[libcxx-commits] [libcxx] [libcxx] Align `__recommend() + 1` by __endian_factor (PR #90292)

[Aliaksiej Kandracienka via libcxx-commits]

alk wrote:

"How" is up to you guys. I know that simple test posted above works on libc++ 17 and fails on libc++ 18. Shouldn't it be enough? Here is link again: https://paste.debian.net/hidden/8edbffc1/

Some kind of crash will happen on any C++ memory allocator that cares to verify sizes. gperftools on master does in debug builds. "Abseil" tcmalloc from github.com/google/tcmalloc has been doing this for ages. So you should be able to verify regression in multiple ways.

At FreeBSD ticket Dimitry Andric claims that he found the commit(s) (kudos and much thanks for this; @DimitryAndric). Here: https://bugs.freebsd.org/bugzilla/show_bug.cgi?id=279560#c10 But it isn't super trivial to me, so I'll leave it to experts.

"how" might be worth mini-retrospective, especially given that I know for sure that Google does sized delete internally on 100% internal workloads.


https://github.com/llvm/llvm-project/pull/90292