all cases with the latest version of the code, but maybe I missed one. Let me see if I can reproduce.

[goodell]

I wasn't building with --enable-debug, so my assertions weren't enabled when I did my testing. I've pushed a fixed version which knows how to handle libfabric returning either FI_SOCKADDR or FI_SOCKADDR_IN.

[mellanox-github]

• Coverity found 880 errors for all: http://bgate.mellanox.com:8888/jenkins/job/gh-ompi-master-pr//ws/cov_build/all_539/output/errors/index.html
• Coverity found 5 errors for oshmem: http://bgate.mellanox.com:8888/jenkins/job/gh-ompi-master-pr//ws/cov_build/oshmem_539/output/errors/index.html
• Coverity found 2 errors for fca: http://bgate.mellanox.com:8888/jenkins/job/gh-ompi-master-pr//ws/cov_build/fca_539/output/errors/index.html
  
  Sent from GH.

[mellanox-github]

Refer to this link for build results (access rights to CI server needed):
http://bgate.mellanox.com/jenkins/job/gh-ompi-master-pr/539/
Test PASSed.

[jsquyres]

Per the assert right above it, does this "if" need to check for FI_SOCKADDR or FI_SOCKADDR_IN?

[jsquyres]

Disregard my previous comment... I see that the purpose of this "if" is mainly just another assert.

However, it'll generate a CID / compiler warning in non-debug builds, because sa will be assigned and not used, right?

[goodell]

However, it'll generate a CID / compiler warning in non-debug builds, because sa will be assigned and not used, right?

I suppose. What's the usual fix for that specific set of warnings? I rather hate sticking more code in just to make various static analysis tools happy when there's no actual problem.

I could also do it a bit less cleanly and exploit the fact that I know struct sockaddr and struct sockaddr_in must use the same layout for sa_family/sin_family. Then I don't need sa at all and I just make the assertion on sin->sin_family instead.

[jsquyres]

You could also just surround the if block with #if OPAL_ENABLE_DEBUG...?