[patch] reread_symbols observer_notify_new_objfile - fix ARM unwinding after reread [Re: [patch+7.4] reread.exp 7.3->7.4 regression]

[Jan Kratochvil <jan dot kratochvil at redhat dot com>]

On Mon, 19 Dec 2011 11:30:54 +0100, Ulrich Weigand wrote:
> In addition, we should probably call observer_notify_new_objfile so that new
> tables can be built up for the re-read file ...

Implemented.

I have tested it really fixes the problem of ARM unwinding after reread on
stripped GDB binary.

Before this fix:

./gdb -nx ./test-gdb-strip -ex 'b gdb_do_one_event' -ex r -ex bt
Breakpoint 1, 0x00372154 in gdb_do_one_event ()
#0  0x00372154 in gdb_do_one_event ()
#1  0x003722c8 in start_event_loop ()
#2  0x003741bc in cli_command_loop ()
#3  0x0036b9e0 in current_interp_command_loop ()
#4  0x0007a054 in ?? ()
#5  0x0036af08 in catch_errors ()
#6  0x0007b58c in ?? ()
#7  0x0036af08 in catch_errors ()
#8  0x0007b5c4 in gdb_main ()
#9  0x00079cd8 in main ()
(gdb) r
The program being debugged has been started already.
Start it from the beginning? (y or n) y
`/opt/jkratoch/redhat/archer-git/gdb/test-gdb-strip' has changed; re-reading symbols.
(no debugging symbols found)
[...]
Breakpoint 1, 0x00372154 in gdb_do_one_event ()
(gdb) bt
#0  0x00372154 in gdb_do_one_event ()
#1  0x003722c8 in start_event_loop ()
#2  0x003741bc in cli_command_loop ()
#3  0x0036b9e0 in current_interp_command_loop ()
#4  0x0007a054 in ?? ()
#5  0x0007a054 in ?? ()
Backtrace stopped: previous frame identical to this frame (corrupt stack?)

After this fix both backtraces are the same.

Testcase is not provided as I do not have some small enough ARM testcase which
unwinds with .ARM.exidx but does not unwind without it.

No regressions on {x86_64,x86_64-m32,i686}-fedora16-linux-gnu and on
arm-fedora13beta-linux-gnueabi.


Thanks,
Jan


gdb/
2011-12-20  Jan Kratochvil  <jan.kratochvil@redhat.com>

	* symfile.c (objfilep): New typedef and new DEF_VEC_P.
	(reread_symbols): Remove variable reread_one, new variables
	new_objfiles, all_cleanups and ix.  Use new_objfiles instead of
	reread_one.  Push changed objfiles to new_objfiles, call
	observer_notify_new_objfile for them later.

--- a/gdb/symfile.c
+++ b/gdb/symfile.c
@@ -2371,15 +2371,22 @@ add_symbol_file_command (char *args, int from_tty)
 }
 
 
+typedef struct objfile *objfilep;
+
+DEF_VEC_P (objfilep);
+
 /* Re-read symbols if a symbol-file has changed.  */
 void
 reread_symbols (void)
 {
   struct objfile *objfile;
   long new_modtime;
-  int reread_one = 0;
   struct stat new_statbuf;
   int res;
+  VEC (objfilep) *new_objfiles = NULL;
+  struct cleanup *all_cleanups;
+
+  all_cleanups = make_cleanup (VEC_cleanup (objfilep), &new_objfiles);
 
   /* With the addition of shared libraries, this should be modified,
      the load time should be saved in the partial symbol tables, since
@@ -2594,21 +2601,33 @@ reread_symbols (void)
 	     and now, we *want* this to be out of date, so don't call stat
 	     again now.  */
 	  objfile->mtime = new_modtime;
-	  reread_one = 1;
 	  init_entry_point_info (objfile);
+
+	  VEC_safe_push (objfilep, new_objfiles, objfile);
 	}
     }
 
-  if (reread_one)
+  if (new_objfiles)
     {
+      int ix;
+
       /* Notify objfiles that we've modified objfile sections.  */
       objfiles_changed ();
 
       clear_symtab_users (0);
+
+      /* clear_objfile_data for each objfile was called before freeing it and
+	 observer_notify_new_objfile (NULL) has been called by
+	 clear_symtab_users above.  Notify the new files now.  */
+      for (ix = 0; VEC_iterate (objfilep, new_objfiles, ix, objfile); ix++)
+	observer_notify_new_objfile (objfile);
+
       /* At least one objfile has changed, so we can consider that
          the executable we're debugging has changed too.  */
       observer_notify_executable_changed ();
     }
+
+  do_cleanups (all_cleanups);
 }