fullscreen/full-screen-plugin.html is very flaky on Yosemite WK2

[Alexey Proskuryakov]

fullscreen/full-screen-plugin.html fails on Yosemite WK2 about 50% of the time: <http://webkit-test-results.appspot.com/dashboards/flakiness_dashboard.html#showAllRuns=true&showExpectations=true&revision=179781&tests=fullscreen%2Ffull-screen-plugin.html>.

This test was added in <http://trac.webkit.org/r170717>.

@@ -1,3 +1,3 @@
 Test that plugin doesn't restart when taking it to full screen and back.
-SUCCESS
+FAIL

[Alexey Proskuryakov]

Created attachment 246230 [details]
speculative fix

[WebKit Commit Bot]

Comment on attachment 246230 [details]
speculative fix

Clearing flags on attachment: 246230

Committed r179794: <http://trac.webkit.org/changeset/179794>

[WebKit Commit Bot]

All reviewed patches have been landed.  Closing bug.

[Alexey Proskuryakov]

This made the test fail differently, shedding more light on why it failed before:

-Test that plugin doesn't restart when taking it to full screen and back.
-SUCCESS
+CONSOLE MESSAGE: line 26: TypeError: undefined is not an object (evaluating 'testObject.property = 'foo'')
+FAIL: Timed out waiting for notifyDone to be called

Apparently, testObject is not necessarily available even by the time the load event fires.

[Alexey Proskuryakov]

Created attachment 246244 [details]
second try

[Sam Weinig]

Comment on attachment 246244 [details]
second try

View in context: https://bugs.webkit.org/attachment.cgi?id=246244&action=review

> LayoutTests/fullscreen/full-screen-plugin.html:65
> +function checkForPlugin()
> +{
> +    var plugin = document.getElementById('plugin');
> +    if (plugin.testObject)
> +        test();
> +    else
> +        setTimeout(checkForPlugin, 100);
> +}

It seems like it might make sense for the test plugin to have a way to indicate it is ready without the need to poll. Maybe the plugin could look for a predefined function name (e.g. function pluginIsAvailable()) and call it when it has been instantiated. I can't imagine this is the only place where this mistake is made.

[Alexey Proskuryakov]

I thought that plug-ins were supposed to be available in onload, however maybe some of the async layout patches broke that. Or maybe it never worked correctly in WebKit.

[WebKit Commit Bot]

Comment on attachment 246244 [details]
second try

Rejecting attachment 246244 [details] from commit-queue.

Failed to run "['/Volumes/Data/EWS/WebKit/Tools/Scripts/webkit-patch', '--status-host=webkit-queues.appspot.com', '--bot-id=webkit-cq-02', 'validate-changelog', '--check-oops', '--non-interactive', 246244, '--port=mac']" exit_code: 1 cwd: /Volumes/Data/EWS/WebKit

/Volumes/Data/EWS/WebKit/LayoutTests/ChangeLog neither lists a valid reviewer nor contains the string "Unreviewed" or "Rubber stamp" (case insensitive).

Full output: http://webkit-queues.appspot.com/results/5013007134359552

[Alexey Proskuryakov]

Committed <http://trac.webkit.org/r179811>.

[Alexey Proskuryakov]

Hmm, it still fails.

@@ -1,3 +1,2 @@
-Test that plugin doesn't restart when taking it to full screen and back.
-SUCCESS
+FAIL: Timed out waiting for notifyDone to be called

[Alexey Proskuryakov]

Marked as flaky in <http://trac.webkit.org/changeset/180170>.

Antti, this is a test for your fix, please make it work correctly if you can.