Russell Connnor said:

I have a Haskell program that defines the value __ as

__ = Prelude.error "Logical or arity value used"

Yhc seems unhappy with this name. Any reason for this? Can Yhc be made to accept this?

Malcolm said:

> Anyone want the -underscore flag to stick around, and not just be the > default which can't be turned off?

It depends whether you want to claim Haskell'98 compatibility or not. Haskell-prime is likely to make yhc's default behaviour with underscores the new standard however.

> For that matter: > Compliance Options: > -underscore Enable H'98 underscore-is-lower-case (default=off) > -puns Enable pre-98 named-field puns (default=on) > -98 Haskell 98 compliance (default=off) > > Why is -98 off, why is -puns on (what are puns!!!) and lets kill this > -underscore option. The n+k option died, others deserve to die too :)

The option -98 is just a collective switch to say -underscore -nopuns in a shorter way. It is off by default because of user demand. Named field puns were a feature of Haskell-1.3 that went away in Haskell'98, but some users liked them and campaigned for their reinstatement.

Issac said:

> Can anyone outline exactly which behaviour is which? What the > differences are? Is one behaviour a superset of the other? Are there > any incompatabilities?

H'98: underscore is purely considered a lowercase character (except for `_' which is basically just a keyword). This allows: _Bob :: Int; _Bob = 3

default for nhc and yhc, and a likely-to-get-in proposal for Haskell-prime, is skipping leading underscores in an identifier for the purpose of determining case. This allows: type _Bob = Int

Each example would be a syntax error in the other underscore-treatment. It makes no difference except for identifiers that start with underscores not followed by a lowercase letter. The proposed/nhc/yhc default behavior is useful because leading underscore also signifies "unused", which should be possible to do even for type/constructor/class names.

and the Haskell-98 treatment of leading underscores is good just because it's simple and it makes it obvious what case a symbol made up of all underscores is. That's all, I think.

Exactly what the "fix" to this bug is isn't clear, but just so we don't forget it.