Lustre configure fixes

[jsquyres]

@edgargabriel This PR is a few commits to update/fix Lustre configure.m4:

1. Ensure that --with-lustre doesn't cause a compile fail (per "../configure --with-lustre" causes build to fail with "yes/lib" error #3447)
2. Ensure that if the user does --with-lustre and we can't find Lustre support, abort
3. Remove some dead code
4. Trivial style updates

Fixes #3447.

Can you review? If it's good, I'll PR to the various 2.x/3.x branches.

[jsquyres]

Yo @bwbarrett: I see in https://jenkins.open-mpi.org/jenkins/job/open-mpi.build.platforms/130/console that it says

open-mpi.build.platforms » amazon_linux_17.03 completed with result FAILURE

How do I see the console/output of that job to see what failed?

[edgargabriel]

@jsquyres thanks for doing that. I will test it out and let you know.

[bwbarrett]

@jsquyres, don't look at the console for the parameter sweep job, but for the individual builds. So in this case, you'd click on "Details" in github, then "open-mpi.build.platforms" in Jenkins (because it has a red ball), then "amazon_linux_17.03" (because it has a red ball), then click on the console. And you'd see:

--> running examples
--> running C examples
Hello, world, I am 0 of 2, (Open MPI v4.0.0a1, package: Open MPI ec2-user@ip-172-31-15-133 Distribution, ident: 4.0.0a1, repo rev: v2.x-dev-4091-geb03679, Unreleased developer copy, 153)
Hello, world, I am 1 of 2, (Open MPI v4.0.0a1, package: Open MPI ec2-user@ip-172-31-15-133 Distribution, ident: 4.0.0a1, repo rev: v2.x-dev-4091-geb03679, Unreleased developer copy, 153)
open-mpi-build-script.sh: line 218: 18466 Killed                  ${exec} ./examples/hello_c
Build step 'Execute shell' marked build as failure
Finished: FAILURE

which means /bin/timeout killed mpirun because it took longer than 2 minutes. This has been a transient problem in the 3.x series, so I doubt it's this patch, but it is something we need to sort out.

[bwbarrett]

bot:ompi:retest

[edgargabriel]

I think it generally works. The only little issue that I found was that it aborts correctly if --with-lustre options without a directory is given , and it couldn't find the lustre headers. If you use however --with-lustre= ( and then nothing), it does not abort, but continues and just says that it did not find lustre support.

--- MCA component fs:lustre (m4 configuration macro)
checking for MCA component fs:lustre compile mode... dso
checking --with-lustre value... simple ok (unspecified)
looking for header without includes
checking lustre/liblustreapi.h usability... no
checking lustre/liblustreapi.h presence... no
checking for lustre/liblustreapi.h... no
checking lustre/liblustreapi.h usability... no
checking lustre/liblustreapi.h presence... no
checking for lustre/liblustreapi.h... no
checking for required lustre data structures... no
checking if MCA component fs:lustre can compile... no

[jsquyres]

@edgargabriel I checked this over the weekend. If you do --with-lustre=, then $with_luster is empty (i.e., no value). So there's no way for us to know that you did that (i.e., it doesn't fill in the same yes value that it does if you --with-lustre without the =). So I think this behavior of --with-lustre= not catching that Lustre support failed is ok.