Guideline for how to declare variables in a range-based for loop

A question which comes up every now and then is what type of loop variable you should use for range-based for, particularly when no mutation takes place.

// case (1) - cheap to copy
for (int x : container) // disallowed in this form by Con.1, should be 'const int'
// vs
for (const int &x : container)

// case (2) - expensive to copy
for (string x : container) // disallowed in this form by ES.71, should be 'const string &x'
// vs
for (const string &x : container)

I think we should recommend to always declare the loop variable as a reference, never as a value, and here's why:

1. The rule is simple and consistent.
2. When mutating a T& variable in the loop, you are already forced to follow it by the language.
3. If you aren't, and you have case (2), then ES.71 already tells you to.
4. Otherwise, if you are following Con.1, then const T x is only one character away from const T &x, so it costs very little effort to follow this rule.
5. It is possible and relatively easy to enforce with automatic tooling.
6. It prevents bugs related to taking the address of the local variable, instead of taking the address of the object within the container.
7. Even for weird iterators like std::ranges::iota_view::iterator, where their reference is actually a value, this method is robust because const& allows for temporary materialization.

In essence, I don't see any benefit to non-reference loop variables, and the alternative is consistent, easy, correct by default, and already recommended in part.

Start by reviewing the existing Con.1 and ES.71 guidance and the five comments on this issue. Determine whether the proposed rule should replace or clarify those recommendations, including the cited range-based-for cases. Done means reaching a decision and documenting the agreed guidance in the relevant C++ Core Guidelines section.