{"id":5886,"date":"2020-04-23T14:58:39","date_gmt":"2020-04-23T14:58:39","guid":{"rendered":"https:\/\/www.modernescpp.com\/index.php\/c-20-coroutine-abstraction-with-cppcoro\/"},"modified":"2023-09-28T07:42:49","modified_gmt":"2023-09-28T07:42:49","slug":"c-20-coroutine-abstraction-with-cppcoro","status":"publish","type":"post","link":"https:\/\/www.modernescpp.com\/index.php\/c-20-coroutine-abstraction-with-cppcoro\/","title":{"rendered":"C++20: Powerful Coroutines with cppcoro"},"content":{"rendered":"

I gave in my last post, “C++20: Coroutines with cppcoro<\/a>“, a basic introduction to the coroutines library from Lewis Baker. This introduction covered the elementary coroutines task and generator. Today, I add threads to tasks and get powerful abstractions.<\/p>\n

<\/p>\n

\"\"<\/p>\n

Do you remember the previous post “C++20: Thread Synchronization with Coroutines<\/a>“? If no, I presented the challenges of a condition variable. A condition variable is a classical thread synchronization method, such as in a sender\/receiver or a producer\/consumer workflow. Condition variables have a big design flaw; they may be invoked without a notification (spurious wakeup) or overhear the notification (lost wakeup). In both cases, you may get a deadlock.\u00a0My following example on thread synchronization based on coroutines didn’t have the inherent risk of condition variables such as spurious or lost wakeup, but the example had another issue. It was too complicated.<\/p>\n

Thanks to cppcoro<\/a>, we can have the best of both worlds.: a straightforward event mechanism that does not have the design flaws of condition variables.
\n<\/span><\/p>\n

single_consumer_event<\/span><\/h2>\n

single_consumer_event<\/span>\u00a0is, according to the documentation, a simple manual-reset event type that supports only a single coroutine awaiting it at a time. This is precisely, what I need:<\/p>\n

