Running out of resources when creating and destroying buffers with the Metal backend

[floooh]

Running out of resources when creating and destroying buffers with the Metal backend

[floooh]

Thanks for the bug report! I'll have a look, resource destruction isn't tested in the sokol samples, so that's likely to have slipped through. The timing is good since today I wanted to work on the Oryol-sokol migration for the Metal backend, Oryol has a resource stress test sample which hopefully would also trigger this bug (I'll test that too).

[floooh]

Ok, it was actually 2 problems: first in the garbage-collect function the release-path was never called because the check was broken (as you suspected), and the second part was that the now free pool index was never put back into the 'free queue'.

I made both those things a bit more robust and cleaner now, and I also added a new stress-test to the samples (https://github.com/floooh/sokol-samples/blob/master/metal/releasetest-metal.c), same as your test case, but I'm also creating and destroying the shader and pipeline (and I reduced the resource pool sizes so that the bug would trigger earlier).

Thanks again for the bug report :)