check for negative ranks in ompi_win_peer_invalid

[jeffhammond]

resolves #3326 (#3326)

[jeffhammond]

@rhc54 If there is an Intel-specific process for contributing to Open-MPI, please let me know and I will follow it.

[bwbarrett]

Wow, that's an old bug... and likely my fault. sigh.

[jeffhammond]

I should have included a test for this. I don't know as much about Open-MPI testing infrastructure. I will try to look it up later but it is likely to fall below ZBB line for me for the foreseeable future.

[ggouaillardet]

@jeffhammond thanks for the PR and the test case, i will merge it and PR to the branches.

our tests are in a private repository, and these tests usually fit in the ibm test suite, i will add it there.
if you want access to this repository, please contact @jsquyres offline.

[ggouaillardet]

@jeffhammond Open MPI moved to Signed-off-by contributions.
basically, that means that when you sign off a commit, you know what you are doing and among other things you already checked this is allowed (or not prohibited, IANAL) by your employer

[jeffhammond]

@ggouaillardet I can't afford to look at proprietary MPI test code given all the open-source MPI work I do but I am very happy that you were willing to adapt my test into the Open-MPI infrastructure.

[ggouaillardet]

@jeffhammond fwiw, this is not proprietary MPI test code strictly speaking. basically, we do not know if we are able to redistribute all our tests repositories (we are not lawyers), that is why the repository is private, and we are generally happy to give access to anyone commited not to redistribute it.