+1, but to quibble, even for trivial things like this we usually try to allow at least 2 weeks for discussion.

Why not just import STRet from GHC.ST and exploit that rather than redefine it? Just curious -- I'm not biased one way or the other.

-Edward

On Thu, Feb 17, 2011 at 8:57 AM, Sebastiaan Visser <haskell@fvisser.nl> wrote:

Hi all,

The STM monad currently has no MonadFix instance, but wonderful things are possible when it has one.

I propose adding the 'MonadFix STM' instance provided by Antoine Latter on the Haskell-Cafe[1] list to the STM package:

> {-# LANGUAGE MagicHash, UnboxedTuples, DoRec #-}
>
> import GHC.Exts
> import GHC.Conc
> import Control.Monad.Fix
>
> data STMret a = STMret (State# RealWorld) a
>
> liftSTM :: STM a -> State# RealWorld -> STMret a
> liftSTM (STM m) = \s -> case m s of (# s', r #) -> STMret s' r
>
> instance MonadFix STM where
>   mfix k = STM $ \s ->
>     let ans        = liftSTM (k r) s
>         STMret _ r = ans
>     in case ans of STMret s' x -> (# s', x #)

Discussion Period: 1 week



-Sebastiaan Visser

[1] http://www.haskell.org/pipermail/haskell-cafe/2011-February/089226.html