Duncan Coutts wrote:
I think it's a perfectly reasonable mental model for people to believe that: data Char = 'a' | 'b' | 'c' | ... data Int = ... -2 | -1 | 0 | 1 | 2 | ...
I don't see why we should remove one and not the other. Students will ask why the can pattern match on strings, characters and booleans but not numbers.
Numbers are special because they are overloaded. A numeric literal is an element of many types. That clearly distinguishes them from other literals.
Perhaps primitive recursion on integers is misleading, but people will still write
foo n | n == 0 = ... | otherwise = ...
where they previously wrote
foo 0 = ... foo n = ...
so what have we gained except less notational convenience?
Discourage anyone from teaching primitive recursion on integers. Recursion on integers then has to be taught as a separate topic, giving opportunity to point out the pitfalls. Sure, it doesn't prevent anyone from writing anything.
Not all pattern matching on numeric literals is involved with recursion on integers, where as virtually all n+k patterns is used for that purpose.
I think there are very few situations where you would use k patterns without recursion.
So there is some distinction between the two forms. n+k patterns are a quirk of the numeric types. k patterns are regular with other types in the language.
As I said above, they are not regular because of overloading.
It's partly the complexity of the language and partly because our latest language spec (H98) is not the language that we all use (H98 + various extensions). I'm sure Haskell-prime will help in this area.
I hope as well that Haskell' will be the language that most people will use and some extensions are certainly required for practical use. I just want to get rid of superfluous features. Ciao, Olaf