Valid trace which crashes both glretrace and qapitrace

[turol]

Using this trace:

https://www.dropbox.com/s/oj0qjzsdi7m0wl2/Overgrowth.bin.x86_64.trace.xz?dl=0

this command:

glretrace -D 8349 --dump-format ubjson Overgrowth.bin.x86_64.trace > /dev/null

crashes with

retrace/glstate_images.cpp:310: bool glstate::getActiveTextureLevelDesc(glstate::Context&, GLenum, GLint, glstate::ImageDesc&): Assertion '0' failed.

This in turn causes qapitrace to crash with:

gui/qubjson.cpp:138: int readSize(QDataStream&, ubjson::Marker): Unimplemented code.
ASSERT: "0" in file gui/qubjson.cpp, line 284

I think this has something to do with texture buffer objects.

[schulmar]

The problem seems to be that the type for GL_RED_INTEGER's InternalFormatDesc is GL_NONE, which leads to the assert.

[schulmar]

More precisely it seems that commit 8374a0d66ca9228269e860089c1535fcda9f9c5f changed the semantics of the if because before it would trigger if the format was not found and now it triggers on type == GL_NONE.

[jrfonseca]

This is a bug in NVIDIA driver.

2575 glBindTexture(target = GL_TEXTURE_BUFFER, texture = 17)
2576 glTexBuffer(target = GL_TEXTURE_BUFFER, internalformat = GL_R32UI, buffer = 7)

but querying GL_TEXTURE_BUFFER_FORMAT_ARB or GL_TEXTURE_INTERNAL_FORMAT returns GL_RED_INTEGER, and not GL_R32UI as it should.

Mesa OpenGL drivers works fine.

I'm afraid there's not much I can do. GL_RED_INTEGER is untyped and does not explain how to interpret the texture buffer (it could be integer, float, 32 bits, 16 bits, etc.)

I'll replace the assertion with a warning, but you won't be able to see the contents of those buffers.

[jrfonseca]

@ahuillet

[turol]

@jrfonseca Not seeing the buffer contents is no big deal, I can always go look at the glBufferData command if I need to. The crashing however IS a big deal since I can't see any states in certain draw calls. Can you fix qapitrace so that it deals gracefully with crashing glretrace?

Thanks for fixing this.

[jrfonseca]

Can you fix qapitrace so that it deals gracefully with crashing glretrace?

Fixing this particular crash is easy.

Fixing generic crashes is harder, but I agree worthwhile. I'll try but I can't guarantee ETA.

BTW, if you apitrace build in release (ie pass -DCMAKE_BUILD_TYPE=Release) it won't assert and neither glretrace or qapitrace should crash.

[ahuillet]

I have filed NVIDIA bug 1733820 to track this issue. If you can provide a repro application, or an update to Piglit's tests/texturing/texture-rg.c to test integer formats, this would be helpful. Thanks

[ahuillet]

The problem in the NVIDIA driver was identified and fixed a while ago.

[jrfonseca]

@ahuillet Thanks!