[Haskell-cafe] Sparks created, but not used (par/pseq, ST monad)

[Daniel Fischer]

Am Donnerstag 04 Februar 2010 17:50:57 schrieb Michael Lesniak:
> Hello haskell-cafe,
>
> I currently have a problem with running par/pseq in the ST monad. The
> function testST is the minimal counterexample that works -- or, to be
> more clear, does not work as expected for me. As a remark, the
> tasks/function calls in my "real application" are much more
> computational expensive, but the code is too long to post here.
>
> > -- File ST.lhs
> > module Main where
> > import Control.Monad.ST
> > import Control.Parallel
> > main = testST
> >
> > testST :: IO ()
> > testST = do
> >    putStrLn "Starting"
> >    (runST $ f 10) `pseq` putStrLn "Stopping"
> >  where
> >        f :: forall s. Int -> ST s ()
> >        f n = do
> >            p $ "\nTask:" ++ show n
> >            if n < 0
> >                then return ()
> >                else do
> >                    (n1,n2) <- return (n-1,n-2)
> >                    q n1
> >                    q n2
> >                    a <- f n1
> >                    b <- f n2
> >                    (a `par` b) `pseq` return a
> >
> > -- Some helper functions
> > p x = unsafeIOToST (putStrLn x)
> > q x = unsafeIOToST (print x)
>
> As far as I understand, compiling with
>
>     ghc --make -O2 -threaded ST.lhs -o st -XRankNTypes
>
> and running with
>
>     ./st +RTS -N -s

You'd need to give a number of capacities, I think (-N2 e.g.).

>
> should create sparks that could be run in parallel. The problem I have
> now, is that sparks are created but not converted, according to -s
> output:
>
>     ... a lot of (uninteresting) stuff
>     SPARKS: 232 (0 converted, 0 pruned)
>     ...
>
> I have not found any information on this type of behaviour on the net
> and would be glad if someone could give me points or hints what's
> happening and how I can improve this.

I think with the strict ST monad, when you have
    a <- f n1
    b <- f n2
they are already evaluated, so there's no point in sparking evaluation in 
parallel.

>
> Cheers,
>   Michael