[PATCH] ld: Properly override the IR definition

[Alan Modra]

On Tue, Jul 21, 2020 at 06:37:31AM -0700, H.J. Lu wrote:
> After all LTO symbols have been read, a new definition in real object
> overrides the previous definition in the IR object.  Before all LTO
> symbols have been read, a new non-weak definition overrides the weak
> definition in the IR object.
> 
> OK for master?

diff --git a/ld/plugin.c b/ld/plugin.c
index b455af6d67..ff9f00e452 100644
--- a/ld/plugin.c
+++ b/ld/plugin.c
@@ -1433,12 +1433,20 @@ plugin_notice (struct bfd_link_info *info,
 	 new value from a real BFD.  Weak symbols are not normally
 	 overridden by a new weak definition, and strong symbols
 	 will normally cause multiple definition errors.  Avoid
-	 this by making the symbol appear to be undefined.  */
-      else if (((h->type == bfd_link_hash_defweak
-		 || h->type == bfd_link_hash_defined)
-		&& is_ir_dummy_bfd (sym_bfd = h->u.def.section->owner))
-	       || (h->type == bfd_link_hash_common
-		   && is_ir_dummy_bfd (sym_bfd = h->u.c.p->section->owner)))
+	 this by making the symbol appear to be undefined.
+
+	 NB: After all LTO symbols have been read, a new definition in
+	 real object overrides the previous definition in the IR object.
+	 Before all LTO symbols have been read, a new non-weak definition
+	 overrides the weak definition in the IR object.  */
+      else if ((((h->type == bfd_link_hash_defweak
+		  || h->type == bfd_link_hash_defined)
+		 && is_ir_dummy_bfd (sym_bfd = h->u.def.section->owner))
+		|| (h->type == bfd_link_hash_common
+		    && is_ir_dummy_bfd (sym_bfd = h->u.c.p->section->owner)))
+	       && (info->lto_all_symbols_read
+		   || ((flags & BSF_WEAK) == 0
+		       && h->type == bfd_link_hash_defweak)))
 	{
 	  h->type = bfd_link_hash_undefweak;
 	  h->u.undef.abfd = sym_bfd;

Can't this just be the following, since non-weak will override weak
without any changes to h->type here?

      else if ((((h->type == bfd_link_hash_defweak
		  || h->type == bfd_link_hash_defined)
		 && is_ir_dummy_bfd (sym_bfd = h->u.def.section->owner))
		|| (h->type == bfd_link_hash_common
		    && is_ir_dummy_bfd (sym_bfd = h->u.c.p->section->owner)))
	       && info->lto_all_symbols_read)

OK with that simplification, if you have tested your patch with a gcc
regression test.

-- 
Alan Modra
Australia Development Lab, IBM