\n
\/\/ cppcoroProducerConsumer.cpp<\/span>\n\n#include <cppcoro\/single_consumer_event.hpp><\/span>\n#include <cppcoro\/sync_wait.hpp><\/span>\n#include <cppcoro\/task.hpp><\/span>\n\n#include <future><\/span>\n#include <iostream><\/span>\n#include <string><\/span>\n#include <thread><\/span>\n#include <chrono><\/span>\n\ncppcoro::<\/span>single_consumer_event event;  \n\ncppcoro::<\/span>task<><\/span> consumer() {\n    \n    auto<\/span> start =<\/span> std::<\/span>chrono::<\/span>high_resolution_clock::<\/span>now();\n    \n    co_await event;  \/\/ suspended until some thread calls event.set()<\/span>\n    \n    auto<\/span> end =<\/span> std::<\/span>chrono::<\/span>high_resolution_clock::<\/span>now();\n    std::<\/span>chrono::<\/span>duration<<\/span>double<\/span>><\/span> elapsed =<\/span> end -<\/span> start;\n    std::<\/span>cout <<<\/span> \"Consumer waited \"<\/span> <<<\/span> elapsed.count() <<<\/span> \" seconds.\"<\/span> <<<\/span> std::<\/span>endl;\n  \n    co_return;\n}\n\nvoid<\/span> producer() {\n\n    using<\/span> namespace<\/span> std::<\/span>chrono_literals;\n    std::<\/span>this_thread::<\/span>sleep_for(2<\/span>s);\n    \n    event.set();  \/\/ resumes the consumer  <\/span>\n    \n}\n\nint<\/span> main() {\n    \n    std::<\/span>cout <<<\/span> std::<\/span>endl;\n    \n    auto<\/span> con =<\/span> std::<\/span>async([]{ cppcoro::<\/span>sync_wait(consumer()); });  \/\/ (1)<\/span>\n    auto<\/span> prod =<\/span> std::<\/span>async(producer);                              \/\/ (2)<\/span>\n    \n    con.get(), prod.get();\n    \n    std::<\/span>cout <<<\/span> std::<\/span>endl;\n    \n}\n<\/pre>\n<\/div>\n
The code should be self-explanatory. The consumer (line 1) and the producer (line 2) run in their thread. The call cppcoro::sync_wait(consumer())<\/span>\u00a0(line 1) serves as a top-level task because the main<\/span> function cannot be a coroutine. The call waits until the coroutine consumer<\/span> is done. The coroutine consumer<\/span> waits in the call co_await event<\/span> until someone calls event.set().<\/span> The function producer<\/span> sends this event after a sleep of two seconds.<\/p>\n
\"cppcoroProducerConsumer\"<\/p>\n
Thanks to the cppcoro library, threads can be canceled.<\/p>\n
Cancellation<\/h2>\n
The caller and the callee communicate with the cppcoro::cancellation_token<\/span>. The callee of the function that gets the request to cancel can respond in two ways.<\/p>\n
    \n
  1. Poll at regular intervals for the request to cancel. The cppcoro::cancellation_token supports two member functions: is_cancellation_requested()<\/span> and throw_if_cancellation_requested()<\/span>.<\/li>\n
  2. Register a callback that is executed in case of a cancellation request.<\/li>\n<\/ol>\n
    The following example exemplifies the first use case.<\/p>\n
    \n
    \/\/ cppcoroCancellation.cpp<\/span>\n\n#include <chrono><\/span>\n#include <iostream><\/span>\n#include <future><\/span>\n\n#include <cppcoro\/cancellation_token.hpp><\/span>\n#include <cppcoro\/cancellation_source.hpp><\/span>\n\nusing<\/span> namespace<\/span> std::<\/span>chrono_literals; \n\nint<\/span> main<\/span>() {\n    \n    std::<\/span>cout <<<\/span> std::<\/span>endl;\n    \n    cppcoro::<\/span>cancellation_source canSource;\n    cppcoro::<\/span>cancellation_token canToken =<\/span> canSource.token();  \/\/ (1)<\/span>\n\n    auto<\/span> cancelSender =<\/span> std::<\/span>async([&<\/span>canSource] { \n        std::<\/span>this_thread::<\/span>sleep_for(2<\/span>s);\n        canSource.request_cancellation();                      \/\/ (3)<\/span>\n        std::<\/span>cout <<<\/span> \"canSource.request_cancellation() \"<\/span> <<<\/span> std::<\/span>endl;\n    });\n        \n    auto<\/span> cancelReceiver =<\/span> std::<\/span>async([&<\/span>canToken] { \n        while<\/span>(true<\/span>) {\n            std::<\/span>cout <<<\/span> \"Wait for cancellation request\"<\/span> <<<\/span> std::<\/span>endl;\n            std::<\/span>this_thread::<\/span>sleep_for(200<\/span>ms);\n            if<\/span> (canToken.is_cancellation_requested()) return<\/span>;  \/\/ (2)<\/span>\n        }\n    });\n\n    cancelSender.get(), cancelReceiver.get();\n    \n    std::<\/span>cout <<<\/span> std::<\/span>endl;\n\n}\n<\/pre>\n<\/div>\n
    Line (1) shows the cancellation_token<\/span>, created by the cancellation_source<\/span>. The caller cancelSender<\/span> gets the cancellation source canSource<\/span>, and the callee cancelReceiver<\/span>\u00a0gets the cancellation token. The callee polls permanently for the cancellation request (line 2), which the caller sends via the call call.request_cancellation()<\/span> (line 3) after two seconds.<\/p>\n
    \"cppcoroCancellation\"<\/p>\n
