Bug 6418 – Incorrect scrollbar when using overflow:auto and word-wrap:break-word; in some cases

• Component: Layout and Rendering
• Status: VERIFIED FIXED

[David Smith]

The URL is a somewhat simplified and fixed up snippet from an Adium chat (which is displayed in a 
WebView). Opera 8.5.1 has the rendering I would expect, with no scrollbar (although it does mess up the 
border slightly). Gecko based browsers don't even get close.

Expected result: no horizontal scrollbar, with the text wrapped if necessary to fit.

Actual result: horizontal scrollbar that can scroll 1px

I originally thought this was another case of http://bugzilla.opendarwin.org/show_bug.cgi?id=5619, but 
discussions on IRC indicated that it may be a separate bug.

[David Smith]

http://trac.adiumx.com/ticket/267 is the relevant ticket in Adium's bug tracking system.

Dave Hyatt has noted in the past that findNextLineBreak contains some suspicious-looking comparisons 
with width+1 instead of width. I think they're what's causing this bug, since it happens only when the text 
is one pixel too wide to fit.

Created attachment 5545 [details]
Reduced testcase

Note how "ipsum" stays on the first line, although this results in a RenderText
wider than its containing RenderBlock (84 and 83 pixels respectively).

Just changing the "width+1" instances to "width" solves this particular
problem, but breaks tons of other things. I have a feeling that one way or
another the rounding hacks are involved in this bug.

Created attachment 5608 [details]
The opposite case

Despite the +1, premature line-breaking can happen too. In this case, "Dolor"
is broken into two lines in TOT and shipping WebKit.

Eliminating the "single character but not single-character run" case from the
WebTextRenderer rounding hack logic when fixing bug 5414 may have increased the
likelihood of this happening, but as you can check, it happened even before.

I'm working on a fix...

Created attachment 5627 [details]
Proposed patch

[Darin Adler]

Comment on attachment 5627 [details]
Proposed patch

Looks great. Nice to get rid of the always-false appliedEndWidth boolean along
with fixing this.

r=me