⚡ Merge Scheduler and Context

[kammce]

Summary

This PR refactors the async_context library by merging the scheduler class into the context base class, improving memory efficiency, simplifying ownership semantics, and reducing indirection. It also introduces a new proxy_context creation method (from) and removes redundant context parameters in scheduling calls.

---

Changes Overview

1. Merged scheduler into context:

   • The context class is now an abstract base class with pure virtual function (do_schedule) that are implemented by derived context classes.
   • This removes the need for a separate scheduler object and reduces context size down to just 64 bytes on 64-bit systems.

2. Stack Memory Ownership:

   • Derived context classes now own their stack memory directly.
   • The context class has a new initialize_stack_memory method to set up the stack.
   • Example:

     class my_context : public async::context {
         std::array<byte, 4096> m_stack;
         my_scheduler& m_scheduler;
     
     public:
         my_context(my_scheduler& sched) : m_scheduler(sched) {
             initialize_stack_memory(m_stack);
         }
     
     protected:
         void do_schedule(blocked_by state, block_info info) override {
             m_scheduler.handle(*this, state, info);
         }
     };

3. Proxy Context Implementation:

   • The proxy_context class now implements the proxy pattern by delegating scheduling to the original context.
   • A new static method from(context& p_parent) is added for constructing proxies.

4. Simplified Scheduling API:

   • Removed the redundant context& p_context parameter from schedule() calls.
   • The new signature is: schedule(blocked_by state, block_info info).

5. Memory Efficiency Gains:

   • Reduced base context size from 64 bytes to 48 bytes.
   • Eliminated pointer indirection and virtual call overhead.

6. Updated Dependencies & Build Files:

   • Removed strong_ptr dependency
   • Updated test and benchmark code to use the new context-based design.

---

Benefits

Benefit	Description
💾 Memory Efficiency	async::context now fits within a single cache line
⚡ Performance	Faster virtual calls and reduced indirection from variant
🔒 Ownership Clarity	Clear separation of stack ownership
🧩 Flexibility	Custom context types for various environments or scheduler types
🧠 Simplicity	Cleaner API and less abstraction overhead

---

Example Usage

struct test_context : public async::context {
    std::array<async::uptr, 8192> m_stack{};

    test_context() {
        this->initialize_stack_memory(m_stack);
    }

protected:
    void do_schedule(blocked_by state, block_info info) noexcept override {
        // Handle scheduling logic here
    }
};

test_context ctx;

---

Testing & Compatibility

• All existing tests pass with updated context construction and proxy usage.
• Benchmarks are adjusted to use the new benchmark_context for performance measurement.
• The proxy context creation now uses proxy_context::from(parent) instead of the previous proxy constructor.

---

Conclusion

This PR significantly refactors async_context to better align with embedded systems’ constraints by reducing memory overhead, simplifying ownership models, and enhancing flexibility for different scheduler implementations.

This PR message was generated by Qwen3-coder:30b using LM studio and based on a diff of this change. The content was read and edited by myself.