Bug 71996 – REGRESSION(r98542): Chromium: Rendering error on Google maps

• Status: RESOLVED
• Resolution: FIXED
• Component: WebCore Misc.

[David Barr]

http://code.google.com/p/chromium/issues/detail?id=102521

What steps will reproduce the problem?
1. Maps.google.com
2. Look for directions
3. By Car and By Transport

What is the expected result?

Lines depicting the path to take 

What happens instead?

Lines are not visible 
Icons are misaligned 

Bisection identified http://trac.webkit.org/changeset/98542 as the change that caused the regression.

[David Barr]

A minimal temporary fix is to simply disable the MatchedDeclarationCache by adding "result.isCacheable = false;" to the end of CSSStyleSelector::matchAllRules().

[David Barr]

Created attachment 114434 [details]
Patch

[Mike Lawther]

I haven't been able to repro the missing direction lines, but I the misaligned icons does occur in WebKit Nightly r99757.

Antti - this looks like it might be a dupe of https://bugs.webkit.org/show_bug.cgi?id=71703, since they both seem to have been caused by the same patch as identified by David?

[Antti Koivisto]

Comment on attachment 114434 [details]
Patch

Disabling the feature stops us from getting testing. I don't think we should do this as the bug does not make the build unlivable.

[Antti Koivisto]

Could some from Google provide a reduced test case for this? maps.google.com is generally difficult to debug due to obfuscation.

[Mike Lawther]

The feature has already been demonstrated to cause at least two regressions, this one on a live site and https://bugs.webkit.org/show_bug.cgi?id=71703.

While we don't have a minimal repro yet, the feature really doesn't seem to be working as intended ie no side effects other than memory/performance improvements.

Wouldn't it be better to roll this back, or effectively disable it as in David's patch so it can be examined offline? Also, the feature was submitted with no new tests - are you confident that all the new code was covered by the existing tests?

[David Barr]

Created attachment 114625 [details]
Small but not minimal repro of maps alignment issue

This is still quite messy but is sufficient to demonstrate a change in behavior.

[David Barr]

Created attachment 114627 [details]
Slightly better repro, visible difference when served by http

[David Barr]

Created attachment 114655 [details]
Minimal reproduction

Should see concentric squares but the middle square moves to the top.

[Antti Koivisto]

Great, thank! I'm on it.

[Eric Seidel (no email)]

Thank you very much for the reduction, david.

[Antti Koivisto]

Created attachment 114830 [details]
fix

Turns out that r98542 exposed that there is a field missing from StyleBoxData copy constructor.

[Antti Koivisto]

http://trac.webkit.org/changeset/100078

[Eric Seidel (no email)]

ouch!  Thanks antti!

[Mike Lawther]

Thanks for the quick fix Antti!

[David Barr]

Thank you, Antti! I've tested rolling the change into Chrome and it's all good news.