Correctness of DefaultReallocFunc about xgl HOT 5 CLOSED

Since there isn't any aligned realloc primitive in POSIX, I think a good workaround would be to prepend a header to each allocation and remember the allocation size there.

from xgl.

That looks fairly broken. I'm not aware of a case where we use realloc but I'm guessing you've hit one.

from xgl.

@JonCampbell407 Are you fine with moving forward with a fix along these lines?

I think a good workaround would be to prepend a header to each allocation and remember the allocation size there.

from xgl.

Doing that would add a small amount of overhead to every allocation. Since realloc is never used in the driver I'd rather not do that. I think we should just change DefaultReallocFunc() to assert with VK_NEVER_CALLED(). What are you doing to hit this? I can't think of a case where we would want to use realloc in the driver.

from xgl.

Like you suggest, bailing out sound more lightweight that doing extra work on every allocation and free. I don't know if there would be any measurable slowdown caused by adding such header to every allocation.
I'll prepare a PR to return a nullptr and assert in reallocation to make sure things don't misbehave in the future..

I hit this when working on cache creator and testing corner cases related to possible error codes after my refactoring changes. The ICD doesn't use realloc but we have a vulkan layer test that does reallocation. I wanted to move default vulkan allocation function implementation to either PAL or the XGL cache support library and use them in the cache creator tool.

from xgl.