The following browser-chrome tests in "toolkit/content" show different behavior if they are run in a different order: * "browser_bug471962.js" (bug 471962 and bug 486342) * "browser_keyevents_during_autoscrolling.js" (bug 501608) The two tests succeed if they are run in the order above. At least on Windows XP, if I run "browser_keyevents_during_autoscrolling.js" alone (passing a --test-path to run_tests.py), or before "browser_bug471962.js" (that can be achieved by renaming the files), the former fails deterministically. If I run "browser_bug471962.js" after "browser_keyevents_during_autoscrolling.js", the former fails too. If I run "browser_bug471962.js" alone, it always succeeds.

I still can't work on this at present, but the following change found in the as-yet-unlanded patch in bug 501608, attachment 391901 [details] [diff] [review], looks promising: - // cleaning-up - gBrowser.loadURI("about:blank"); + // cleaning-up + gBrowser.addTab(); + gBrowser.removeCurrentTab(); Maybe the final loadURI call starts an asynchronous load that ends while the next test is running, causing it to fail if it attempts another load. It's just an hypothesis, though. This doesn't explain why the test fails if run alone, which can actually be a different problem than the failure of the next test.

Portions of Masayuki's patch from bug 501608, plus waiting for all the events to be dispatched after the clean-up phase, seem to solve the interdependency problem on my machine.

Paolo, is this still an issue?

(In reply to comment #3) > Paolo, is this still an issue? This is fixed by your changes, thanks! Out of curiosity, can you explain why calling stop() on the newly opened tab actually fixes the dependency issue? In "browser_bug471962.js", I now moved the code that performs the clean-up to a common location, so that tests that use the library don't need to worry about the implementation details. Do you think you can review the patch on bug 521831?

Comment on attachment 396191 [details] [diff] [review] The patch Previous comment seems to suggest this patch isn't needed any more.