Sven Panne writes:
> Ralf Hinze wrote:
> > [...] Rewriting the definition to
> >
> > > (@@) f g a = g a >>= f
> >
> > avoids the problem.
>
> I discovered something similar yesterday. Things are even more funny:
>
> > (f @@ g) = \a -> g a >>= f -- Does not work
>
> > f @@ g = \a -> g a >>= f -- Does work!
>
Weird as it may seem, the former version of (@@) above and Ralf's
example
@@ :: (Monad m) => (a -> m b) -> (c -> m a) -> c -> m b
(f @@ g) a = g a >>= f
are both examples of value definitions with invalid LHSes.
The LHS for a valdef in Haskell is either:
- a pattern, e.g., "(x:xs)"
- a var followed by a series of patterns, e.g., "f (n+1) (x:xs)"
- the infix binary application of a varop, e.g., "f @@ [g,h]".
The LHS "(f @@ g) =" doesn't fall into any of these since (bracketed)
patterns cannot contain infix applications of varops. If @@ is
substituted for a constructor, everything is cool.
"(f @@ g) a =" fails because of the same, but has the additional
problem of mixing infix varop application with a trailing apat ("a"),
which just ain't legal.
Don't let any other implementations of the Haskell grammar tell you
otherwise :-)
--Sigbjorn
