Avoids hung subprocess on macOS 10.12.4 when running a background task

[idpaterson]

As discussed in #111 there is a problem on macOS 10.12.4 with running a subprocess from a double-forked daemon. Alfred-Workflow follows the classic recommendation for how to do this in Python so unless it is specific to how the Alfred app runs the workflow the problem may affect other python code as well.

This workaround is not predicated on the OS version. If the immediate exit of the parent process is essential to prevent zombie processes and other issues related to forking only once then this may introduce a regression.

The unit tests work fine in 10.12.4 so I think this may be related to how Alfred runs the workflow.

[idpaterson]

Pressing pause on this for a moment, deploying the workaround in my own workflow did not fix the problem.

[deanishe]

deploying the workaround in my own workflow did not fix the problem

Bugger 😞

You could try removing the forks and using nohup or plain subprocess.Popen instead. The child process must disconnect from IO streams, however, as Alfred doesn't consider a script finished till it closes STDOUT and STDERR.

With native Mac apps, you should always follow a fork with an exec. I'm not sure if that applies to the UNIX layer, however.

[idpaterson]

So glad you found it, well done!

[deanishe]

I think I did. I couldn't get the updated code to hang in 100+ tries across multiple workflows, at any rate.

All I did was compare the code with python-daemon. The use of os._exit() rather than sys.exit() was the only substantial difference, so I tried that.