Configure logic updates

[rhc54]

No description provided.

[ggouaillardet]

@rhc54 are you sure flex is needed to build PMIx ?

In the case of Open MPI, flex is checked at autogen.pl time, and the generated *.l files are included in the dist tarball (and so does PMIx, at least in 3.1.2 I just checked)

Long story short, developers do need flex (and they run autogen.pl that checks for it), and other folks (who build from tarballs) might not need it at all.

About C++ compilers, Open MPI still require them because we build the mpic++ wrapper that uses a C++ compiler. There is no such thing in pmix (but there is in prte).
I have no objection not requiring a C++ compiler any more (even if we might spot in advance something that could go wrong if the pmix library is to be used by a C++ linker).

[rhc54]

@rhc54 are you sure flex is needed to build PMIx ?

I cloned the repo, ran autogen.pl, ran configure, and then typed "make all" - and that's where I hit the error that "flex" wasn't found. Now it is possible that the tarball might not require it. I can perhaps spend a little more time looking at it again. This was on a virgin SLES15 system and was the first time I had ever actually had a system that didn't include flex - I'm willing to bet that nobody tests our build system to check any more.

About C++ compilers, Open MPI still require them because we build the mpic++ wrapper that uses a C++ compiler.

I understand why C++ is required by OMPI, but the problem here is that we had stale m4 code checking for C++ and Fortran pthread support. This failed for me because I didn't have a C++ compiler on the system - which is absurd. This change doesn't prevent someone from using a C++ compiler - it just removes the check for C++ thread support as we don't use it.

[ggouaillardet]

unlike what I wrote earlier, autogen.pl does not check for flex (and to my surprise, the same thing applies to Open MPI).
flex is used at compile time if the *.l files are not present (e.g. you are building from a git tree and not from a tarball).
So I guess the better fix is to test flex in autogen.pl (so you can build PMIx from a tarball on a virgin system that has no flex).

I am fine with the changes related to C++.

[rhc54]

I can find no mention of "flex" anywhere in autogen.pl - can you explain why you believe autogen.pl is supposed to run flex?

[ggouaillardet]

autogen.pl does not run flex, but since configure might not run it either (e.g. when invoked on a tarball) I believe the best place to check for flex is in autogen.pl. Sorry I did not make that clear earlier.

[rhc54]

I really prefer to let autoconf do the work, so I instead moved the check for flex so it is only done in a git directory tree. That should solve the problem, I think.

[ggouaillardet]

@rhc54 fair enough, I issued #1182 in order to support VPATH and git worktrees.
If/when merged, it should be backported into this PR.