Question

I want to invoke a process from within a haskell program and capture stdout as well as stderr.

What I do:

(_, stdout, stderr) <- readProcessWithExitCode "command" [] ""

The problem: This way, stdout and stderr are captured separately, however I want the messages to appear in the right place (otherwise I would simply stdout ++ stderr which separates error messages from their stdout counterparts).

I do know that I could achieve this if I’d pipe the output into a file, i.e.

tmp <- openFile "temp.file" ...
createProcess (proc "command" []) { stdout = UseHandle tmp,
                                    stderr = UseHandle tmp }

So my current workaround is to pipe outputs to a tempfile and read it back in. However I’m looking for a more direct approach.

If I was on unix for sure I’d simply invoke a shell command á la

command 2>&1

and that’s it. However, I’d like to have this as portable as possible.

What I need this for: I’ve built a tiny haskell cgi script (just to play with it) which invokes a certain program and prints the output. I want to html-escape the output, thus I can’t simply pipe it to stdout.

I was thinking: Maybe it’s possible to create an in-memory-handle, like a PipedInputStream/PipedOutputStream in Java, or ArrayInputStream/ArrayOutputStream which allows for processing IO streams within memory. I looked around for a function :: Handle on hoogle, but did not find anything.

Maybe there is another Haskell module out there which allows me to merge two streams?

Answer 1

You can use pipes to concurrently merge two input streams. The first trick is to read from two streams concurrently, which you can do using the stm package:

import Control.Applicative
import Control.Proxy
import Control.Concurrent
import Control.Concurrent.STM
import System.Process

toTMVarC :: (Proxy p) => TMVar a -> () -> Consumer p a IO r
toTMVarC tmvar () = runIdentityP $ forever $ do
    a <- request ()
    lift $ atomically $ putTMVar tmvar a

fromTMVarS :: (Proxy p) => TMVar a -> () -> Producer p a IO r
fromTMVarS tmvar () = runIdentityP $ forever $ do
    a <- lift $ atomically $ takeTMVar tmvar
    respond a

I will soon provide the above primitives in a pipes-stm package, but use the above for now.

Then you just feed each Handle to a separate MVar and read from both concurrently:

main = do
    (_, mStdout, mStderr, _) <- createProcess (proc "ls" [])
    case (,) <$> mStdout <*> mStderr of
        Nothing               -> return ()
        Just (stdout, stderr) -> do
            out <- newEmptyTMVarIO
            err <- newEmptyTMVarIO
            forkIO $ runProxy $ hGetLineS stdout >-> toTMVarC out
            forkIO $ runProxy $ hGetLineS stderr >-> toTMVarC err
            let combine () = runIdentityP $ forever $ do
                    str <- lift $ atomically $
                        takeTMVar out `orElse` takeTMVar err
                    respond str
            runProxy $ combine >-> putStrLnD

Just change out putStrLnD with however you want to process the input.

To learn more about the pipes package, just read Control.Proxy.Tutorial.