testsuite: fixes for Lassen

[wihobbs]

Problem: Lassen doesn't like the words login, workflow, or infeasible, and it causes our CI spellcheck tests to break on that machine.

Add those words to the exempt word list in the docs.

Looking at previous issues, I thought that #3071 might be at play since LC_ALL=C is set on Lassen but not on TOSS 4 machines. Changing that to LC_ALL="" didn't solve the issue.

I also tried using SHARNESS_SKIP_TESTS=spellcheck command listed in the testsuite readme, but it still ran the test. Do we support that command? When I git grep'd for it in core I couldn't find anywhere in our shell scripts it's used. I did find SKIP_TESTS in t/sharness.sh but setting that variable to the same thing also didn't exempt the spellcheck test.

Also, just if anyone was curious ;) according to the OED "log in" or "log-in" are the official words. "workflow" and "infeasible" are also words.

[grondo]

I also tried usingSHARNESS_SKIP_TESTS=spellcheck command listed in the testsuite readme, but it still ran the test. Do we support that command? When I git grep'd for it in core I couldn't find anywhere in our shell scripts it's used. I did find SKIP_TESTS in t/sharness.sh but setting that variable to the same thing also didn't exempt the spellcheck test.

The problem is that the spellcheck test is not a sharness test, it is under doc and all sharness tests are under t.

The README should indeed reference SKIP_TESTS not SHARNESS_SKIP_TESTS, want to add a patch to this PR to change that?

[wihobbs]

Thanks @grondo for the workaround on the bit mapping/affinity tests. The testsuite runs in lassen on a compute node using make check. I'm looking into some failures on the runner now, although I think it might be a problem with lassen's hwloc, because the tests are taking a while to run like they used to on tioga... looking at it more.

[wihobbs]

All right, it's down to one consistent failure.

FAIL: t5000-valgrind.t 1 - valgrind reports no new errors on 2 broker run
ERROR: t5000-valgrind.t - exited with status 1

And some inconsistent failures in t2233. I'll see if these keep showing up.

One thing (not to take this offtrack, just noting to myself) is that HWLOC_COMPONENTS=x86 does not need to be set in Lassen's gitlab CI configuration. It was causing a lot of junk output, and the number of failing tests decreased with it removed.

[grondo]

Want to resolve conflicts here and rebase on current master?

[wihobbs]

Yeah, let's do that and get this closed. We can come back to t5000 later

[grondo]

Nice, thanks!

[wihobbs]

Ughhhhh, fixing

[wihobbs]

Thanks @grondo!

[codecov]

Codecov Report

Merging #5551 (9a85ddf) into master (41bc9ab) will increase coverage by 0.02%.
The diff coverage is n/a.

Additional details and impacted files

@@            Coverage Diff             @@
##           master    #5551      +/-   ##
==========================================
+ Coverage   83.42%   83.44%   +0.02%     
==========================================
  Files         487      487              
  Lines       82872    82872              
==========================================
+ Hits        69138    69156      +18     
+ Misses      13734    13716      -18     

see 10 files with indirect coverage changes