Issue 12495: Rewrite InterProcessSignalTests - Python tracker (original) (raw)

InterProcessSignalTests uses 4 different signal handlers (SIGUSR1, SIGUSR2, SIGALRM, SIGHUP) and uses 2 methods to raise a signal (a subprocess executing "kill -SIG pid" or signal.alarm(1)). It uses signal.pause() and/or time.sleep(1) to wait the signal, or just nothing for SIGUSR2 (configured to be ignored).

The testcase tests too many unrelated things.

signal.alarm(1) is not an interprocess signal: it is a signal send to the process itself.

Why using two different signal handlers? A modifies a_called, B modifies b_called but raise also a Python exception. How is it related to interprocess signal handling?

Why checking that the handler A (SIGHUP) has not been called when the signal handler B (SIGUSR1) is called?

Why is the garbage collector disabled?

I propose to write a new simple testcase: install a signal handler raising a Python exception, send a signal using a child process, ensure that the signal has been received (wait for the exception). Pseudo-code:

s = signal.SIGUSR1 def handler(signum, frame): 1/0 signal.signal(s, handler) try: subprocess.call([sys.executable, '-c', '... kill(%s, %s)' % (os.getpid(), s)) ... wait the signal ... except ZeroDivisionError: pass else: raise Exception("ZeroDivisionError not raised"

The whole test has to be run in a subprocess.

The new test may pass on freebsd 6, it should be checked (see issue #12469).