Strange behavior while piping process stdout

I am making a program which shows stdout of my haskell program on UI.
I simply hooked it up with some text area, so I expected it to run normally other than potential crashes by mishaps.
However, I got waitForProcess: does not exist (No child processes) on the child program.
Furthermore, the direct output of the child comes after the output of child of child.

To illustrate, let’s say the UI program spawned a (haskell) program A.
The program A spawns another program B through callCommand.
Now, I expect the output like this:

A: Begin
B: Begin
(Further output of B)
B: End
A: End

and then program A should not emit waitForProcess: does not exist (No child processes) unless something is messed up.

However, what I got is this:

B: Begin
(Further output of B)
B: End
A: waitForProcess: does not exist (No child processes)
A: Begin
My UI Program: waitForProcess: does not exists (No child processes)

I’m sure that the part printing A: Begin should be running, given that there is a noticeable pause (~1s) until B: Begin is printed - likely when A starts running and calls the command.

Perhaps relevant: I inspect the stdout line-by-line like this

actOnLine outp act =
  hIsEOF outp >>= \case
    True -> pure ()
    False -> do
              line <- T.hGetLine outp
              T.putStrLn line
              act line
              actOnLine outp act

Which resides inside the bracket from withCreateProcess.

I am honestly lost on how to start debugging the problem. Would appreciate any idea on what is going on!

What kind of UI are we talking about?

Can you show more of the code - for example the way you span and wait for the process?

Anyway just to be sure I’d start by disabling the output- and input-buffers ( hSetBuffering).
Especially on Windows input-/output-buffers are tricky from Haskell.

Oh, I mean GUI with label showing the current console output.
I am on Linux, so it’s likely not a platform issue.

Here’s the surrounding code relevant to the process creation:

  do
      forkIO . withCreateProcess (proc "xmonad-manage" ["build", "pulpmonad"]){std_out = CreatePipe} $
        \_ (Just outp) _ _ -> actOnLine outp myAct
      pure ()
      where
        actOnLine outp act =
          hIsEOF outp >>= \case
            True -> #setVisibleChildName stack (T.pack "main")
            False -> do
              line <- T.hGetLine outp
              T.putStrLn (T.pack "Got: " <> line)
              act line
              actOnLine outp act

Where do I set the buffering, the UI program or program A?
I’d like to avoid touching program A, it was left untouched for quite a while.

So just to make sure:

  • the code here is in the UI program
  • so the UI calls A with withCreateProcess
  • A calls B with callCommand?

in this case B's stdout should be inherited from A so I think this should be allright.

Never done anything like this so this is just what I’d try next but try

forkIO . withCreateProcess (proc "xmonad-manage" ["build", "pulpmonad"]){std_out = CreatePipe} $
    \_ (Just outp) _ _ -> do
        hSetBuffering outp NoBuffering
        actOnLine outp myAct

and see if this changes anything - it probably won’t as the default should be LineBuffering - meaning the buffer should get flushed after each newline - so putStrLn should flush it

PS: where does the waitForProcess happen?

1 Like

Uhm, I never wait for the process from withCreateProcess - as you could see, I currently ignore the ProcessHandle.
Should I? I thought withCreateProcess will wait for the process to clean it up properly.

I was asking for the error you mentioned - if you don’t do it it’ll most likely be your callCommand (as this uses waitForProcess) - but it’s really strange that this fails within

Yep, the callCommand one does not do any other waiting business.
EDIT: Just tried alternative buffering. Does not change anything.

Hi,

I dried to recreate this situation with this small program:

{-# LANGUAGE LambdaCase #-}

module Main where

import Control.Concurrent (forkIO)
import System.Environment (getArgs)
import System.IO (hGetLine, hIsEOF)
import System.Process

main :: IO ()
main = do
  name : path : cmds <- getArgs
  let go outp =
        hIsEOF outp >>= \case
          True -> putStrLn $ name ++ " done"
          False -> do
            line <- hGetLine outp
            putStrLn $ name ++ " received " ++ line
            go outp
  putStrLn $ name ++ " starting.."
  res <- withCreateProcess (proc path cmds) {std_out = CreatePipe} $
    \_ (Just outp) _ _ -> go outp
  putStrLn $ name ++ " ended."
  pure ()

this is not exactly the same (as it does not use callCommand and I opted out of Text for now) but this works for me as expected:

$ ./hsRun A echo "Hi"
A starting..
A received Hi
A done
A ended.

$ ./hsRun A ./hsRun B echo "Hi"
A starting..
A received B starting..
A received B received Hi
A received B done
A received B ended.
A done
A ended.

can you come up with a small example that demonstrates your problem I can try locally?

Trying several things with this example, I found another potential problem.
While the order itself is retained, the timing is weird.

