Adrian Hey wrote:
> I've read this paper, and it is good, but I still see no reason to
> change my opinion regarding referential transparency and IO.
> But I am terribly confused about this issue, there seems to be no real
> consensus about this in the FP world. I just can't see how if the
> result of an action is dependent on unknown external world agents
> (as it is with the IO monad) it could be described as referentially
> transparent. I can accept that we have referential transparency for
> user defined monads, but not the IO monad.
Consider this:
> type IO a = StateOfUniverse -> (a, StateOfUniverse)
> -- Not syntactically correct, but you know what I mean.
So anything that is declared, say `IO Int', means that it is actually a
function that reads in the state of the universe, potentially modifies
it, and then returns an Int value along with the new state of the
universe. The interesting thing to note is that the state of the
universe never changes between calls that are strung together using the
`>>=' operator. That is, the StateOfUniverse that is returned by the
first monad is exactly same state that is fed into the second. Whether
or not you want to call this "referentially transparent", well I guess
that's up to your own philosophic bias.
- Michael Hobbs
