It looks good to me.
I don't understand why my change fixes the problem, given that the
destroy_tex_sampler_cb()function is supposed to release all sampler views created by that context.
Maybe the texture attachment that still holds a sampler view should have been cleaned earlier (and since I can't reproduce the issue with piglit tests, maybe it's a multithreading issue)
(I've edited the log to add a bit more information)