glretrace crashes

[apitrace]

glretrace crashes

[jrfonseca]

Please say which OS you have, and GL driver you have installed.

If you can provide the trace, that would be the best. If the trace is big then please upload to some web server. You can send it to my privately (to jose dot r dot fonseca at gmail dot com) and package inside a password protected.

Otherwise, if you're on Linux/Darwin, please invoke:

$ gdb --args glretrace -v /path/to/trace/that/causes/crash
(gdb) run
...
crash will happen
(gdb) bt

[jrfonseca]

And paste the final output of that.

[jrfonseca]

This was a bug on the r600 gallium GL driver.

These were the last calls when retracing:

glTexCoordPointer(size = 2, type = GL_FLOAT, stride = 24, pointer = blob(2576))
glColorPointer(size = 4, type = GL_UNSIGNED_BYTE, stride = 24, pointer = blob(2572))
glVertexPointer(size = 3, type = GL_FLOAT, stride = 24, pointer = blob(2580))
glDrawElementsBaseVertex(mode = GL_TRIANGLE_STRIP, count = 110, type = GL_UNSIGNED_SHORT, indices = blob(220), basevertex = 54)

the biggest index in the "indices" buffer is 53.

apitrace recorded precisely the amount of data necessary:

blob size = (maxIndex + basevertex)*stride + size * sizeType
>>> (53 + 54)*24 + 2 * 4
2576
>>> (53 + 54)*24 + 4 * 1
2572
>>> (53 + 54)*24 + 3 * 4
2580

but the GL driver tried to read more that.

This is likely because the GL driver is mis-calculating the vertex data size, using more the approximate bug wrong formula of:

blob size = (maxIndex + basevertex + 1)*stride
>>> (53 + 54 +1)*24
2592

which ends up causing reading a few extra bytes. This is a common and recurrent mistake on Mesa.

You'll need to file a bug within the GL driver.