Add generalized flip

[joneshf]

Closes #82

The name of this function along with its operator alias can change.
I don't know a better name for either and have no strong feelings.

[texastoland]

Wow cool!

instance ff ::             Functor (a -> _)
flip        :: ∀ a b c   .                     (a -> (b -> c)) -> b -> (a -> c)
flip'       :: ∀   b c f . Functor  f       =>  f    (b -> c)  -> b ->  f    c
flip'       :: ∀ a b c   . Functor (a -> _) =>  a -> (b -> c)  -> b -> (a -> c)

Do we need to keep the specialized version? Does it need an operator?

[paf31]

Yes, we should keep the specialized (optimized) version, I think.

[texastoland]

That's going down the Haskell map versus fmap road. In that case call it fflip. Rather than naming conventions for specialized/default variants of functions there should a way to provide implementations in type classes (e.g. DefaultSignatures) or without (as in this case). But adding TIMTOWTDI seems undesirable.

[joneshf]

Don't let the name flip' sway your judgement. const and pure live in harmony. compose and map live in harmony.

Call it some other name if that makes it easier to deal with. switch? surface? embed? It takes a function in some computational context and returns a function that stuffs the result in the computational context.

Again, I don't care the name, so use what you like best.

TIMTOWTDI

wat?

[michaelficarra]

"There is more than one way to do it".

[texastoland]

const and pure live in harmony. compose and map live in harmony.

Excellent points. I still think it'd be better to address the user facing issue of specializing implementations than keep renaming things.

There is more than one way to do it.

(Perl) cf. There should be one—and preferably only one—obvious way to do it. (Python)

[paf31]

"There is more than one way to do it".

This doesn't bother me so much.

[joneshf]

What's the verdict here?

[texastoland]

Sounds like it just needs a name. I'd like any function to be able to be specialized but it's mostly unrelated.

[garyb]

fapply perhaps? Not a great name, but it's definitely closest in signature to apply, but only requires the Functor constraint. It seems more apply-y than flip-y to me anyway, even though when dealing with (->) a it is indeed flip.

[garyb]

flap? 😄

[garyb]

^ in all seriousness I do actually like flap, it doesn't really mean anything (useful for an abstract function), and if you try really hard I guess it kind works:

1. it's part flip, part apply
2. it takes a function in a context, opens a flap and throws a in

[texastoland]

It comes up in Hayoo but no solid convention.

[texastoland]

unapply? It also looks vaguely (less than apply) like unfoldr and sieve in Profunctor.

[paf31]

To me, one of the most useful places for this function is alongside <$> and <*>, since you can use it to apply a function when one of the arguments is not wrapped in an f, without using pure.

So I suggest it should be called applied, with an infix alias of some kind.

[epost]

flap = abstract flip. I love it!

[joneshf]

an infix alias of some kind.

One that comes to mind is (<@>). to pun off of @garyb, it's got an a inside a "flap":

it takes a function in a context, opens a flap and throws a in

😉

[joneshf]

So, we're at flap and <@>. Bueno, or different name?

[paf31]

Sounds good, thanks! Do the precedence levels work out so that this can be used with <*>?

[joneshf]

Do you have an example of how you'd like it to parse?

[texastoland]

Perhaps it should be a test case?

[paf31]

Something like threeve <$> Just 1 <@> 'a' <*> Just true I think.

[paf31]

If the operator can be made to work like that, I think it would be nice to put something about it in the comment as well.

[joneshf]

Updated. Didn't add a test since there's no precedent for what a ps test looks like here.

[paf31]

Thanks, and sorry this took such a long time to merge!

[joneshf]

No worries. Thanks for merging!