CWG2966 [basic.fundamental] Alignment and value representation of nullptr_t

[frederick-vs-ja]

Full name of submitter (unless configured in github; will be published with the issue): Jiang An

Reference (section label): [basic.fundamental]

Link to reflector thread (if any):

Issue description:

nullptr_t has several properties in C23 that is either not specified or different in C++.

WG14 N3220 7.21.2 p3 reads:

3. The size and alignment of nullptr_t is the same as for a pointer to character type. An object representation of the value nullptr is the same as the object representation of a null pointer value of type void*. An lvalue conversion of an object of type nullptr_t with such an object representation has the value nullptr; if the object representation is different, the behavior is undefined.³⁰³⁾

In C++ it's only required that sizeof(std::nullptr_t) == sizeof(void*). Do we want to specify that alignof(std::nullptr_t) == alignof(void*)? (The superficial wording difference from C doesn't matter due to [basic.compound] p7.)

The UB specified in that paragraph isn't present in C++. In C++, conversions from nullptr_t effectively touch no bit, so the value representation of a nullptr_t object, which is currently unspecified in C++, doesn't matter in most uses. However, the decision on value representation of nullptr_t can lead to different results via bit_cast, and there're implementation divergence (https://gcc.gnu.org/bugzilla/show_bug.cgi?id=117727).

Perhaps it's better to specifiy that the value representation (and the valid set of value bits) of nullptr_t is implementation-defined.

Suggested resolution:

Modify [basic.fundamental] p16 as indicated:

[...] sizeof(std​::​nullptr_t) shall be equal to sizeof(void*) and alignof(std::nullptr_t) shall be equal to alignof(void*). The value representation of std::nullptr_t and the non-empty set of value representations of std::nullptr_t that correspond to values are implementation-defined. [Note ?: An implementation can decide that the value representation of std::nullptr_t contains no bits. — end note]

[jensmaurer]

I agree about the alignment question, I think we should specify this to be the same as void*, too.

Since C++ opted to never read the bits of the value representation of std::nullptr_t for the lvalue-to-rvalue conversion, I don't think we need to talk about its value representation, and there is no need to require an implementation to document its choice ("implementation-defined"). The results from bit_cast are highly implementation-dependent anyway, so I'm not seeing a problem with the implementation divergence.

[jensmaurer]

CWG2966