981240 - GL_DEPTH_ATTACHMENT gets stuck (can't unset it) in WebGL

Status: RESOLVED FIXED

(Core :: Graphics: CanvasWebGL, defect)

Description

[Greg]

User Agent: Mozilla/5.0 (Windows NT 6.3; WOW64; Trident/7.0; Touch; .NET4.0E; .NET4.0C; .NET CLR 3.5.30729; .NET CLR 2.0.50727; .NET CLR 3.0.30729; Tablet PC 2.0; ASU2JS; rv:11.0) like Gecko

Steps to reproduce:

I'm a game engine developer currently using Emscripten to port my engine to web.
I've noticed that there's a bug in Firefox (it works on Chrome and Opera) in the following scenario.

Setting a depth render buffer makes it stuck, and next call to unset it probably doesn't work, which then causes issue of GL_FRAMEBUFFER_INCOMPLETE_DIMENSIONS_EXT (when trying to set a color render buffer without the depth buffer).

Here's the C++ code:
void test()
{
   GLuint FBO, _txtr, _small, _depth;
   glGenFramebuffers(1, &FBO);
   glBindFramebuffer(GL_FRAMEBUFFER, FBO);

   glGenTextures(1, &_txtr);
   glBindTexture(GL_TEXTURE_2D, _txtr);
   glTexImage2D(GL_TEXTURE_2D, 0, GL_RGBA, 150, 150, 0, GL_RGBA, GL_UNSIGNED_BYTE, NULL);

   glGenTextures(1, &_small);
   glBindTexture(GL_TEXTURE_2D, _small);
   glTexImage2D(GL_TEXTURE_2D, 0, GL_RGBA, 75, 75, 0, GL_RGBA, GL_UNSIGNED_BYTE, NULL);

   glGenRenderbuffers(1, &_depth);
   glBindRenderbuffer   (GL_RENDERBUFFER, _depth);
   glRenderbufferStorage(GL_RENDERBUFFER, GL_DEPTH_COMPONENT16, 150, 150);

   if(glGetError()!=GL_NO_ERROR)printf("glGetError\n");
   glFramebufferTexture2D   (GL_FRAMEBUFFER, GL_COLOR_ATTACHMENT0, GL_TEXTURE_2D  , _txtr, 0);
   glFramebufferRenderbuffer(GL_FRAMEBUFFER, GL_DEPTH_ATTACHMENT  , GL_RENDERBUFFER, _depth);
   glFramebufferRenderbuffer(GL_FRAMEBUFFER, GL_STENCIL_ATTACHMENT, GL_RENDERBUFFER, 0);

   if(glGetError()!=GL_NO_ERROR)printf("glGetError\n");
   GLenum error =glCheckFramebufferStatus(GL_FRAMEBUFFER);
   if(    error!=GL_FRAMEBUFFER_COMPLETE)
   {
      printf("err %d \n", error);
   }

   if(glGetError()!=GL_NO_ERROR)printf("glGetError\n");
   glBindFramebuffer(GL_FRAMEBUFFER, FBO);
   glFramebufferTexture2D   (GL_FRAMEBUFFER, GL_COLOR_ATTACHMENT0, GL_TEXTURE_2D  , _small, 0);
   glFramebufferRenderbuffer(GL_FRAMEBUFFER, GL_DEPTH_ATTACHMENT  , GL_RENDERBUFFER, 0);
   glFramebufferRenderbuffer(GL_FRAMEBUFFER, GL_STENCIL_ATTACHMENT, GL_RENDERBUFFER, 0);

   if(glGetError()!=GL_NO_ERROR)printf("glGetError\n");
   error =glCheckFramebufferStatus(GL_FRAMEBUFFER);
   if(    error!=GL_FRAMEBUFFER_COMPLETE)
   {
      printf("err %d \n", error);
   }
}


Actual results:

after trying to set the framebuffer for the 2nd time with GL_DEPTH_ATTACHMENT  cleared to 0 I get GL_FRAMEBUFFER_INCOMPLETE_DIMENSIONS_EXT 


Expected results:

I should get GL_FRAMEBUFFER_COMPLETE

This works ok on Chrome/Opera but not on FF

Comment 1

[Greg]

You can see this bug live at this page:
http://esenthel.com/site/webgl_test/Project.html
(it's a temporary link for the work-in-progress demo that I'm making).

You can see that it's working differently on Chrome/Opera and on Firefox due to this bug.
On Chrome there's a nice bloom and glow effect, which you can't see on Firefox.

the bloom and glow are done by downsampling the graphics to a smaller render target, because on FF it fails to setup the framebuffer, the entire downsampled data is set to zero, resulting in no bloom/glow.

Hope this can be resolved quickly, as in a few days I'll be realeasing the Web support for my engine and make the demo public.
Wish that Firefox users would have the same experience as Chrome users when trying the demo.

Comment 2

[Kelsey Gilbert [:jgilbert]]

Neat. I'll check.

Comment 3

[Kelsey Gilbert [:jgilbert]]

Attachment: testcase

Comment 4

[Kelsey Gilbert [:jgilbert]]

Yep, this is pretty simple. Fix incoming.

Comment 5

[Kelsey Gilbert [:jgilbert]]

Strangely, this doesn't reproduce on my linux desktop.

Comment 6

[u480271]

I checked Chromium Canary vs Nightly and I could see very little graphical difference.

Comment 7

[Kelsey Gilbert [:jgilbert]]

Attachment: patch

Comment 8

[Kelsey Gilbert [:jgilbert]]

https://tbpl.mozilla.org/?tree=Try&rev=650651527290

Comment 9

[u480271]

Comment on attachment 8388935 [details] [diff] [review]
patch

Review of attachment 8388935 [details] [diff] [review]:
-----------------------------------------------------------------

::: content/canvas/src/WebGLFramebuffer.h
@@ +57,5 @@
>          bool HasAlpha() const;
>          bool IsReadableFloat() const;
>  
>          void SetTexImage(WebGLTexture* tex, GLenum target, GLint level);
> +        void SetRenderbuffer(WebGLRenderbuffer* rb);

Good!

Comment 10

[Kelsey Gilbert [:jgilbert]]

remote:   https://hg.mozilla.org/integration/mozilla-inbound/rev/a6fc178ac9a1

Comment 11

[Greg]

Thanks for the progress.

I've moved the demo with the issue over here:
http://www.esenthel.com/?id=live_demo

(in case you need it for tests)

Comment 12

[Carsten Book [:Tomcat]]

https://hg.mozilla.org/mozilla-central/rev/a6fc178ac9a1

Comment 13

[Kelsey Gilbert [:jgilbert]]

I'm going to REOPEN this, since I think it's important we add the testcase here.