Created attachment 596512 [details] testcase (load about:memory to abort) ###!!! ABORT: bad nsCSSUnit: 'false', file layout/style/nsCSSValue.cpp, line 1247 History: * The memory reporter was added in bug 671299 part 3 * The turn unit was added in bug 716628
Created attachment 596514 [details] [diff] [review] patch
Looks like the fuzzer just used the value from https://hg.mozilla.org/mozilla-central/file/a096800c8f6b/layout/style/test/property_database.js#l1173 and then loaded about:memory.
Comment on attachment 596514 [details] [diff] [review] patch Yeah, looks like this code appeared after I wrote the patch but before I landed it... and we had no automated tests that would catch this. :( r=me
> Yeah, looks like this code appeared after I wrote the patch but before I > landed it... and we had no automated tests that would catch this. :( Bug 699307 is open for improving testing of memory reporters. It has an r- patch that runs all the memory reporters on a blank tab, but I guess it would be much better to load some page that does a ton of stuff first. Any suggestions for such a page? I assume accessing random internet sites is not good in tests, i.e. the pages should be local?
The test machines black-hole external IPs, as I recal. So if you check in a test that touches random stuff on the internet, it should always go orange with a test timeout. So yes, test pages should be local....
Could you run all the memory reporters after each mochitest?
(In reply to Jesse Ruderman from comment #8) > Could you run all the memory reporters after each mochitest? I don't know how to do that... it sounds expensive -- it'd probably add a second or so to every mochitest.