P2502R2 `<generator>`: Synchronous Coroutine Generator For Ranges

[StephanTLavavej]

WG21-P2502R2 <generator>: Synchronous Coroutine Generator For Ranges
WG21-P2787R1 pmr::generator
LWG-3762 generator::iterator::operator== should pass by reference
LWG-3826 Redundant specification [for overload of yield_value]
LWG-3894 generator::promise_type::yield_value(ranges::elements_of<Rng, Alloc>) should not be noexcept

Feature-test macro:

#define __cpp_lib_generator 202207L

The IDE already knows about this extensionless header.

Todo

According to Casey Carter (@CaseyCarter) in #4342 (comment) and as observed by A. Jiang (@frederick-vs-ja) in #4342 (comment) , the feature/generator branch needs:

• More thorough testing
• A quick audit to ensure it meets all the requirements
• A review of the allocator handling
• Thought put into what to do when __cpp_lib_byte is unavailable

[StephanTLavavej]

I believe that Casey Carter (@CaseyCarter) is working on his feature.

[CaseyCarter]

I haven't had time for STL work, and this has sat idle for so long that it's embarrassing. If someone wants to pick this up, feel free to do so. My prototype implementation is in a single commit at a5f4f74, which I just rebased on current main. It should provide a good starting point - if nothing else, it illustrates how the "stack of coroutine handles" the specification describes should actually be implemented via links in promises and awaiters and not a concrete std::stack requiring external memory allocation. The test is currently more of a demonstration than a true test, as fits a reference implementation for a WG21 proposal, it needs completion.

I'd also recommend auditing to ensure the implementation meets all of the requirements in the Standard wording. I think I implemented changes as they were made to the proposal, but suspect I did something hack-ish storing non-trivially-copyable allocator types.