问题
It is a well known fact that (>>=) can be implemented using fmap and join while join can be implemented using >>=. Is there any reason we don't define the Monad class with join included and using the following default definitions?
join x = x >>= id
x >>= f = join $ f <$> x
This would allow a minimal definition to include either just (>>=) or join, instead of forcing (>>=). Might be a bit helpful considering category theory tends to favour join.
The usual argument against modifying classes is that we break backwards compatibility. However, in this case, that wouldn't happen - we only add the possibility of defining Monad using join.
回答1:
That was meant to happen with the Applicative-Monad proposal (which has made it to GHC 7.10). However, there is a technical issue involving type roles in GHC which has postponed indefinitely the implementation of what you suggest.
来源：https://stackoverflow.com/questions/31552064/why-isnt-join-part-of-the-monad-class