Possible UB in ByteString hash function

[MarkCMann]

The ByteString hash function caches previously calculated values. The member variable that stores the cached value is a non-atomic Int.

I believe that kotlinx-ios hash cache logic was inspired by the google protobuf ByteString hash implementation. But that hash caching has a disclaimer that the non-atomic Int backing the cache is safe because of JVM memory model semantics.

Is it safe to use a non-atomic Int for Kotlin native builds? Does the hash cache member variable need to be converted to an Atomic<Int> for non-JVM platforms?

[fzhinkin]

Thanks for reporting it!

There should be no issues on currently supported targets: on both X86_64 and Aarch64 (in fact, Aarch32, too) word-sized accesses are atomic as long as an address is word-aligned (which should be true for a field).

But the fact that things work correctly right now should not be taken for granted, I'll check what could be done here.