[Haskell-cafe] Re: (flawed?) benchmark : sort
gtener at gmail.com
Mon Mar 10 16:33:31 EDT 2008
No, '=' should not mean an identity but any equivalence relation. Therefore,
we can use whatever equivalence relation suits us. The reasoning you
provided is IMHO rather blur. Anyway, having possibility of using different
equivalence relations is great because they mean different abstraction
classes - and not all of them are isomorphic.
On Mon, Mar 10, 2008 at 9:09 PM, Daniel Fischer <daniel.is.fischer at web.de>
> But antisymmetry means that (x <= y) && (y <= x) ==> x = y, where '='
> identity. Now what does (should) 'identity' mean?
> Depends on the type, I dare say. For e.g. Int, it should mean 'identical
> pattern', shouldn't it? For IntSet it should mean 'x and y contain exactly
> the same elements', the internal tree-structure being irrelevant. But that
> means IntSet shouldn't export functions that allow to distinguish (other
> by performance) between x and y.
> In short, I would consider code where for some x, y and a function f we
> (x <= y) && (y <= x) [or, equivalently, compare x y == EQ] but f x /= f y
> broken indeed.
> So for
> data Foo = Foo Int (Int -> Int),
> an Ord instance via compare (Foo a _) (Foo b _) = compare a b
> is okay if Foo is an abstract datatype and outside the defining module
> guaranteed that
> compare (Foo a f) (Foo b g) == EQ implies (forall n. f n == g n), but not
> the data-constructor Foo is exported.
-------------- next part --------------
An HTML attachment was scrubbed...
More information about the Haskell-Cafe