[Haskell-cafe] Re: FRP for game programming / artifical life simulation

[Daniel Fischer]

Am Freitag 30 April 2010 17:23:19 schrieb Antoine Latter:
> On Fri, Apr 30, 2010 at 3:37 AM, Daniel Fischer
>
> <daniel.is.fischer at web.de> wrote:
> > Am Donnerstag 29 April 2010 20:08:00 schrieb Ben:
> >> A technical question: it seems like the instance of ArrowLoop is too
> >> strict (this is something I've wondered about in Liu's paper too.)
> >> Shouldn't it be
> >>
> >>  instance ArrowLoop SFAuto where
> >>      loop (SFAuto s f) = SFAuto s f'
> >>          where
> >>            f' (b, s1) = let (~(c, d), s2) = f ((b, d), s1)
> >>                         in (c, s2)
> >
> > Let-bindings are already lazy, so the '~' makes no difference here.
> > Apart from the readability, both are the same as
> >
> > where
> >  f' (b,s1) = let x = f ((b, snd $ fst x),s1) in (fst $ fst x, snd x)
>
> Ben's version is slightly lazier - even though the let binding is
> lazy, pattern matching is strict.
>
> so (let ((x,y).z) = (undefined, "hello") in z) will exception out, but
> (let (~(x,y),z) = (undefined, "hello") in z) will not.
>
> I don't know if you need that level of laziness, though.

Probably not. Although, you're right, if only s2 is ever looked at and not 
c, Ben's version can give a result where the library instance throws an 
exception.
Was fooled by the use of c in the result.

>
> Antoine