[Haskell-cafe] Announcing OneTuple-0.1.0

[Luke Palmer]

On Fri, Oct 3, 2008 at 7:26 PM, Tim Chevalier <catamorphism at gmail.com> wrote:
> On Fri, Oct 3, 2008 at 2:29 PM, Jason Dusek <jason.dusek at gmail.com> wrote:
>> Lennart Augustsson <lennart at augustsson.net> wrote:
>>> But (a) is not a lifted version of a, whereas (a,b) is a lifted
>>> version of the a b product.
>>> So it's not consistent, and thereby wrong.
>>
>>  Well, we can't represent the unlifted product in Haskell,
>>  right? You have to use some constructor. So if we just say we
>>  are using tuples to represent unlifted products, what's so bad
>>  about that?
>>
>
> Unless I'm confused, unboxed tuples represent unlifted products. In a
> sense this is "[using] some constructor", but in a sense not, since an
> unboxed tuple constructor has no runtime representation.

Well, unboxed tuples are not really lifted nor unlifed, since you
can't even pass one to a function.

I like to pretend tuples are unlifted.  Here's how I do it:

* Never use seq on tuples (or functions).  I could make this precise
by putting seq in a typeclass (like it used to be - like it should
be), and not having instances for tuples.
* Never do a strict pattern match on a tuple.  I.e. instead of writing
f (x,y) = ..., I will write f ~(x,y) =... everywhere.

Then (_|_,_|_) might as well be _|_, we have no way to tell them apart.

I like to pretend functions are unlifed the same way; i.e. const _|_ = _|_.

There are apparently occasions where lazily matching on a tuple will
introduce a space leak.  I am not 1337 enough to recognize them yet.

Luke