Bug 11350

Summary:

http/tests/incremental/slow-utf8-css sometimes coalesces RenderTexts

Product:

WebKit

Reporter:

Adam Roben (:aroben) <aroben>

Component:

Tools / Tests

Assignee:

Geoffrey Garen <ggaren>

Status:

RESOLVED FIXED

Severity:

Normal

CC:

ap, aroben, timothy

Priority:

Keywords:

LayoutTestFailure

Version:

420+

Hardware:

Mac

OS:

OS X 10.4

URL:

http://build.webkit.org/results/post-commit-powerpc-mac-os-x/3898/http/tests/incremental/slow-utf8-css-diffs.txt

Attachments:

Description	Flags
tcpdump of pass then fail	none
patch-o-rama	ap: review+

Adam Roben (:aroben)

Reported 2006-10-18 13:01:29 PDT

http/tests/incremental/slow-utf8-css is intermittently failing. The symptom is that adjacent RenderTexts are sometimes being coalesced into one RenderText. It has been theorized that this is due to variances in the speed that chunks of data are sent to DumpRenderTree by Apache.

Attachments
tcpdump of pass then fail (44.17 KB, text/plain) 2006-10-18 21:51 PDT, Geoffrey Garen	no flags	Details
patch-o-rama (2.38 KB, patch) 2006-10-19 12:12 PDT, Geoffrey Garen	ap: review+	Details Formatted Diff Diff
View All Add attachment proposed patch, testcase, etc.

Geoffrey Garen

Comment 1 2006-10-18 21:50:58 PDT

To verify, I ran this command: run-webkit-tests http/tests/incremental/slow-utf8-css.pl http/tests/incremental/slow-utf8-css.pl The first test passed. The second test failed, with identical results to the buildbot. I've attached a tcpdump of the back-to-back transactions.

Geoffrey Garen

Comment 2 2006-10-18 21:51:59 PDT

Created attachment 11144 [details] tcpdump of pass then fail

Geoffrey Garen

Comment 3 2006-10-19 08:09:14 PDT

The traffic is slightly different in the two transactions but, as far as I can tell, the chunks sent by the server are identical.

Alexey Proskuryakov

Comment 4 2006-10-19 11:50:59 PDT

As discussed on IRC, coalesced RenderTexts mean that all data was processed in one pass, so the test isn't actually performed (it checks for a bug where Decoder only processed one chunk correctly for CSS). If the solution with inserting a small timeout won't work, we can also try converting the test into an HTML+CSS one - then timing won't matter.

Geoffrey Garen

Comment 5 2006-10-19 12:12:46 PDT

Created attachment 11151 [details] patch-o-rama

Geoffrey Garen

Comment 6 2006-10-19 12:15:10 PDT

I suspect the issue was that the test depended on a WebCore timer. Inserting a delay ensure that the timer fires consistently between the chunks of data.

Geoffrey Garen

Comment 7 2006-10-19 13:48:24 PDT

r17141

Note You need to log in before you can comment on or make changes to this bug.