Re: [Haskell-cafe] Re: (flawed?) benchmark : sort

Aaron Denney wrote:

On 2008-03-11, Adrian Hey wrote:

...

Having tried this approach myself too (with the clone) I can confirm that *this way lies madness*, so in future I will not be making any effort to define or respect "sane", unambiguous and stable behaviour for "insane" Eq/Ord instances for any lib I produce or hack on. Instead I will be aiming for correctness and optimal efficiency on the assumption that Eq and Ord instances are sensible.

Good. But sensible only means that the Eq and Ord instances agree, not that x == y => f x == f y.

So can I assume that max x y = max y x? If not, how can I tell if I've made the correct choice of argument order. If I can't tell then I guess I have no alternative but document my arbitrary choice in the Haddock, and probably for the (sake of completeness) provide 2 or more alternative definitions of the "same" function which use a different argument order. Regards -- Adrian Hey