On 07 January 2005 22:17, Tomasz Zielonka wrote:

On Fri, Jan 07, 2005 at 11:00:27PM +0100, Tomasz Zielonka wrote:

...

Hmmm, TMVar's seem to be significantly (ie. 6 times) faster than MVars in some simple tests :)

Is it expected?

If it is, we can reimplement MVars using STM, when STM becomes stable :)

No, TMVars are quite a bit slower than MVars if you use the basic takeMVar/putMVar ops. However, if you use the safer withMVar/modifyMVar combinators, then you'll probably find that performance is about the same. A simple test I did comparing TChan with Chan came up with roughly the same performance, and STM isn't really tuned yet. Cheers, Simon