Alternative MaybeT instance might not be distributive

[LukaJCB]

Imagine this scenario:

f :: MaybeT Effect (a -> b)
f = empty 
g :: MaybeT Effect (a -> b)
x :: MaybeT Effect a

-- Will run the effects of x only once
(f <|> g) <*> x

-- Will run the effects of x twice
 (f <*> x) <|> (g <*> x)

[hdgarrood]

Thanks for the report. This does indeed seem problematic. Perhaps we should just remove the Alternative instance in the next breaking release.

[triallax]

After this library is updated to 0.14, can I make a PR for this? Or do we not breaking release for now?

[JordanMartinez]

We're doing breaking changes now, not later.

[hdgarrood]

Actually, I think we should leave this as-is. Many Alternative instances in the wild obey a "left catch" law:

pure a <|> b = pure a

instead of the documented distributivity law, and MaybeT is one example. Other examples include parsers and Aff. See also purescript/purescript-control#63 - I think we should wait until we've made progress there before removing any instances here.

[hdgarrood]

I think this is just one example of the issue described in purescript/purescript-control#63, so I think we should close this for now and continue tracking the issue there.