Bash – Why do ‘nohup’ and ‘disown’ not work on SoX (invoked as ‘play’)

Let's first look at what happens if a program is started from an interactive shell (connected to a terminal) without & (and without any redirection). So let's assume you've just typed foo:

• The process running foo is created.
• The process inherits stdin, stdout, and stderr from the shell. Therefore it is also connected to the same terminal.
• If the shell receives a SIGHUP, it also sends a SIGHUP to the process (which normally causes the process to terminate).
• Otherwise the shell waits (is blocked) until the process terminates or gets stopped.

Now, let's look what happens if you put the process in the background, that is, type foo &:

• The process running foo is created.
• The process inherits stdout/stderr from the shell (so it still writes to the terminal).
• The process in principle also inherits stdin, but as soon as it tries to read from stdin, it is halted.
• It is put into the list of background jobs the shell manages, which means especially:
  • It is listed with jobs and can be accessed using %n (where n is the job number).
  • It can be turned into a foreground job using fg, in which case it continues as if you would not have used & on it (and if it was stopped due to trying to read from standard input, it now can proceed to read from the terminal).
  • If the shell received a SIGHUP, it also sends a SIGHUP to the process. Depending on the shell and possibly on options set for the shell, when terminating the shell it will also send a SIGHUP to the process.

Now disown removes the job from the shell's job list, so all the subpoints above don't apply any more (including the process being sent a SIGHUP by the shell). However note that it still is connected to the terminal, so if the terminal is destroyed (which can happen if it was a pty, like those created by xterm or ssh, and the controlling program is terminated, by closing the xterm or terminating the SSH connection), the program will fail as soon as it tries to read from standard input or write to standard output.

What nohup does, on the other hand, is to effectively separate the process from the terminal:

• It closes standard input (the program will not be able to read any input, even if it is run in the foreground. it is not halted, but will receive an error code or EOF).
• It redirects standard output and standard error to the file nohup.out, so the program won't fail for writing to standard output if the terminal fails, so whatever the process writes is not lost.
• It prevents the process from receiving a SIGHUP (thus the name).

Note that nohup does not remove the process from the shell's job control and also doesn't put it in the background (but since a foreground nohup job is more or less useless, you'd generally put it into the background using &). For example, unlike with disown, the shell will still tell you when the nohup job has completed (unless the shell is terminated before, of course).

So to summarize:

• & puts the job in the background, that is, makes it block on attempting to read input, and makes the shell not wait for its completion.
• disown removes the process from the shell's job control, but it still leaves it connected to the terminal. One of the results is that the shell won't send it a SIGHUP. Obviously, it can only be applied to background jobs, because you cannot enter it when a foreground job is running.
• nohup disconnects the process from the terminal, redirects its output to nohup.out and shields it from SIGHUP. One of the effects (the naming one) is that the process won't receive any sent SIGHUP. It is completely independent from job control and could in principle be used also for foreground jobs (although that's not very useful).

For your command to detect eof, it has to read from stdin. So presumably it is expecting some input. So it sounds like what you need is not an empty input (/dev/null is exactly meant for that), but input that never comes.

It can be simulated with a pipe where nobody is ever going to write on the other end like:

sleep 999999999 | the-command

Or to avoid having to run that extra sleep command, it could be done with a named pipe:

fifo=$(mktemp -u) &&
  mkfifo "$fifo" &&
  (rm "$fifo" && the-command <&3 3<&- &) 3<> "$fifo"

Here using an intermediary file descriptor to work around the fact that the shell connects stdin to /dev/null implicitely when you start a command with & (unless you add an explicit stdin redirection like our <&3 here).

On Linux (and probably on Linux only), you can also do:

the-command < /dev/fd/1 3>&1 > /dev/null | :

/dev/fd/1 where fd 1 is connected to a pipe, on Linux, behaves like a named pipe. That is, when you open it in read mode, you get the reading end of the pipe.

So above, fd 0 will be connected to the reading end of a pipe whose other end is on the fd 3 of the-command. Because the-command is not going to write anything on its fd 3, any read attempt on fd 0 will block (or a non-blocking read will return with there's nothing to read yet, or a select/poll will return nothing to read either as the-command is probably doing if it's doing anything else than waiting for input that never comes).