Hi,
in this [4.8/4.9] diagnostic regression the gcc_assert in
check_member_templates trips:
/* The parser rejects any use of virtual in a function template. */
gcc_assert (!(TREE_CODE (decl) == FUNCTION_DECL
&& DECL_VIRTUAL_P (decl)));
the ultimate reason being that in r187587 we inadvertently (I suppose,
because the change has no relation to the rest of the commit and isn't
explained) changed cp_parser_function_specifier_opt to always set
ds_virtual in decl_specs, upon error too. Thus the below first hunk
simply reverts that change and is enough to fix the primary bug.
Then, in mainline only, we have a variant of the issue for our implicit
function templates extension, which requires a little more work, because
when we parse 'virtual' we don't know yet that all the function
parameters will be auto. To handle that I'm adding a check in
finish_fully_implicit_template.
Tested x86_64-linux.
Thanks,
Paolo.
/////////////////////////////
/cp
2013-10-02 Paolo Carlini <paolo.carl...@oracle.com>
PR c++/58565
* semantics.c (potential_constant_expression_1): Handle LABEL_EXPR.
/testsuite
2013-10-02 Paolo Carlini <paolo.carl...@oracle.com>
PR c++/58565
* g++.dg/parse/crash64.C: New.
Index: cp/semantics.c
===================================================================
--- cp/semantics.c (revision 203101)
+++ cp/semantics.c (working copy)
@@ -8422,6 +8422,7 @@ potential_constant_expression_1 (tree t, bool want
case OVERLOAD:
case TEMPLATE_ID_EXPR:
case LABEL_DECL:
+ case LABEL_EXPR:
case CONST_DECL:
case SIZEOF_EXPR:
case ALIGNOF_EXPR:
Index: testsuite/g++.dg/parse/crash64.C
===================================================================
--- testsuite/g++.dg/parse/crash64.C (revision 0)
+++ testsuite/g++.dg/parse/crash64.C (working copy)
@@ -0,0 +1,7 @@
+// PR c++/58565
+// { dg-options "" }
+
+void foo()
+{
+ int i = ({ L: ; }); // { dg-error "void value not ignored" }
+}
