Non-H98 crusade, contd.

[ross at soi.city.ac.uk]

On Sun, Feb 27, 2005 at 11:47:21AM +0100, Sven Panne wrote:
> OK, let's try to summarize what seems to be the current consensus about
> more or less conservative & settled extensions to Haskell98:
> 
>   a) rank 2 type signatures
>   b) polymorphic components for data constructors
>   c) pattern guards
>   d) data declarations with no constructors
>   e) constraints on typeclass methods
>   f) instances on type synonyms
>   g) relaxation of polymorphic recursion

I take back what I said about instances on type synonyms: they are tied
up with generalized instance heads, which are certainly a major change
from H98, as they break the termination guarantee.

> Probably a) and b) should be considered together(?). Furthermore, currently
> only GHC implements c), but a simple desugaring transformation should be
> enough to implement it in the other Haskell systems and I don't see any
> hard semantic issues here.

All these but pattern guards are implemented in Hugs.  Is such a
desugaring transformation written down anywhere?  That's essential for
describing the extension, but wouldn't be usable as an implementation:
users will be confused by type errors that relate to desugared forms.
(I'm also not sure about the use of monadic binding notation for the
non-monadic matches).

e) is in GHC and Hugs without special flags.
g) is in GHC, Hugs and nhc98, all without any special flags.

I think we should be biased in favour of extensions that give extra
expressiveness, like a), b) and e), over those that simplify things you
could have written anyway.