Double close

[humanthrope]

The PidFile class registers close() to be called at process exit time, but if the user manually calls close(), this can lead to a race condition where Instance A of a script closes the pid file and instance Instance B of the same script is allowed to run, but A delete's B's pid file.

Pseudo Code

pid_file = PidFile(...)
try:
    pid_file.create()
except Exception:
    if not pid_file.filename or not os.path.exists(pid_file.filename):
        raise
    raise MyAlreadyRunningError

try:
    do_something()
finally:
    pid_file.close()
    # A hasn't called at exit yet, but B is allowed to run
    # Shortly later, A exits and A's `atexit` closes the pid file that B creates

It would be possible to not call close(), but we have tests to check for MyAlreadyRunningError and pytest does not exit (therefore never calls the registered atexit handlers) until after the full suite of test has been run.

[trbs]

I see... good catch !

Could you please double check if aad0c9c02d416f08a18a25cf6c77ad1b2e2c8b53 properly solves this ?

[humanthrope]

Looks good to me. Thanks!

[trbs]

Awesome, thanks for reporting the issue.