They\u2019re not the easiest to learn and the tooling is still evolving, but for high performance use cases like game engines, servers, and real time systems, coroutines unlock faster execution, lower memory usage, and far more maintainable code. In short, they make async programming finally feel <\/span>human<\/span>.<\/span><\/p>\n C++ coroutines let you write high-performance async code that pauses and resumes naturally, eliminating the complexity of traditional multitasking approaches.\u00a0<\/span><\/p>\n C++ coroutines make this possible by letting you pause and resume functions naturally, almost like hitting pause on a video and picking up exactly where you left off. What makes this exciting isn’t just cleaner code (though that’s huge), but the fact that your programs can handle thousands of tasks at once while using far less memory than traditional approaches\u2014without drowning in callback hell or complicated state machines.<\/span><\/p>\n Think of it as getting the best of both worlds: code that’s easy to read and understand, plus performance that doesn’t make you choose between simplicity and speed.<\/span><\/p>\n With <\/span>C++20<\/span><\/a>, <\/span>C++ Coroutines<\/b> arrived to make <\/span>asynchronous programming<\/span><\/a> cleaner and more intuitive. They allow you to write code that looks sequential but runs asynchronously under the hood. It reduces boilerplate and keeps your logic easy to follow.<\/span><\/p>\n But what exactly are <\/span>C++ Coroutines<\/span><\/a>, and can they truly change the way we write async code? Or are they just another niche feature? Let\u2019s break it down.<\/span><\/p>\n At their core, <\/span>C++ Coroutines<\/b> are a C++20-added feature that lets you pause and resume a function\u2019s execution without blocking the main thread. Think of them like hitting \u201cpause\u201d on a task. You stop at a certain point, go do something else, and then \u201cresume\u201d right where you left off.<\/span><\/p>\n For instance, when you\u2019re preparing food, you stop and hop to other tasks. You may start boiling water, then while waiting for it to complete, you chop some onions, but then you also start marinating your meat\u2014you get the point.<\/span><\/p>\n In programming terms, this means you can write <\/span>asynchronous code<\/b> that looks and feels like synchronous code. Instead of chaining <\/span>callbacks<\/span><\/a> or building complex <\/span>state machines<\/span><\/a>, coroutines allow your logic to flow naturally while the compiler handles the heavy lifting.<\/span><\/p>\n A coroutine in C++ is just a function that uses one or more of these special keywords:<\/span><\/p>\n Here\u2019s a minimal <\/span>coroutine example in C++<\/b>:<\/span><\/p>\n In this example, the function <\/span>sayHello()<\/b> starts execution, prints \u201cHello, \u201c, and then <\/span>suspends<\/b> itself with co_await. It only resumes when explicitly told to, allowing other operations to run in between.<\/span><\/p>\n Unlike <\/span>coroutines in C<\/b>, which often require manual stack management or custom state handling, <\/span>C++ coroutines<\/b> are built into the language and supported directly by the compiler. This makes them far more ergonomic for modern <\/span>async programming in C++<\/b>.<\/span><\/p>\n To really see the difference, let\u2019s compare how the same async task is written using <\/span>C++14 threads and futures<\/b> versus a <\/span>C++20 coroutine<\/b>. The logic is identical, but the coroutine version reads more like straightforward, top-to-bottom code.<\/span><\/p>\n Goal: download ? process ? save without blocking the main thread.<\/span><\/p>\n Takeaways<\/b><\/p>\n Same workflow, written as one sequential coroutine. Readers can run this as-is.<\/span><\/p>\n Takeaways<\/b><\/p>\n As you can see, coroutines let us express the same workflow with far less boilerplate, setting the stage for why modern C++ needed them in the first place.<\/span><\/p>\n In the next section, we\u2019ll explore why C++ needed coroutines in the first place and how they improve the developer experience over older async approaches like callbacks and goto-style state jumps.