Re: rawhide's gdb segfaults, w/patch

[Jan Kratochvil <jan.kratochvil@redhat.com>]

Hi Jim,

thanks but your patch needed a second part.  It is a regression from:
http://sourceware.org/ml/gdb-cvs/2008-04/msg00136.html

On Sun, 28 Dec 2008 18:30:14 +0100, Jim Meyering wrote:
> Here's an untested and quite possibly-wrong patch.
> I.e., if the warning should be given even when "thisfun" is NULL,
> it would have to be different.

I find it right as gdbarch_return_value_ftype even has a comment:
FUNCTYPE may be NULL in which case the return convention is computed based
only on VALTYPE.

This new argument and functype|func_type is only used in sh-tdep.c
http://sourceware.org/ml/gdb-patches/2008-04/msg00277.html
and even there it can be safely NULL.

This new argument and functype|func_type was added by
http://sourceware.org/ml/gdb-patches/2008-04/msg00276.html


I find just questionable whether the new testcase should be in
gdb.base/nodebug.exp or gdb.base/return*.exp but may be any way is OK.


Thanks,
Jan


gdb/
2009-01-09  Jim Meyering  <meyering@redhat.com>
	    Jan Kratochvil  <jan.kratochvil@redhat.com>

	Avoid NULL dereference.
	* stack.c (return_command): Guard use of SYMBOL_TYPE (thisfun).
	New variable func_type.

gdb/testsuite/
2009-01-09  Jan Kratochvil  <jan.kratochvil@redhat.com>

	* gdb.base/nodebug.exp (return from function with no debug info): New.

--- gdb/stack.c	3 Jan 2009 05:57:53 -0000	1.183
+++ gdb/stack.c	9 Jan 2009 20:30:25 -0000
@@ -1823,7 +1823,8 @@ return_command (char *retval_exp, int fr
            is discarded, side effects such as "return i++" still
            occur.  */
 	return_value = NULL;
-      else if (using_struct_return (SYMBOL_TYPE (thisfun), return_type))
+      else if (thisfun != NULL
+	       && using_struct_return (SYMBOL_TYPE (thisfun), return_type))
 	{
 	  query_prefix = "\
 The location at which to store the function's return value is unknown.\n\
@@ -1856,10 +1857,12 @@ If you continue, the return value that y
     {
       struct type *return_type = value_type (return_value);
       struct gdbarch *gdbarch = get_regcache_arch (get_current_regcache ());
-      gdb_assert (gdbarch_return_value (gdbarch, SYMBOL_TYPE (thisfun),
-      					return_type, NULL, NULL, NULL)
+      struct type *func_type = thisfun == NULL ? NULL : SYMBOL_TYPE (thisfun);
+
+      gdb_assert (gdbarch_return_value (gdbarch, func_type, return_type, NULL,
+					NULL, NULL)
 		  == RETURN_VALUE_REGISTER_CONVENTION);
-      gdbarch_return_value (gdbarch, SYMBOL_TYPE (thisfun), return_type,
+      gdbarch_return_value (gdbarch, func_type, return_type,
 			    get_current_regcache (), NULL /*read*/,
 			    value_contents (return_value) /*write*/);
     }
--- gdb/testsuite/gdb.base/nodebug.exp	3 Jan 2009 05:58:03 -0000	1.10
+++ gdb/testsuite/gdb.base/nodebug.exp	9 Jan 2009 20:30:26 -0000
@@ -215,5 +215,12 @@ if [runto inner] then {
     if [runto middle] then {
 	gdb_test "backtrace 10" "#0.*middle.*#1.*top.*#2.*main.*" \
 	    "backtrace from middle in nodebug.exp"
+
+	# Test return from a function with no debug info (symbol; still it may
+	# have a minimal-symbol).  In gdb.base/return*.exp we would need to
+	# build a separate executable with no "debug" option.
+	gdb_test "return 0" "#0 .* top \\(.*"				     \
+		 "return from function with no debug info"		     \
+		 "Make selected stack frame return now\\? \\(y or n\\) " "y"
     }
 }