[PATCH] Fix PR c/5163

[Jakub Jelinek]

Hi!

Assigning one extern declaration as abstract origin to another one certainly
cannot work.  This patch fixes the testcase (and bootstrapped with no make
check regressions), though I wonder whether the test shouldn't be extended
to && ! DECL_EXTERNAL (newdecl) too (or why is actually a DECL_EXTERNAL ever
marked DECL_INLINE).

2001-12-21  Jakub Jelinek  <jakub@redhat.com>

	* c-decl.c (duplicate_decls): Don't set abstract origin to origin
	if olddecl from different binding level is external.
	Fixes PR c/5163.

	* gcc.dg/20011221-1.c: New test.

--- gcc/c-decl.c.jj	Tue Dec 18 01:28:02 2001
+++ gcc/c-decl.c	Fri Dec 21 18:52:00 2001
@@ -2019,7 +2019,7 @@ duplicate_decls (newdecl, olddecl, diffe
 	  DECL_ARGUMENTS (newdecl) = DECL_ARGUMENTS (olddecl);
 	  if (DECL_INLINE (newdecl))
 	    DECL_ABSTRACT_ORIGIN (newdecl)
-	      = (different_binding_level
+	      = ((different_binding_level && ! DECL_EXTERNAL (olddecl))
 		 ? DECL_ORIGIN (olddecl)
 		 : DECL_ABSTRACT_ORIGIN (olddecl));
 	}
--- gcc/testsuite/gcc.dg/20011221-1.c.jj	Sat Dec 22 00:18:38 2001
+++ gcc/testsuite/gcc.dg/20011221-1.c	Sat Dec 22 00:24:10 2001
@@ -0,0 +1,14 @@
+/* Origin: PR c/5163 from aj@suse.de.  */
+/* { dg-do compile } */
+/* { dg-options "-O3 -g" } */
+
+extern int bar (int);
+
+int
+foo (void)
+{
+  extern int bar (int);
+  
+  return bar (5);
+}
+    

	Jakub