[expr.const] Definition domain fixes for "usable in constant expressions"

[hubert-reinterpretcast]

• Fix index entries to identify the domains of the definitions
• Fix wording error that restricted the domain of the definition for variables

[eisenwave]

It's not obvious that the new restriction that $V$ is constant-initialized is editorial. What does this do? Also, doesn't this conclude from the first bullet that $V$ is constexpr or something else?

[hubert-reinterpretcast]

It's not obvious that the new restriction that V is constant-initialized is editorial.

That $V$ is constant-initialized is, in the strict reading of the old wording, a prerequisite for the definition to apply.

Also, doesn't this conclude from the first bullet that V is constexpr or something else?

It does not, because in

extern int &r0;
int &r = r0;
constexpr int &r2 = r;

r is potentially-constant and has an initializing declaration reachable from, and in the same translation unit as, the definition of r2.

[eisenwave]

P2 because we have two identical \defn entries, and that's probably not how those macros should be used.

[jensmaurer]

This is a good change.

@hubert-reinterpretcast , could you please rebase + force-push?

[hubert-reinterpretcast]

could you please rebase + force-push?

@jensmaurer: done

[jensmaurer]

@tkoeppe , you had concerns here, so I'll leave the merge to you. We need a squash.

[tkoeppe]

@jensmaurer I don't quite recall -- are you referring to the indexing perhaps, or the normative wording?

[jensmaurer]

Ah, sorry, I confused the author of the (still open) comment here. @tkoeppe , just go ahead if you feel this is ok.