Re: Signal handling bugaboo in command substitution

[Peter Stephenson <p.stephenson@xxxxxxxxxxx>]

On Mon, 07 Mar 2016 18:44:06 -0800
Bart Schaefer <schaefer@xxxxxxxxxxxxxxxx> wrote:
> At a PS1 prompt:
> 
> torch% print $(sleep 3; echo foo)
> 
> Press ctrl+z during the sleep.  Zsh is now hung, because the command
> substitution is occuring in prefork() so there's nothing to handle the
> stopped children and the parent itself ignores the signal.  Zsh is hung;
> it won't return to a prompt, the command substitution will never produce
> the awaited output, and nothing (except "kill -CONT" from another shell)
> will wake it back up.
> 
> I'm not sure what to do here.  In other circumstances it's OK to stop a
> command substitution with a ctrl+z, and in any kind of non-interactive
> shell or even in a subshell the parent would handle the signal.
> 
> Bash appears to leave TSTP blocked here when interactive.  I don't think
> testing for interactivity is sufficient in zsh context, though.

How about something like this?

pws

diff --git a/Src/exec.c b/Src/exec.c
index b60fc90..caeb461 100644
--- a/Src/exec.c
+++ b/Src/exec.c
@@ -1001,9 +1001,21 @@ entersubsh(int flags)
 	 * signals.  If it is, we need to keep the special behaviour:
 	 * see note about attachtty() above.
 	 */
-	signal_default(SIGTTOU);
-	signal_default(SIGTTIN);
-	signal_default(SIGTSTP);
+	if (flags & ESUB_NOMONITOR)
+	{
+	    /*
+	     * Allowing any form of interactive signalling here is
+	     * actively harmful as we are in a context where there is no
+	     * control over the process.
+	     */
+	    signal_ignore(SIGTTOU);
+	    signal_ignore(SIGTTIN);
+	    signal_ignore(SIGTSTP);
+	} else {
+	    signal_default(SIGTTOU);
+	    signal_default(SIGTTIN);
+	    signal_default(SIGTSTP);
+	}
     }
     if (interact) {
 	signal_default(SIGTERM);