pyopengl-users

[PyOpenGL-Users] glDeleteBuffers() and arrays.vbo

[Dan H. <Dan...@no...>]

Hi all,

I may have found a bug in PyOpenGL's arrays.vbo module, but it could be 
a problem with my code or system. When a vertex or other buffer is 
automatically deleted by PyOpenGL's deleter, here's the error that I get:

Exception OpenGL.error.NullFunctionError: NullFunctionError('Attempt to 
call an undefined function glDeleteBuffers, check for 
bool(glDeleteBuffers) before calling',) in <function doBufferDeletion at 
0x0A8892B0> ignored

To me, that sounds like my OpenGL implementation (from Nvidia) doesn't 
implement glDeleteBuffers() for whatever reason. The deleter code in 
vbo.py appears to check for this, but it checks for an AttributeError 
rather than an OpenGL.error.NullFunctionError, or better yet, doing a 
protective bool(glDeleteBuffers) before actually calling glDeleteBuffers().

By the way, is it problematic that when doing a manual VBO.delete(), the 
automatic deleter is still invoked? Shouldn't VBO.delete() pop the 
deleter so glDeleteBuffers() isn't called twice for a particular VBO? Or 
is that not a problem?

Dan

Re: [PyOpenGL-Users] glDeleteBuffers() and arrays.vbo

[Mike C. F. <mcf...@vr...>]

Dan Helfman wrote:
> Hi all,
>
> I may have found a bug in PyOpenGL's arrays.vbo module, but it could be 
> a problem with my code or system. When a vertex or other buffer is 
> automatically deleted by PyOpenGL's deleter, here's the error that I get:
>
> Exception OpenGL.error.NullFunctionError: NullFunctionError('Attempt to 
> call an undefined function glDeleteBuffers, check for 
> bool(glDeleteBuffers) before calling',) in <function doBufferDeletion at 
> 0x0A8892B0> ignored
>   
This sounds like the VBO is getting deleted after module cleanup?  You
should always have a glDeleteBuffers implementation if you have a
glGenBuffers implementation.  Basically this code is getting called
after the system has done so much cleanup that the functions aren't
available any more (at least, that's my interpretation).
> To me, that sounds like my OpenGL implementation (from Nvidia) doesn't 
> implement glDeleteBuffers() for whatever reason. The deleter code in 
> vbo.py appears to check for this, but it checks for an AttributeError 
> rather than an OpenGL.error.NullFunctionError, or better yet, doing a 
> protective bool(glDeleteBuffers) before actually calling glDeleteBuffers().
>   
That's now fixed in bzr head, will show up in the 3.0.1a2 release.  i.e.
we catch the NullFunctionError, so you shouldn't see the ignored error
messages any more.
> By the way, is it problematic that when doing a manual VBO.delete(), the 
> automatic deleter is still invoked? Shouldn't VBO.delete() pop the 
> deleter so glDeleteBuffers() isn't called twice for a particular VBO? Or 
> is that not a problem?
>   
The auto-deleter uses the same list (self.buffers of the vbo), so the
delete method's pop() from that list meant there was nothing left in the
auto-deleter.  It shouldn't be possible for the deleter to get called
before the delete method, but just in case I've altered it to also use
pop() to retrieve values (rather than simple iteration).  Calling
glDeleteBuffers() twice for the same buffer would be a problem if, for
instance, another vbo had already been allocated that ID.

That said, you *shouldn't* ever need to explicitly call delete on the
VBO instance unless for some reason you want to "reuse" the VBO object
but not the GL-side buffer (I can't think of a real use-case for that).

Thanks for the bug report,
Mike

-- 
________________________________________________
  Mike C. Fletcher
  Designer, VR Plumber, Coder
  http://www.vrplumber.com
  http://blog.vrplumber.com

Re: [PyOpenGL-Users] glDeleteBuffers() and arrays.vbo

[Dan H. <Dan...@no...>]

Mike C. Fletcher wrote:
> Dan Helfman wrote:
>> Exception OpenGL.error.NullFunctionError: NullFunctionError('Attempt to 
>> call an undefined function glDeleteBuffers, check for 
>> bool(glDeleteBuffers) before calling',) in <function doBufferDeletion at 
>> 0x0A8892B0> ignored
>>   
> This sounds like the VBO is getting deleted after module cleanup?  You
> should always have a glDeleteBuffers implementation if you have a
> glGenBuffers implementation.  Basically this code is getting called
> after the system has done so much cleanup that the functions aren't
> available any more (at least, that's my interpretation).

This is happening when a VBO instance is finalized after its containing 
object goes away. And as far as I can tell, the deleter is only being 
called once for each VBO. So I'm not sure what else could be doing the 
cleanup unless it's in the underlying C code. I put some prints in the 
explicit VBO.delete() call, so I've confirmed that that's not getting 
called.

I should mention that this is all with the 3.0.0 release.

> That's now fixed in bzr head, will show up in the 3.0.1a2 release.  i.e.
> we catch the NullFunctionError, so you shouldn't see the ignored error
> messages any more.

Great, thanks!

> That said, you *shouldn't* ever need to explicitly call delete on the
> VBO instance unless for some reason you want to "reuse" the VBO object
> but not the GL-side buffer (I can't think of a real use-case for that).

Gotcha.

Dan

Re: [PyOpenGL-Users] glDeleteBuffers() and arrays.vbo

[Dan H. <Dan...@no...>]

Dan Helfman wrote:
> Mike C. Fletcher wrote:
>> Dan Helfman wrote:
>>> Exception OpenGL.error.NullFunctionError: NullFunctionError('Attempt to 
>>> call an undefined function glDeleteBuffers, check for 
>>> bool(glDeleteBuffers) before calling',) in <function doBufferDeletion at 
>>> 0x0A8892B0> ignored
>>>   
>> This sounds like the VBO is getting deleted after module cleanup?  You
>> should always have a glDeleteBuffers implementation if you have a
>> glGenBuffers implementation.  Basically this code is getting called
>> after the system has done so much cleanup that the functions aren't
>> available any more (at least, that's my interpretation).
> 
> This is happening when a VBO instance is finalized after its containing 
> object goes away. And as far as I can tell, the deleter is only being 
> called once for each VBO. So I'm not sure what else could be doing the 
> cleanup unless it's in the underlying C code. I put some prints in the 
> explicit VBO.delete() call, so I've confirmed that that's not getting 
> called.
> 
> I should mention that this is all with the 3.0.0 release.

Follow-up: Upon further investigation, turns out that this is a 
threading issue. GL and the VBOs were initialized in one thread, and 
then VBO finalization was getting triggered from a different thread. As 
soon as I changed it so that the VBO deleter only got triggered from the 
correct thread, the NullFunctionErrors went away.

Dan