[libcxx-commits] [libcxx] [libc++] Merge the private iterator_traits aliases with their ranges versions (PR #162661)

[Louis Dionne via libcxx-commits]

================
@@ -71,15 +65,9 @@ struct _ClassicAlgPolicy {};
 
 template <>
 struct _IterOps<_ClassicAlgPolicy> {
-  template <class _Iter>
-  using __value_type _LIBCPP_NODEBUG = typename iterator_traits<_Iter>::value_type;
----------------
ldionne wrote:

Why is it okay to make this (general) change? I understand that we are probably just accepting more code than previously, but it is possible for conforming C++20 code to break when we do this?

A few questions:
- In C++20, is it possible for an iterator to have different types for `std::iter_value_t<It>` and `std::iterator_traits<It>::value_type`?
- If so, is it possible for these two types to differ to the extent that a classic algorithm (which should in principle use `iterator_traits`, or at least did historically) would break if we start using `std::iter_value_t<It>` instead?
- If it's possible for these two typedefs to differ *and* to the extent of code breaking, are these cases so far fetched that we don't care to support them?
- The same set of questions holds for every nested typedef we're modifying here.

Also CC @frederick-vs-ja since I think I remember having that discussion in a previous patch.

https://github.com/llvm/llvm-project/pull/162661