> > data RegExp sigma = Empty | Epsilon | Sym sigma
> > | Seq (RegExp sigma,RegExp sigma)
>
> BTW, why not Seq (RegExp sigma) (RegExp sigma)?
Oops, I suppose the slogan is "Who uses Haskell has to Curry"... :-)
(I changed this).
> > data (Fin sigma) => NFA' sigma = forall q. (Fin q) => NFA' (NFA q sigma)
>
> > re2nfa:: (Fin sigma) => (RegExp sigma) -> (NFA' sigma)
>
> > re2nfa (Seq (e1,e2)) = let NFA' a1 = re2nfa e1
> > NFA' a2 = re2nfa e2
> > in NFA' (a_seq a1 a2)
>
> Try:
>
> re2nfa (Seq (e1,e2)) = case re2nfa e1 of
> NFA' a1 -> case re2nfa e2 of
> NFA' a2 -> NFA' (a_seq a1 a2)
>
> or even case (re2nfa e1, re2nfa e2) of ...
Thanks a lot , Marcin. This does the trick.
> > RegExp.hs:57:
> > My brain just exploded.
> > I can't handle pattern bindings for existentially-quantified constructors.
>
> > Did I encounter a limitation of the implementationbs of existential
> > types or is there just a stupid bug in my code?
>
> The former. Values with existential qualification must be "opened"
> by either explicit case or a pattern in a parameter. They cannot be
> opened by a let-binding.
Aha! Now I'd like to know whether there is a deeper reason behind
this? Or is it just due to the fact that existential types are not yet
properly integrated in the Haskell type system?
Cheers,
Thorsten
--
Dr. Thorsten Altenkirch phone : (+49 89) 2178-2209
Theoretical Computer Science fax : (+49 89) 2178-2238
LMU, D-80538 Munich, Germany Oettingenstr 67, room D105
http://www.tcs.informatik.uni-muenchen.de/~alti
