[Haskell-cafe] Re: type families and type signatures

[Mark P Jones]

The surprising thing about this example is the fact that
the definition of foo is accepted, and not the fact that
the definition of foo' is rejected.  At least in Manuel's
"equivalent" program using functional dependencies, both
functions have ambiguous types, and hence both would be
rejected.  It sounds like your implementation may be using
a relaxed approach to ambiguity checking that delays
ambiguity checking from the point of definition to the
point of use.  Although this is probably still sound, it
can lead to puzzling error diagnostics ...

All the best,
Mark

Tom Schrijvers wrote:
>>>> type instance Id Int = Int
>>>>
>>>> foo :: Id a -> Id a
>>>> foo = id
>>>>
>>>> foo' :: Id a -> Id a
>>>> foo' = foo Is this expected?
>>>
>>> Yes, unfortunately, this is expected, although it is very 
>>> unintuitive. This is for the following reason.
>>
>> Huh? This sounds very wrong to me, simply because  foo  and  foo'  
>> have the very same type.
> 
> Type systems reject programs that don't go wrong. It's hard to 
> understand on the basis of such a single program why it should be 
> rejected. The problem is decidability. There is no algorithm that 
> accepts all well-behaved programs and rejects all ill-behaved programs. 
> There probably is an algorithm that accepts a particular program.
> 
> So far we haven't found an algorithm that accepts this example, that is 
> decidable and sufficiently general to cover many other useful cases. 
> This is our motivation for rejecting this program.
> 
> Consider it a challenge to find a better algorithm in the design space.
> 
> Cheers,
> 
> Tom
> 
> -- 
> Tom Schrijvers
> 
> Department of Computer Science
> K.U. Leuven
> Celestijnenlaan 200A
> B-3001 Heverlee
> Belgium
> 
> tel: +32 16 327544
> e-mail: tom.schrijvers at cs.kuleuven.be
> url: http://www.cs.kuleuven.be/~toms/
> _______________________________________________
> Haskell-Cafe mailing list
> Haskell-Cafe at haskell.org
> http://www.haskell.org/mailman/listinfo/haskell-cafe