http://gcc.gnu.org/bugzilla/show_bug.cgi?id=51573
Richard Guenther <rguenth at gcc dot gnu.org> changed:
What |Removed |Added
----------------------------------------------------------------------------
CC| |dnovillo at gcc dot
| |gnu.org, jason at gcc dot
| |gnu.org
Blocks| |48508, 48437
--- Comment #5 from Richard Guenther <rguenth at gcc dot gnu.org> 2011-12-16
10:21:16 UTC ---
Because of the way C++ treats things (look at PR48508) it is impossible to
move fixing PRs 48508 and 48437 to the point where we handle decl merging.
Thus, while
Index: gcc/lto-symtab.c
===================================================================
--- gcc/lto-symtab.c (revision 182398)
+++ gcc/lto-symtab.c (working copy)
@@ -785,6 +785,12 @@ lto_symtab_prevailing_decl (tree decl)
if (TREE_CODE (decl) == FUNCTION_DECL && DECL_ABSTRACT (decl))
return decl;
+ /* Extern decls with function context should not be merged, they
+ appear in BLOCK_VARS. */
+ if (DECL_EXTERNAL (decl)
+ && decl_function_context (decl))
+ return decl;
+
/* Ensure DECL_ASSEMBLER_NAME will not set assembler name. */
gcc_assert (DECL_ASSEMBLER_NAME_SET_P (decl));
Index: gcc/lto/lto.c
===================================================================
--- gcc/lto/lto.c (revision 182398)
+++ gcc/lto/lto.c (working copy)
@@ -867,9 +867,13 @@ uniquify_nodes (struct data_in *data_in,
if (t == NULL_TREE)
continue;
- if (TREE_CODE (t) == VAR_DECL)
+ if (TREE_CODE (t) == VAR_DECL
+ && (!DECL_EXTERNAL (t)
+ || !decl_function_context (t)))
lto_register_var_decl_in_symtab (data_in, t);
- else if (TREE_CODE (t) == FUNCTION_DECL && !DECL_BUILT_IN (t))
+ else if (TREE_CODE (t) == FUNCTION_DECL && !DECL_BUILT_IN (t)
+ && (!DECL_EXTERNAL (t)
+ || !decl_function_context (t)))
lto_register_function_decl_in_symtab (data_in, t);
else if (TYPE_P (t) && !TYPE_CANONICAL (t))
TYPE_CANONICAL (t) = gimple_register_canonical_type (t);
works great for C it does not work for C++ at all because in
static void
bar (void)
{
extern void foo (int);
foo (0);
}
the function-local extern declaration of 'foo' has a toplevel DECL_CONTEXT.
Thus the only way I see to fix this bug is to introduce either a new
streaming hook or add an argument to the stream_write_tree that
indicates whether the actual tree should be streamed by reference,
not affecting its siblings.
