Issue with `pyopengl-accelerate` and `glGenVertexArrays`/`glDeleteVertexArrays`

[pauldmccarthy]

Howdy, I've just hit an issue with pyopengl-accelerate - it seems that glGenVertexArrays returns a scalar np.uint32, but gl.glDeleteVertexArrays requires an array. I'm not sure if other GL functions are affected.

This code works fine when pyopengl-accelerate is not installed (this code was run within an environment where a GL context had already been successfully created):

import OpenGL.GL as gl
vao = gl.glGenVertexArrays(1)
gl.glDeleteVertexArrays(1, vao)

However, with pyopengl-accelerate installed:

In [1]: import OpenGL.GL as gl

In [2]: vao = gl.glGenVertexArrays(1)

In [3]: vao
Out[3]: np.uint32(23)

In [4]: gl.glDeleteVertexArrays(1, vao)
---------------------------------------------------------------------------
ArgumentError                             Traceback (most recent call last)
Cell In[4], line 1
----> 1 gl.glDeleteVertexArrays(1, vao)

File src/latebind.pyx:40, in OpenGL_accelerate.latebind.LateBind.__call__()

File src/wrapper.pyx:315, in OpenGL_accelerate.wrapper.Wrapper.__call__()

File src/wrapper.pyx:312, in OpenGL_accelerate.wrapper.Wrapper.__call__()

File ${PREFIX}/lib/python3.12/site-packages/OpenGL/platform/baseplatform.py:487, in _NullFunctionPointer.__call__(self, *args, **named)
    485 def __call__(self, *args, **named):
    486     if self.load():
--> 487         return self(*args, **named)
    488     else:
    489         try:

ArgumentError: ("argument 2: TypeError: Numpy format handler passed a non-numpy-array object 23 (of type <class 'numpy.uint32'>)", (1, np.uint32(23)))

The error originates from NumpyHandler.c_check_array. So far my naive attempts to patch this file have not worked, as I don't understand how the pyopengl code works. I'll keep poking, but it would be great if somebody with more knowledge of the code base could take a look.

The call to glDeleteVertexArrays does work if I change the code to gl.glDeleteVertexArrays(np.array([vao]). But I believe that this coding pattern is used extensively, so it would be a shame for downstream code to be required to pass arrays, particularly as this code works just fine without pyopengl-accelerate.

Thanks!

[mcfletch]

I've added a test case to test_core, but so far it's worked fine everywhere I've tried it with or without accelerate.

test_scalars_mapped_to_arrays

[pauldmccarthy]

Hi @mcfletch, thanks for taking a look! I evidently missed a critical piece of information in my initial report - I have set OpenGL.ERROR_ON_COPY = True in my code, which seems to be necessary to trigger the exception.

[pauldmccarthy]

Hmm, there's something different in the way that the unit tests are executed - I added print(type(buffer)) to test_scalars_mapped_to_arrays, and it prints <class 'int'>. However if I try to simulate the test environment interactively, I get <class 'numpy.uint32'> (and gl.glDeleteBuffers raises the reported ArgumentError):

import pygame, pygame.display

import numpy as np

import OpenGL

OpenGL.ERROR_ON_COPY = True

import OpenGL.GL as gl

pygame.display.init()
screen = pygame.display.set_mode((300, 300),
    pygame.OPENGL | pygame.DOUBLEBUF)

buffer = gl.glGenBuffers(1)
assert np.isscalar(buffer), type(buffer)
print(type(buffer))
gl.glDeleteBuffers(1, buffer)

[pauldmccarthy]

I tried to add OpenGL.ERROR_ON_COPY to the top of test_core.py (with the other flags), but it had no effect.

What I think is happening is that, during test discovery, pytest is importing all of the test modules, and an import OpenGL.GL statement in another test module is being executed before the configuration flags in test_core/basetestcase are being set (which subsequently have no effect, as the GL functions have already between instantiated and frozen).

In my case, I was able to get test_scalars_mapped_to_arrays to fail by adding OpenGL.ERROR_ON_COPY = True at the top of accelerate/tests/test_arraydatatypeaccel.py (which happens to be the first test module that pytest discovers, and thus the initial invocation of from OpenGL import GL in a test run).

I'm not sure if there is a simple solution to ensure that test modules are executed without import side-effects from other modules.

[pauldmccarthy]

In my case, I was able to get test_scalars_mapped_to_arrays to fail ...

More easily:

PYOPENGL_ERROR_ON_COPY=1 pytest -k test_scalars_mapped_to_arrays