<\/span><\/p>\n Before <\/span>C++20<\/b>, asynchronous programming in C++ was often messy. Developers relied on <\/span>callback hell<\/span><\/i>, long <\/span>promise chains<\/span><\/a>, or manual state machines to keep tasks running without blocking the main thread. A seemingly simple workflow\u2014like reading a file, processing it, and sending the results over a network\u2014required multiple callback functions, manual state passing, and careful thread-safety management. The result was verbose, error-prone code scattered across files, making debugging feel like watching a movie out of order.<\/span><\/p>\n C++ Coroutines<\/b> change this by letting you write asynchronous logic in a clean, sequential style. The compiler automatically handles the suspend and resume points, so you can focus on <\/span>what<\/span><\/i> should happen instead of <\/span>how<\/span><\/i> to wire it all together.<\/span><\/p>\n In practice, this brings big benefits:<\/span><\/p>\nWhat are C++ Coroutines, really?<\/span><\/h2>\n
\n
#include <iostream>\r\n\r\n#include <coroutine>\r\n\r\nstruct SimpleTask {\r\n\r\n\u00a0\u00a0\u00a0\u00a0struct promise_type {\r\n\r\n\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0SimpleTask get_return_object() { return {}; }\r\n\r\n\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0std::suspend_never initial_suspend() { return {}; }\r\n\r\n\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0std::suspend_never final_suspend() noexcept { return {}; }\r\n\r\n\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0void return_void() {}\r\n\r\n\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0void unhandled_exception() {}\r\n\r\n\u00a0\u00a0\u00a0\u00a0};\r\n\r\n};\r\n\r\nSimpleTask sayHello() {\r\n\r\n\u00a0\u00a0\u00a0\u00a0std::cout << \"Hello, \";\r\n\r\n\u00a0\u00a0\u00a0\u00a0co_await std::suspend_always{};\r\n\r\n\u00a0\u00a0\u00a0\u00a0std::cout << \"World!\\n\";\r\n\r\n}\r\n\r\nint main() {\r\n\r\n\u00a0\u00a0\u00a0\u00a0auto task = sayHello(); \/\/ Starts coroutine\r\n\r\n\u00a0\u00a0\u00a0\u00a0\/\/ At this point, execution is paused after \"Hello, \"\r\n\r\n\u00a0\u00a0\u00a0\u00a0\/\/ In real use, you\u2019d resume it when ready.\r\n\r\n}<\/pre>\n<\/div>\nC++14 vs C++20 side-by-side example<\/span><\/h3>\n
1) C++14 Threads\/Futures (verbose wiring)<\/span><\/h4>\n
\/\/ C++14: compile with -std=c++14\r\n\r\n#include <iostream>\r\n\r\n#include <thread>\r\n\r\n#include <future>\r\n\r\n#include <chrono>\r\n\r\nusing namespace std::chrono_literals;\r\n\r\n\/\/ --- Simulated work ---------------------------------------------------------\r\n\r\nstd::string download() {\r\n\r\n\u00a0\u00a0\u00a0\u00a0std::this_thread::sleep_for(500ms);\r\n\r\n\u00a0\u00a0\u00a0\u00a0return \"data\";\r\n\r\n}\r\n\r\nstd::string process(const std::string& in) {\r\n\r\n\u00a0\u00a0\u00a0\u00a0std::this_thread::sleep_for(400ms);\r\n\r\n\u00a0\u00a0\u00a0\u00a0return \"processed(\" + in + \")\";\r\n\r\n}\r\n\r\nvoid save(const std::string& in) {\r\n\r\n\u00a0\u00a0\u00a0\u00a0std::this_thread::sleep_for(300ms);\r\n\r\n\u00a0\u00a0\u00a0\u00a0std::cout << \"Saved: \" << in << \"\\n\";\r\n\r\n}\r\n\r\n\/\/ --- Pipeline using threads + futures --------------------------------------\r\n\r\nint main() {\r\n\r\n\u00a0\u00a0\u00a0\u00a0\/\/ 1) Set up futures\/promises to shuttle results between stages\r\n\r\n\u00a0\u00a0\u00a0\u00a0std::promise<std::string> p_download;\r\n\r\n\u00a0\u00a0\u00a0\u00a0std::future<std::string> f_download = p_download.get_future();\r\n\r\n\u00a0\u00a0\u00a0\u00a0std::promise<std::string> p_process;\r\n\r\n\u00a0\u00a0\u00a0\u00a0std::future<std::string> f_process = p_process.get_future();\r\n\r\n\u00a0\u00a0\u00a0\u00a0\/\/ 2) Each stage runs on its own thread, waiting on the previous stage\r\n\r\n\u00a0\u00a0\u00a0\u00a0std::thread t1([&] { p_download.set_value(download()); });\r\n\r\n\u00a0\u00a0\u00a0\u00a0std::thread t2([&] {\r\n\r\n\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0auto d = f_download.get(); \u00a0 \u00a0 \u00a0 \u00a0 \/\/ wait for download\r\n\r\n\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0p_process.set_value(process(d)); \u00a0 \/\/ pass to processing\r\n\r\n\u00a0\u00a0\u00a0\u00a0});\r\n\r\n\u00a0\u00a0\u00a0\u00a0std::thread t3([&] {\r\n\r\n\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0auto r = f_process.get();\u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \/\/ wait for processing\r\n\r\n\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0save(r);\r\n\r\n\u00a0\u00a0\u00a0\u00a0});\r\n\r\n\u00a0\u00a0\u00a0\u00a0\/\/ 3) Don\u2019t forget to join\u2014easy to leak or forget in real code\r\n\r\n\u00a0\u00a0\u00a0\u00a0t1.join(); t2.join(); t3.join();\r\n\r\n}<\/pre>\n<\/div>\n\n
