bug#30762: 27.0.50; Apparent regression in process.c caused by checking pty_flag before calling setsid

bug#30762: 27.0.50; Apparent regression in process.c caused by checking pty_flag before calling setsid

[Thomas Fitzsimmons]

Hi,

On x86_64, CentOS 7.3.1611, I noticed that on master tip (I tested
461e6818...), eudc-expand-inline was hanging.  It calls the ldapsearch
command line utility (by way of ldap-search-internal) using
start-process, and waits for a password prompt:

  (let* ((process-connection-type nil)
	 (proc-args (append arglist ldap-ldapsearch-args
			    filter))
	 (proc (apply #'start-process "ldapsearch" buf
		      ldap-ldapsearch-prog
		      proc-args)))
    (while (null (progn
		   (goto-char (point-min))
		   (re-search-forward
		    ldap-ldapsearch-password-prompt-regexp
		    (point-max) t)))
      (accept-process-output proc 1))
      [...])

master tip gets stuck in that while loop, whereas Emacs 25.1 is fine.

If I print the contents of the buffer within the loop, it shows:

Process ldapsearch stopped (tty output)

I isolated the working-to-non-working transition to this commit:

commit 8e5a769965313a7a1c42b5992ed24e8b0ea71ead
Author: Paul Eggert <eggert@cs.ucla.edu>
Date:   Tue Dec 6 23:41:45 2016 -0800

    Clean out some IRIX cruft
    
    * etc/MACHINES: Remove obsolete discussion of IRIX.
    * src/process.c (allocate_pty) [__sgi]: Remove SGI-specific code.
    (create_process) [HAVE_PTYS]: Don't worry about IRIX.
    * src/syntax.c (scan_sexps_forward): Remove obsolete comment.
    * src/unexelf.c [__sgi]: Don't include <syms.h>.
    (unexec) [__sgi]: Remove SGI-specific code.

and specifically the addition of this check in src/process.c:

-      /* We tried doing setsid only if pty_flag, but it caused
-	 process_set_signal to fail on SGI when using a pipe.  */
-      setsid ();
+      if (pty_flag)
+	setsid ();

If I remove the if (pty_flag) line, then eudc-expand-inline works again.

It may be that ldap-search-internal should change how it's invoking
ldapsearch, but I thought I would report the change in process.c
behavior anyway.

Thomas

bug#30762: 27.0.50; Apparent regression in process.c caused by checking pty_flag before calling setsid

[Eli Zaretskii]

> From: Thomas Fitzsimmons <fitzsim@fitzsim.org>
> Date: Fri, 09 Mar 2018 23:18:51 -0500
> Cc: Paul Eggert <eggert@cs.ucla.edu>
> 
> master tip gets stuck in that while loop, whereas Emacs 25.1 is fine.
> 
> If I print the contents of the buffer within the loop, it shows:
> 
> Process ldapsearch stopped (tty output)
> 
> I isolated the working-to-non-working transition to this commit:
> 
> commit 8e5a769965313a7a1c42b5992ed24e8b0ea71ead
> Author: Paul Eggert <eggert@cs.ucla.edu>
> Date:   Tue Dec 6 23:41:45 2016 -0800
> 
>     Clean out some IRIX cruft
>     
>     * etc/MACHINES: Remove obsolete discussion of IRIX.
>     * src/process.c (allocate_pty) [__sgi]: Remove SGI-specific code.
>     (create_process) [HAVE_PTYS]: Don't worry about IRIX.
>     * src/syntax.c (scan_sexps_forward): Remove obsolete comment.
>     * src/unexelf.c [__sgi]: Don't include <syms.h>.
>     (unexec) [__sgi]: Remove SGI-specific code.
> 
> and specifically the addition of this check in src/process.c:
> 
> -      /* We tried doing setsid only if pty_flag, but it caused
> -	 process_set_signal to fail on SGI when using a pipe.  */
> -      setsid ();
> +      if (pty_flag)
> +	setsid ();
> 
> If I remove the if (pty_flag) line, then eudc-expand-inline works again.

This change is on the emacs-26 branch as well, so if we want to
restore previous code, we should do it on emacs-26, not only on
master.

Thanks.

bug#30762: 27.0.50; Apparent regression in process.c caused by checking pty_flag before calling setsid

[Paul Eggert]

Eli Zaretskii wrote:
> This change is on the emacs-26 branch as well, so if we want to
> restore previous code, we should do it on emacs-26, not only on
> master.

I installed the fix into emacs-26 and merged it into master. Thanks, Thomas, for 
reporting it and for tracking down the fix.