https://gcc.gnu.org/bugzilla/show_bug.cgi?id=127456
Jakub Jelinek <jakub at gcc dot gnu.org> changed:
What |Removed |Added
----------------------------------------------------------------------------
CC| |jakub at gcc dot gnu.org,
| |jason at gcc dot gnu.org
--- Comment #2 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
Guess the problem is that we do:
/* if (auto v = x<=>y, v == 0); else return v; */
if (TREE_CODE (comp) == SPACESHIP_EXPR)
TREE_TYPE (comp) = rettype;
else
comp = build_static_cast (input_location, rettype, comp,
complain);
info.check (comp);
if (defining)
{
tree var = create_temporary_var (rettype);
DECL_NAME (var) = get_identifier ("retval");
pushdecl (var);
cp_finish_decl (var, comp, false, NULL_TREE, flags);
comp = retval = var;
}
eq = build_new_op (info.loc, EQ_EXPR, flags, comp,
integer_zero_node, NULL_TREE, NULL_TREE,
NULL, complain);
so don't actually implement what the comment says, but instead emit
if (R v = x<=>y; v == 0); else return v;
where R is the return type of the defaulted operator<=>.
Now, if R is one of std::strong_ordering, std::weak_ordering or
std::partial_ordering, guess it makes no difference. But when it is some
arbitrary other type, in this case const std::strong_ordering &, already the
TREE_TYPE (comp) = rettype; doesn't look right. So wonder if the above code
except for the final eq = build_new_op shouldn't be guarded on if (cat_tag_for
(rettype) != cc_last) and otherwise it shouldn't implement what the comment
says (except , vs. ;),
so actually perform auto deduction.