379437 - 3-14% Txul regression from bug 368247

Status: RESOLVED INCOMPLETE

(Core :: Layout, defect, P3)

Description

[Andrew Schultz]

Bug 368247 landing regressed argos' Txul numbers 2-3%.  The Mac/FF box regressed 8%.  nye's numbers regressed by 14%.  KaiRo's Mac box (tpol) regressed by 11%.

When loading up today's SeaMonkey nightly, it takes 1-2 seconds to paint the chrome (which I verified was caused by 368247 by backing it out locally).

Comment 1

[Andrew Schultz]

Attachment: seamonkey stack during paint

org is actually the process using CPU during the paint.  The stack is the same as in bug 371867 comment 7.  The attachment is the seamonkey side.

Comment 2

[Andrew Schultz]

Attachment: html testcase

Load the testcase and drag the scrollbar up and down.  With current trunk, the page scrolls smoothly at first (for <1 second) but then grinds to a halt and takes at least a second for each repaint.

If I delete the border style rule, scrolling is quick and smooth.

Comment 3

[Martijn Wargers (dead)]

Andrew, that testcase looks more or less the same as the one from bug 379834.

Comment 4

[Andrew Schultz]

Yes.  The Txul regression was conceivably due to use of CSS borders in chrome.

Comment 5

[Vladimir Vukicevic [:vlad] [:vladv] (needinfo me, slow to respond)]

Yup, it almost certainly is.  I've been fixing some actual rendering regressions, but this is still very high on my todo list.

Comment 6

[Gábor Stefanik]

Er, really? This bug seems not to affect Windows. (Scrolling the HTML testcase is fast as hell.) On the other hand, bug 379834 is cross-platform.

Comment 7

[Vladimir Vukicevic [:vlad] [:vladv] (needinfo me, slow to respond)]

379834 is mostly cairo, and should be fixed with cairo 1.4.6 once I do that update.  The problem there is that cairo is doing a whole bunch of work for trapezoids that are way outside the surface bounds.

Comment 8

[Robert O'Callahan (:roc) (email my personal email if necessary)]

Vlad, have you got an update here?

Comment 9

[Shawn Wilsher :sdwilsh]

I'm going through and marking old performance regression bugs as INCOMPLETE that are likely too old to be valid or get any traction on them.

Please re-open if you have more information or can demonstrate the regression still exists.