On Wednesday 22 February 2006 15:53, Malcolm Wallace wrote:

"Simon Peyton-Jones" wrote:

...

| I don't seriously propose for haskell-prime that signatures should | be required on exports. Just permitting them would be a large and | useful step up already.

If this is to be a serious proposal, someone had better think what to do about classes, data types, instances and so on.

As far as I can see, there is very little to change. Here is a concrete proposal. A qvar in an export list may optionally have a type signature. A qtycon or qtycls in an export list may optionally have an annotation saying whether it is a type, newtype, data, or class. Instances cannot be mentioned in export lists, and this does not change.

export -> qvar

| qtycon [ (..) | ( cname_1, ... , cname_n ) ] (n>=0) | qtycls [ (..) | ( var_1, ... , var_n ) ] (n>=0) | module modid

becomes

export -> qvar [ :: type ]

| [type] qtycon | [newtype] qtycon [ (..) | ( cname_1, ... , cname_n ) (n>=0) | [data] qtycon [ (..) | ( cname_1, ... , cname_n ) (n>=0) | [class] qtycls [ (..) | ( var_1, ... , var_n ) (n>=0) | module modid

As far as import entity lists are concerned, we permit an optional annotation for type synonyms, newtypes, datatypes, and classes:

import -> var

| tycon [ (..) | ( cname_1, ... , cname_n ) ] (n>=0) | tycls [ (..) | ( var_1, ... , var_n ) ] (n>=0)

becomes

import -> var

| [type] tycon | [newtype] tycon [ (..) | ( cname_1, ... , cname_n ) ] (n>=0) | [data] tycon [ (..) | ( cname_1, ... , cname_n ) ] (n>=0) | [class] tycls [ (..) | ( var_1, ... , var_n ) ] (n>=0)

Anyone see any difficulties? No, but one question: If the type signature is given in the export lists, is it then necessary (or even allowed) later on? I would vote for _not_ having it twice in the file.

Cheers, Georg

On Wed, Feb 22, 2006 at 04:38:57PM +0000, Jon Fairbairn wrote:

I can't think of why off-hand, but it seems conceivable that one might want to use a function polymorphically within a module but export a monomorphic version.

This would be tricky, functions have different representations depending on their type, they would effectivly be different entities with the same name which would cause a whole host of complications. (for instance does Self.foo refer to the polymorphic or monomorphic version). I would think that if multiple type signatures were allowed, they would need to be identical. (modulo beta blah) John -- John Meacham - ⑆repetae.net⑆john⑈