Good point. This brings us back to Russell's original suggestion of swap and swap' to cover both cases. That module has all of ~4 methods in it right now. -- I'm evaluating that lazily -- so its not exactly like adding both will break the complexity budget of a self-contained module designed to deal with tuples. ;)

-Edward Kmett

On Thu, Jun 18, 2009 at 1:27 AM, Jonathan Cast <jonathanccast@fastmail.fm> wrote:
On Thu, 2009-06-18 at 01:24 -0400, Edward Kmett wrote:
> If we look at swap from the standpoint of the laws/RULES it should
> support, viewing Hask over (,) as a symmetric monoidal category you
> get something like:
>
>
> {-# RULES
> "swap . swap = id" forall x. swap (swap x) = x
> "fst . swap = snd" forall x. fst (swap x) = snd x
> "snd . swap = fst" forall x. snd (swap x) = fst x
>  #-}
>
>
> That seems to argue for the lazy definition being the default to avoid
> the strict pattern match in swap breaking the latter very pleasing
> equalities.

Lazy swap fails the swap . swap rule:

 swap (swap undefined)
= swap (snd undefined, fst undefined)
= swap (undefined, undefined)
= (undefined, undefined)

which is distinct from undefined because Haskell tuples are lifted.

jcc