413583 - cairo xlib buggy_repeat not detected correctly

Status: RESOLVED FIXED

Description

[Matthew Middleton (:zzxc)]

Created attachment 298628 [details]
xdpyinfo output

With certain background images, including the "master-vfl31713.gif" image on Youtube, garbage from elsewhere on the screen results on Linux instead of the actual image.  See the URL for an example broken page - it references the broken image both on youtube.com and a backup on my server.  The header with the search box on www.youtube.com is also broken.

Note that the issue doesn't occur with the mozilla.org gif at the bottom.  The issue also does not occur with a "background-repeat: no-repeat" rule.

This regressed between the 2007-01-17 and 2007-01-19 nightly builds, when Cairo was upgraded.

Comment 1

[Matthew Middleton (:zzxc)]

Created attachment 298629 [details]
Screenshot of page in URL

Adding screenshot of URL

Comment 2

[David Baron :dbaron: ⌚️UTC-8]

I see this as well.  I'm using the radeon driver (that's the free one) on Fedora 8.

Comment 3

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

Matthew says he's using the 'ati' driver (which I assume is 'radeon' by another name?).

What changed in cairo is that the X server started reporting a different version number, without changing the vendor string -- so for a while there it was reporting version 1.x and cairo was setting buggy_repeat on for all servers under 6.8; this was recently fixed to only look at servers > 6.0.

So we have a problem; this may be driver-specific.  I have a radeon machine at work now with ubuntu on it, i'll check this tomorrow.

If this -is- driver specific, I have no idea what to do.  The buggy_repeat path is slow.

Comment 4

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

Note that we have an easy workaround, if necessary, for the beta: we go back to the old 'broken' variant and unconditionally set broken_repeat to TRUE.

I don't think we have a good way of detecting which driver is in use at runtime (maybe we can fudge it if GLX is enabled and getting the OpenGL renderer string?).

Comment 5

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

Yeah, I can reproduce this with the ati/radeon driver.  Ugh.

Comment 8

[John Kloosterman]

I was asked to describe my configuration: I run X.org 7.2 with the sis display
driver, version 0.9.3.

Comment 9

[John Kloosterman]

Sorry, that should be X.org 1.3.0, X11R7.2.

Comment 10

[Andrew Schultz]

I'm also using the radeon driver with xorg 1.3.0.0

Comment 11

[Andrew Schultz]

Oh, and the radeon driver is apparently version 6.7.196

Comment 12

[dolphinling]

I get this with xorg 1.3.0.0 and the intel i810 driver. 

Comment 13

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

Created attachment 299854 [details] [diff] [review]
beta bandaid

Okay; I do not want to ship a beta with this bug, because in the worst case it can cause X server crashes or other massive system instability.  Instead, this patch enabled the old workaround at the cost of some performance.

Comment 15

[Behdad Esfahbod]

We're working on fixing this in cairo very soon.

Comment 16

[Matěj Cepl]

This is on Fedora/Rawhide with the latest xorg-x11-drv-ati and firefox3
packages (upgraded daily)

Comment 18

[WayneGretzky]

Created attachment 302403 [details]
Example page: teletext.ch (correct display with FF2)

Comment 19

[WayneGretzky]

Created attachment 302404 [details]
Testcase pages: see Comment field

Correct display (shots taken with FF2):
- http://img206.imageshack.us/img206/6909/nhlcomcorrectff2xs5.png
- http://img401.imageshack.us/img401/9867/teletextchcorrectff2ku0.png

Incorrect display (shots taken with FF3):
- http://img401.imageshack.us/img401/673/nhlcomfaultyff3qt6.png
- http://img232.imageshack.us/img232/8900/teletextchfaultyff3bs0.png

Comment 20

[WayneGretzky]

Incorrect display (shots taken with FF3 & the default theme):
- http://img401.imageshack.us/img401/2695/nhlcomscoresab5.png
- http://img401.imageshack.us/img401/4930/teletextdm5.png

Comment 21

[Fabien Tassin]

vlad, is there a bug opened on the cairo and/or X.org side ?

Comment 23

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

As is, this bug is fixed (AFAIK); there is still an underlying issue here, but we can't fix that on the mozilla side.