[Webkit-unassigned] [Bug 65277] Provide flexible (in terms of position relative to space) logical word breaker to improve performance of --webkit-visual-word

[bugzilla-daemon at webkit.org]

https://bugs.webkit.org/show_bug.cgi?id=65277





--- Comment #12 from Xiaomei Ji <xji at chromium.org>  2011-08-09 11:02:49 PST ---
(In reply to comment #10)
> When I profiled the release build of DumpRenderTree with --repeat-each 5 on LayoutTests/editing/selection/move-by-word-visually*, I got the results (only showing relevant part):
> 
> Before change:
> 3.8% rightWordPosition
>    1.9% collectWordBreaksInBox
>    1.7% rightWordBoundary
>    0.1% nextBoundary
>    ...
> 
> After change:
> 3.4% rightWordPosition
>    1.7% rightWordBoundary
>    1.2% collectWordBreaksInBox
>    0.3% nextBoundary
>    ...
> 
> Results for leftWordPosition are similar. This makes sense as, before the change, calls to collectWordBreaksInBox occurred when going right in LTR blocks and left in RTL blocks (roughly accounts for half of test cases). Now word breaks don't need to be collected for those cases; but they still need to be collected when the correct word break is at the logical end of line or when moving from one box to another.

rightWordBoundary (and leftWordBoundary) could be improved using the same technic. 
currently, word breaks are collected in the box for 50% of the cases.

Then, nextBoundary() and previousBoundary() will take even higher percentage of time, and we can continue work on reducing constructing of VisiblePosition.

-- 
Configure bugmail: https://bugs.webkit.org/userprefs.cgi?tab=email
------- You are receiving this mail because: -------
You are the assignee for the bug.