actually, GHC gives me "could not deduce Blah f a from Blah f1 a"
first. It seems that desugaring type function notation into an additional
constraint helps, so there's something odd going on:

Silly me, I didn't paste the whole type error. Yes, GHC gives both. I should
add that I tested this under GHC 6.8.2. But this is known not to work with a
(one/two months old) GHC head.

yes, I tested with 6.9.20080514. And just in case my phrasing
was unclear: desugaring the type function application makes the
error go away, so you have a workaround.

It is just confusing that this example shows that the "desugaring"
is not a desugaring, in the current implementation..

Claus