With the following code,

{-# LANGUAGE LambdaCase #-}

module Main where

import Control.Concurrent
import System.Environment (getArgs)
import System.IO (hGetLine, hIsEOF)
import System.Process

main :: IO ()
main = do
  name : path : cmds <- getArgs
  let go outp =
        hIsEOF outp >>= \case
          True -> putStrLn $ name ++ " done"
          False -> do
            line <- hGetLine outp
            putStrLn $ name ++ " received " ++ line
            go outp
  putStrLn $ name ++ " starting.."
  threadDelay 1000000
  withCreateProcess (proc path cmds){std_out = CreatePipe} $
    \_ (Just outp) _ _ -> go outp
  threadDelay 1000000
  putStrLn $ name ++ " ended."
  pure ()

Time delays are like this:

A starting..
    (waits 3 seconds)
A received B starting..
A received B received Hi
A received B done
A received B ended.
A done
    (waits 1 second)
A ended.

It seems like the inputs are received “after” B finishes. Which means, the outputs are somehow clumped? I don’t get why. Isn’t the default LineBuffering?

Hi I added the NoBuffering and this seems to fix the “delay” for me (testing this on windows though)

{-# LANGUAGE LambdaCase #-}

module Main where

import Control.Concurrent (forkIO, threadDelay)
import System.Environment (getArgs)
import System.IO (BufferMode (NoBuffering), hGetLine, hIsEOF, hSetBuffering, stdout)
import System.Process

main :: IO ()
main = do
  hSetBuffering stdout NoBuffering
  name : path : cmds <- getArgs
  let go outp =
        hIsEOF outp >>= \case
          True -> putStrLn $ name ++ " done"
          False -> do
            line <- hGetLine outp
            putStrLn $ name ++ " received " ++ line
            go outp
  putStrLn $ name ++ " starting.."
  threadDelay 1000000
  putStrLn $ name ++ " waited"
  res <- withCreateProcess (proc path cmds) {std_out = CreatePipe} $
    \_ (Just outp) _ _ -> do
      hSetBuffering outp NoBuffering
      go outp
  putStrLn $ name ++ " ended."
  threadDelay 1000000
  putStrLn $ name ++ " waited"
  pure ()

I see, so that was the intended behavior (although I do not know why).

I cannot reproduce it easily, it seems like program A emits waitForProcess: does not exist when e.g. terminated via Ctrl+C. I wonder why.

Would Text.Printf be relevant? I use it in the program A.

for the CTRL+C part you might want to read the documentation around delegate_ctlc here

if this is the intended behaviour? Don’t know (don’t see anything from the docs).

Also don’t know if Text.Printf would help (don’t see why it should but cannot say I’m sure)

I got the strange ordering now when using in combination with callCommand.

Code without problem:

{-# LANGUAGE LambdaCase #-}

module Main where

import Control.Concurrent
import System.Environment (getArgs)
import System.IO
import System.Process

main :: IO ()
main = do
  name : path : cmds <- getArgs
  hSetBuffering stdout NoBuffering
  let go outp =
        hIsEOF outp >>= \case
          True -> putStrLn $ name ++ " done"
          False -> do
            line <- hGetLine outp
            putStrLn $ name ++ " received " ++ line
            go outp
  putStrLn $ name ++ " starting.."
  threadDelay 1000000
  withCreateProcess (proc path cmds){std_out = CreatePipe} $
    \_ (Just outp) _ _ -> go outp
  threadDelay 1000000
  callProcess path cmds
  threadDelay 1000000
  putStrLn $ name ++ " ended."
  pure ()

gives order

A starting..
A received B starting..
A received B received Hi
A received B done
A received Hi
A received B ended.
A done
B starting..
B received Hi
B done
Hi
B ended.
A ended.

Now, if I comment out hSetBuffering stdout NoBuffering like this:

...
  name : path : cmds <- getArgs
  hSetBuffering stdout NoBuffering
  let go outp =
...

I got

A starting..
A received Hi
A received B starting..
A received B received Hi
A received B done
A received B ended.
A done
B starting..
B received Hi
B done
Hi
B ended.
A ended.

which doesn’t make sense to me, as the program should be fairly sequential.
It almost seems like the buffer from the different processes interfere each other…

With the default buffering, I also got waitForProcess: does not exist (No child processes) on Ctrl+C.

Any idea what is happening? I am at a loss now.

The spurious waitForProcess error message is probably a bug in the process package, fixed in version 1.6.14.0 via https://github.com/haskell/process/pull/231.

2 Likes

Thanks to @robx for fixing this and also improving cabal-install process handling.

1 Like

Oh, so I need to update the process package. Could take long, but better do it now than later.

By the way, does interrupting the current thread impact the process call?
Haddock documentation of callProcess has the following line:

If an asynchronous exception is thrown to the thread executing callProcess,
the forked process will be terminated and callProcess will wait (block)
until the process has been terminated.

…which makes me worried, because I routed the thread to block and send the child process output to another thread.

EDIT: the bugfix did not fix my problem. Uh, should I not block the thread?
…strange, waitForProcess seem to always throw in particular circumstances - even without the blocking calls. Even more strange that the error bubbles up from the process calling cabal.
Sadly, cannot reproduce it in simple examples.