On Tue, 2007-09-04 at 23:03 +0200, Peter Verswyvelen wrote:
Jonathan Cast wrote:
I don't think this has been mentioned explicitly yet, but the discrepancy is purely for pedagogical purposes.
In Gofer, list comprehensions (and list syntax, IIRC) /was/ generalized (to an arbitrary instance of MonadPlus). But that means that any mistake in your syntax likely brings up a type error mentioning MonadPlus. This confuses CS freshmen (who are easily confused anyway); thus, Haskell restricts list syntax to lists so the type errors are simpler.
By contrast, most CS freshman have already used languages with multiple number types, so all you have to do is explain that type errors involving Num are Haskell's way of dealing with them. So the syntax can be generalized to the type class in that case without confusing freshmen as much.
jcc
Well, that is a very good reason, but for newbies (or should I say, for me ;-) ), most error messages are very confusing anyway! An since Haskell is really an advanced language, why not go all the way? It feels to me that for learning FP to newbies, a small subset of Haskell would be more suitable to get started, so really easy error messages can be given (Helium does that already?)
Exactly. But the Haskell 98 standard pre-dates Helium. I think Haskell' could be made more complicated now that Helium exists, but don't know whether that's in the cards.
Otherwise you would need a very clever compiler/editor machine learning system, that looks at how a class of users fixes a certain error, so the compiler can adapt its error message the next time a similar pattern occurs (which is science fiction right now I think...)
GHC I think tries to mediate this through the minds of its developers.
Now, these complex error messages are not just for Haskell; when I did complicated C++ template programming, the error message sometimes became as long as a full page when printed, and it took me more time to decipher the error than to fix the code ;-)
PS: Gofer, is that an existing language? As far as some googling can tell me, it's dead?
Right. Gofer died the quick death of most research languages, although it influence Hugs. jcc