Bug 150542

Summary:	REGRESSION: fast/forms/state-restore-per-form.html is a flaky timeout
Product:	WebKit	Reporter:	Alexey Proskuryakov <ap>
Component:	Page Loading	Assignee:	Nobody <webkit-unassigned>
Status:	NEW ---
Severity:	Normal	CC:	beidson, jhoneycutt, jonlee, ryanhaddad, sam, wenson_hsieh
Priority:	P2	Keywords:	LayoutTestFailure
Version:	WebKit Nightly Build
Hardware:	Unspecified
OS:	Unspecified

Description Alexey Proskuryakov 2015-10-24 20:50:39 PDT

The first time this test timed out was on September 22, r190111. The timeouts are not frequent enough to pinpoint the culprit, but it was almost certainly no earlier than 24 hours before that.

When the test times out, there is no output other than "FAIL: Timed out waiting for notifyDone to be called", meaning that we don't get into the post-form submit part of it.

Comment 1 Alexey Proskuryakov 2015-10-24 21:02:59 PDT

I wonder how this is supposed to work - we have a lot of tests where we check whether a control named "emptyOnFirstVisit" is autofilled, but we don't seem to reset autofill state between tests (or maybe I just didn't find it?)

These tests are all in fast/forms and in imported/blink/fast/forms.

Comment 2 Alexey Proskuryakov 2015-10-24 23:52:30 PDT

"emptyOnFirstVisit" reuse is probably red herring.

Anyway, I can reproduce locally like this:

run-webkit-tests fast/forms/state-restore-per-form.html -f --repeat 100 --no-retry

Comment 3 Alexey Proskuryakov 2015-10-25 00:06:16 PDT

Sadly, this reproduces locally at least as far back as r189500, so the change that triggered the regression was probably something that changed timing, and the root cause isn't new.

Comment 4 Alexey Proskuryakov 2015-10-25 11:33:11 PDT

When the test fails, form submission just doesn't happen somehow. We remove original page content and send a pagehide event, but we never load anything after submitting the form!

This is pretty incredible.

Comment 5 Ryan Haddad 2015-11-30 14:15:57 PST

Marked as flaky on mac-wk2 in <http://trac.webkit.org/changeset/192827>