Re: Constraints on definition of `length` should be strengthened

There's a way to get unbiased products in Haskell. The trick is to make their kind [Type] -> Type: data family Product (xs :: [Type]) data instance Product '[] = Unit data instance Product '[a] = Id a data instance Product '[a,b] = Pair a b data instance Product '[a,b,c] = Triple a b c ... We could make (a,b) syntactic sugar for Product '[a,b] at type level and for Pair at term level - this would invalidate all asymmetric instances. Unfortunately, this is would break an immense amount of code and violate the Haskell standard. (Another variation is to define Product inductively as a GADT with constructors Cons and Nil, but the result would be a HList that has a different memory representation than tuples). On Tue, Apr 4, 2017 at 12:14 AM, Nathan Bouscal wrote:

On Mon, Apr 3, 2017 at 1:48 PM, Sven Panne wrote:

...

2017-04-03 22:29 GMT+02:00 Nathan Bouscal :

...

I expect most people probably agree that it'd be nice to have tuples be an unbiased cartesian product, but the actual fact of the matter is that tuples as they exist in Haskell are biased.

Tuples *are* unbiased, the bias is just an artifact of seeing them as a curried function, where the positions are "eaten" from left to right. Again, this mathematically correct, but more often than not the main intent of using a tuple-

… no, they're not. What other type of correct is there than mathematically correct? "Zero isn't an integer, that's just an artifact of *seeing* it as an integer."

...

...

We can't just ignore that and pretend they're unbiased.

We *can* ignore that, just use Henning's Decorated for an isomorphic variant.

...

It definitely sucks that the answer people would naively give to "what is a tuple in Haskell" is not the correct answer, but we're stuck in that situation.

See above, we are not stuck: We *can* get back normal people's intuition and Haskell's semantics back in line by removing the tuple instances and adding something like Decorated. It is just a matter of priorities: This will temporarily damage the Haskell ecosystem a bit, but in the long run it will be the nicer, more explicit, more intuitive way.

You can't get tuples to behave like they're unbiased. You can try to hide the fact that they're biased by getting rid of the only possible instances they can support, but that doesn't magically make them unbiased. It sounds like you just want to rename tuples to Decorated. Maybe that's a good idea, but call it what it is.

...

...

The question is how to make the best of it.

If the tuple instances are removed and Decorated is added, things are easy to fix: The compiler will tell you exactly the places where you were too lazy to define and use a custom data type, and the fix is mechanical. The current situation is quite the opposite: People silently get totally unexpected behavior.

_______________________________________________ Libraries mailing list Libraries@haskell.org http://mail.haskell.org/cgi-bin/mailman/listinfo/libraries

_______________________________________________ Libraries mailing list Libraries@haskell.org http://mail.haskell.org/cgi-bin/mailman/listinfo/libraries