244582 - Slow cursor when beginning to hover over certain gif images where the <a> tag is given a hover style.

Status: RESOLVED FIXED

(Core Graveyard :: GFX: Win32, defect)

Description

[Kevin Ar18]

User-Agent:       Mozilla/5.0 (Windows; U; Windows NT 5.1; en-US; rv:1.8a2) Gecko/20040524 Firefox/0.8.0+
Build Identifier: Mozilla/5.0 (Windows; U; Windows NT 5.1; en-US; rv:1.8a2) Gecko/20040524 Firefox/0.8.0+

This issue occurs with certain gif images used on hardocp.com.
Whenever you begin to move your mouse over the image, the cpu useage rises and
the mouse lags.
This problem only seems to occur with certain gif image used on hardocp.com. 
Also, it only occurs when those image are placed inside of an <a> tag and given
some hover style information.

Reproducible: Always
Steps to Reproduce:
1. Go to http://www.hardocp.com/article.html?art=NjExLDY=
2. Move your cursor over the first benchmark image.
Actual Results:  
The mouse lags for a little while and cpu useage rises.


If someone can redirect this bug to the proper component (instead of
browser-general), that would be appreciated.

Also, see the comment below for a little more information on this bug.  It
includes a testcase based on hardocp's code.

Comment 1

[Kevin Ar18]

Attachment: Testcase based on HardOCP's code

This is a fairly simplified testcase of the issue.
I noticed that by using a different image, the issue went away.
I also tried opening the problem image in an image editor, saving as a bitmap,
and then saving the bitmap as a gif.  The problem went away, thus the issue
must be something specific in the gif images.

It should also be noted that removing the <a href=""> from around the <img> tag
caused the problem to go away.
Putting an img:hover style does not cause the problem.	Only when the image is
made into a link using the <a href=""> tag and then the <a> tag is given a
hover style does the problem occur.
I noticed that several different styles that I gave to a:hover will cause the
problem (such as color, background-color, and cursor).

Comment 2

[Kevin Ar18]

Keywords: perf, regression

This bug is not present in the following builds:
1.6	March 11
1.6b	February 6
1.7a	March 4

However it is present in these builds:
1.7b	March 24
1.7rc2	May 17

Comment 3

[Martijn Wargers (dead)]

Yes, I can see the problem.

The bug occurs in:
Mozilla/5.0 (Windows; U; Windows NT 5.0; en-US; rv:1.7b) Gecko/2004-03-17
Firefox/0.8.0+

But it doesn't occur in:
Mozilla/5.0 (Windows; U; Windows NT 5.0; en-US; rv:1.7b) Gecko/2004-03-11

Comment 4

[Martijn Wargers (dead)]

Well, found another build:
Mozilla/5.0 (Windows; U; Windows NT 5.1; en-US; rv:1.7b) Gecko/20040316
Firefox/0.8.0+
In this build I can also see the problem.

I think this bug is also related to bug 244555, since they seem to have the same
period in where the performance decrease occurs.

Comment 5

[Martijn Wargers (dead)]

I think the final patch of bug 205893 is causing the problem.
Backing that patch out seems to make the performance of Mozilla in this testcase
normal again.
This is also the case with the testcase in bug 244555.
I think you'll see the performance increase/decrease more likely on somewhat
slower computers. I use a 800MHz Duron.

Comment 6

[Aaron Kaluszka]

I don't see this at all on 20040524.  What graphics card do you use?

Comment 7

[Kevin Ar18]

Nvidia Vanta - it doesn't support per pixel alpha blending.
Perhaps this bug applies to all video cards that don't support this feature? 
(which would be pre-geforce video cards for Nvidia's line).

Comment 8

[Aaron Kaluszka]

Well I'm not sure if it's so much that it's old, but that it's an NVidia.  I
have a feeling there's something wrong with the way painting-- maybe DIB
functions-- are done (that's why I opened bug 194627). What we really need are a
set of test cases to quantify the behavior.

Comment 9

[Martijn Wargers (dead)]

(In reply to comment #7)
> Nvidia Vanta - it doesn't support per pixel alpha blending.
> Perhaps this bug applies to all video cards that don't support this feature? 
> (which would be pre-geforce video cards for Nvidia's line).

Ah, that would certainly make sense. 
I have a WinXP machine (Duron 600MHz, NVidia GeForce2 MX200), that doesn't
really slow down on the testcases I mentioned.
But I also have Win2000 machine (Duron 800MHz, Nvidia Riva TNT2 Model 64), which
is really slowed down on the testcases (except when I back out that patch which
mentioned earlier, then it gets better).

Comment 10

[Kevin Ar18]

Based on a comment in Bug 227260, I did some testing again, and noticed that the
bug goes away when you disable mouse shadows.  Perhaps this problem and the
problem in Bug 227260 are related?

Comment 11

[Boris Zbarsky [:bzbarsky]]

Stefan, assigning to you because it looks like your patch caused this... Please
let me know if you won't be able to look into this, ok?  Confirming based on
minimal testcase comments.

Also ccing reviewers from bug 205893.

Comment 12

[Asa Dotzler [:asa]]

Stefan, are you around? can you take a look at this?

Comment 13

[Martijn Wargers (dead)]

I think this is also fixed by the fix for bug 284716. 
I can't see the bug anymore with a 2005-03-6 build. 
Tested with 800MHz Duron/NVidia TNT2 M64 16 bit color depth.

Comment 14

[Kevin Ar18]

The testcase and original page did not show the bug anymore, however, I was able
to reproduce the bug on a cached version of the site:
http://64.233.187.104/search?q=cache:vacNyldjttwJ:www.hardocp.com/article.html%3Fart%3DNjExLDY%3D+%22This+graph+shows+how+powerful%22&hl=en

After I saved the page to my hard drive, however, I was unable to continue to
reproduce the bug.  Can anyone else see it on their system?

Comment 15

[Boris Zbarsky [:bzbarsky]]

Marking fixed per comment 13.