Re: [RFA/dwarf-2] Fix for the null record problem

[Joel Brobecker <brobecker@gnat.com>]

[-- Attachment #1: Type: text/plain, Size: 823 bytes --]

Hello Jim,

Good catch about the is_declaration() case.

> Joel, how about dropping the DW_AT_specification test from
> die_is_declaration, and then using it in the new test in your patch?

Based on Jason's answer, it looks like the DW_AT_specification check
was indeed necessary. I have added a comment in that respect in the
little cleanup patch you suggested earlier. See:

    http://sources.redhat.com/ml/gdb-patches/2004-04/msg00325.html

Here a new version of the initial patch that follows your suggestion
of using die_is_declaration(). It's indeed more correct, and even
easier to read.

2004-04-15  Joel Brobecker  <brobecker@gnat.com>

        * dwarf2read.c (read_structure_scope): Identify stub types
        using die_is_declaration() only.

Tested on x86-linux, no regression.
OK to apply?

Thanks,
-- 
Joel

[-- Attachment #2: dwarf2read.c.null-record.diff --]
[-- Type: text/plain, Size: 764 bytes --]

--- dwarf2read.c.cleanup	2004-04-15 13:23:39.000000000 -0700
+++ dwarf2read.c	2004-04-15 13:36:10.000000000 -0700
@@ -3144,6 +3144,9 @@ read_structure_type (struct die_info *di
       TYPE_LENGTH (type) = 0;
     }
 
+  if (die_is_declaration (die, cu))
+    TYPE_FLAGS (type) |= TYPE_FLAG_STUB;
+
   /* We need to add the type field to the die immediately so we don't
      infinitely recurse when dealing with pointers to the structure
      type within the structure itself. */
@@ -3240,11 +3243,6 @@ read_structure_type (struct die_info *di
 
       do_cleanups (back_to);
     }
-  else
-    {
-      /* No children, must be stub. */
-      TYPE_FLAGS (type) |= TYPE_FLAG_STUB;
-    }
 
   processing_current_prefix = previous_prefix;
   if (back_to != NULL)