Re: skipping parallel autotest tests

[Paolo Bonzini]

Eric Blake wrote:
> I'm interested in getting a beta-quality 2.64 release out the door soon.
> There are still known issues (primarily coverage of parallel autotest, but
> also a growing backlog of suggested improvements in my inbox) that prevent
> this from being a stable release.

Can you summarize these improvements?  Otherwise, parallel autotest can
be deemed experimental and not prevent the release to be considered stable.

> But there has been enough of a lull in
> fast-and-furious development lately that I think we would benefit from the
> extra exposure of having a release available for wider testing.
> 
> I'm thinking of committing this patch to counteract the fact that Ralf
> (and others) have still not had enough time to hammer out whether it is
> possible to work around dash and ksh bugs when it comes to parallel
> testsuites.  The intent here is that you can still test these shells, by
> adding TEST_PARALLEL_AUTOTEST to your environment, but that default
> out-of-the-box testing will skip rather than risk hanging or dropping a
> fork bomb on insufficiently tested platforms.  I tested that
> TESTSUITEFLAGS='-k parallel' skipped all these tests on OpenBSD's sh until
> I added the extra envvar, but that TESTSUITEFLAGS='-k autotest' still ran
> and passed all tests under bash without any extra effort.

All good ideas in here, though.

Paolo