Thank you so much @lorentey for putting together such a well thought through pitch. The functionality this will offer is a major building block for high-performance applications and libraries and is crucial to make Swift more useful as a systems programming language.
In SwiftNIO we have been missing this sort of functionality since forever. To SwiftNIO specifically this will improve two things:
- minor: we can delete a lot of C code (
CNIOAtomics) - major: because of the memory ordering guarantees proposed, we will be able to remove a number of locks from some hot code paths (for example in the
EventLooptask list) which should provide really nice speedups in high-load scenarios
+1 for leaving out the double-words for now. Although I already can’t wait for that proposal 
+1 also on offering the low-level APIs first. They might not look as pretty as other things in the stdlib but it’s IMHO exactly the right call: it allows library authors to build the lock-free/atomic things that users can then use with a nice API that fits whatever is being built.
Overall +💯