On Tue, Aug 12, 2014 at 08:02:13PM -0400, David Feuer wrote:
On Aug 12, 2014 7:50 PM, "Bertram Felgenhauer" < bertram.felgenhauer@googlemail.com> wrote:
I tend to agree that those MonadFix instances are not immensely useful.
From my perspective as a generally ignorant person, I think these should be kept as long as
1. They satisfy the required laws and 2. No more useful instances can be written.
That said, it might make sense to remove them if
1*. You can formulate a new law that distinguishes the instances you consider sensible from the ones you consider insensible.
There is a law that discriminates against the instances on strict monads, known as right shrinking or right tightening: mfix (\ (x,_) -> f x >>= rightM g) = mfix f >>= rightM g where rightM :: Monad m => (a -> m b) -> a -> m (a, b) rightM f x = f x >>= \ y -> return (x, y) A special case is mfix (const e) = e In do-notation, these become do {rec {ss; x <- e}; ss'} = do {rec {ss}; x <- e; ss'} (x not free in ss or e) do {rec {x <- e}; ss} = do {x <- e; ss} (x not free in e) This law is pretty handy, but it was deliberately omitted from the axioms for the MonadFix class, to permit the axioms for Maybe, Either, [] and IO. The transformers package is just following those.