https://gcc.gnu.org/bugzilla/show_bug.cgi?id=118292
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,
| |redi at gcc dot gnu.org
--- Comment #1 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
The reason why it fails and doesn't fail with -Dno_unique_address=
is that in the latter case when checking if B has
"has all non-static data members and bit-fields in the class and its base
classes first declared in the same class"
2110 /* ...has all non-static data members and bit-fields in the
class
2111 and its base classes first declared in the same class.
*/
2112 for (basefield = TYPE_FIELDS (basetype); basefield;
2113 basefield = DECL_CHAIN (basefield))
2114 if (TREE_CODE (basefield) == FIELD_DECL
2115 && !(DECL_FIELD_IS_BASE (basefield)
2116 && is_empty_field (basefield)))
2117 {
2118 if (field)
2119 CLASSTYPE_NON_STD_LAYOUT (t) = 1;
2120 else
2121 field = basefield;
2122 break;
2123 }
sees basefield with A type and is_empty_field is false because of the e member
in A,
so B is marked as non-std-layout.
In the no_unique_address case, is_empty_field is true for the basefield with A
type,
I think there was the intent that classes with [[no_unique_address]] empty
non-static data member at the beginning are treated ABI-wise the same as if it
was a base class instead. And if it is instead
struct A : E1 {};
then both GCC and clang agree B is a standard layout class.
