[dcl.type.pack.index] Clarify pack-index-specifier rules

[brevzin]

Currently, [dcl.type.pack.index] reads:

pack-index-specifier:
    typedef-name ... [ constant-expression ]

The typedef-name P in a pack-index-specifier shall denote a pack.

A typedef-name is either an identifier or a simple-template-id, which suggests that maybe something<Ts>...[0] is valid in addition to Ts...[0], but the first sentence here restricts that case, since a simple-template-id can never denote a pack.

But to figure that out is kind of a logic puzzle. Can we just say that more directly?

The typedef-name P in a pack-index-specifier shall be an identifier that denotes a pack.

[pdimov]

FWIW, I'd prefer something<Ts>...[0] to be valid, and GCC currently accepts it.

[brevzin]

Actually, the expression wording explicitly uses the wording I'm suggesting here — which suggests maybe that something<Ts>...[0] is supposed to work. Although I don't see how something<Ts> can be said to denote a pack.

Do we just need a different sentence entirely here?

[brevzin]

Given that p2 currently talks about sizeof...(P) which is only valid for names, I think that we definitely mean to only support names. Otherwise, p2 would also need to change.

[jensmaurer]

We definitely want an identifier here; nothing else can "denote a pack". (Pack expansions can appear in select places, but this is different.)

Parallelism with [expr.prim.pack.index] p1 suggests the amended phrasing is a consistency improvement.