cpp-coroutines
C++20 coroutines skill for understanding coroutine mechanics and debugging. Use when working with co_await, co_yield, co_return, implementing promise_type, understanding coroutine frame layout, debugging suspended coroutines in GDB, or inspecting frame allocation with Compiler Explorer. Activates on queries about C++20 coroutines, co_await, co_yield, promise_type, coroutine_handle, coroutine suspension, or coroutine frame.
Works with
---
name: cpp-coroutines
description: C++20 coroutines skill for understanding coroutine mechanics and debugging. Use when working with co_await, co_yield, co_return, implementing promise_type, understanding coroutine frame layout, debugging suspended coroutines in GDB, or inspecting frame allocation with Compiler Explorer. Activates on queries about C++20 coroutines, co_await, co_yield, promise_type, coroutine_handle, coroutine suspension, or coroutine frame.
license: MIT
---
# C++20 Coroutines
## Purpose
Guide agents through C++20 coroutine mechanics: `co_await`, `co_yield`, `co_return`, implementing the required `promise_type`, understanding coroutine frame memory layout, debugging suspended coroutines in GDB, and reducing frame allocation overhead.
## Triggers
- "How do co_await, co_yield, and co_return work?"
- "How do I implement promise_type for a coroutine?"
- "How does a coroutine suspend and resume?"
- "How do I debug a suspended coroutine in GDB?"
- "How much memory does a coroutine frame use?"
- "How do I write a generator with co_yield?"
## Workflow
### 1. The three coroutine keywords
```cpp
// co_return — return a value and end the coroutine
co_return value;
// co_yield — produce a value, suspend, resume later
co_yield value;
// co_await — suspend until an awaitable completes
auto result = co_await some_awaitable;
```
A function is a coroutine if it contains any of these three keywords. Its return type must be a coroutine type with a `promise_type`.
### 2. Minimal coroutine type — Task
```cpp
#include <coroutine>
#include <stdexcept>
#include <optional>
template <typename T>
struct Task {
struct promise_type {
std::optional<T> value;
std::exception_ptr exception;
Task get_return_object() {
return Task{std::coroutine_handle<promise_type>::from_promise(*this)};
}
std::suspend_always initial_suspend() { return {}; } // lazy start
std::suspend_always final_suspend() noexcept { return {}; }
void return_value(T v) { value = std::move(v); }
void unhandled_exception() { exception = std::current_exception(); }
};
std::coroutine_handle<promise_type> handle;
explicit Task(std::coroutine_handle<promise_type> h) : handle(h) {}
Task(Task&&) = default;
Task& operator=(Task&&) = default;
~Task() { if (handle) handle.destroy(); }
T get() {
handle.resume(); // resume to completion
if (handle.promise().exception)
std::rethrow_exception(handle.promise().exception);
return std::move(*handle.promise().value);
}
};
// Usage
Task<int> compute() {
co_return 42;
}
int main() {
auto task = compute();
int result = task.get(); // 42
}
```
### 3. Generator with co_yield
```cpp
template <typename T>
struct Generator {
struct promise_type {
T current_value;
Generator get_return_object() {
return Generator{std::coroutine_handle<promise_type>::from_promise(*this)};
}
std::suspend_always initial_suspend() { return {}; }
std::suspend_always final_suspend() noexcept { return {}; }
void return_void() {}
void unhandled_exception() { throw; }
std::suspend_always yield_value(T value) {
current_value = value;
return {}; // suspend after yielding
}
};
std::coroutine_handle<promise_type> handle;
explicit Generator(std::coroutine_handle<promise_type> h) : handle(h) {}
~Generator() { if (handle) handle.destroy(); }
struct iterator {
std::coroutine_handle<promise_type> handle;
bool done;
iterator& operator++() {
handle.resume();
done = handle.done();
return *this;
}
T operator*() const { return handle.promise().current_value; }
bool operator!=(std::default_sentinel_t) const { return !done; }
};
iterator begin() {
handle.resume(); // advance to first yield
return {handle, handle.done()};
}
std::default_sentinel_t end() { return {}; }
};
// Usage
Generator<int> iota(int start, int end) {
for (int i = start; i < end; ++i)
co_yield i;
}
for (int x : iota(0, 5)) {
std::cout << x << ' '; // 0 1 2 3 4
}
```
### 4. Awaitable — custom co_await target
```cpp
// An awaitable has three methods:
// await_ready() — true means don't suspend
// await_suspend(handle) — suspend: store handle, schedule resume
// await_resume() — return value of co_await expression
struct TimerAwaitable {
int delay_ms;
bool await_ready() const noexcept { return delay_ms <= 0; }
void await_suspend(std::coroutine_handle<> h) {
// Schedule h.resume() to be called after delay
std::thread([h, this]() {
std::this_thread::sleep_for(std::chrono::milliseconds(delay_ms));
h.resume();
}).detach();
}
void await_resume() const noexcept {} // no return value
};
// suspend_always and suspend_never are built-in awaitables
std::suspend_always{}; // always suspends
std::suspend_never{}; // never suspends (no-op)
```
### 5. Coroutine frame layout and memory
The compiler allocates a coroutine frame (heap object) containing:
- Local variables that live across suspension points
- The promise object
- The current suspension state (where to resume)
- A pointer to the resumption/destruction functions
```cpp
// Inspect frame size with Compiler Explorer (godbolt.org)
// Compile with: g++ -std=c++20 -O2 -S
// Look for: operator new call size in the generated asm
// Or: clang -std=c++20 -O2 -emit-llvm -S | grep "coro.size"
