----- Original Message ----- > On 6/17/2014 2:22 PM, Philipp Kewisch wrote: > > Hey Kent, > > > > reading bug 995417, the crash only happens if > > MOZ_DISABLE_NONLOCAL_CONNECTIONS is set, which is set only in the > > testing infrastructure. > > Which is actually not ideal because it means we can end up with failures > on TBPL that aren't hit during the course of normal local testing.
If we do, that is a bug; MOZ_DISABLE_NONLOCAL_CONNECTIONS is setup in the Python modules that both TBPL and local testing use. So there shouldn't be any problems with locally-unreproducible failures. A related problem is that MOZ_DISABLE_NONLOCAL_CONNECTIONS may or may not be set for all the testsuites we run. It's not set for Talos, for instance, and perhaps the C++ unittests and a few others. And there's no guarantee that a new testsuite will have MOZ_DISABLE_NONLOCAL_CONNECTIONS set, which is an unfortunate loophole (though if they use common infrastructure, the chances go up signficantly). It's a reasonable fix given the constraints, but it's not a perfect one. -Nathan _______________________________________________ dev-platform mailing list dev-platform@lists.mozilla.org https://lists.mozilla.org/listinfo/dev-platform
