On Wed, Feb 01, 2006 at 02:51:08PM +0000, Philippa Cowderoy wrote:
I'm not convinced on that. You'd have to specify a surprisingly low-level language to allow that to the extent the real optimisation nuts want, and that's something that really should be beyond the scope of the standard. Even if we stick with something simple it's extremely likely that we'd end up specifying a dictionary-passing implementation of typeclasses - something that seriously disadvantages some valuable extensions and implementation techniques (it'd really mess up JHC from what I can tell, for example).
I am thinking we don't specify any particular translation scheme. just a sudset of the language that is considered 'core' that every haskell program could _potentially_ be reduced to. whether compilers actually take the 'example' route given in the report is a different manner. for example jhc might leave in typeclasses because they can't be desugared into pure haskell without GADTs. I wouldn't want to see a dictionary passing implementations of type-classes prescribed either :) John -- John Meacham - ⑆repetae.net⑆john⑈