[libc++] Should we backport `hash<nullptr_t>` and conditionally enabled `hash` from P0513R0?

[frederick-vs-ja]

P0513R0 resolved LWG2543, LWG2791, LWG2809, and LWG2817. Among these LWG issues, LWG2543 (conditionally enabled hash, affecting hash<unique_ptr>) and LWG2817 (hash<nullptr_t>) affect pre-C++17 utilities.

Currently there's implementation divergence:

• libc++ doesn't backport LWG2543 or LWG2817.
• libstdc++ backports LWG2543 but not LWG2817.
• MSVC STL backports both LWG2543 or LWG2817.

Should we backport LWG2543 and/or LWG2817?

[philnik777]

I'm not sure I understand. hash<nullptr_t> is provided from C++17 onwards. hash<unique_ptr<T, D>> is only enabled if unique_ptr<T, D>::pointer is enabled.

[frederick-vs-ja]

Oh, sorry, the original description was wrong. Now fixed.

hash<nullptr_t> is provided from C++17 onwards.

Yeah. But it was originally requested via LWG2817. I was asking whether we should backport it to old modes as resolution of usual LWG issues.

hash<unique_ptr<T, D>> is only enabled if unique_ptr<T, D>::pointer is enabled.

Currently, the conditionally enabling mechanism for hash<unique_ptr<T, D>> is only available since C++17 in libc++. I wonder whether we should make this available since C++11 (or even C++03 as we've already provided hash<unique_ptr<T, D>> in C++03 mode).

[philnik777]

Ah, I didn't see that __enable_hash_helper is defined differently based on version. That's quite surprising. Is there a good reason to backport them? AFAIK MSVC's C++14 mode isn't really C++14, so I don't think that really counts.

[frederick-vs-ja]

For LWG2543 (conditionally enable hash specializations), I think "keeping behavior consistent between standard modes" is a good reason for backport. I'm not sure whether it's better to backport LWG2817 (hash<nullptr_t>), but we've done similar thing for hash<filesystem::path> (LWG3657).

[ldionne]

Our general policy is to backport LWG issues as they're seen as bugs in the original standard version that introduced whatever component we're fixing.

So based on that reasoning, I would backport both LWG2543 and LWG2817.