Re: [trac@galois.com: Re: [GHC] #1218: Add sortNub and sortNubBy to Data.List]

On Mar 20, 2007, at 6:53 PM, Duncan Coutts wrote:

On Tue, 2007-03-20 at 14:11 +0100, Nils Anders Danielsson wrote:

...

On Tue, 20 Mar 2007, Duncan Coutts wrote:

...

H98 requires that Eq and Ord be equality classes and total orders

Does it? I was under the impression that a compiler can never assume that any laws hold of any user-defined instances (to do optimisations, for instance).

The report states the following:

"The Eq class provides equality (==) and inequality (/=) methods." "The Ord class is used for totally ordered datatypes."

I interpret this as being mere usage guidelines, and nothing more.

Aye.

Hmm, I've always taken these to be binding contracts. For example, do I get to assume either == or <= are transitive? But I admit this is at the level of writing libraries---so that, as another poster noted, I'm allowed to use a sort function which yields a different result (possibly _|_) if I don't obey the contract. But I'd often like my compiler to assume that <= is transitive... -Jan-Willem Maessen