Bug 33342 – http/tests/xmlhttprequest/event-listener-gc.html is flaky

• Status: RESOLVED
• Resolution: FIXED
• Component: Tools / Tests

[Eric Seidel (no email)]

http/tests/xmlhttprequest/event-listener-gc.html failed on GTK Linux 64-bit Debug bot

http://build.webkit.org/results/GTK%20Linux%2064-bit%20Debug/r52942%20(2212)/results.html

--- layout-test-results/http/tests/xmlhttprequest/event-listener-gc-expected.txt	2010-01-07 13:01:07.450743702 -0800
+++ layout-test-results/http/tests/xmlhttprequest/event-listener-gc-actual.txt	2010-01-07 13:01:07.450743702 -0800
@@ -4,7 +4,4 @@
 
 If the test passes, you'll see a series of 'PASS' messages below.
 
-PASS: event handler fired after garbage collection.
-PASS: event handler fired after garbage collection.
-PASS: event handler fired after garbage collection.

[Eric Seidel (no email)]

Another failure, just now.  Same bot:
http://build.webkit.org/results/GTK%20Linux%2064-bit%20Debug/r53052%20(2272)/http/tests/xmlhttprequest/event-listener-gc-diffs.txt

[Eric Seidel (no email)]

Two in a row, actually:
http://build.webkit.org/results/GTK%20Linux%2064-bit%20Debug/r53053%20(2274)/http/tests/xmlhttprequest/event-listener-gc-diffs.txt

[Csaba Osztrogonác]

Fail again, but on Qt bot:

http://build.webkit.org/results/Qt%20Linux%20Release/r53268%20%285990%29/results.html

[Eric Seidel (no email)]

Another failure:
http://build.webkit.org/results/GTK%20Linux%2064-bit%20Debug/r53335%20(2435)/http/tests/xmlhttprequest/event-listener-gc-diffs.txt

This test was last touched by Sinchiro 4 months ago, and looks to be originally from Geoff.
http://trac.webkit.org/browser/trunk/LayoutTests/http/tests/xmlhttprequest/event-listener-gc.html

I'm not sure I really understand the failure condition.  It's possible that sometimes these listeners are getting incorrectly destroyed?

[Alexey Proskuryakov]

The test still fails in the same way, and on Mac too.

[Alexey Proskuryakov]

Created attachment 245960 [details]
proposed fix

[WebKit Commit Bot]

Comment on attachment 245960 [details]
proposed fix

Clearing flags on attachment: 245960

Committed r179572: <http://trac.webkit.org/changeset/179572>

[WebKit Commit Bot]

All reviewed patches have been landed.  Closing bug.

[Alexey Proskuryakov]

That didn't fix it.

[Alexey Proskuryakov]

Created attachment 246091 [details]
second attempt

[WebKit Commit Bot]

Comment on attachment 246091 [details]
second attempt

Clearing flags on attachment: 246091

Committed r179696: <http://trac.webkit.org/changeset/179696>

[WebKit Commit Bot]

All reviewed patches have been landed.  Closing bug.

[Alexey Proskuryakov]

Aaand.. That also didn't fix it!

[Alexey Proskuryakov]

Tweaked test output a little bit in <http://trac.webkit.org/r179749>, hoping that this will shed some light on the issue.

[Alexey Proskuryakov]

Created attachment 246216 [details]
third attempt

[Darin Adler]

Comment on attachment 246216 [details]
third attempt

As more and more tests depend on this, we should probably add a way to force layout to internals, and then only use "document.body.offsetTop" when outside the test runners.

[WebKit Commit Bot]

Comment on attachment 246216 [details]
third attempt

Clearing flags on attachment: 246216

Committed r179789: <http://trac.webkit.org/changeset/179789>

[WebKit Commit Bot]

All reviewed patches have been landed.  Closing bug.

[Alexey Proskuryakov]

It still fails :)

[Alexey Proskuryakov]

Looks like even with the forced layout, there is enough work on the main thread that has higher priority than the timer.

I could finally reproduce locally, and have a patch that fixes that, but want to do a bit more investigation into why timers are so low priority.

[Alexey Proskuryakov]

Created attachment 246507 [details]
fourth attempt

I couldn't find any relevant mistakes in how timers function. The work that delays our zero delay timer is from run loop observers that are serviced earlier, and IPC messages are also serviced earlier.

So, let's just stop using a timer here.

[Darin Adler]

Comment on attachment 246507 [details]
fourth attempt

View in context: https://bugs.webkit.org/attachment.cgi?id=246507&action=review

> LayoutTests/http/tests/xmlhttprequest/event-listener-gc.html:41
> +window.addEventListener("load", test);
> +window.addEventListener("load", collect);

No need for "window." on these lines of code.

[WebKit Commit Bot]

Comment on attachment 246507 [details]
fourth attempt

Clearing flags on attachment: 246507

Committed r180105: <http://trac.webkit.org/changeset/180105>

[WebKit Commit Bot]

All reviewed patches have been landed.  Closing bug.