From mboxrd@z Thu Jan  1 00:00:00 1970
Received: from math.gatech.edu (euclid.skiles.gatech.edu [130.207.146.50]) by werple.net.au (8.7/8.7.1) with SMTP id FAA09955 for <mason@werple.mira.net.au>; Tue, 14 Nov 1995 05:04:10 +1100 (EST)
Received: by math.gatech.edu (5.x/SMI-SVR4)
	id AA03105; Mon, 13 Nov 1995 12:42:13 -0500
Resent-Date: Mon, 13 Nov 1995 18:42:05 +0100
Old-Return-Path: <pws@ifh.de>
Message-Id: <9511131742.AA02362@sgi.ifh.de>
To: zsh-workers@math.gatech.edu (Zsh hackers list)
Subject: Yet another fake subshell fix
Date: Mon, 13 Nov 1995 18:42:05 +0100
From: Peter William Stephenson <pws@ifh.de>
Resent-Message-Id: <"oGgnS2.0.Om.pDufm"@euclid>
Resent-From: zsh-workers@math.gatech.edu
X-Mailing-List: <zsh-workers@math.gatech.edu> archive/latest/604
X-Loop: zsh-workers@math.gatech.edu
Precedence: list
Resent-Sender: zsh-workers-request@math.gatech.edu

More ramifications to do with exec'ing the last command in a shell
that's about to exit:  going into a subshell messes up the shell
terminal file descriptor (plus other things), so that builtins are
unhappy.  You can see it with things like

% zsh -fc 'vared DISPLAY'

(useless example, obviously) which I scrupulously made sure worked
when I wrote the patch to delay setting up the terminal in
non-interactive shells.

The answer is simple, correct, and I should have done it when I first
thought about exec'ing the final command: there's no point in trying
to do the fake-exec thing for a builtin or shell command anyway; it
simply removes a bit of post processing by the shell (which you might
be glad of someday).

If there are any problems with this, they must have been there for a
long time, since this is what zsh always used to do with builtins (as
well as every other command).

This would actually fix the previous problems with exit instead of
_exit in 99.99% of cases (we return to, say, getoutput, which handles
$(...), and _exit from there), but the second patch I sent for that
does get everything right so we still need it.  The remaining .01% of
cases are where a nervous user asks for a real `exec read' or what
have you inside a subshell.

*** Src/exec.c~	Wed Nov  8 10:05:22 1995
--- Src/exec.c	Mon Nov 13 17:58:16 1995
***************
*** 1266,1274 ****
       * in a subshell pipeline, or if the shell is no longer needed    *
       * (e.g. the last command after a `-c').  However, if the command *
       * is going into the background, or if we need to trap EXIT or    *
!      * ZERR, then we can't do this.                                   */
  
!     if (last1 && (how & Z_SYNC)
  	&& !sigtrapped[SIGZERR] && !sigtrapped[SIGEXIT]
  	&& execok()) {
  	cmd->flags |= CFLAG_FAKE_EXEC;
--- 1266,1276 ----
       * in a subshell pipeline, or if the shell is no longer needed    *
       * (e.g. the last command after a `-c').  However, if the command *
       * is going into the background, or if we need to trap EXIT or    *
!      * ZERR, then we can't do this.  Also, if it's getting run in the *
!      * current shell, there's no point (and we unnecessarily mess up  *
!      * terminal settings etc.)                                        */
  
!     if (last1 && (how & Z_SYNC) && !is_cursh
  	&& !sigtrapped[SIGZERR] && !sigtrapped[SIGEXIT]
  	&& execok()) {
  	cmd->flags |= CFLAG_FAKE_EXEC;

-- 
Peter Stephenson <pws@ifh.de>       Tel: +49 33762 77366
WWW:  http://www.ifh.de/~pws/       Fax: +49 33762 77330
Deutches Electronen-Synchrotron --- Institut fuer Hochenergiephysik Zeuthen
DESY-IfH, 15735 Zeuthen, Germany.