Koen Claessen wrote:
Imagine a commutative monad, CIO. Commutative monads have the property that it does not matter in what order actions are performed, they will have the same effect. In other words, for all m1 :: CIO A, m2 :: CIO B, k :: A -> B -> CIO C, it should hold that:
do a <- m1 do b <- m2 b <- m2 === a <- m1 k a b k a b
... provided 'a' and 'b' are distinct variables, right? The following is legal in Haskell
do x <- ... let foo = ... x ... x <- ... let foo = ... x ...
That shows that variables add a bit more complexity to the question. Not only the dependency in actions should be considered, but also data dependency. In the above code, if action 'm2' happens to refer to the variable 'a', does commutativity still hold? Here are a few more similar example, assuming the proposed 'global level <-' syntax and the commutative nature of the action of creating an IORef:
z <- newIORef x x = [y1,y2] y1 <- newIORef t y2 <- newIORef t t <- newIORef True
Should this be accepted? Should the actions be executed as written? What if it were written
y1 <- (return $! t) >>= newIORef
A similar example:
class C a where op :: a -> a y <- newIORef (op (undefined::IORef Bool)) x <- newIORef True instance C (IORef Bool) where op _ = x
which shows that the data dependency analysis is a bit trickier than expected, even if the definitions appear to be `well-ordered'. Template Haskell may bring some new complications. Suppose module A imports module B and uses some of the functions of B at TH time. Module A is also free to use bindings of B at run time. Now, if module B has initializing actions, would they be executed once or twice? Would created CIORefs be shared across the phases? Would that behavior be consistent with respect to module A being compiled or interpreted? These questions do arise in Scheme -- where the agreement seems to be to have two kinds of imports: regular import and import for-syntax. [regarding partial signatures]
Ah! I had forgotten about that. See also:
http://www.mail-archive.com/haskell@haskell.org/msg05186.html
Incidently, some of that is already available in Haskell, http://pobox.com/~oleg/ftp/Haskell/types.html#partial-sigs