CWG3058 "Program point" is not defined

[t3nsor]

Full name of submitter: Alisdair Meredith

Link to reflector thread: https://lists.isocpp.org/core/2025/08/18396.php

Issue description: See reflector thread. Alisdair may decide to submit a NB comment for this, so I'm posting it here to ensure that there's an easily accessible permalink.

[jensmaurer]

Since we're dealing with tokens (not individual characters) in phase 7, as long as the "point" we're talking about is a single token (e.g. an identifier), I don't think the difference between a zero-length point and a token matters at all when considering "program points".

[t3nsor]

I can see how it might not matter whether a point is an entire token or a single point within the token, but I think it might matter whether a point is between two tokens (or at the beginning or end of a TU).

[jensmaurer]

It'd be helpful to have examples where our use of "program point" is unclear even given the "token" interpretation.

[t3nsor]

[temp.point] sometimes specifies that a point of instantiation immediately precedes or follows another point of instantiation. This can be chained:

template <int x> struct A;

template <> struct A<0> {};

template <int x> struct A { A<x - 1> a; };

A<100> a;

The point of instantiation of A<100> is specified to immediately precede the namespace-scope definition of a. The point of instantiation of A<99> is specified to immediately precede that of A<100>, and so on and so on. This doesn't make sense if points are at tokens: it would imply that every time you go to the immediately preceding point, you go back one token. That would mean eventually you get to a point where the specialization A<0> isn't even reachable, and then eventually to imaginary tokens before the beginning of the TU.

I think this makes it sufficiently obvious that there must be an infinite number of program points between each pair of tokens, but there should be a definition somewhere that explicitly says this; readers should not have to infer it based on the intent of [temp.point]. I don't think we need program points at tokens, but I'm not sure.

[jensmaurer]

CWG3058