Patch: PR 17408/17409

[Dale Johannesen]

On Sep 11, 2004, at 12:45 PM, Joseph S. Myers wrote:

> On Sat, 11 Sep 2004, Dale Johannesen wrote:
>
>> Having now read what the standard requires (which seems poorly chosen,
>> but it's Too Late Now), getting all cases right seems messy.  Setting
>> DECL_EXTERN on an initialized object obviously leads to trouble, and
>> is contrary to the meaning of that flag.  I'm now thinking in terms of
>> adding a bit, which tells whether an initialized object was defined 
>> using
>> 'extern' or not.  The code that resolves
>
> For the present bugs, there is probably a reasonably straightforward 
> local
> fix.  DECL_EXTERNAL should end up getting reset in start_decl.
> Investigate what's going wrong.

Right you are.  The Static bit was getting clobbered by pushdecl, 
apparently the
Extern bit confused it.   The following fixes pr15360-1.c and the 
examples in both
newer PRs.   I'll start a test run,  OK if that looks all right?

Index: c-decl.c
===================================================================
RCS file: /cvs/gcc/gcc/gcc/c-decl.c,v
retrieving revision 1.579
diff -u -d -b -w -p -r1.579 c-decl.c
--- c-decl.c    10 Sep 2004 23:56:24 -0000      1.579
+++ c-decl.c    13 Sep 2004 17:44:37 -0000
@@ -2981,8 +2981,11 @@ start_decl (struct c_declarator *declara
       TEM may equal DECL or it may be a previous decl of the same name. 
  */
    tem = pushdecl (decl);

-  if (initialized)
+  if (initialized && DECL_EXTERNAL (tem))
+    {
      DECL_EXTERNAL (tem) = 0;
+      TREE_STATIC (tem) = 1;
+    }

    return tem;
  }