Give more-useful info on a fatal error (Bug#12328).

* doc/emacs/trouble.texi (Crashing): New section, documenting this.
* etc/NEWS: Document the change.
* src/alloc.c [ENABLE_CHECKING]: Do not include <execinfo.h>.
(die) [ENABLE_CHECKING]: Call fatal_error_backtrace instead
of doing the work ourselves.
* src/emacs.c (fatal_error_signal): Let fatal_error_backtrace
do most of the work.
(fatal_error_backtrace): New function, taken from the guts
of the old fatal_error_signal, but with a new option to output
a backtrace.
(shut_down_emacs) [!DOS_NT]: Use strsignal to give more-useful
info about the signal than just its number.
* src/lisp.h (fatal_error_backtrace, emacs_backtrace): New decls.
* src/sysdep.c: Include <execinfo.h>
(emacs_backtrace): New function, taken partly from the previous
code of the 'die' function.
(emacs_abort): Call fatal_error_backtrace rather than abort.

10 files changed, 111 insertions, 11 deletions

diff --git a/doc/emacs/ChangeLog b/doc/emacs/ChangeLog
index a4f9985ad85..78f1d2c8e3b 100644
--- a/doc/emacs/ChangeLog
+++ b/doc/emacs/ChangeLog
@@ -1,3 +1,8 @@
+2012-09-04  Paul Eggert  <eggert@cs.ucla.edu>
+
+        Give more-useful info on a fatal error (Bug#12328).
+        * trouble.texi (Crashing): New section, documenting this.
+
 2012-08-24  Michael Albinus  <michael.albinus@gmx.de>
 
         * cmdargs.texi (General Variables): Setting
diff --git a/doc/emacs/emacs.texi b/doc/emacs/emacs.texi
index 6357aebc6ff..192a9a2bb28 100644
--- a/doc/emacs/emacs.texi
+++ b/doc/emacs/emacs.texi
@@ -1136,6 +1136,7 @@ Dealing with Emacs Trouble
 * Screen Garbled::      Garbage on the screen.
 * Text Garbled::        Garbage in the text.
 * Memory Full::         How to cope when you run out of memory.
+* Crashing::            What Emacs does when it crashes.
 * After a Crash::       Recovering editing in an Emacs session that crashed.
 * Emergency Escape::    What to do if Emacs stops responding.
 
@@ -1320,7 +1321,7 @@ when you get it, not just free for the manufacturer.
 If you find GNU Emacs useful, please @strong{send a donation} to the
 Free Software Foundation to support our work.  Donations to the Free
 Software Foundation are tax deductible in the US.  If you use GNU Emacs
-at your workplace, please suggest that the company make a donation. 
+at your workplace, please suggest that the company make a donation.
 For more information on how you can help, see
 @url{http://www.gnu.org/help/help.html}.
 
diff --git a/doc/emacs/trouble.texi b/doc/emacs/trouble.texi
index 025185c583f..1806339e45d 100644
--- a/doc/emacs/trouble.texi
+++ b/doc/emacs/trouble.texi
@@ -149,6 +149,7 @@ Emacs.
 * Screen Garbled::        Garbage on the screen.
 * Text Garbled::          Garbage in the text.
 * Memory Full::           How to cope when you run out of memory.
+* Crashing::              What Emacs does when it crashes.
 * After a Crash::         Recovering editing in an Emacs session that crashed.
 * Emergency Escape::      What to do if Emacs stops responding.
 @end menu
@@ -277,6 +278,44 @@ editing in the same Emacs session.
 out of memory, because the buffer menu needs a fair amount of memory
 itself, and the reserve supply may not be enough.
 
+@node Crashing
+@subsection When Emacs Crashes
+
+  Emacs is not supposed to crash, but if it does, before it exits it
+reports some information about the crash to the standard error stream
+@code{stderr}.  This report may be useful to someone who later debugs
+the same version of Emacs on the same platform.  The format of this
+report depends on the platform, and some platforms support backtraces.
+Here is an example, generated on x86-64 GNU/Linux with version 2.15 of
+the GNU C Library:
+
+@example
+Fatal error 11: Segmentation fault
+Backtrace:
+emacs[0x5094e4]
+emacs[0x4ed3e6]
+emacs[0x4ed504]
+/lib64/libpthread.so.0[0x375220efe0]
+/lib64/libpthread.so.0(read+0xe)[0x375220e08e]
+emacs[0x509af6]
+emacs[0x5acc26]
+emacs[0x5adbfb]
+emacs[0x56566b]
+emacs[0x59bac3]
+emacs[0x565151]
+...
+@end example
+
+@noindent
+The number @samp{11} is the system signal number that corresponds to
+the problem, a segmentation fault here.  The hexadecimal program
+addresses can be useful in debugging sessions.  For example, the GDB
+command @samp{list *0x509af6} prints the source-code lines
+corresponding to the @samp{emacs[0x509af6]} entry in the backtrace.
+
+The three dots at the end indicate that Emacs suppressed further
+backtrace entries, in the interest of brevity.
+
 @node After a Crash
 @subsection Recovery After a Crash
 
