Re: [PATCH] travis-ci: run previously failed tests first, then slowest to fastest

From: Jeff King <peff@peff.net>
Date: Jan 22, 2016, 05:52 UTC
Message-ID: <20160122055255.GA14657@sigill.intra.peff.net>
In-Reply-To: <20160122023359.GA686558@vauxhall.crustytoothpaste.net>

On Fri, Jan 22, 2016 at 02:33:59AM +0000, brian m. carlson wrote:

Show 12 quoted lines

> > Especially t0025-crlf-auto.sh failed multiple times ([2], [3]) on the OS X builds
> > when I increase the parallelism:
> > 
> > not ok 4 - text=true causes a CRLF file to be normalized
> > not ok 9 - text=auto, autocrlf=true _does_ normalize CRLF files
> > 
> > Anyone an idea why that might be the case?
> 
> I've seen this on my personal box too[0] when running make -j4 all test.
> I wasn't able to pin down why it was occurring, but if we're going to
> run the tests in parallel, it's probably worth spending some time
> figuring it out.

Interesting. I run the test suite in parallel probably a dozen times per day, and I've never seen this. However, I was able to trigger it eventually with:

  for i in 1 2 3 4 5 6 7 8
  do
    (while ./t0025-crlf-auto.sh --root=/var/ram/git-tests/foo-$i -v -i >/tmp/foo-$i 2>&1
    do
      : nothing
    done
    echo FAILED $i
    ) &
  done

I get a few of the threads failing (in test 4) after 2-3 minutes. The "-v" output is pretty unenlightening, though. I don't see anything racy-looking in the test unless it is something with "read-tree" and stat mtimes.

-Peff