[Haskell-cafe] Monad transformers [Stacking monads]

[Daniel Fischer]

Am Dienstag, 7. Oktober 2008 20:27 schrieb Andrew Coppin:
> The good news: I managed to turn ResultSet into a monad transformer.
> Yay, me!
>
> The bad news: It generates the entire result set before returning
> anything to the caller.
>
> In other words, it works perfectly for finite result sets, and locks up
> forever on infinite result sets. Since the entire *point* of the monad
> is to handle infinite result sets correctly, that's kind-of a problem.
> And one that I see absolutely no way of fixing. :-(
>
> Basically, the core code is something like
>
>   raw_bind :: (Monad m) => [[x]] -> (x -> m (ResultSet y)) -> m
> (ResultSet y)
>   raw_bind [] f = return empty
>   raw_bind (xs:xss) f = do
>     rsYs <- mapM f xs
>     rsZ <- raw_bind xss f
>     return (foldr union (cost rsZ) rsYs)
>
> As you can see, this generates all of rsZ before attempting to return
> anything to the caller. And I'm really struggling to see any way to
> avoid that.
>

Maybe it is as simple as

raw_bind (xs:xss) f = do
     rsYs <- mapM f xs
     ~rsZ <- raw_bind xss f
     return (foldr union (cost rsZ) rsYs)

then rsZ should only be evaluated when it's needed