diff --git a/etc/ChangeLog b/etc/ChangeLog
index f3f244c7e39..aa788275256 100644
--- a/etc/ChangeLog
+++ b/etc/ChangeLog
@@ -1,3 +1,8 @@
+2012-09-04  Paul Eggert  <eggert@cs.ucla.edu>
+
+        Give more-useful info on a fatal error (Bug#12328).
+        * NEWS: Document the change.
+
 2012-09-01  Paul Eggert  <eggert@cs.ucla.edu>
 
         Better seeds for (random).
diff --git a/etc/NEWS b/etc/NEWS
index e4a612f6a9e..9a38deef05c 100644
--- a/etc/NEWS
+++ b/etc/NEWS
@@ -97,6 +97,11 @@ machines.  Other functions that use this format, such as
 file-attributes and format-time-string, have been changed accordingly.
 Old-format time stamps are still accepted.
 
+** Emacs now generates backtraces on fatal errors.
+On encountering a fatal error, Emacs now outputs a textual description
+of the fatal signal, and a short backtrace on platforms like glibc
+that support backtraces.
+
 ** New functions `system-users', `system-groups' return lists of the user
 name, group names known to the system (where possible).
 
diff --git a/src/ChangeLog b/src/ChangeLog
index b2634c4fdc4..ea3b3a92e01 100644
--- a/src/ChangeLog
+++ b/src/ChangeLog
@@ -1,3 +1,22 @@
+2012-09-04  Paul Eggert  <eggert@cs.ucla.edu>
+
+        Give more-useful info on a fatal error (Bug#12328).
+        * alloc.c [ENABLE_CHECKING]: Do not include <execinfo.h>.
+        (die) [ENABLE_CHECKING]: Call fatal_error_backtrace instead
+        of doing the work ourselves.
+        * emacs.c (fatal_error_signal): Let fatal_error_backtrace
+        do most of the work.
+        (fatal_error_backtrace): New function, taken from the guts
+        of the old fatal_error_signal, but with a new option to output
+        a backtrace.
+        (shut_down_emacs) [!DOS_NT]: Use strsignal to give more-useful
+        info about the signal than just its number.
+        * lisp.h (fatal_error_backtrace, emacs_backtrace): New decls.
+        * sysdep.c: Include <execinfo.h>
+        (emacs_backtrace): New function, taken partly from the previous
+        code of the 'die' function.
+        (emacs_abort): Call fatal_error_backtrace rather than abort.
+
 2012-09-04  Stefan Monnier  <monnier@iro.umontreal.ca>
 
         * lread.c (readevalloop): Call internal-macroexpand-for-load to perform
diff --git a/src/alloc.c b/src/alloc.c
index 2a831cea62a..157d3a1d799 100644
--- a/src/alloc.c
+++ b/src/alloc.c
@@ -6685,21 +6685,14 @@ which_symbols (Lisp_Object obj, EMACS_INT find_max)
 
 #ifdef ENABLE_CHECKING
 
-# include <execinfo.h>
-
 bool suppress_checking;
 
 void
 die (const char *msg, const char *file, int line)
 {
-  enum { NPOINTERS_MAX = 500 };
-  void *buffer[NPOINTERS_MAX];
-  int npointers;
   fprintf (stderr, "\r\n%s:%d: Emacs fatal error: %s\r\n",
            file, line, msg);
-  npointers = backtrace (buffer, NPOINTERS_MAX);
-  backtrace_symbols_fd (buffer, npointers, STDERR_FILENO);
-  emacs_abort ();
+  fatal_error_backtrace (SIGABRT, INT_MAX);
 }
 #endif
 
diff --git a/src/emacs.c b/src/emacs.c
index b242cbf8844..18f249dc899 100644
--- a/src/emacs.c
+++ b/src/emacs.c
@@ -298,6 +298,14 @@ void
 fatal_error_signal (int sig)
 {
   SIGNAL_THREAD_CHECK (sig);
+  fatal_error_backtrace (sig, 10);
+}
+
+/* Report a fatal error due to signal SIG, output a backtrace of at
+   most BACKTRACE_LIMIT lines, and exit.  */
+_Noreturn void
+fatal_error_backtrace (int sig, int backtrace_limit)
+{
   fatal_error_code = sig;
   signal (sig, SIG_DFL);
 
