https://gcc.gnu.org/bugzilla/show_bug.cgi?id=88865
Bug ID: 88865
Summary: [[no_unique_address]] leads to sizeof(T) == 0, which
cannot be
Product: gcc
Version: unknown
Status: UNCONFIRMED
Severity: normal
Priority: P3
Component: c++
Assignee: unassigned at gcc dot gnu.org
Reporter: redbeard0531 at gmail dot com
Target Milestone: ---
https://godbolt.org/z/rJT5X9
struct B {};
struct A {
[[no_unique_address]] B a;
[[no_unique_address]] B b;
[[no_unique_address]] B c;
[[no_unique_address]] B d;
};
int f() {
return sizeof(A);
}
f():
push rbp
mov rbp, rsp
mov eax, 0
pop rbp
ret
In addition to the major issue that sizeof(A) must not be 0, it also must not
be 1 either. It must be (at least) 4.
http://eel.is/c++draft/intro.object#9.sentence-2 is very clear that
[[no_unique_address]] (which clauses 7 and 8 define to mean a "subobject of
zero size") only allows members of *different types* to overlap. a,b,c, and d
are all distinct objects of the same type B, and therefore must have distinct
addresses.
> Two objects with overlapping lifetimes that are not bit-fields may have the
> same address if one is nested within the other, or if at least one is a
> subobject of zero size and they are of different types; otherwise, they have
> distinct addresses and occupy disjoint bytes of storage.
https://godbolt.org/z/160XGN shows that some parts of gcc seem to understand
this rule, some something very strange must be going on.
