On Dec 28, 2005, at 1:38 PM, Tomasz Zielonka wrote:
On Wed, Dec 28, 2005 at 05:28:28PM +0000, Chris Kuklewicz wrote:
But STM, wrapped in small pieces, makes for interesting IO commands (untested):
waitForZero :: (Num a, Ord a) => (TVar a) -> IO () waitForZero tv = atomically $ do v <- readTVar tv when (v>0) retry
This function is rather useless in IO - you will get race conditions. That's what you get for leaving the STM wonderland ;-)
Actually, I think it works in the particular instance given of constructing read-write locks because the call to waitForZero is guarded by a mutex. But it would certainly be easy to introduce a race condition using constructs like this. Given the alternatives {use STM fully, use STM some, don't use STM}, I would have a hard time justifying the "use STM some" alternative (at least for new programs). If you are OK with introducing a dependency on STM, why not go whole hog? Rob Dockins Speak softly and drive a Sherman tank. Laugh hard; it's a long way to the bank. -- TMBG