@@ -312,6 +320,7 @@ fatal_error_signal (int sig)
         Fkill_emacs (make_number (sig));
 
       shut_down_emacs (sig, Qnil);
+      emacs_backtrace (backtrace_limit);
     }
 
   /* Signal the same code; this time it will really be fatal.
@@ -323,6 +332,9 @@ fatal_error_signal (int sig)
 #endif
 
   kill (getpid (), fatal_error_code);
+
+  /* This shouldn't be executed, but it prevents a warning.  */
+  exit (1);
 }
 
 #ifdef SIGDANGER
@@ -1992,7 +2004,7 @@ shut_down_emacs (int sig, Lisp_Object stuff)
       {
         reset_all_sys_modes ();
         if (sig && sig != SIGTERM)
-          fprintf (stderr, "Fatal error (%d)", sig);
+          fprintf (stderr, "Fatal error %d: %s", sig, strsignal (sig));
       }
   }
 #else
diff --git a/src/lisp.h b/src/lisp.h
index 0835795a26d..9ee9cd74b56 100644
--- a/src/lisp.h
+++ b/src/lisp.h
@@ -3266,6 +3266,7 @@ extern Lisp_Object Qfile_name_handler_alist;
 #ifdef FLOAT_CATCH_SIGILL
 extern void fatal_error_signal (int);
 #endif
+extern _Noreturn void fatal_error_backtrace (int, int);
 extern Lisp_Object Qkill_emacs;
 #if HAVE_SETLOCALE
 void fixup_locale (void);
@@ -3412,6 +3413,7 @@ extern int set_window_size (int, int, int);
 extern EMACS_INT get_random (void);
 extern void seed_random (void *, ptrdiff_t);
 extern void init_random (void);
+extern void emacs_backtrace (int);
 extern _Noreturn void emacs_abort (void) NO_INLINE;
 extern int emacs_open (const char *, int, int);
 extern int emacs_close (int);
diff --git a/src/sysdep.c b/src/sysdep.c
index 78164a8f02a..1f4de194c64 100644
--- a/src/sysdep.c
+++ b/src/sysdep.c
@@ -21,6 +21,7 @@ along with GNU Emacs.  If not, see <http://www.gnu.org/licenses/>.  */
 
 #define SYSTIME_INLINE EXTERN_INLINE
 
+#include <execinfo.h>
 #include <signal.h>
 #include <stdio.h>
 #include <setjmp.h>
@@ -1856,12 +1857,30 @@ snprintf (char *buf, size_t bufsize, char const *format, ...)
 }
 #endif
 
+/* If a backtrace is available, output the top lines of it to stderr.
+   Do not output more than BACKTRACE_LIMIT or BACKTRACE_LIMIT_MAX lines.
+   This function may be called from a signal handler, so it should
+   not invoke async-unsafe functions like malloc.  */
+void
+emacs_backtrace (int backtrace_limit)
+{
+  enum { BACKTRACE_LIMIT_MAX = 500 };
+  void *buffer[BACKTRACE_LIMIT_MAX + 1];
+  int bounded_limit = min (backtrace_limit, BACKTRACE_LIMIT_MAX);
+  int npointers = backtrace (buffer, bounded_limit + 1);
+  if (npointers)
+    ignore_value (write (STDERR_FILENO, "\nBacktrace:\n", 12));
+  backtrace_symbols_fd (buffer, bounded_limit, STDERR_FILENO);
+  if (bounded_limit < npointers)
+    ignore_value (write (STDERR_FILENO, "...\n", 4));
+}
+
 #ifndef HAVE_NTGUI
 /* Using emacs_abort lets GDB return from a breakpoint here.  */
 void
 emacs_abort (void)
 {
-  abort ();
+  fatal_error_backtrace (SIGABRT, 10);
 }
 #endif