I've looked at the possibly-spurious single test failures on the queue for
the past week. There's no obvious single test causing trouble. One time it
was a data-cleaning test that failed with an HTTP error, probably due to
the webservice being weird; that's annoying, but if it only happened once
we should probably let it slide. Twice it was swig_xqj segfaulting, and
that was for the same merge proposal so I'm assuming it wasn't "spurious"
at all. The other three that I found also seemed either completely random
(cosmic rays?) or possibly indicative of a real problem with the merge
proposal.
For the time being, there's not anything obvious to do. If you have a test
run that fails with one or two test failures that seem completely unrelated
to your changes, please do two things:
1. Re-submit your changes by marking the proposal "Approved" again.
2. Email the CDash page for the failing test run to me - and let me know if
the second test run with the same changes does NOT have that test fail.
If any test proves to be unreliable repeatedly, I'll disable it.
Ceej
aka Chris Hillery