I couldn’t find anything in the standard saying that the pointer cast to atomic type is defined, or that the atomic typed are the same in c and c++ which was why I did not want to use them. Otoh, c++20 has https://en.cppreference.com/w/cpp/atomic/atomic_ref which should be well defined to do what we want here. I don’t know if there’s something like that added/proposed for c though…

This is finally specified in C++23:

_Atomic(T) is identical to std::atomic while both are well-formed

This came out of http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2020/p0943r6.html, and the existing bugs you mentioned are slowly being addressed in GCC: https://gcc.gnu.org/bugzilla/show_bug.cgi?id=65146. Though those corner cases are relevant for common struct objects, but not for the primitive types we usually annotate here.

OTOH, atomic_ref<T> is not compatible with C11, as it does have different alignment expectations. We currently have the static analyzer disabled for these checks, since we weren't correctly annotating our fields to ensure the expected alignment.

This needs a manual backport against #42255.

Just point out that this has still not been backported since someone needs to do it manually (@barche, if you are interested in this, perhaps you can help out?)

OK, how does this work in practice? Make a separate PR to merge into the release-1.7 branch?

againt backports-release-1.7 yeah