280982 - Redraw is inefficent when two or more animated elements are visible in the viewport.

Status: RESOLVED FIXED

(Camino Graveyard :: General, defect, P3)

Description

[Wevah (Nate Weaver)]

When two or more animated elements (e.g., animated GIFs) are visible in the
viewport, the entire rectangle bounding the animated parts is redrawn, instead
of just the parts themselves. This can be seen easily in the linked testcase
using Quartz Debug (compare with Firefox, which does not have this problem).

This bug also appears to manifest with just one animation when the page is
scrolled with a scroll wheel; in this case, the entire viewport is redrawn
instead of just the animation and the page area that was just scrolled into view.

Much of this might be due to Cocoa combining drawing rects on non-opaque views,
but perhaps a workaround can be found.

Comment 1

[Simon Fraser [no longer active]]

Doesn't 10.2 have some new NSView updating methods that might help here?

Comment 2

[Simon Fraser [no longer active]]

I wonder if fixing this would help with typing in pages with animations?

Comment 3

[Jasper]

It's probably very closely related.

Comment 4

[L. H.]

*** Bug 287609 has been marked as a duplicate of this bug. ***

Comment 5

[Simon Fraser [no longer active]]

I've verified that [getRectsBeingDrawn:count:] in [ChildView drawRect:] does
return the 2 rects that need updating. Somewhere (maybe in cocoa?) something is
causing the union of those two rects to be flushed to the screen.

Comment 6

[Smokey Ardisson (offline for a while; not following bugs - do not email)]

In addition to animated GIFs, there's another demonstration of this with JS(?)
animated text on the Gmail login page (no gmail account needed to see it):

https://www.google.com/accounts/ServiceLogin?service=mail&passive=true&continue=https%3A%2F%2Fgmail.google.com%2Fgmail%3Fui%3Dhtml%26zy%3Dl&hl=en

Should this block 0.9?  The bug it blocks is approved for blocking 0.9 (will
both be fixed at once?)

Comment 7

[Jasper]

*** Bug 297260 has been marked as a duplicate of this bug. ***

Comment 8

[Christopher Petrilli]

From what I can tell, even editing an email in Gmail triggers this.  The JS on
the front page (which is now changed) would cause near 100% CPU usage, but even
composing a message would cause (on a 1.8GHz G5 iMac) 1-2 second lags between
typing and echoing of *each character*.  THis makes Camino useless for Gmail. 
Note that this problem does not happen in Firefox.

Comment 9

[Smokey Ardisson (offline for a while; not following bugs - do not email)]

(In reply to comment #8)
> From what I can tell, even editing an email in Gmail triggers this.

Typing issues are bug 272954, although the two are related.

Comment 10

[Jasper]

Lot of people seem to have forgotten about this bug, eben while it's the most
obvious speed bumb we could get when fixed. I still think this is something we
should fix asap i possible do I'll be putting some priority on it.

Comment 11

[Jasper]

*** Bug 299742 has been marked as a duplicate of this bug. ***

Comment 12

[Jasper]

*** Bug 299569 has been marked as a duplicate of this bug. ***

Comment 13

[Jasper]

More and more websites start to use more and more animation crap because most
browsers can handle it. This needs to get fixed soon, users are suffering some
nasty stuff because of this.

Comment 14

[stefan livens]

regarding bug 299742
I don't think it's ****, I think this Ajax stuff is pretty interesting: websites
don't have to update/load the whole page. 
Unfortunately: that's just what Camino does. (i know: reloading != redrawing)

Maybe there should be a META bug for these bugs, where al the excess drawing
bugs are put together. (bug 299742 (excess drawing thru JS), 272954 (excess
drawing in TEXTAREA), 280982 (excess drawing of too much of the page with
animated gif's)

Or change the title of this bug to include "dynamic elements" and "excess
drawing" (as this bug is referenced like in the description of bug 290234)

oh: this excess drawing (of bug 299742) is also visible in Deer Park, so we have
to mark bug 299742 as NOT a duplicate of this one.

Comment 15

[stefan livens]

camino 20050307: only paragraph of text gets updated
camino 20050308: whole page gets updated

both from the "nighly builds"

Comment 16

[stefan livens]

ignore previous post, should have been in bug 299742.
Sorry.

Comment 17

[Simon Fraser [no longer active]]

Can someone pin down a regression window for this (maybe look around 20050311).

Comment 18

[Simon Fraser [no longer active]]

Attachment: Testcase with 2 changing text divs

This patch has 2 changing text divs. It does not suffer the unionized redraws,
suggesting that those are image-only, so affectged by the Quartz image drawing
(bug 245407).

Comment 19

[Jasper]

I don't know how you are looking simon. But when I look with quartz debug I can
clearly see that one big rectangel including the two divs is painted and not 2
small rectagles of only the divs.

Comment 20

[Jasper]

Ok well on the other hand Safari shows the same thing. So please tell me with
what I should look ;)

Comment 21

[Simon Fraser [no longer active]]

(In reply to comment #19)
> I don't know how you are looking simon. But when I look with quartz debug I can
> clearly see that one big rectangel including the two divs is painted and not 2
> small rectagles of only the divs.

You're right, I was too hasty. I see a unified update rect on 10.3 with Quartz
Debug.

Comment 22

[Simon Fraser [no longer active]]

Attachment: Patch

The CHBrowserView (which is the ancestor of all the gecko views) was using the
non-optimal view repainting, which caused the update area to be larger. This
fixes that, and fixes both image and text testcases.

Comment 23

[Mark Mentovai]

Comment on attachment 190079 [details] [diff] [review]
Patch

r=mark with preprocessor conditionals to avoid [self respondsToSelector:] and
fallback when MAC_OS_X_VERSION_MIN_REQUIRED >= 10.3

Comment 24

[Simon Fraser [no longer active]]

Checked in.

Comment 25

[Smokey Ardisson (offline for a while; not following bugs - do not email)]

Clearing the ? flag since this was fixed....