On 10/25/07, Brandon S. Allbery KF8NH
My point is that there's no promise for that one *even in C*. (The equivalent construct being adding the new row before nextKey has failed.)
Sure, but in C, it's highly likely that the full evaluation of the key list happens in one place, that's just how code tends to get written. That said, in C code, I've often seen bugs where code called during the iteration of a collection modifies that collection; that's something that has been really refreshing to get away from when writing Haskell. With the unsafeInterleaveIO example, the "pure" keylist could get deferred indefinitely, stored in a data structure, partially evaluated at many different times against many different versions of the database, etc., and it's not necessarily clear to the person who just has a [Key] that they are doing deferred calls to nextKey like it tends to be in C. It's safe if you use it in a predictable fashion, and in a real API I'd probably provide "getKeys" and "unsafeLazyGetKeys" and let the programmer decide. -- ryan