While updating the C++ DR table, I noticed that one of the new DRs, DR 2413,
is trivial:
---
The “Down with typename!” paper, P0634R3, overlooked the case of a
conversion-type-id
in a conversion-function-id:
template<class T> struct S {
operator typename T::X(); // typename is not helpful here.
};
This context should be added to the list of contexts in which a qualified-id is
assumed
to name a type.
---
It's still in "drafting", but it looks like a shoo-in to me.
Bootstrapped/regtested on x86_64-linux, ok for trunk?
2019-08-05 Marek Polacek <pola...@redhat.com>
DR 2413 - typename in conversion-function-ids.
* parser.c (cp_parser_conversion_type_id): Call
cp_parser_type_specifier_seq with CP_PARSER_FLAGS_TYPENAME_OPTIONAL
instead of CP_PARSER_FLAGS_NONE.
* g++.dg/cpp2a/typename17.C: New test.
diff --git gcc/cp/parser.c gcc/cp/parser.c
index 86857e7d468..83e6d24a9c1 100644
--- gcc/cp/parser.c
+++ gcc/cp/parser.c
@@ -14844,8 +14844,9 @@ cp_parser_conversion_type_id (cp_parser* parser)
parser->type_definition_forbidden_message
= G_("types may not be defined in a conversion-type-id");
- /* Parse the type-specifiers. */
- cp_parser_type_specifier_seq (parser, CP_PARSER_FLAGS_NONE,
+ /* Parse the type-specifiers. DR 2413 clarifies that `typename' is
+ optional in conversion-type-id. */
+ cp_parser_type_specifier_seq (parser, CP_PARSER_FLAGS_TYPENAME_OPTIONAL,
/*is_declaration=*/false,
/*is_trailing_return=*/false,
&type_specifiers);
diff --git gcc/testsuite/g++.dg/cpp2a/typename17.C
gcc/testsuite/g++.dg/cpp2a/typename17.C
new file mode 100644
index 00000000000..bf534f1717f
--- /dev/null
+++ gcc/testsuite/g++.dg/cpp2a/typename17.C
@@ -0,0 +1,6 @@
+// DR 2413 - typename in conversion-function-ids.
+// { dg-do compile { target c++2a } }
+
+template<class T> struct S {
+ operator T::X();
+};
