618848 - Post process shaders and texture mapping crash OS X 10.5

Status: RESOLVED WONTFIX

(Core :: Graphics: CanvasWebGL, defect)

Description

[Corban Brook]

User-Agent:       Mozilla/5.0 (Macintosh; Intel Mac OS X 10.5; rv:2.0b8pre) Gecko/20101211 Firefox/4.0b8pre
Build Identifier: 4.0b8pre (2010-12-11)

The combination of post process shaders and texture mapping halt OS X 10.5 on a mac book pro. When script is run, Minefield document window will flash blue or display multi-colored artifacts, a few seconds later entire OS X GUI will flicker once or twice before freezing up, usually see strange rendering artifacts across entire screen (windows rendered strangely with no regards to the z-depth), mouse still tracks but GUI does not respond to clicks. I have left the machine running for up to 20 minutes to see if it would recover from frozen state but never does.

Post process shaders turned OFF and texture mapping ON: Works fine. 
Post process shaders turned ON and texture mapping OFF: Works fine.



Reproducible: Always

Comment 1

[David Humphrey (:humph)]

This bug is a bad one, cause it's killing Corban's ability to work on all our WebGL code with us.  I suspect it's driver related, but it only seemed to start a week ago or so.  Corban, do you remember the dates?

Comment 2

[David Humphrey (:humph)]

12:55 < corban> i was running it off my laptop in the mozilla offices
13:01 < corban> seems to me in the most recent nightly they made DEPTH_COMPONENT synonymous for DEPTH_COMPONENT16
13:01 < corban> they both do the same thing now
13:01 < ccliffe> DEPTH_COMPONENT isn't valid
13:01 < ccliffe> it should produce an invalid enum error
13:01 < corban> it doesnt tho
13:01 < ccliffe> and cause the depth buffer to break
13:01 < ccliffe> yeah that's why I've had that patch in for you for awhile
13:02 < ccliffe> doesn't make sense why that enum would be available on one OS but not another

Comment 3

[Charles J. Cliffe]

Some more info:

From what I understand there seems to be a bad interaction when using
Render-To-Texture (FBO) and textures on corban's system.  Our shader pipeline
just bounces several RTT buffers back and forth to add each layer of effects --
doesn't make much sense that just loading in the surface texture data as well
is causing a fail this hard (sounds like an internal driver issue to me).

And by 'Texture Mapping Off' we just move the image URL -- so all the texture
references are generated but texImage2D is never called to upload because the
image 'onload' never fires. 

So the only real difference that causes the crash is the textures actually
being loaded into memory and some minor mipmap/clamping parameters that may
change after 'onload' checks the image dimensions.

Comment 4

[Corban Brook]

Tested in Chromium, It works fine with post-process off and texture maps properly just like in minefield. Yet with post process on, just renders a black screen, No crash.

Comment 5

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

Can you get a stack and run in the gl sync debug mode that bjacob put in?  You'll probably have to ssh in to get a stack.

Comment 6

[Benoit Jacob [:bjacob] (mostly away)]

To do that:
 * get a debug build
 * run with this environment variable defined: MOZ_GL_DEBUG

Comment 7

[Jeff Muizelaar [:jrmuizel]]

Does this bug still exist. Can we get a reproducible testcase?

Comment 8

[Kelsey Gilbert [:jgilbert]]

OS X 10.5 is no longer supported.