Re: [PATCH] msvcp90: Reimplement condition variables on top of RtlWaitOnAddress().

[Piotr Caban]

On 3/11/22 17:04, Henri Verbeet wrote:
> This is largely the same as the condition variable implementation in ntdll.
> There are essentially two reasons for doing this:
>  - RtlWaitOnAddress() avoids server calls, compared to the existing keyed
>    event implementation.
>  - The NtReleaseKeyedEvent() call in _Cnd_signal() can hang if a thread was
>    killed inside _Cnd_wait(), for example by an application calling exit().
>    Piotr Caban reports this scenario is supposed to work with msvcp140,
>    although not necessarily with earlier versions of msvcp. (And indeed,
>    native msvcp140 is a workaround for the affected applications.)

I've sent a set of patches that tries to fix it differently. It's using 
_Condition_variable to implement _Cnd_* functions in msvcp110/msvcp120. 
In msvcp140 CONDITION_VARIABLE and SRWLock is used instead.

It should fix the problem you're describing in msvcp140. 
_Condition_variable implementation is still to be fixed (currently 
Wine's implementation hangs while native crashes).

Thanks,
Piotr