Hi Mihai, maybe the term "thread" in my mail is not correct. What I mean is that a value gets stored by f and discovered by g. *f,g :: IO () f = withFile "toto" WriteMode (flip hPutStr "42") g = withFile "toto" ReadMode hGetLine >>= (\s -> putStrLn $ "Answer:" ++ s) main = f >> g* Is it possible to do the same without files (the types must remain IO())? On Wed, Aug 29, 2012 at 11:04 AM, Mihai Maruseac wrote:

On Wed, Aug 29, 2012 at 10:58 AM, Corentin Dupont wrote:

...

Hi all, there is something very basic that it seems escaped me. For example with the following program f and g have type IO () and I can thread a value between the two using a file. Can I do the exact same (not changing the types of f and g) without a file?

f,g :: IO () f = withFile "toto" WriteMode (flip hPutStr "toto") g = withFile "toto" ReadMode hGetLine >>= putStrLn main = f >> g

Of course:

f,g :: IO () f = putStr "Answer: " g = print 42

Main> f >> g Answer: 42

The () is threaded by >>, not the file content

-- MM

On 29 August 2012 10:21, Corentin Dupont wrote:

*f,g :: IO () f = withFile "toto" WriteMode (flip hPutStr "42") g = withFile "toto" ReadMode hGetLine >>= (\s -> putStrLn $ "Answer:" ++ s) main = f >> g*

Is it possible to do the same without files (the types must remain IO())?

One can use an IORef to get a similar effect. import Data.IORef import System.IO.Unsafe {-# NOINLINE toto #-} toto :: IORef String toto = unsafePerformIO (newIORef "") f,g :: IO () f = writeIORef toto "42" g = readIORef toto >>= (\s -> putStrLn $ "Answer:" ++ s) main = f >> g HTH, Ozgur

On Wed, Aug 29, 2012 at 12:00:50PM +0200, Corentin Dupont wrote:

Thanks Eugene and Ozgur. I also looked on the side of IORef, but it doesn't looked to be a good solution since we have to use unsafePerformIO.

I have a big program to modify, and I want to pass some new data to existing functions of type IO(). I'd like to avoid changing all the function's types down the chain... What is the best way to do that?

There is no good way. IORef is probably the best, but as you correctly point out, it is quite ugly. The fact that you have to change your program's types is a feature, not a bug. One of Haskell's great strengths is that the types make plain any communication which can take place between different components. If you want to go behind the type system's back and add hidden communication channels, you might as well be writing in C, and should not be surprised that there is no good way to do it. -Brent