Re: [reactive] Feedback on my first Reactive program: Monad Laws

31 Mar 2009

      I believe this is the right identity law (m >>= return === m), though it may
have the same root cause.  The annoying thing is that this and a probably
related bug affect all functions (I know of) with type:
Event (Maybe a) -> Event a
, namely justE, joinMaybes & the somewhat artificial "fmap fromJust .
filterE (not . isNothing)".  An example is:--

module Main
    where

import FRP.Reactive
import FRP.Reactive.LegacyAdapters
import Control.Concurrent
import Control.Applicative
import Control.Monad
import Data.Monoid
import Data.Maybe

aFilter :: String -> Maybe String
aFilter (_:_) = Just "A Non-Empty Line!"
aFilter _     = Nothing

justEventFilterFunction :: (Show a) => Event (Maybe a) -> Event a
--justEventFilterFunction = justE
justEventFilterFunction = fmap fromJust . filterE (not.isNothing)
--justEventFilterFunction = joinMaybes

main = do
        clock        <- makeClock
        (sink, evnt) <- makeEvent clock
        forkIO $ forever (getLine >>= sink)
        adaptE $ fmap putStrLn (pure "The First Line!" `mappend` (justE .
fmap aFilter $ evnt))

which again delays the `pure "The First Line!"` until the first line of
input is scanned, with all suitable uncommentings of the code.  The problem
is definitely in justEventFilterFunction; removing it (sensibly) removes the
delay.  The joinMaybes version clearly suffers from the Monad problem I
mentioned before; the fact that justE does the same thing leads me to
suspect the problem is adjustE not doing its job or being too strict, but I
know far too little about the internals to have a clue.

Freddie

2009/3/30 Conal Elliott <conal@conal.net>
...
We discovered during quickchecking that a monad associativity law can
fail.  And that it's a semantic bug, not an implementation bug.  I don't
know whether Freddie is running into this same issue or something else.  -
Conal
On Sun, Mar 29, 2009 at 8:27 PM, Jake McArthur <jake.mcarthur@gmail.com>wrote:
...
Freddie Manners wrote:
...
I've hit a rather fundamental problem: it seems Event doesn't obey the
Monad laws.
Yes, this is a known problem. Event is not a Monad. I don't think there
are plans to correct it any time soon. My personal preference is to pretend
that Event has no Monad instance.
- Jake
_______________________________________________
Reactive mailing list
Reactive@haskell.org
http://www.haskell.org/mailman/listinfo/reactive