Re: Is wait not interruptable?

[Peter Stephenson <pws@xxxxxxx>]

Dave Yost <Dave@xxxxxxxx> wrote:
> This program isn't working as I would expect:
> 
> #!/bin/zsh
> 
> (sleep 3 ; echo A) &
> proc1=$!
> (sleep 4 ; echo B) &
> proc2=$!
> trap 'echo kill $proc2 $proc1 ; kill $proc2 $proc1' INT
> wait $proc1
> kill $proc2
> #ps
> 
> 316 Z% wait-kill-test
> ^Ckill 4099 4098
>
> The ^C was typed 1 second after the command was 
> started, and the first sleep plodded along to completion for its full 
> 3 seconds.

There are two things here, I think.  Unfortunately I'm not a signal expert;
we could do with one to maintain this part of the code.  Bart will probably
be able to shed some more light on the following.

The first is that the subshell itself (the stuff in parentheses) doesn't
get killed because it's waiting for the child process.  This seems to be a
feature of SIGTERM.  If you "setopt trapsasync" or send a different signal
such as SIGHUP it does get killed.  I'm not really sure why SIGTERM is
different.

The second is that even if the subshell gets killed, the sleep processes
themselves don't.  This seems to be because (a) the NOHUP option is on by
default (b) even with "setopt hup" we (i.e. in this case the subshell that
is exiting after the first problem is worked around) don't send HUP to
processes when exiting unless the exiting shell is interactive.

I make no warranty that any of the behaviour I've found is correct, since
signal handling confuses me.

I *think* the patch below fixes (b) by sending SIGHUP to our own process
group when exiting from a non-interactive shell, but I'll take advice on
this.  (I can at least then get the test program to work with "setopt hup
trapsasync".)  The manual isn't clear over whether this is supposed to
happen non-interactively, though it would probably be fair to assume that
if HUP is set in a non-interactive shell you'd expect the SIGHUP to be
sent.  I don't know if I should signal_ignore(SIGHUP) before killpg(0,
SIGHUP) but it didn't seem to be necessary.

The Src/jobs.c hunk is a separate thing I noticed and will commit anyway:
the error message for "kill -" (no option or signal) is confusing.


Index: Src/builtin.c
===================================================================
RCS file: /cvsroot/zsh/zsh/Src/builtin.c,v
retrieving revision 1.171
diff -u -r1.171 builtin.c
--- Src/builtin.c	10 Dec 2006 23:27:03 -0000	1.171
+++ Src/builtin.c	15 Dec 2006 11:44:38 -0000
@@ -4423,10 +4423,9 @@
      */
     errflag = 0;
 
-    if (isset(MONITOR)) {
-	/* send SIGHUP to any jobs left running  */
-	killrunjobs(from_where == 1);
-    }
+    /* send SIGHUP to any jobs left running  */
+    killrunjobs(from_where == 1);
+
     if (isset(RCS) && interact) {
 	if (!nohistsave) {
 	    int writeflags = HFILE_USE_OPTIONS;
Index: Src/jobs.c
===================================================================
RCS file: /cvsroot/zsh/zsh/Src/jobs.c,v
retrieving revision 1.50
diff -u -r1.50 jobs.c
--- Src/jobs.c	6 Nov 2006 12:49:21 -0000	1.50
+++ Src/jobs.c	15 Dec 2006 11:44:38 -0000
@@ -2027,6 +2027,10 @@
 		    return 1;
 		} else
 		    signame = *argv;
+		if (!*signame) {
+		    zwarnnam(nam, "-: signal name expected");
+		    return 1;
+		}
 		signame = casemodify(signame, CASMOD_UPPER);
 		if (!strncmp(signame, "SIG", 3))
 		    signame+=3;
Index: Src/signals.c
===================================================================
RCS file: /cvsroot/zsh/zsh/Src/signals.c,v
retrieving revision 1.41
diff -u -r1.41 signals.c
--- Src/signals.c	30 May 2006 22:35:03 -0000	1.41
+++ Src/signals.c	15 Dec 2006 11:44:38 -0000
@@ -636,9 +636,15 @@
 killrunjobs(int from_signal)
 {
     int i, killed = 0;
- 
+
     if (unset(HUP))
         return;
+
+    if (!isset(MONITOR))
+    {
+	killpg(0, SIGHUP);
+	return;
+    }
     for (i = 1; i <= maxjob; i++)
         if ((from_signal || i != thisjob) && (jobtab[i].stat & STAT_LOCKED) &&
             !(jobtab[i].stat & STAT_NOPRINT) &&




To access the latest news from CSR copy this link into a web browser:  http://www.csr.com/email_sig.php