585502 - Crash [@ @0x0 strlen | nsICanvasRenderingContextWebGL_CompileShader] [@ strlen | nsICanvasRenderingContextWebGL_CompileShader ]

Status: RESOLVED FIXED

(Core :: Graphics: CanvasWebGL, defect)

Description

[katestarbird]

User-Agent:       Mozilla/5.0 (Macintosh; Intel Mac OS X 10.5; rv:2.0b4pre) Gecko/20100808 Minefield/4.0b4pre
Build Identifier: Mozilla/5.0 (Macintosh; Intel Mac OS X 10.5; rv:2.0b4pre) Gecko/20100808 Minefield/4.0b4pre

This webgl o3d development site, which has been functioning pretty well, now crashes every time. All you have to do is navigate there. Though I'm getting lots of crashes with today's build - so this may be a more general issue than o3d webgl.

Reproducible: Always

Steps to Reproduce:
1. http://www.cs.colorado.edu/~starbird/avatar_code/avatar_webgl.html
2.
3.
Actual Results:  
crash

Expected Results:  
functional o3d demo

Comment 1

[Makoto Kato [:m_kato]]

Repro on Mozilla/5.0 (Windows NT 6.0; WOW64; rv:2.0b4pre) Gecko/20100809

bp-f65872a4-a7e2-4b5d-966c-584082100809

Comment 2

[Alex]

I've been crashing whenever I load a page using WebGL, here's one stack.

bp-585b45ab-2524-4f52-b291-48d0e2100808

Comment 3

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

No crash on linux x86-64 / nvidia driver / mozilla-central from today

Comment 4

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

Strange, the back-trace looks a lot like bug 579191 which was fixed 2 weeks ago.

Comment 5

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

ok, this crash is on ANGLE-using code, so I guess the reason why I'm not getting it is that ANGLE is not enabled on linux x86-64. See bug 578877.

if anyone wants to help debug this: at WebGLContextGL.cpp:2867, please print the value of |ShGetInfoLog(compiler)|. I would guess that it returns a null pointer for some reason...

Comment 7

[Saint Wesonga]

Attachment: Patch v1

The retval from ShGetInfoLog needs to be checked since nsDependentCString does not accept NULL buffers - we end up calling strlen(NULL).

ShGetInfoLog appears to be failing because the process hasn't been initialized...

Comment 8

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

So, do you confirm it was actually returning NULL?

Next question: if the reason why it's returning NULL is that something earlier hasn't been initialized, perhaps *this* is what should be fixed?

Comment 9

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

adding Vlad in CC.

Comment 10

[Aaron Train [:aaronmt]]

Another crash URL

http://www.peternitsch.net/demo/webgl/index.html

Comment 11

[Saint Wesonga]

(In reply to comment #8)
> So, do you confirm it was actually returning NULL?
> 

Yes, it was actually returning NULL.

> Next question: if the reason why it's returning NULL is that something earlier
> hasn't been initialized, perhaps *this* is what should be fixed?


The problem here is that http://mxr.mozilla.org/mozilla-central/source/content/canvas/src/WebGLContextValidate.cpp#450 checks the mShaderValidation while it is still false and therefore doesn't call ShInitialize before returning to http://mxr.mozilla.org/mozilla-central/source/content/canvas/src/WebGLContext.cpp#299 which then proceeds to set mShaderValidation to true (from the pref in http://mxr.mozilla.org/mozilla-central/source/content/canvas/src/WebGLContext.cpp#350).

This needs to be fixed, but it's probably better to leave the null check in as well.

Comment 12

[Saint Wesonga]

Attachment: Patch

Comment 13

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

Attachment: fix

So, we really want validation enabled everywhere, even on mobile (because mobile might accept a more liberal syntax due to extensions).  This simplifies things a bit, and we check the pref in Initialize etc.

Comment 14

[Marcia Knous [:marcia]]

Adding the other stack so it gets picked up in crash stats.

Comment 15

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

http://hg.mozilla.org/mozilla-central/rev/8767a9c9a4bd