https://gcc.gnu.org/bugzilla/show_bug.cgi?id=127807
--- Comment #4 from Sam James <sjames at gcc dot gnu.org> --- I should say: the reason I assume(d) there was a problem here, is that I only had 4 running when there were many more tokens available and quite a lot of work left to do. But maybe that was a bogus assumption, as the machine only has 32 tokens available and it looks like the fallback logic at least tries to do something here (but maxes out unless GCC_TEST_PARALLEL_SLOTS is set), though that doesn't help for Ada or libgomp.
