Bug 101540

Summary:

Attached vs detached tree performance

Product:

WebKit

Reporter:

Philip Rogers <pdr>

Component:

Layout and Rendering

Assignee:

Nobody <webkit-unassigned>

Status:

NEW

Severity:

Normal

CC:

abarth, ggaren

Priority:

Version:

420+

Hardware:

Unspecified

OS:

Unspecified

Attachments:

Description	Flags
Testcase showing performance badness	none
Profile	none

Philip Rogers

Reported 2012-11-07 19:40:34 PST

Created attachment 172919 [details] Testcase showing performance badness In the attached repro, appending a bunch of nodes to a detached element is much slower than appending a bunch of nodes to an attached element. I did a quick trace of this and it looks like the time is spent in walking the render tree (ComposedShadowTreeWalker::traverseSiblingOrBackToInsertionPoint), which doesn't seem right. We may have some easy performance wins here. I've attached a screenshot of the trace comparison as well (left: not detached, right: detached). Original bug: http://code.google.com/p/chromium/issues/detail?id=155907

Attachments
Testcase showing performance badness (1.40 KB, text/html) 2012-11-07 19:40 PST, Philip Rogers	no flags	Details
Profile (1.15 MB, image/png) 2012-11-07 19:41 PST, Philip Rogers	no flags	Details
View All Add attachment proposed patch, testcase, etc.

Philip Rogers

Comment 1 2012-11-07 19:41:00 PST

Created attachment 172920 [details] Profile

Philip Rogers

Comment 2 2012-11-07 19:42:23 PST

Additional details from the original bug: "In particular, I would have expected the non-detach mode to be much slower if I did anything to invalidate layout, and I would have expected the slowdown to be in layout. As expected, both approaches spend the same amount of time in JS as well as in layout. Unexpectedly, adding the nodes while the elements are attached causes one very long style recalc before layout and paint."

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