Coming from a position of extreme Rust ignorance & C++ brain poisoning:
Here’s the issue: waiting_for_elements is a Vec<Waker>. The channel cannot know how many tasks are blocked, so we can’t use a fixed-size array. Using a Vec means we allocate memory every time we queue a waker. And that allocation is taken and released every time we have to wake.
Why isn't a structure that does amortized allocation an option here? I appreciate the design goal was "no allocations in steady-state", but that's what you'd expect if you were using C++'s std::vector: After a while the reserved space for the vector gets "big enough".
Here’s the issue: waiting_for_elements is a Vec<Waker>. The channel cannot know how many tasks are blocked, so we can’t use a fixed-size array. Using a Vec means we allocate memory every time we queue a waker. And that allocation is taken and released every time we have to wake.
Why isn't a structure that does amortized allocation an option here? I appreciate the design goal was "no allocations in steady-state", but that's what you'd expect if you were using C++'s std::vector: After a while the reserved space for the vector gets "big enough".