Yves, that is exactly how I designed my program so far.
Human player needs a monad IO, AI needs just a monad, whatever it is, and I
make both run in IO.
And, as you said, the type of the ai (bot :: Monad m => Player m) contains
no IO, so I know that, even if I make it run in IO, it won't make any
side-effect.
My problem was, for example, if I want a player to run in its OWN monad.
Human uses IO, which is unique and shared by all the human players in the
program.
But what if I want an AI that remember every former opponent's move, so that
it could adapt its reflexion all along the game?
Then this AI would have to run in its own State monad, for instance.
2010/4/13 Tillmann Rendel
Yves Parès wrote:
data Player m = Player { plName :: String, -- unique for each player plTurn :: GameGrid -> m Move -- called whenever the player must play }
What I try to avoid is having every player running in IO monad.
One could define the following players.
human :: MonadIO m => Player m human = ...
bot :: Monad m => Player m bot = ...
Note that these players are polymorphic in m, only assuming some minimal interface.
Now you can run both players in a single monad which is good enough for both, because it supports the union of the interfaces assumed by the various players. In this case, this monad could be IO, since it supports both MonadIO and Monad.
changeBoard :: Board -> Move -> board changeBoard = ...
play :: Player IO -> Player IO -> GameGrid -> IO GameResult play p1 p2 board do move <- plTurn p1 board play p2 p1 (changeBoard board move)
While this is probably not as expressive as what you want, it is reasonably simple, and it has the property that bot is not in the IO monad.
I have first seen this pattern in monadic interpreters, where you could have types like the following.
eval :: (MonadReader Env m) => Expression -> m Value exec :: (MonadReader Env m, MonadIO m) => Statement -> m ()
These types reflect that the interpreted language permits side-effects only in statements, but not in expressions. This is similar to your situation: You want your types to reflect that your game permits side-effects only in human players, not in artifical intelligences.
Tillmann