Chris Smith wrote:
apfelmus wrote:
For 1), it's enough to have a primitive
scheduleWriteTVar :: TVar a -> a -> STM ()
that ensures to write the TVar at the very end of the atomically block..
Unfortunately, though, this breaks the very thing that makes STM attractive: namely, composability. Now in order to work with a TVar, I need to know whether anything that came before me might have modified it, and if so take the current value as a parameter instead of reading it like normal.
Or am I misunderstanding something?
You're correct, that's what I meant. But it's nothing more and nothing less than the purely functional way of dealing with "mutable" state, isn't it? And you need a parameter anyway, namely the TVar a itself. I mean, when it's in scope like in do a <- readTVar v writeTVar v (a+1) readTVar v you don't need a parameter. But if the do-block is broken up into functions, you need a parameter foo v = do a <- readTVar v writeTVar v (a+1) bar v bar v = readTVar v and you may as well supply its value instead of the reference v . Regards, apfelmus