LWG4530 Unclear specification for formatter<charT[N], charT>

[JeffGarland]

https://cplusplus.github.io/LWG/issue4530

LWG would like LEWG to look at this issue if it should be treated as an array or a string if it is explicitly instantiated. More details in the issue.

@inbal2l

[jwakely]

For additional context from the LWG telecon, the formatter<C[N], C> specialization is not used when formatting a string literal; that uses the formatter<const C*, C> specialization. The char[N] specialization is used explicitly in the code example, with a wrapper that preserved the array type (instead of letting it decay to a pointer) and then explicitly formats using the formatter<char[N], char> specialization. This doesn't happen "by mistake" if a user is just trying to format strings.

The GCC interpretation is that if you are explicitly formatting an array, then that should not be assumed to be a NTBS, unlike const char* which is a NTBS. So GCC formats all N characters in the array. The alternative (used by {fmt} and other std::lib implementations) appears to make an unsafe assumption about the array.

If the user wants to treat the array as an NTBS, they can just use the formatter<const C*, C> specialization, which has exactly that behaviour.

But we have implementation divergence, so need to decide which interpretation we want, and then specify the behaviour of the char[N] specialization.