// Reduce frame size:
// 1. Don't keep large objects alive across co_await
struct Bad {
std::vector<char> large_buf; // whole vector lives in frame
co_return large_buf.size(); // large_buf crosses suspension
};
// 2. Move data out before suspending
std::vector<char> buf = get_data();
size_t sz = buf.size(); // capture only what's needed
buf.clear(); // release before suspension
co_await next_event;
// sz still valid; buf released
```
### 6. Debugging suspended coroutines in GDB
```bash
# Coroutines appear as regular stack frames after resume()
# To inspect a suspended coroutine:
(gdb) info locals
# Look for coroutine_handle variables
# Print the promise object
(gdb) p *(promise_type*)(handle.__handle_)
# GDB 14+ has coroutine-specific support
(gdb) info coroutines # GCC coroutine support (experimental)
# Step through coroutine execution
(gdb) step # enters co_await implementation
(gdb) finish # returns from coroutine frame function
(gdb) next # step over suspension point
# View all threads (coroutines running on thread pool)
(gdb) info threads
(gdb) thread 2
(gdb) bt
```
### 7. Boost.Asio `co_spawn` and `co_await`
```cpp
#include <boost/asio.hpp>
#include <boost/asio/co_spawn.hpp>
#include <boost/asio/awaitable.hpp>
namespace net = boost::asio;
net::awaitable<void> echo_session(net::ip::tcp::socket socket) {
char buf[1024];
for (;;) {
std::size_t n = co_await socket.async_read_some(net::buffer(buf));
co_await net::async_write(socket, net::buffer(buf, n));
}
}
int main() {
net::io_context io;
net::co_spawn(io, listen_accept(io), net::detached);
io.run();
}
```
`co_spawn` launches coroutines on an executor; `co_await` chains completion tokens without callback nesting.
### 8. `std::generator` (C++23)
```cpp
#include <generator>
#include <ranges>
std::generator<int> fibonacci() {
int a = 0, b = 1;
while (true) {
co_yield a;
auto next = a + b;
a = b;
b = next;
}
}
// Usage
for (int v : fibonacci() | std::views::take(10))
printf("%d\n", v);
```
Lazy sequences without manual coroutine handle management — compiler provides `std::generator` promise type.
### 9. Coroutine frame layout in GDB
```bash
# Compile with debug info
g++ -std=c++20 -g -O0 -o app app.cpp
gdb ./app
```
```gdb
(gdb) break my_coro
(gdb) run
(gdb) info frame # current stack frame
(gdb) info locals # promise, handle in scope
# Inspect coroutine frame pointer (compiler-specific mangling)
(gdb) p *(MyPromise*)h.address() # h = coroutine_handle
# GCC coroutine support (GCC 14+)
(gdb) info coroutines
# Pretty-print promise state
(gdb) set print pretty on
(gdb) p promise
```
Suspended coroutines may not appear on stack until resumed — trace via stored `coroutine_handle`.
### 10. Compilation time impact
Coroutines increase template instantiation and header parsing cost:
| Mitigation | Effect |
|------------|--------|
| `-O2` HALO | Reduces generated frame glue |
| Out-of-line `co_await` in `.cpp` | Cuts recompilation cascade |
| Pimpl for coroutine return types | Hides `awaitable` templates from headers |
| `ccache` / modules | See `skills/rust/rust-build-times` patterns for C++ |
Measure with `g++ -ftime-report` or `clang -ftime-trace`. Coroutine-heavy headers (Asio) benefit from unity builds sparingly — balance with RAM use.
### 11. Common pitfalls
| Issue | Cause | Fix |
|-------|-------|-----|
| `co_await` in a non-coroutine | Function missing coroutine return type | Change return type to a coroutine type |
| Dangling handle after `co_return` | Using handle after coroutine finishes | Check `handle.done()` before resume |
| Double-resume | Resuming an already-resumed coroutine | Track state; only resume when suspended |
| Coroutine frame never freed | Forgot `handle.destroy()` | Use RAII wrapper (Task, Generator) |
| Heap allocation overhead | New frame per coroutine call | Enable HALO (Heap Allocation eLision Optimization) with `-O2` |
| Recursive co_await depth | Stack overflow from deep chains | Use `std::coroutine_handle<>` tail-call pattern |
## Related skills
- Use `skills/compilers/cpp-templates` for other advanced C++20 features
- Use `skills/rust/rust-async-internals` for Rust's equivalent Future/Poll model
- Use `skills/debuggers/gdb` for GDB session managementMore Debugging skills
diagnosing-bugs
mattpocock/skills
Diagnosis loop for hard bugs and performance regressions. Use when the user says "diagnose"/"debug this", or reports something broken/throwing/failing/slow.
explore-code
lllllllama/rigorpilot-skills
Rigor Improve implementation leaf skill for auditable candidate implementation in deep learning research repositories. Use when the researcher explicitly authorizes exploratory work on an isolated branch or worktree to transplant modules, adapt a backbone, add LoRA or adapter layers, replace a head, or stitch together meaningful low-risk migration ideas with rollback-aware records in `explore_outputs/`. Do not use for end-to-end exploration orchestration on top of `current_research`, trusted baseline reproduction, conservative debugging, environment setup, verified contribution claims, or default repository analysis.
safe-debug
lllllllama/rigorpilot-skills
Rigor Debug / Rigor Audit skill for deep learning research work. Use when the user pastes a traceback, terminal error, CUDA OOM, checkpoint load failure, shape mismatch, NaN loss symptom, or training failure and wants conservative diagnosis before any patching, with debug fixes clearly separated from research contributions. Do not use for broad refactoring, speculative adaptation, automatic exploratory patching, or general repository familiarization.