    I want to make two interesting observations.<\/p>\n
      \n
    1. The cancellation is cooperative. If the callee ignores that cancellation request, nothing happens.<\/li>\n
    2. We get with C++20 an improved std::thread: std::jthread. std::jthread<\/span>\u00a0joins automatically in its destructor and can be in interrupted via an interrupt token. Read more details about the improved std::thread in my previous post: “A new Thread with C++20: std::jthread<\/a>“.<\/li>\n<\/ol>\n
      cppcoro even supports a mutex.<\/p>\n
      async_mutex<\/span><\/h2>\n
      A mutex such as cppcoro::async_mutex<\/span> is a synchronization mechanism to protect shared data from being accessed by multiple threads simultaneously.<\/p>\n
      \n
      \/\/ cppcoroMutex.cpp<\/span>\n\n#include <cppcoro\/async_mutex.hpp><\/span>\n#include <cppcoro\/sync_wait.hpp><\/span>\n#include <cppcoro\/task.hpp><\/span>\n\n#include <iostream><\/span>\n#include <thread><\/span>\n#include <vector><\/span>\n\n\ncppcoro::<\/span>async_mutex mutex;\n\nint<\/span> sum{};                                                                  \/\/ (2)<\/span>\n\ncppcoro::<\/span>task<><\/span> addToSum(int<\/span> num) {\n    cppcoro::<\/span>async_mutex_lock lockSum =<\/span> co_await mutex.scoped_lock_async(); \/\/ (3)<\/span>  \n    sum +=<\/span> num;\n  \n}                                                                           \/\/ (4)<\/span>\n\nint<\/span> main() {\n    \n    std::<\/span>cout <<<\/span> std::<\/span>endl;\n    \n    std::<\/span>vector<<\/span>std::<\/span>thread<\/span>><\/span> vec(10<\/span>);                                       \/\/ (1)<\/span>\n    \n    for<\/span>(auto<\/span>&<\/span> thr:<\/span> vec) {\n        thr =<\/span> std::<\/span>thread<\/span>([]{ for<\/span>(int<\/span> n =<\/span> 0<\/span>; n <<\/span> 10<\/span>; ++<\/span>n) cppcoro::<\/span>sync_wait(addToSum(n)); } );\n    }\n    \n    for<\/span>(auto<\/span>&<\/span> thr:<\/span> vec) thr.join();\n    \n    std::<\/span>cout <<<\/span> \"sum: \"<\/span> <<<\/span> sum <<<\/span> std::<\/span>endl;\n    \n    std::<\/span>cout <<<\/span> std::<\/span>endl;\n    \n}\n<\/pre>\n<\/div>\n
      Line (1) creates ten threads. Each thread adds the numbers 0 to 9 to the shared sum<\/span> (line 2). The function addToSum<\/span> is the coroutine. The coroutine waits in the expression co_await mutex.scoped_lock_async() (line 3) <\/span>until the mutex is acquired. The coroutine that waits for the mutex is not blocked but suspended. The previous lock-holder resumes the waiting coroutine in its unlock<\/span> call.\u00a0 \u00a0As its name suggests, the mutex stays locked until the end of the scope (line 4).<\/p>\n
      \"cppcoroMutex\"<\/p>\n
      What’s next?<\/h2>\n
      Thanks to the function cppcoro::when_all<\/span>, you can wait on one and more coroutines. I use\u00a0cppcoro::when_all<\/span> with cppcoro::static_thread_pool<\/span>\u00a0in my next post<\/a> to compose powerful workflows.<\/p>\n","protected":false},"excerpt":{"rendered":"
      I gave in my last post, “C++20: Coroutines with cppcoro“, a basic introduction to the coroutines library from Lewis Baker. This introduction covered the elementary coroutines task and generator. Today, I add threads to tasks and get powerful abstractions.<\/p>\n","protected":false},"author":21,"featured_media":8406,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[375],"tags":[445],"class_list":["post-5886","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-c-20","tag-coroutines"],"_links":{"self":[{"href":"https:\/\/www.modernescpp.com\/index.php\/wp-json\/wp\/v2\/posts\/5886","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/www.modernescpp.com\/index.php\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/www.modernescpp.com\/index.php\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/www.modernescpp.com\/index.php\/wp-json\/wp\/v2\/users\/21"}],"replies":[{"embeddable":true,"href":"https:\/\/www.modernescpp.com\/index.php\/wp-json\/wp\/v2\/comments?post=5886"}],"version-history":[{"count":3,"href":"https:\/\/www.modernescpp.com\/index.php\/wp-json\/wp\/v2\/posts\/5886\/revisions"}],"predecessor-version":[{"id":8420,"href":"https:\/\/www.modernescpp.com\/index.php\/wp-json\/wp\/v2\/posts\/5886\/revisions\/8420"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/www.modernescpp.com\/index.php\/wp-json\/wp\/v2\/media\/8406"}],"wp:attachment":[{"href":"https:\/\/www.modernescpp.com\/index.php\/wp-json\/wp\/v2\/media?parent=5886"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.modernescpp.com\/index.php\/wp-json\/wp\/v2\/categories?post=5886"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.modernescpp.com\/index.php\/wp-json\/wp\/v2\/tags?post=5886"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}