maybe, either: define "fantasy-land/concat" conditionally

[davidchambers]

Closes #346

Before:

S.concat(S.Just(1), S.Just(1));
// ! TypeError: Semigroup.methods.concat(...) is not a function

After:

S.concat(S.Just(1), S.Just(1));
// ! TypeError: Type-class constraint violation
//
//   concat :: Semigroup a => a -> a -> a
//             ^^^^^^^^^^^    ^
//                            1
//
//   1)  Just(1) :: Maybe Number, Maybe FiniteNumber, Maybe NonZeroFiniteNumber, Maybe Integer, Maybe ValidNumber
//
//   ‘concat’ requires ‘a’ to satisfy the Semigroup type-class constraint; the value at position 1 does not.

Losing the ability to filter(odd, Just(4)) is a real shame. This happens because Just(4) does not satisfy the requirements of Monoid as it does not satisfy the requirements of Semigroup.

It seems we need mfilter. Fantasy Land does not yet specify MonadPlus. We could define mfilter and hard-code support for the Maybe type, while also pushing to have MonadPlus added to Fantasy Land. With luck we would later be able to move the implementation from mfilter to Maybe#fantasy-land/mplus. Is this reasonable?

[safareli]

MonadPlus is just Monad + Alternative with some extra laws(on which there still is debate), and even if it would be added to FL one couldn't distinguish between some (Apply m, Chain m, Plus m) => m and (MonadPlus m) => m as it's not adding any new methods.

so you can write:

// :: (Alternative m, Monad m) => (a -> Boolean, m a) -> m a
// -- or:
// :: (Plus m, Chain m, Apply m) => (a -> Boolean, m a) -> m a
// --  but I think first one is much better
const mfilter = (p,ma) => {
  const T = ma.constructor
  return ma.chain(a => p(a) ? T.of(a) : T.zero()
}

[davidchambers]

Terrific suggestion, @safareli! The Sanctuary projects—sanctuary-type-classes in particular—would not be where they are today without your involvement. Thank you!

I just opened sanctuary-js/sanctuary-type-classes#37. Have a look. :)

[safareli]

@davidchambers Welcome! ;)

[davidchambers]

I rebased on #370 to confirm that values of type Maybe a will remain compatible with filterM. :)

[Avaq]

LGTM - Seems tests experience IE10 iFrame resource error again. Probably some race condition.