cp_parser_simple_type_specifier tries a variety of different things that
might qualify as a user-defined type: an actual type-name, a constrained
auto, a CTAD placeholder. In a context where a type-specifier is optional,
this is all tentative. With -std=c++14 -fconcepts, we try type-name and
constrained auto in sub-tentative parses, and when we run out of things to
try we haven't found anything but also haven't failed the outer tentative
parse, so parse_definitely succeeds, discarding the nested-name-specifier.
Fixed by failing if we didn't find anything.
I said in r14-3203 that we should disable this combination of flags if
further problems arise, but this seems like a more general problem that only
happened to occur with just this combination of flags. So it lives on.
PR c++/116071
gcc/cp/ChangeLog:
* parser.cc (cp_parser_simple_type_specifier): Call
cp_parser_simulate_error if nothing worked.
gcc/testsuite/ChangeLog:
* g++.dg/parse/pr116071.C: New test.
}
/* If it didn't work out, we don't have a TYPE. */
- if ((flags & CP_PARSER_FLAGS_OPTIONAL)
- && !cp_parser_parse_definitely (parser))
- type = NULL_TREE;
+ if (flags & CP_PARSER_FLAGS_OPTIONAL)
+ {
+ if (!type)
+ cp_parser_simulate_error (parser);
+ if (!cp_parser_parse_definitely (parser))
+ type = NULL_TREE;
+ }
/* Keep track of all name-lookups performed in class scopes. */
if (type
--- /dev/null
+// PR c++/116071
+// { dg-options "-std=c++14 -fconcepts" }
+
+template<class T> struct S { ~S(); };
+template<class T> S<T>::~S() { }
+
+template<typename MemPtr>
+struct result_of;
+
+template<typename Res, typename Class>
+struct result_of<Res Class::*>
+{
+ using type = void;
+};
+
+struct thread {
+ void join() { };
+};