⚠️ Note to self:

This will need the usual MSVC-internal changes to add a new header.

One of scary things here is locking down the _Cnd_t and _Mtx_t.
We cannot further change them after this change, as the assumption on their structure are embedded into import library.

One of scary things here is locking down the _Cnd_t and _Mtx_t. We cannot further change them after this change, as the assumption on their structure are embedded into import library.

The alternative could be a satellite DLL, this makes the change possible without core header extraction, but this can be more cumbersome overall.

I think I'm fine with that limitation - the headers (not just the import lib) already know about the mutex representation, and the condition variable representation has been boiled down to its essentials now.

Satellite DLLs are really annoying. I don't think we'll ever use them again unless we're absolutely forced to.

There's one more transformation I want to perform (I want to get rid of the forward declaration squirrelliness) but that won't affect representation.

I'm mirroring this to the MSVC-internal repo - please notify me if any further changes are pushed.

Thanks for fixing this notorious bug that we previously thought was unfixable! ⭐ 😻 🎉

_Cnd_timedwait_for is missing _CRTIMP2_PURE, forcing static linkage always. This is problematic, since the function is missing from the DLL version of the runtime library. This is also issue with the current replacement _Cnd_timedwait_for_unchecked

This is intentional; we routinely inject new functions (non-dllexported) into the STL's import lib, effectively statically linking them. We require that the final link be performed by a toolset that's at least as new as any of the toolsets used to produce OBJs/LIBs participating in that link.