On Fri, Nov 18, 2016 at 7:51 PM, dieter <die...@handshake.de> wrote: > Pedro Franco de Carvalho <redd...@gmail.com> writes: > >> Assume a python program sets a handler function for the >> `signal.SIGALRM` signal, and then schedules an alarm in T seconds with >> `signal.alarm(T)`. >> >> Assume the program later cancels any scheduled alarm with `signal.alarm(0)`. > > I do not think so. You might have a "race condition". > > I think of the following (hypothetical) situation: the alarm > arrives while "signal.alarm(0)" processing has started but not yet finished. > In this case, the alarm arrival will schedule the alarm handler > activation - but, as "signal.alarm" is "C" implemented, the handler > will not be activated immediately (but delayed until the next > execution of Python code).
That situation would be safe if the cancelling of the timer is in a sense of "okay, don't bother" rather than "this must now not happen". For example, you could set a timeout on some operation (say, a subprocess execution) by setting an alarm. When the process terminates, you cancel the alarm. If, just after you cancel the alarm, the signal comes through, that's fine - the handler will already need to cope with the process having just finished. So I would say that it's safe IFF you accept this restriction. ChrisA -- https://mail.python.org/mailman/listinfo/python-list
