bug#7225: 23.2.50; [PATCH] C-c C-c doesn't work in shell mode (Cygwin)

[Ken Brown]

On 10/17/2010 6:08 PM, Eli Zaretskii wrote:
> > Date: Sun, 17 Oct 2010 17:40:39 -0400
> > From: Ken Brown<kbrown@cornell.edu>
> > CC: "7225@debbugs.gnu.org"<7225@debbugs.gnu.org>
> >
> > I haven't yet found an alternative to TIOCGPGRP that works for Cygwin,
> > and it may well be that there isn't one.  But I think there's a separate
> > issue that is not specific to Cygwin.  It seems to me that line 6233 of
> > process.c is simply wrong, and not just for Cygwin.
> >
> > When we reach that line, we want the process group ID of the foreground
> > process group of the terminal associated with p (which is a shell in the
> > most common use case).  We don't have TIOCGPGRP, so we don't know how to
> > do this.  We therefore give up and set gid = p->pid.  Is there any
> > situation in which this is the right thing to do?  It means that (in the
> > common use case) we'll send the signal to the shell instead of to the
> > process running in the shell.  Wouldn't it be better to just return at
> > that point and issue a warning message saying that we can't send the signal?
>
> The problem that code is trying to solve is how to send a signal to
> the whole process group starting at the shell (or whatever process is
> the group leader).  Failure to do so could mean that the immediate
> subprocess of Emacs will get the signal, but its children will not.
> If the signal kills the subprocess, its children may remain behind as
> orphans.

Am I misunderstanding the comment preceding the definition of 
emacs_get_tty_pgrp?  Here's what it says:

/* Return the foreground process group for the tty/pty that
   the process P uses.  */

That's not the same as the process group of the shell, at least in 
Cygwin.  See below.  You seem to be assuming that the process group of 
the shell will include all of the shell's children.  Is that what 
happens in GNU/Linux?

> > What this would mean for Cygwin, once I make the change I proposed (and
> > assuming I don't find a better solution), is that the only signals we'll
> > be able to send to the foreground process of a shell are SIGINT,
> > SIGQUIT, and SIGTSTP, and we'll see a failure message if we try to send
> > a different signal.  That's better than sending a signal to the wrong
> > process.
>
> It's not the wrong process.  The problem is that its children might
> not get the signal.

It is the wrong process on Cygwin.  Here's an example.  I start a shell 
and then run 'cat'.  In another shell I run 'ps'.  The relevant part of 
the output is:

      PID    PPID    PGID     WINPID  TTY  UID    STIME COMMAND
      508     920     508       3552    1 1009 18:50:15 /usr/bin/sh
I     916     508     916       1832    1 1009 18:50:18 /usr/bin/cat

508 is the PID as well as the PGID of the shell.  But cat is in a 
different process group, even though it's a child of the shell (as we 
see from the PPID).  On the other hand, it has the same controlling tty 
as the shell, so we'd be in business if we could get the PGID of the 
foreground process of the controlling tty of the shell.  That's what I 
thought emacs_get_tty_pgrp was doing (according to its comment).  When I 
type C-c C-c, emacs sends a signal to PGID 508, which is not the right 
thing to do.

> So the question is: how to send a signal to the whole process group
> with Cygwin?  Does Cygwin support the "minus pgid" method of calling
> `kill'?

Yes.