Relevant discussion in #799. Seems Travis is broken on my fork, so submitting this to trigger an official build.
Ben, did you manage to get this working?
Run valgrind on Travis CI
@leejarvis Looks good-ish now. Some failures due to slow XPath, but I can't do much about that here.
Thanks, looks good. However I'm not sure this is worth it if we're going to constantly get these kind of failures. I think if something needs debugging we can run valgrind on local machines to avoid the long waits and failures. I'm open to suggestions, though.
The failures are in the rspec test suite.
Right! I misread the output, I'll check this out tomorrow. Cheers!
@benlangfeld and @leejarvis -- I'd love to get valgrind running on Travis, except that the only assertion I'm currently making on the valgrind run is with my eyeballs.
So an interesting prerequisite is to wire up an assertion of some sort into https://github.com/jbarnette/hoe-debugging, so that the build fails if unexpected output is emitted. Difficulty level: medium.
Also, it would be great if we could with- and without-valgrind as separate builds. I know we can do this, because I'm looking at jruby's travis config and jruby's build and they're clearly doing something amazing with the matrix bit that I can't quite grok because I don't fully understand the fully operational battle station that is TravisCI in 2015. Difficulty level: medium for nitwits like me, probably low for both of you.
I still think this would be interesting to do. Happy to chat about my opinions if someone has time on their hands.
Specifically, I think it would also be great to add the test:libxml-ruby and test:valgrind:libxml-ruby suites to Travis.