We need a test for the bug 44153.
Created attachment 139050 [details] Adds a regression test
Comment on attachment 139050 [details] Adds a regression test View in context: https://bugs.webkit.org/attachment.cgi?id=139050&action=review > LayoutTests/editing/selection/deleteFromDocument-after-document-open-crash.html:25 > + layoutTestController.notifyDone(); This test doesn't appear to need waitUntilDone/notifyDone - all the work is done from unload.
Comment on attachment 139050 [details] Adds a regression test View in context: https://bugs.webkit.org/attachment.cgi?id=139050&action=review >> LayoutTests/editing/selection/deleteFromDocument-after-document-open-crash.html:25 >> + layoutTestController.notifyDone(); > > This test doesn't appear to need waitUntilDone/notifyDone - all the work is done from unload. document.open() sometimes ends DRT so without this, the test will be flaky.
> document.open() sometimes ends DRT so without this, the test will be flaky. How can this possibly be non-deterministic? Is there an uninitialized variable somewhere? Looks like you're covering up a real bug.
(In reply to comment #4) > > document.open() sometimes ends DRT so without this, the test will be flaky. > > How can this possibly be non-deterministic? Is there an uninitialized variable somewhere? > > Looks like you're covering up a real bug. Maybe. The last time I checked, this was expected behavior per spec and non-determinism was due to some race conditions. But then I'm not too familiar with this area.
Hm... it appears that this problem no longer reproduces at least on this test. I'm gonna just remove waitUntilDone/notifyDone for now.
Committed r115360: <http://trac.webkit.org/changeset/115360>
> it appears that this problem no longer reproduces at least on this test Perhaps it's more complicated when the script runs during load, as opposed to unload handler.
(In reply to comment #8) > > it appears that this problem no longer reproduces at least on this test > > Perhaps it's more complicated when the script runs during load, as opposed to unload handler. Hm... that's possible. I guess Chromium WebKit gardeners will notice if this test becomes flaky :)
> unload onload, thank you autocorrection :)