The current code base uses atomic operations through platform specific APIs available on all supported systems. The current code mixes atomic and non-atomic operations according to context the variable is accessed in, and whether or not historical values yield acceptable behavior.

Mixing atomic/non-atomic operations is a data race and as such is undefined behavior.

Insights gained from studying the semantics of the C/C++ memory model could be applied to the existing code.

This arose from a discussion around WT-11460; and the purpose of this ticket is to account for effort in facilitating a discussion.