[PATCH] fix error sending signal to dead process

"Patrick Monnerat" <[email protected]> Thu, 5 Apr 2012 14:45:49 +0200

Newsgroups	gmane.comp.debugging.insight
Message-ID	<AB5E58B87EB73C46A38073D8F459F113D9F882@dataspheresrv01>

 
Using insight-7.4.50 (cvs snapshot 20120403), continuing or stepping
after a fault (i.e. segfault) occurred results in sending a signal to
the dead process.
This is due to hook gdbtk_annotate_signal() trying to get thread_info
structure with a null pid.
Continuing in the same insight session (by re-running the program)
finally ends in insight segfaulting.
 
The attached patch fixes this problem. It simply do not pass signal to
tcl if the pid is null.
 
Cheers,
Patrick
 
P.S.: the bug report system linked from your web site
(http://sources.redhat.com/cgi-bin/gnatsweb.pl?database=insight&user=gue
st&password=guest&cmd=login) is down (HTTP error 500).

insight-7.4.50-sig2dead.patch (application/octet-stream, 912 B)

diff -Naur insight-7.4.50.orig/gdb/gdbtk/generic/gdbtk-hooks.c insight-7.4.50.new/gdb/gdbtk/generic/gdbtk-hooks.c
--- insight-7.4.50.orig/gdb/gdbtk/generic/gdbtk-hooks.c	2012-03-28 15:09:12.000000000 +0200
+++ insight-7.4.50.new/gdb/gdbtk/generic/gdbtk-hooks.c	2012-04-05 12:44:20.284306992 +0200
@@ -804,7 +804,7 @@
 gdbtk_annotate_signal (void)
 {
   char *buf;
-  struct thread_info *tp = inferior_thread ();
+  struct thread_info *tp;
 
   /* Inform gui that the target has stopped. This is
      a necessary stop button evil. We don't want signal notification
@@ -812,6 +812,11 @@
      timeout. */
   Tcl_Eval (gdbtk_interp, "gdbtk_stop_idle_callback");
 
+  if (ptid_equal(inferior_ptid, null_ptid))
+    return;
+
+  tp = inferior_thread ();
+
   buf = xstrprintf ("gdbtk_signal %s {%s}",
 	     target_signal_to_name (tp->suspend.stop_signal),
 	     target_signal_to_string (tp->suspend.stop_signal));