shell/ash_doc.txt

	Wait + signals

We had some bugs here which are hard to test in testsuite.

Bug 1280 (http://busybox.net/bugs/view.php?id=1280):
was misbehaving in interactive ash. Correct behavior:

$ sleep 20 &
$ wait
^C
$ wait
^C
$ wait
^C
...


Bug 1984 (http://busybox.net/bugs/view.php?id=1984):
traps were not triggering:

trap_handler_usr () {
    echo trap usr
}
trap_handler_int () {
    echo trap int
}
trap trap_handler_usr USR1
trap trap_handler_int INT
sleep 3600 &
echo "Please do: kill -USR1 $$"
echo "or: kill -INT $$"
while true; do wait; echo wait interrupted; done


Bug 189 (https://bugs.busybox.net/show_bug.cgi?id=189)

func() {
    sleep 1
}
while (true); do
    func
    echo Looping
done

^C was observed to make ash processes geometrically multiply (!) instead
of exiting. (true) in subshell does not seem to matter, as another user
reports the same with:

trap "echo USR1" USR1
while true; do
    echo Sleeping
    sleep 5
done

Compat note.
Bash version 3.2.0(1) exits this script at the receipt of SIGINT
_only_ if it had two last children die from it.
The following trace was obtained while periodically running
"killall -SIGINT sleep; sleep 0.1; kill -SIGINT <bash_PID>":

23:48:32.376707 clone(...) = 13528
23:48:32.388706 waitpid(-1, 0xffc832ec, 0) = ? ERESTARTSYS (To be restarted)
23:48:32.459761 --- SIGINT (Interrupt) @ 0 (0) ---
    kill -SIGINT <bash_PID> is ignored, back to waiting:
23:48:32.463706 waitpid(-1, [{WIFEXITED(s) && WEXITSTATUS(s) == 0}], 0) = 13528
    sleep exited with 0
23:48:37.377557 --- SIGCHLD (Child exited) @ 0 (0) ---
23:48:37.378451 clone(...) = 13538
23:48:37.390708 waitpid(-1, [{WIFSIGNALED(s) && WTERMSIG(s) == SIGINT}], 0) = 13538
    sleep was killed by "killall -SIGINT sleep"
23:48:38.523944 --- SIGCHLD (Child exited) @ 0 (0) ---
23:48:38.524861 clone(...) = 13542
23:48:38.538706 waitpid(-1, 0xffc832ec, 0) = ? ERESTARTSYS (To be restarted)
23:48:38.624761 --- SIGINT (Interrupt) @ 0 (0) ---
    kill -SIGINT <bash_PID> is ignored, back to waiting:
23:48:38.628706 waitpid(-1, [{WIFEXITED(s) && WEXITSTATUS(s) == 0}], 0) = 13542
    sleep exited with 0
23:48:43.525674 --- SIGCHLD (Child exited) @ 0 (0) ---
23:48:43.526563 clone(...) = 13545
23:48:43.538709 waitpid(-1, [{WIFSIGNALED(s) && WTERMSIG(s) == SIGINT}], 0) = 13545
    sleep was killed by "killall -SIGINT sleep"
23:48:44.466848 --- SIGCHLD (Child exited) @ 0 (0) ---
23:48:44.467735 clone(...) = 13549
23:48:44.481706 waitpid(-1, 0xffc832ec, 0) = ? ERESTARTSYS (To be restarted)
23:48:44.567757 --- SIGINT (Interrupt) @ 0 (0) ---
    kill -SIGINT <bash_PID> is ignored, back to waiting:
23:48:44.571706 waitpid(-1, [{WIFEXITED(s) && WEXITSTATUS(s) == 0}], 0) = 13549
    sleep exited with 0
23:48:49.468553 --- SIGCHLD (Child exited) @ 0 (0) ---
23:48:49.469445 clone(...) = 13551
23:48:49.481708 waitpid(-1, [{WIFSIGNALED(s) && WTERMSIG(s) == SIGINT}], 0) = 13551
    sleep was killed by "killall -SIGINT sleep"
23:48:50.515837 --- SIGCHLD (Child exited) @ 0 (0) ---
23:48:50.516718 clone(...) = 13555
23:48:50.530706 waitpid(-1, 0xffc832ec, 0) = ? ERESTARTSYS (To be restarted)
23:48:50.615761 --- SIGINT (Interrupt) @ 0 (0) ---
    kill -SIGINT <bash_PID> is ignored, back to waiting:
23:48:50.619705 waitpid(-1, [{WIFSIGNALED(s) && WTERMSIG(s) == SIGINT}], 0) = 13555
    sleep was killed by "killall -SIGINT sleep".
    This is the second one in a row. Kill ourself:
23:48:51.504604 kill(13515, SIGINT)     = 0
23:48:51.504689 --- SIGINT (Interrupt) @ 0 (0) ---
23:48:51.504915 +++ killed by SIGINT +++

As long as there is at least one "sleep 5" which exited successfully
(not killed by SIGINT), bash continues. This is not documented anywhere
AFAIKS.

Why keyboard ^C acts differently?

00:08:07.655985 clone(...) = 14270
00:08:07.669707 waitpid(-1, [{WIFEXITED(s) && WEXITSTATUS(s) == 0}], 0) = 14270
00:08:12.656872 --- SIGCHLD (Child exited) @ 0 (0) ---
00:08:12.657743 clone(...) = 14273
00:08:12.671708 waitpid(-1, [{WIFSIGNALED(s) && WTERMSIG(s) == SIGINT}], 0) = 14273
00:08:13.810778 --- SIGINT (Interrupt) @ 0 (0) ---
00:08:13.818705 kill(14269, SIGINT)     = 0
00:08:13.820103 --- SIGINT (Interrupt) @ 0 (0) ---
00:08:13.820925 +++ killed by SIGINT +++

Perhaps because at the moment bash got SIGINT it had no children?
(it did not manage to spawn new sleep yet, see the trace)