502754 - crash when reloading a page with a canvas3d element with an empty fragment shader [@ ... | glClear | nsCanvasRenderingContextGLWeb20::Clear]

Status: RESOLVED WORKSFORME

(Core :: Graphics: CanvasWebGL, defect)

Description

[Dana Keeler (she/her) [:keeler]]

User-Agent:       Mozilla/5.0 (Macintosh; U; Intel Mac OS X 10.5; en-US; rv:1.9.1) Gecko/20090624 Firefox/3.5
Build Identifier: mozilla-central revision f46e6aee1335

Firefox crashes when reloading the attached page (or sometimes even on load).  I believe it has something to do with the empty fragment shader (starting around line 250 of test.html), because if the commented-out section is un-commented-out, it works just fine.
Also, this may be a graphics card issue, because the crash location appears to be in graphics card code.

Reproducible: Always

Steps to Reproduce:
1. Compile a canvas3d-enabled build ('ac_add_options --enable-canvas3d' in .mozconfig)
2. Enable canvas3d (user_pref("extensions.canvas3d.enabled_for_all_sites", true); in prefs.js)

3. Open the attached test.html
4. If firefox doesn't crash, hit reload once (or maybe twice) (you might need an ATI RadeonX1600 graphics card...)
Actual Results:  
firefox crashes

Expected Results:  
a 3D box should be displayed

Comment 1

[Dana Keeler (she/her) [:keeler]]

Attachment: test file that causes crash

Comment 2

[Dana Keeler (she/her) [:keeler]]

Attachment: dependency for test.html

Comment 3

[Dana Keeler (she/her) [:keeler]]

Attachment: dependency for test.html

Comment 4

[Dana Keeler (she/her) [:keeler]]

Attachment: crash report

Comment 5

[timeless]

don't suppose you have a system that wouldn't use ATIRadeonX1000GLDriver? :)

Comment 6

[Dana Keeler (she/her) [:keeler]]

Attachment: standalone c program that triggers this bug

Comment 7

[Dana Keeler (she/her) [:keeler]]

The c program does not trigger the bug on an nvidia GeForce Go 7300 under linux nor on another nvidia card (don't know exactly what type) under OS X.
(A note about the program: the "-lGLUT" part might have to be "-lglut" in some circumstances, such as under linux.)

Comment 8

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

Sounds like another shader validation bug.

Comment 9

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

Can you check if this is still a problem?  Leaving open for now to verify, not sure if empty programs are valid or not.

Comment 10

[Dana Keeler (she/her) [:keeler]]

I suspect that this is a driver issue, and not a Firefox/canvas3d issue.  I don't have access to a computer with an ATI RadeonX1600 (or any ATI card, for that matter), so I don't know if an update has fixed it.  I can say I can't reproduce it with the Intel card I currently have.  If someone has the time and can find a MacBook Pro that is 2-4 years old, it might have the right card.  In any case, this specific bug is probably not Mozilla's responsibility.  However, I would say it is worth thinking about how canvas3d might expose more of these bugs.

Comment 11

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

This should be fixed, since ANGLE will fail the translation step on this shader.