[compiler-rt] [win/asan] Improve SharedReAlloc with HEAP_REALLOC_IN_PLACE_ONLY. (PR #132558)

[Thurston Dang via llvm-commits]

thurstond wrote:

> > Would this approach in general make bug detection worse? The existing behavior of realloc always returning a new pointer (with the old memory marked inaccessible) can catch erroneous code that assumes the realloc is in place (or worse, inconsistently uses both the old pointer and the return value of realloc).
> 
> I am a little confused now - when I read the [documentation to HeapReAlloc](https://learn.microsoft.com/en-us/windows/win32/api/heapapi/nf-heapapi-heaprealloc) I understand the paragraph of HEAP_REALLOC_IN_PLACE_ONLY as it is not allowed to return a different pointer. And if resize cannot be done in place it has to fail e.g return NULL. This patch should just modify behaviour when HEAP_REALLOC_IN_PLACE_ONLY is given.

Sorry, my bad. I wasn't familiar with the HEAP_REALLOC_IN_PLACE_ONLY API. Thanks for the pointer!

After reading it, I still wonder whether implementing this will reduce bug detection. For example, some code might incorrectly not expect NULL to be returned (even though the allocator is allowed to do so), or perhaps returning NULL will force the user code to call realloc without HEAP_REALLOC_IN_PLACE_ONLY (thereby allowing stronger use-after-free detection).


https://github.com/llvm/llvm-project/pull/132558