DData revision

[Ketil Malde]

JP Bernardy <jyp_7 at yahoo.com> writes:

>> Would this make sense?  Would it be useful?

> I'm not sure that I understand either :)

Yeah, my definitions weren't exactly clear. I'll try to elucidate:

> Let me give an example...

>> type MyString = MyString String 

>> instance Eq MyString where
>>   (MyString x) == (MyString y) = length x == length y 

I'm sure you'll agree that this is not something anybody sane would
do.  (Actually, I did something remotely similar, and I've regretted
it ever since; it gave me some nasty, hard-to-track down bugs.  I
always derive Eq since then.)

Which is why I think Bag perhaps should, instead of relying on Eq (or
actually Ord, I suppose), be provided with a separate parameter
specifying how to compare items for equality for bagging purposes.

>> Bag.toString $ Bag.fromString $ [ms "x", ms "y"]

So instead of doing this, I would like to be able to do

        mkBag myComp [ms "x", ms "y"]
           where myComp x y = compare (length x) (length y)

which would then bag all equal-length strings (possibly storing them
in multisets using regular Eq).

> That is why I choose to keep the "MultiSet" name. 

Just to make that clear, I've no issue with the naming.

> If Eq is defined as structural equality, MultiSet and
> Bag are equivalent.

And since Eq "should" define structural equality, I propose a
different interface for Bags (if we need them at all, that is)

Was that a bit clearer?

-kzm
-- 
If I haven't seen further, it is by standing in the footprints of giants