In order to turn on e10s mochitests on for OSX, we have to figure out why the apz tests aren't passing. As far as I can tell, there's a problem with scroll events on OSX, but only in iframes. When I ran the tests in this directory by themselves, they passed; it's only in the frame, while running the directory that they fail.

We should use this bug to track turning the tests back on.

Do you have the output from a run of the tests, or a try push showing the failures? When I run this locally on my OS X machine it passes: mach mochitest -f plain --e10s gfx/layers/apz/test/ I'm not opposed to disabling the tests temporarily but I'd like to have some idea of what the problem is at least.

Comment on attachment 8647820 [details] [diff] [review] Disable the tests Clearing review for now.

I'll try again and debug a little. It looks like it'll be a bit before we can get these tests enabled on treeherder, so no rush.

Ok, thanks. Also FYI I would expect test_layerization.html might fail, but that should be fixed in bug 1177018. The rest of the tests should be passing.

Hijacking this bug as per IRC discussion yesterday to investigate why the gfx/layers/apz/test/mochitest/test_bug1151667.html is failing with e10s on windows. The try push showing the failure is at https://treeherder.mozilla.org/#/jobs?repo=try&revision=6fe342dcb884 I wasn't able to reproduce locally (I'll try harder) but in the meantime I did a try push with more logging: https://treeherder.mozilla.org/#/jobs?repo=try&revision=cc85eb2665b8

The try push I did above still hasn't been scheduled, and from trychooser it looks like the test queue on windows machines is climbing towards infinity... :(

The jobs finally finished; I looked at the logs and it appears to be the same problem as in bug 1203901 (i.e. the layer tree on the compositor side is not fully updated before we start sending the wheel events). I'll do a try push with a fix.

Screwed up that try push, here's a better one: https://treeherder.mozilla.org/#/jobs?repo=try&revision=3ad9682fa544

That try push has test_bug1151667.html passing, it seems. However test_layerization is suffering from the same problem. I'll fix that one too (and any others...) while I'm at it.

Thanks for figuring this out!

Might as well get this landed. I'm still debugging the test_wheel_transactions failures which seem to have a different cause. (https://treeherder.mozilla.org/#/jobs?repo=try&revision=970f97a1e94e&group_state=expanded)

What seems to be happening in test_wheel_transaction.html is that near the end of the test there are two loops with wheel events, and a timeout in between. The intent is that the timeout ends the transaction so the second set of wheel events end up going to the outer frame. Instead it looks like the last event from the first loop is delayed a tiny bit, and starts a new transaction. The first event from the second loop ends up getting added to that transaction, and goes to the inner APZC instead of the outer APZC, never triggers a scroll event, and causes the test to hang.

(In reply to Kartikaya Gupta (email:kats@mozilla.com) from comment #14) > What seems to be happening in test_wheel_transaction.html is that near the > end of the test there are two loops with wheel events, and a timeout in > between. The intent is that the timeout ends the transaction so the second > set of wheel events end up going to the outer frame. Instead it looks like > the last event from the first loop is delayed a tiny bit, and starts a new > transaction. The first event from the second loop ends up getting added to > that transaction, and goes to the inner APZC instead of the outer APZC, > never triggers a scroll event, and causes the test to hang. I wonder if the problem is that synthesizeNativeWheelAndWaitForScrollEvent() only waits for the _first_ scroll event triggered by the wheel event, while the delayed one is a subsequent one (and thus the first loop doesn't wait for it). It would be nice to wait for all scroll events triggered by a wheel event, I'm just not sure how to do that.

A bit of discussion on IRC led us to conclude that this is most likely a timestamp resolution problem; otherwise it should be impossible for those two events to end up in the same transaction. Increasing the timeout seems to fix it, per https://treeherder.mozilla.org/#/jobs?repo=try&revision=506cd6278b0a&group_state=expanded

Comment on attachment 8682549 [details] [diff] [review] Patch for test_bug1151667 and test_layerization Review of attachment 8682549 [details] [diff] [review]: ----------------------------------------------------------------- ::: gfx/layers/apz/test/mochitest/test_layerization.html @@ +7,5 @@ > <title>Test for layerization</title> > <script type="application/javascript" src="/tests/SimpleTest/SimpleTest.js"></script> > + <script type="application/javascript" src="/tests/SimpleTest/EventUtils.js"></script> > + <script type="application/javascript" src="/tests/SimpleTest/paint_listener.js"></script> > + <script type="application/javascript" src="apz_test_utils.js"></script> Did alpha ordering go out of vogue? :p

Comment on attachment 8683202 [details] [diff] [review] Patch for test_wheel_transaction Review of attachment 8683202 [details] [diff] [review]: ----------------------------------------------------------------- ::: gfx/layers/apz/test/mochitest/test_wheel_transactions.html @@ +82,5 @@ > // Scroll up a bit more. It's still |outer| scrolling because > // |inner| is still scrolled all the way to the top. > yield scrollWheelOver(outer, 10); > > // Wait for the transaction timeout to elapse. // timeout * 5 is used to make it less likely that the timeout is less // than the system timestamp resolution

(In reply to Botond Ballo [:botond] from comment #17) > > + <script type="application/javascript" src="apz_test_utils.js"></script> > > Did alpha ordering go out of vogue? :p Whoops, fixed :) (In reply to Botond Ballo [:botond] from comment #18) > // timeout * 5 is used to make it less likely that the timeout is less > // than the system timestamp resolution Added (to both sites, for consistency).

remote: https://hg.mozilla.org/integration/mozilla-inbound/rev/a2e622ecf00f remote: https://hg.mozilla.org/integration/mozilla-inbound/rev/dc4dd8cbc003

https://hg.mozilla.org/mozilla-central/rev/a2e622ecf00f https://hg.mozilla.org/mozilla-central/rev/dc4dd8cbc003