Wait on child PID after failed exec

[DaveCTurner]

On Linux, the following program generates 10 zombie processes, one for each time that callProcess tries to call a nonexistent process.

import Control.Concurrent
import Control.Exception
import Control.Monad
import System.IO
import System.Process

ignoreIOException :: IOException -> IO ()
ignoreIOException _ = return ()

main = do
  _ <- forkIO $ replicateM_ 10 $ callProcess "does-not-exist" [] `catch` ignoreIOException
  _ <- getLine
  return ()

The reason seems to be that the child PID leaks because the parent does not call waitpid() if the child has failed. This simple change seems to fix it.

[dcoutts]

After looking at it for a while, I agree.

Minor point, as far as I can see (pid > 0) is necessarily true at that point in the code, and I'd add a comment, e.g.

     // We forked the child, but the child had a problem and stopped
     // so it's our responsibility to reap here as nobody else can.
     waitpid(pid, &unused, 0);

[DaveCTurner]

Seems reasonable. Do you want to make that change yourself or do you want another PR?

[DaveCTurner]

Ok, 77dc6dc takes your suggestions into account.

The docs for waitpid() say "If status is not NULL,..." which sort of implies that NULL is acceptable when you don't care about the status. So I've used that.

[dcoutts]

Thanks!