Test ap with fantasy-land and use consistent ordering of combination

[LiamGoodacre]

This PR is assuming that fantasy-land's ap is supposed to be like Haskell's (<**>) :: Applicative f => f a -> f (a -> b) -> f b, and not like flip (<*>) :: Applicative f => f a -> f (a -> b) -> f b (which is how Ramda currently treats it).

My test case attempts to show the difference by using Either. The important case is the first one:

eq(R.ap(leftFn, leftVal).simple(), leftFn.simple());

// i.e: forall e x. R.ap(Left(e), x) = Left(e)

[kedashoe]

Thanks for the PR @LiamGoodacre . To comply with FL, we just pass along the first argument to the second if the latter has fantasy-land/ap. So for example, if sanctuary's Either were passed in to us, we would get here:

https://github.com/sanctuary-js/sanctuary/blob/master/index.js#L2519

. In the instance you highlighted, other is not a Right and we would get back other, ie Left(e) in your example. Some quick tests seem to confirm this.

Ramda doesn't really have much to say about it, it is up to the instance to implement ap correctly.

[LiamGoodacre]

So you're saying that for FL, ma['fantasy-land/ap'](mf) is supposed to be implemented like mf <*> ma and not ma <**> mf?

I haven't been able to see on the FL spec that it is supposed to be one or the other. Given the type signature I was expecting <**>. Conversations in the Gitter channel also suggest it was supposed to be <**>.

[kedashoe]

You probably know more about this than me :) There have been many discussions about ap in FL, see fantasyland/fantasy-land#145 as a starting point. FL also has a gitter that is quite active, so you can ask there as well.

[LiamGoodacre]

Closing because: lack of interest; I no-longer have the need to use Ramda; I'm not particularly invested in the outcome.