Debian Bug report logs - #501133
please add patch to allow keeping fds open

Toggle useless messages

Message #5 received at submit@bugs.debian.org (full text, mbox, reply):

From: Thomas Viehmann <tv@beamnet.de>

To: Debian Bug Tracking System <submit@bugs.debian.org>

Subject: please add patch to allow keeping fds open

Date: Sat, 04 Oct 2008 15:50:15 +0200

[Message part 1 (text/plain, inline)]

Package: vte
Version: 1:0.16.14-3
Severity: important
X-Debbugs-CC: 493353@bugs.debian.org

Hi Loic, Sebastian, Gnome Maintainers,

as witnessed by #493353, the GTK frontend to gdebi is completely broken
because it relies on a patch to vtk that provides a way to keep dpkg's
status-fd open. The patch looks fairly sane to me.
In terms of the lenny release, we seem to have three options
1) patch vtk as attached,
2) remove gdebi-gtk (or gdebi),
3) work around the problem (e.g. open a named pipe and invoke a dpkg
   wrapper that proxies that to the dpkg status-fd).

At this point, the easiest path seems to be 1), with the main argument
against it being that upstream 0.17 does not seem to have incorporated
the patch.

Opinions? I'd be happy to actually do the upload, but I would prefer an
explicit OK for it, either here or on IRC.

Kind regards

T.
-- 
Thomas Viehmann, http://thomas.viehmann.net/

[vte-keep-fds.diff (text/x-patch, inline)]

diff -u vte-0.16.14/debian/changelog vte-0.16.14/debian/changelog
--- vte-0.16.14/debian/changelog
+++ vte-0.16.14/debian/changelog
@@ -1,3 +1,10 @@
+vte (1:0.16.14-3.1) unstable; urgency=low
+
+  * Non-maintainer upload.
+  * Add patch from Ubuntu to allow children to keep fds open.
+
+ -- Thomas Viehmann <tv@beamnet.de>  Sat, 04 Oct 2008 15:23:04 +0200
+
 vte (1:0.16.14-3) unstable; urgency=low
 
   * Fix 1:0.16.14-2 changelog entry.
only in patch2:
unchanged:
--- vte-0.16.14.orig/debian/patches/91_keep_fds.patch
+++ vte-0.16.14/debian/patches/91_keep_fds.patch
@@ -0,0 +1,110 @@
+diff -Nur vte-0.16.0/src/pty.c vte-0.16.0.new/src/pty.c
+--- vte-0.16.0/src/pty.c	2007-02-26 22:05:14.000000000 +0100
++++ vte-0.16.0.new/src/pty.c	2007-03-16 10:15:20.000000000 +0100
+@@ -160,7 +160,27 @@
+ 		const char *name;
+ 		int fd;
+ 	} tty;
++	int keep_fd;
+ };
++
++static int
++_vte_pty_keep_fd(char **env_add)
++{
++   int i, res = -1;
++   if(env_add == NULL)
++      return res;
++
++   const gchar *needle="VTE_PTY_KEEP_FD=";
++   for(i=0; env_add[i] != NULL; i++) {
++      gchar *s = strstr(env_add[i],needle);
++      if(s != NULL) {
++	 res = atoi(&s[strlen(needle)]);
++	 break;
++      }
++   }
++   return res;
++}
++
+ static void
+ vte_pty_child_setup (gpointer arg)
+ {
+@@ -185,7 +205,6 @@
+ 			"Setting up child pty: name = %s, fd = %d\n",
+ 				tty ? tty : "(none)", fd);
+ 
+-
+ 	/* Start a new session and become process-group leader. */
+ #if defined(HAVE_SETSID) && defined(HAVE_SETPGID)
+ 	_vte_debug_print (VTE_DEBUG_PTY, "Starting new session\n");
+@@ -253,6 +272,20 @@
+ 		close(fd);
+ 	}
+ 
++	int i;
++	int keep_fd = data->keep_fd;
++	if(keep_fd > 0) {
++	   /* Close most descriptors. */
++	   for (i = 0; i < sysconf(_SC_OPEN_MAX); i++) {
++	      if (i != keep_fd &&
++		  i != fd && 
++		  i != STDOUT_FILENO && 
++		  i != STDIN_FILENO && 
++		  i != STDERR_FILENO) {
++		 close(i);
++	      }
++	   }
++	}
+ 
+ 	/* Reset our signals -- our parent may have done any number of
+ 	 * weird things to them. */
+@@ -345,13 +378,16 @@
+ 			g_printerr ("    directory: %s\n",
+ 					directory ? directory : "(none)");
+ 		}
+-
++ 		GSpawnFlags flags = (G_SPAWN_CHILD_INHERITS_STDIN |
++  				     G_SPAWN_FILE_AND_ARGV_ZERO |
++  				     G_SPAWN_SEARCH_PATH |
++  				     G_SPAWN_DO_NOT_REAP_CHILD);
++		if(argv)
++		   flags |= G_SPAWN_FILE_AND_ARGV_ZERO;
++  		if (data->keep_fd > 0)
++  		   flags |= G_SPAWN_LEAVE_DESCRIPTORS_OPEN;
+ 		ret = g_spawn_async_with_pipes (directory,
+-				arg2, envp2,
+-				G_SPAWN_CHILD_INHERITS_STDIN |
+-				G_SPAWN_SEARCH_PATH |
+-				G_SPAWN_DO_NOT_REAP_CHILD |
+-				(argv ? G_SPAWN_FILE_AND_ARGV_ZERO : 0),
++				arg2, envp2, flags,
+ 				vte_pty_child_setup, data,
+ 				pid,
+ 				NULL, NULL, NULL,
+@@ -407,6 +443,10 @@
+ 	return ret;
+ }
+ 
++
++
++
++
+ /* Open the named PTY slave, fork off a child (storing its PID in child),
+  * and exec the named command in its own session as a process group leader */
+ static gboolean
+@@ -419,6 +459,7 @@
+ 
+ 	data.mode = TTY_OPEN_BY_NAME;
+ 	data.tty.name = path;
++	data.keep_fd = _vte_pty_keep_fd(envp);
+ 
+ 	if (!_vte_pty_run_on_pty(&data,
+ 			command, argv, envp, directory,
+@@ -443,6 +484,7 @@
+ 
+ 	data.mode = TTY_OPEN_BY_FD;
+ 	data.tty.fd = fd;
++	data.keep_fd = _vte_pty_keep_fd(envp);
+ 
+ 	if (!_vte_pty_run_on_pty(&data,
+ 				command, argv, envp, directory,

Message #10 received at 501133-done@bugs.debian.org (full text, mbox, reply):

From: Thomas Viehmann <tv@beamnet.de>

To: 501133-done@bugs.debian.org

Subject: on a second thought...

Date: Sat, 04 Oct 2008 16:37:34 +0200

... fixing gdebi to not rely on funny stuff seems the better solution.
In particular, it avoids a versioned dependency of gdebi on libvteX.

Kind regards

T.
-- 
Thomas Viehmann, http://thomas.viehmann.net/

Send a report that this bug log contains spam.