RE: Parsing kinds in external core

Sorry for a slow reply.

You definitely cannot infer kinds for an arbitrary FC program.  The right thing to do is to *stop* using Type inside a KindedTyVar (always a hack) and instead use HsType.  That'll require some work to make the existing setup work again, but I think it's the Right Thing.

Simon

| -----Original Message-----
| From: [email protected] [mailto:[email protected]]
| On Behalf Of Aaron Tomb
| Sent: 14 December 2006 23:52
| To: [email protected]
| Subject: Parsing kinds in external core
|
| Hi,
|
| I encountered what might be a problem with parsing External Core
| using HsType to represent types, though it might be possible to work
| around it. It comes up when reading in a type variable with a kind
| annotation which may be a type equality kind. These are represented
| in HsType as the KindedTyVar alternative of HsTyVarBndr, which takes
| a Kind as its second argument. However, Kind is just a name for Type
| (not HsType), and thus it contains Names. Before type equality kinds
| existed, this wasn't a problem, because no kind could contain type
| variables. But now they can.
|
| So, to properly construct an appropriate Kind seems like it would
| require running the renamer early on (probably in the parser), which
| sounds like a headache, if its even feasible at all.
|
| Alternatively, we could ignore kind annotations in the parser (at
| least for equality kinds) and create UserTyVars instead of
| KindedTyVars. I believe it would then be possible to use existing
| machinery to infer them later on (and, if I understand correctly, the
| old implementation did this). Is it true that we can always infer
| kinds?
|
| One problem with that approach, even if inference is no problem, is
| that errors in kind annotations would be ignored. Maybe that doesn't
| really matter.
|
| In any case, it seems to make sense to leave kind annotations in the
| concrete syntax for External Core, because even if GHC can infer
| them, they might make other implementations that use External Core
| somewhat easier.
|
| Aaron
|
| _______________________________________________
| Cvs-ghc mailing list
| [email protected]
| http://www.haskell.org/mailman/listinfo/cvs-ghc