https://gcc.gnu.org/bugzilla/show_bug.cgi?id=69694
Patrick Palka <ppalka at gcc dot gnu.org> changed:
What |Removed |Added
----------------------------------------------------------------------------
Status|UNCONFIRMED |NEW
Last reconfirmed| |2016-02-11
CC| |ppalka at gcc dot gnu.org
Known to work| |4.9.1
Ever confirmed|0 |1
--- Comment #1 from Patrick Palka <ppalka at gcc dot gnu.org> ---
Reduced test case:
template<typename... Ts> using void_t = void;
extern void *declval_void;
template<typename, typename> struct Fun { };
template<typename Desc>
struct Fun<Desc, void_t<decltype (declval_void = Desc::name)>>
{
void fun();
};
struct Tag { static constexpr void* name = 0; };
template<typename> void a()
{
Fun<Tag, void>{}.fun();
}
g++ 4.9 compiles this successfully, g++ 5.1 doesn't, neither does trunk.
The problem is that type_dependent_expression_p considers the MODOP_EXPR within
the decltype to be dependent even after substitution because its TREE_TYPE is
NULL. Since the decltype is considered dependent even after substitution, it
never gets resolved to a concrete type. This stops void_t<decltype (...)> from
getting resolved to "void". And ultimately it makes partial-specialization
selection not work properly.
So build_x_modify_expr probably should not build a MODOP_EXPR with NULL
TREE_TYPE if the operands are not dependent.
