[Jason approved this patch off-line, and I am committing now that tests have successfully run.]

This is a bug I found while investigating early dwarf generation, but that is broken mainline as well.

For the following code:

namespace S
{
  int i=777;
  int
  f()
  {
    int i = 42;
    {
      extern int i;
      return i;
    }
  }
}

...we end up emitting an extern declaration for "i" twice, once in the innermost lexical scope, and one in the namespace. The one in the namespace is unnecessary (and incorrect), although it is really not impacting any testcases since the second one is the one that takes precedence within the innermost scope.

With this patch, we get one less DIE for this scenario on mainline, while fixing other problems in early dwarf. Double yay!

Aldy
commit 81de7d658e94160426b38e6c43c09a484fb530ba
Author: Aldy Hernandez <al...@redhat.com>
Date:   Fri Oct 24 18:00:49 2014 -0600

        * dwarf2out.c (declare_in_namespace): Only emit external
        declarations in the local scope once.

diff --git a/gcc/dwarf2out.c b/gcc/dwarf2out.c
index a87f9c0..3bce20f 100644
--- a/gcc/dwarf2out.c
+++ b/gcc/dwarf2out.c
@@ -20476,6 +20476,26 @@ declare_in_namespace (tree thing, dw_die_ref 
context_die)
   if (debug_info_level <= DINFO_LEVEL_TERSE)
     return context_die;
 
+  /* External declarations in the local scope only need to be emitted
+     once, not once in the namespace and once in the scope.
+
+     This avoids declaring the `extern' below in the
+     namespace DIE as well as in the innermost scope:
+
+          namespace S
+         {
+            int i=5;
+            int foo()
+           {
+              int i=8;
+              extern int i;
+             return i;
+           }
+          }
+  */
+  if (DECL_P (thing) && DECL_EXTERNAL (thing) && local_scope_p (context_die))
+    return context_die;
+
   /* If this decl is from an inlined function, then don't try to emit it in its
      namespace, as we will get confused.  It would have already been emitted
      when the abstract instance of the inline function was emitted anyways.  */

Reply via email to