312419 - we could optimize clip:rect(0,0,0,0) better

Status: RESOLVED WORKSFORME

(Core :: Web Painting, defect)

Description

[David Baron :dbaron: (⌚️UTC-4, no longer working on Mozilla)]

It was pointed out to me that the S5 slide system is quite slow in Mozilla when
there are a lot of slides.  A profile shows that the performance problem is in
painting code, including both frame paint methods and the view manager's display
list construction.

Given the current source code of S5, I don't see something that's easily
fixable.  However, I tried to fix the performance problem by making S5 do hiding
using 'clip:rect(0,0,0,0)' instead of 'visibility:hidden', and that didn't help
the performance problem.  (Using 'display:none' did help.)

If the view manager treated 'clip:rect(0,0,0,0)' a bit like view-visibility
(which is not 'visibility') being hidden, then I'd think this problem would go
away.  I'm not sure that that's sufficient, but I think it's something we ought
to be able to optimize based on.

This needs a minimized testcase.  So there's something, I'll attach a .tar.gz
file of a modified S5 slideshow that shows the problem that I think we should be
able to fix.

I'll have a blog about this shortly at http://dbaron.org/log/2005-10#e20051013a

Comment 1

[David Baron :dbaron: (⌚️UTC-4, no longer working on Mozilla)]

Attachment: .tar.gz file containing complicated testcase

Comment 2

[Eli Friedman]

Is this any better on trunk?  We should be clipping stuff out properly now.