This bug still exists in current xterm, as far as I can tell.
The following untested patch implements Richard Braakman's suggestion from 1997,
slightly more portably.
Open questions:
(1) Where should the logging be closed?
(2) Is any additional portability goo needed to send this upstream (e.g. for
systems without termios.h)?
(3) Is anyone interested in testing it?
I assign these changes to the public domain, since they're trivial.
--- main.c 2007-04-10 04:39:13.000000000 -0400
+++ main.c.new 2007-04-10 04:56:45.000000000 -0400
@@ -4613,11 +4613,9 @@
#endif /* USE_SYSV_UTMP */
#endif /* HAVE_UTMP */
- close(screen->respond); /* close explicitly to avoid race with slave
side */
-#ifdef ALLOWLOGGING
- if (screen->logging)
- CloseLog(screen);
-#endif
+ /* Flush pending data before releasing ownership, so nobody else can
+ grab it */
+ tcflush(screen->respond, TCOFLUSH);
if (am_slave < 0) {
/* restore ownership of tty and pty */
@@ -4625,6 +4623,16 @@
#if (defined(USE_PTY_DEVICE) && !defined(__sgi) && !defined(__hpux))
set_owner(ptydev, 0, 0, 0666U);
#endif
+
+ /* Close after releasing ownership to avoid race condition: other programs
+ grabbing it, and *then* having us release ownership... */
+
+ close(screen->respond); /* close explicitly to avoid race with slave
side */
+#ifdef ALLOWLOGGING
+ if (screen->logging)
+ CloseLog(screen);
+#endif
+
}
#ifdef NO_LEAKS
if (n == 0) {
--
Nathanael Nerode <[EMAIL PROTECTED]>
This space intentionally left blank.
--
To UNSUBSCRIBE, email to [EMAIL PROTECTED]
with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED]
