Chapel: Add more smoke testing

[ronawho]

We currently have linux64, linux32, osx, and intel-whitebox post-commit smoke testing. It'd be beneficial to add smoke tests for:

• [x] llvm
• [x] linux64 gasnet
• [x] cygwin? (punted based on https://github.com/chapel-lang/chapel/issues/5484#issuecomment-320768168)
• [x] other whitebox configurations? (currently just test icc) (punting based on low-priority/impact)
• [ ] cray module build
• [ ] valgrind

[ronawho]

The llvm config seems like the most important once since there's a lot of nightly configs that test llvm.

Seems like this should be trivial to add at least for llvm and gasnet. @awallace-cray would it be easy to add llvm and gasnet configurations to the smoketest, or is it difficult for some reason?

[awallace-cray]

Only problems would be competition for machines and/or waiting for results. Suggest we add those two configs w "hellos" only, run them on the stronger machines, and see what happens.

[ronawho]

The smoke test only runs hello6-taskpar-dist

[ronawho]

Added llvm and gasnet testing.

I think we should punt on cygwin testing. Anything with cygwin is a pain to setup, won't generate much noise if nightly testing fails, and at least in the past year hasn't broken in a way that regular linux64 testing hasn't caught.

It might be worth adding a cce whitebox config, but I don't think that's very pressing since that also won't generate much mail.

[ronawho]

Should we add valgrind smoke testing too?

make check (what we use as part of smoke testing) doesn't have a way to trigger valgind and I'm not sure we want to add that so we'd probably have to expand our internal smoke test to run the hellos under start_test (which might not be a bad idea anyways)

[mppf]

A recent experience indicates it'd be useful to have smoke tests for --no-local and/or gasnet run more tests (e.g. all of release/examples/primers). Arguably this could apply to all smoke tests.

[mppf]

A recent experience indicates it might be worthwhile building mason in the smoke test.