Wow, lots of great ideas. Thanks, guys. Lemmih, I worry about the uncertainty in the semantics that seems to be introduced by the unsafe stuff. But I actually hadn't noticed GHC.Conc.unsafeIOToSTM before, so it's good to know it's there. Rich, Even if I use randomIO outside the STM code, I don't know of a (safe) way to bring it in. Anyway, the number of random values needed depends on other stuff going on within the STM part. Christian, I think setStdGen has the same problem as just using randomIO and ignoring the initial seed - there's no nice way of moving the (IO a) value generated into the STM monad. Rob, Henning, I think I'll take this approach, or something similar. Thanks for the pointers. -Chad

On 1/12/07, Rich Neswold wrote:

On 1/12/07, Chad Scherrer wrote:

...

Even if I use randomIO outside the STM code, I don't know of a (safe) way to bring it in.

Define your STM action to be (Int -> STM s). Generate the random number and then pass it in:

mySTM :: Int -> STM a mySTM n = do { ... }

To use it:

do { n <- getStdRandom (...) ; atomically (mySTM n) }

...

Anyway, the number of random values needed depends on other stuff going on within the STM part.

Ah. This detail removes my suggestion. But how about this?

randomizer :: TMVar Int -> IO () randomizer v = do { n <- getStdRandom (...) ; atomically (putTMVar v n) ; randomizer v }

Start the randomizer action using forkIO. This gives you a steady supply of random numbers in the STM monad just by reading the TMVar (via takeTMVar).

Rather than having a separate thread computing the random numbers using IO, why not just stick an StdGen in a TVar and write a function like: type RandomVar = TVar StdGen rnd :: RandomVar -> STM a rnd var = do g <- readTVar var let (r,g') = random g writeTVar var g' return r -- Sebastian Sylvan +46(0)736-818655 UIN: 44640862

On Sat, Jan 13, 2007 at 01:49:36PM +1000, Matthew Brecknell wrote:

...

Rather than having a separate thread computing the random numbers using IO, why not just stick an StdGen in a TVar and write a function like:

type RandomVar = TVar StdGen

rnd :: RandomVar -> STM a rnd var = do g <- readTVar var let (r,g') = random g writeTVar var g' return r

The user of this approach should be aware that it may lead to non-determinism. That is, the sequence of psuedo-random numbers extracted by any one thread will depend on those extracted by other threads, which may in turn depend on the scheduling of those threads.

The TVar approach might also lead to excessive STM transaction abort-retry cycles, if multiple threads are retrieving many pseudo-random numbers via a single TVar.

As a workaround for these problems you can create a separate RandomVar for each "atomically" block. Best regards Tomasz