I quite like the argument that take is a total function and as such all its return values are from teh specificed range. I can also see the logic in take n [] = [] where n > 0 taking n from nothing, or the empty set, returns nothing! The same should apply to head and tail. head or tail of [] should be []. What does the list think? Cheers, Paul At 16:43 12/09/2007, you wrote:
Hi
On 12 Sep 2007, at 11:44, ChrisK wrote:
Conor McBride wrote:
I'd like operations to complain about bogus input, rather than producing bogus output.
Then you want a runtime assertion checking error helpful Data.List replacement.
Could you use Control.Exception.assert and make a wrapper for Data.List?
http://www.haskell.org/ghc/docs/latest/html/users_guide/sec- assertions.html
Hmmm. It might be quite annoying to make it a wrapper if it's just a question of appealing to error rather than returning dummy values in failure cases. Defining the domain of a function can be quite like defining the function itself. Also, sometimes there can be problems arriving at a Bool. For example, zipping colists is a productive coprogram, and it can raise errors in off-diagonal cases, but you can't compute in advance whether you're on the diagonal.
A more serious point is that in some cases we might want take to underapproximate, or zip to truncate (or tail [] = [] ?). I don't think there's always a clear "library" choice here. I tend to be pragmatic about it. I was just pointing out that it does sometimes make sense to use less defined functions when one has a more specific domain in mind. I'm usually happy to write the fussy variants myself, if I'm that agitated.
Funny old world
Conor
_______________________________________________ Haskell-Cafe mailing list Haskell-Cafe@haskell.org http://www.haskell.org/mailman/listinfo/haskell-cafe