2) C++20 Coroutine (linear, top?to?bottom)<\/span><\/h4>\n
\/\/ C++20: compile with -std=c++20\r\n\r\n#include <iostream>\r\n\r\n#include <coroutine>\r\n\r\n#include <thread>\r\n\r\n#include <chrono>\r\n\r\n#include <semaphore>\r\n\r\nusing namespace std::chrono_literals;\r\n\r\n\/\/ --- Tiny awaitable to \"sleep\" without blocking the coroutine ---------------\r\n\r\nstruct sleep_awaitable {\r\n\r\n\u00a0\u00a0\u00a0\u00a0std::chrono::milliseconds dur;\r\n\r\n\u00a0\u00a0\u00a0\u00a0bool await_ready() const noexcept { return dur.count() == 0; } \u00a0 \/\/ run-through?\r\n\r\n\u00a0\u00a0\u00a0\u00a0void await_resume() const noexcept {} \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \/\/ nothing to return\r\n\r\n\u00a0\u00a0\u00a0\u00a0\/\/ Schedule resume on a background thread after 'dur'\r\n\r\n\u00a0\u00a0\u00a0\u00a0void await_suspend(std::coroutine_handle<> h) const {\r\n\r\n\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0std::thread([h, d = dur]{\r\n\r\n\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0std::this_thread::sleep_for(d);\r\n\r\n\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0h.resume(); \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \/\/ continue pipeline()\r\n\r\n\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0}).detach();\r\n\r\n\u00a0\u00a0\u00a0\u00a0}\r\n\r\n};\r\n\r\n\/\/ --- Minimal Task so main() can wait for completion -------------------------\r\n\r\nstruct Task {\r\n\r\n\u00a0\u00a0\u00a0\u00a0struct promise_type {\r\n\r\n\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0std::binary_semaphore done{0};\r\n\r\n\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0Task get_return_object() {\r\n\r\n\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0return Task{ std::coroutine_handle<promise_type>::from_promise(*this) };\r\n\r\n\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0}\r\n\r\n\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0std::suspend_never initial_suspend() noexcept { return {}; }\u00a0 \/\/ start immediately\r\n\r\n\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0std::suspend_always final_suspend() noexcept { \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \/\/ signal on finish\r\n\r\n\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0done.release();\r\n\r\n\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0return {};\r\n\r\n\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0}\r\n\r\n\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0void return_void() {}\r\n\r\n\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0void unhandled_exception() { std::terminate(); }\r\n\r\n\u00a0\u00a0\u00a0\u00a0};\r\n\r\n\u00a0\u00a0\u00a0\u00a0std::coroutine_handle<promise_type> h;\r\n\r\n\u00a0\u00a0\u00a0\u00a0void wait() { h.promise().done.acquire(); h.destroy(); }\r\n\r\n};\r\n\r\n\/\/ --- The async pipeline, but written linearly --------------------------------\r\n\r\nTask pipeline() {\r\n\r\n\u00a0\u00a0\u00a0\u00a0std::cout << \"Downloading...\\n\";\r\n\r\n\u00a0\u00a0\u00a0\u00a0co_await sleep_awaitable{500ms}; \u00a0 \/\/ pause here, resume later\r\n\r\n\u00a0\u00a0\u00a0\u00a0std::cout << \"Processing...\\n\";\r\n\r\n\u00a0\u00a0\u00a0\u00a0co_await sleep_awaitable{400ms};\r\n\r\n\u00a0\u00a0\u00a0\u00a0std::cout << \"Saving...\\n\";\r\n\r\n\u00a0\u00a0\u00a0\u00a0co_await sleep_awaitable{300ms};\r\n\r\n\u00a0\u00a0\u00a0\u00a0std::cout << \"Saved: processed(data)\\n\";\r\n\r\n}\r\n\r\nint main() {\r\n\r\n\u00a0\u00a0\u00a0\u00a0auto t = pipeline();\u00a0 \/\/ starts and runs until first co_await\r\n\r\n\u00a0\u00a0\u00a0\u00a0t.wait(); \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \/\/ wait for coroutine to finish\r\n\r\n}<\/pre>\n<\/div>\n\n
What pipeline() does<\/span><\/h4>\n
![\"diagram](\"https:\/\/i0.wp.com\/www.wholetomato.com\/blog\/wp-content\/uploads\/2025\/08\/diagram-to-show-what-pipeline-does.png?resize=572%2C207&ssl=1\")
<\/a><\/p>\nWhy async development needed C++ coroutines and their key benefits<\/span><\/h2>\n