[Haskell-cafe] Re: do

[Brandon S. Allbery KF8NH]

On Oct 15, 2007, at 21:01 , Ryan Ingram wrote:

> Oops, I read too fast, you mentioned that as #1.
>
> On 10/15/07, ChrisK <haskell at list.mightyreason.com> wrote:
>> Also you need to get you hand on State# RealWorld either
>>  (1) Honestly, by wrapping your code in IO again and using it  
>> normally
>
> Silly me.

That makes two of us; my (4) is in fact his (1), I misunderstood the  
"wrapping your code in IO again".

For those trying to follow along with all this silliness, the secret  
to making this work is defining a runIO function --- which is *not*  
the same as unsafePerformIO, because that invokes the internal  
RealWorld# constructor to inject a "fresh" State# RealWorld, whereas  
we're using the one that we were legitimately passed:

   runIO          :: IO a -> State# RealWorld -> (# State#  
RealWorld,a #)
   runIO (IO f) s =  lazy (f s)

(See libraries/base/GHC/IOBase.lhs for why we need the GHC builtin  
"lazy" there.)

The reason for this is that IO is a "state-like" type:

   newtype IO a = IO (State# RealWorld -> (# State# RealWorld,a #))

Which is to say, it is a wrapper around a function which accepts a  
state and returns a "modified" (in our case, actually just passed  
through without even looking at it) state along with the result of a  
computation.  The runIO function is just runState specialized to this  
newtype.  (Because you can't pattern match on (IO (s -> (# s,a #)))  
in a function definition.)

I'm working on a "simple" example of how to do this without treating  
IO as a monad.  So far it's proving a very concrete demonstration to  
me of why you do *not* want to do it this way, but instead should  
consider the monad your friend.  :)

-- 
brandon s. allbery [solaris,freebsd,perl,pugs,haskell] allbery at kf8nh.com
system administrator [openafs,heimdal,too many hats] allbery at ece.cmu.edu
electrical and computer engineering, carnegie mellon university    KF8NH