806289 - Mesa assertion `src_att->Renderbuffer != ((void *)0)' failed triggered by WebGL test case

Status: RESOLVED WORKSFORME

(Core :: Graphics: CanvasWebGL, defect)

Description

[Christoph Diehl [:posidron]]

Attachment: testcase

Adapter Description: VMware, Inc. -- Gallium 0.4 on llvmpipe (LLVM 0x300)
Device ID: Gallium 0.4 on llvmpipe (LLVM 0x300)
Driver Version: 2.1 Mesa 8.0.4
GPU Accelerated Windows: 0/1 Basic 
Vendor ID: VMware, Inc.
WebGL Renderer: VMware, Inc. -- Gallium 0.4 on llvmpipe (LLVM 0x300)

Comment 1

[Christoph Diehl [:posidron]]

Attachment: callstack

Comment 2

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

Seems to be an assertion that a framebuffer attachment point doesn't have both a texture and a renderbuffer attached to it. Should not be an assertion.

Need to:
 - file bug at freedesktop.org
 - investigate if this requires us to make a work-around. If this is only an assert in debug builds, we don't need to work around it. Does it crash release builds?

Comment 3

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

(I know I said that asserts in debug builds "probably" indicated bad crashers. But this one looks like a really abusive assert).

Comment 4

[Christoph Diehl [:posidron]]

Yes, it does crash my release build Version 16.0.1 on Linux as well.

Comment 5

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

I mean does it crash with a release build of _Mesa_ ?

Comment 6

[Christoph Diehl [:posidron]]

This is the default driver on my Ubuntu 12.04

Comment 7

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

Thanks, I was confused by it being an assertion. So it's really serious and we need a work-around.

Comment 8

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

Let's make this a mentored bug. The goal is to find a workaround so that we don't crash.

 - reproduce on your machine. This is a Mesa bug so you need a Mesa-based driver (e.g. not the NVIDIA proprietary driver). The best is to build llvmpipe manually with http://www.mesa3d.org/llvmpipe.html and run against it with LD_PRELOAD= .  (make a debug build)

 - look at the testcase. it has numeric constants, map them back to named webgl constants using the values in the spec

 - understand what's going wrong in Mesa

 - make a work-around so we don't crash when we run on Mesa.


Ignore what I wrote in comment 2.

Comment 9

[Dan Princ]

(In reply to Benoit Jacob [:bjacob] from comment #8)
> Let's make this a mentored bug. The goal is to find a workaround so that we
> don't crash.

Hey, I'd like to fix this (or at least try :). Can you assign me on it?

Comment 10

[Dan Princ]

Can anyone confirm, that this bug is still happening? I am unable to reproduce it. 

I'm using Firefox 22.0a1 with llvm (latest stable Mesa 9.0.3) on 64bit openSUSE and nothing seems to be wrong when running the testcase. (WebGL otherwise works.)

Comment 11

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

I don't have an old mesa around to test, so let's just take your word --- let's mark this as WORKSFORME. It does happen that Mesa bugs get fixed in a new version. Thanks for checking.