microsoft#4351)

Merge `main` to `feature/generator`

Resolved conflict in yvals_core.h, removing all guards around `__cpp_lib_generator` except `_HAS_CXX23`.

* `__cpp_lib_concepts` is now always available in C++20 and later.
* `<generator>` doesn't care about `std::byte` now.
* C1XX, Clang, and EDG all define `__cpp_impl_coroutine` in C++20 and later, which wasn't the case earlier.

Also add a missing semicolon - this emits a `_Pragma`.

VSO-1951821 "EDG instantiates the wrong overload of `promise_type::operator new` for `generator` machinery"

Merge `main` to `feature/generator`

Merge `main` to `feature/generator`

…4464)

Merge `main` to `feature/generator`

Co-authored-by: Stephan T. Lavavej <[email protected]>

Resolved a trivial merge conflict in `<ranges>` with the `inline constexpr` change for `_Compile_time_max_size`.

Merge `main` to `feature/generator`

Co-authored-by: Casey Carter <[email protected]>

Merge `main` to `feature/generator`

…crosoft#4621)

Co-authored-by: Stephan T. Lavavej <[email protected]>
Co-authored-by: Alex Guteniev <[email protected]>

Merge `main` to `feature/generator`

Merge `main` to `feature/generator`

Resolved an adjacent-add conflict in yvals_core.h.

Resolves a stealth merge conflict with GH 4782.

Merge `main` to `feature/generator`

microsoft#4827)

Merge `main` to `feature/generator`

…ator

Merge `main` to `feature/generator`

This should complete the `generator` feature branch and get it ready to merge. There are very few product code changes here, it's mostly expanded test coverage with some cleanup and reorganization. There's also some documentation of the "stack of coroutines" and the memory layout of allocations that will make it easier for folks to get up to speed on how the code works under the covers.

It's broken down nicely by commit for ease of review; I'll replicate the commit messages here as an overview. Comments without specific context refer to `P2502R2_generator` where most changes were made. The `P2502R2_generator_iterator` and `P2502R2_generator_promise` tests were already very nicely complete.

* Update `generator` template argument mandates
  * Update citations to WG21-N4988
  * Reorder checks to specification order
  * Make `_RRef` more obviously reflect the wording for `RRef` in WG21-N4988 [coro.generator.class]/1.4.
* Allocator testing updates
  * Simplify and correct `StatelessAlloc`
    * Deriving publicly from `std::allocator` not a great idea, as witnessed by the recent addition of `allocate_at_least`.
    * `deallocate` should be `noexcept` for a `Cpp17Allocator`.
    * We don't need to reimplement `std::allocator` when we can simply use it.
    * The domain of equality for `Cpp17Allocators` is an entire `rebind` family, i.e., `StatelessAlloc<T>` and `allocator_traits<StatelessAlloc<T>>::rebind_alloc<U>` must be comparable.
  * PascalCase `stateful_alloc` for consistency and move into the header with `StatelessAlloc`.
* Promise test tweaks
  * "whose member `await_ready` returns `false`" in WG21-N4988 [coro.generator.promise]/11 implies the return type is exactly `bool`, not "convertible to `bool`". Update `test_yield_elements_of_range` accordingly.
  * Move non-portable size check from `P2502R2_generator` into `P2502R2_generator_promise`
* Expand and complete `static_checks`: Implement a complete `generator` traits in the header to use in both `P2502R2_generator_iterator` and `P2502R2_generator`'s `static_checks`.
* Create generic `test_one` template which takes a generator, a description of its static properties, and the expected result of piping the generator through a provided range adaptor. `test_one` validates the static properties with `static_checks`, and confirms the output is as expected.
* Extract test cases for weird reference types (mutable lvalue and rvalue references) from `main` into new function `test_weird_reference_types`.
* Several small tweaks:
  * It's no longer significant that `co_upto` wasn't an example in the proposal; strike the comment.
  * Enforce `co_upto`'s precondition so it's nicely documented.
  * Consistently prefer `same_as` to `is_same_v`.
  * Regroup calls to test cases in `main` topically and title each category.
* Reorganize test code to agree with call order in `main`. [This is the largest individual commit; it is a pure reordering.]
* All product code assertions now depend on `_CONTAINER_DEBUG_LEVEL`. These are all simple O(1) checks, there's no reason not to promote them from `_DEBUG` to `_CONTAINER_DEBUG_LEVEL`.
* Clarify allocation mechanisms, including some fancy memory layout diagrams.
  * Rename `_Promise_allocator` to `_Coro_promise_allocator` to avoid any confusion with `_Promise`. Rename its template parameter `_Allocator` to `_Proto_allocator` to avoid confusion with the rebound allocator type `_Alloc`.
  * Expand `static_assert` message in `operator new`. If and when `_Coro_promise_allocator` is reused by other coroutine types we can worry about making the message more generic.
  * Extract block size computations.
* `Element_awaiter` must direct-non-list-initialize its stored object, per WG21-N4988 [coro.generator.promise]/7.
* Document the "stack of coroutine handles" in a code comment with another work of art.

Addresses microsoft#4952 (comment).

…ource>` after GH 4952.

* `const`-ify members of awaitable objects that can be `const`.
* Annotate that the iterator's `operator*` has a `/* strengthened */` _noexcept-specifier_.

Several minor nits, and I burned away all the unnecessary parts of "mutable rvalue reference type". Don't know where I was heading with it, but it never made it there.