[M3devel] Modula-3 condition variables vs. pthread/Win32 condition variables?

[Tony Hosking]

On 18 Jul 2010, at 02:36, Jay K wrote:

> 
> I was thinking of writing "ThreadWin6.m3".
> We'd check GetVersion and if it is >= 6.0, use it instead.
> This would give us:
>   smaller locks
>   condition variables
>   likely isomorphic code with ThreadPThread.m3
> 
> 
> Question, probably a repeat: (I'll check the archives)
>   Why aren't Modula-3 condition variables a thinner wrapper over pthread condition variables?

Because there is no reliable way to alert a thread waiting on a condition variable.  And we need a uniform way to handshake with a thread for GC, etc.  Better just to build on the per-thread CV.

> One difference I see is alert. Can't alert just signal the condition being waited on?
> 
> 
> Might need interlocked operations against alertable and alerted, and maybe also thread.cond?
>   (Should be viable, even within one word, use the lower two bits of thread.cond?)

Again, probably not a good idea going forward -- we will ultimately build M3 mutex/CV with non-blocking primitives where possible and inflate to pthread mutex only rarely.

> 
> ?
> 
> 
>  - Jay
> 
> 
> 
> 
>