Mateusz Kowalczyk <fuuze...@fuuzetsu.co.uk> writes:
> Thanks for the investigation. While I have not gotten a chance to play
> with this myself, it does seem like your MVar implementation has a sound
> basis.
That's good! Thanks for making sure we understood the requirements.
> While it'd be great to have someone familiar with the inner-workings to
> step in and confirm your findings, it seems that your implementation
> should work, at least from the scheduling perspective. I can not guess
> what the actual performance might be as I'm not familiar with Guile's
> performance profile but I suppose that's another issue.
I wouldn't expect very high performance from the current implementation,
which was written in Scheme in the most straightforward manner possible,
and built upon fairly heavy "fat" mutexes and condition variables. I
could write a much faster implementation in C, based on other pthread
primitives and perhaps optimized using primitives from libatomicops, and
perhaps we should do that at some point.
Mark
