{"id":103210,"date":"2019-12-13T07:00:00","date_gmt":"2019-12-13T15:00:00","guid":{"rendered":"http:\/\/devblogs.microsoft.com\/oldnewthing\/?p=103210"},"modified":"2019-12-12T21:41:18","modified_gmt":"2019-12-13T05:41:18","slug":"20191213-00","status":"publish","type":"post","link":"https:\/\/devblogs.microsoft.com\/oldnewthing\/20191213-00\/?p=103210","title":{"rendered":"C++ coroutines: Short-circuiting suspension, part 1"},"content":{"rendered":"

At the start of this series<\/a>, I gave the basic idea for how the compiler generates code for co_await<\/code>, but I left out some details for expository simplicity. There are some mysterious steps called “We’re not ready to talk about this step yet.”<\/p>\n

Now it’s time to talk about one of those steps.<\/p>\n

It may be the case that when you get around to doing the await_<\/code>suspend<\/code>, the thing your custom awaiter is waiting for has already completed. It could be that the operation completed synchronously, or that it was so fast that it finished even before you could schedule the completion.<\/p>\n

You don’t want to invoke the handle directly from your await_<\/code>suspend<\/code>, because that would run the coroutine continuation as a subroutine inside the suspension:<\/p>\n\n\n\n\n\n\n\n
\n
calculate x<\/code><\/div>\n
obtain awaiter<\/code><\/div>\n<\/td>\n<\/tr>\n
\n
save state for resumption<\/div>\n
awaiter.await_suspend(handle);<\/code><\/div>\n
handle()<\/code><\/div>\n<\/td>\n<\/tr>\n
\u00a0<\/td>\n\n
restore state after resumption<\/div>\n
result = awaiter.await_resume();<\/code><\/div>\n<\/td>\n<\/tr>\n
\u00a0<\/td>\n\n
execution continues<\/div>\n
coroutine finishes<\/div>\n<\/td>\n<\/tr>\n
\n
handle()<\/code> returns<\/div>\n
return to caller<\/div>\n<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n

This can result in quite a significant accumulation of stack frames if there are a lot of consecutive co_await<\/code>s of already-completed operations.<\/p>\n

To avoid this problem, you can change your await_<\/code>suspend<\/code> member to return bool<\/code>. Your implementation should check whether the operation has already completed. If so, then do not<\/i> schedule the handle for execution, but instead just return false<\/code>, to indicate that suspension should be abandoned and that execution should resume immediately. Otherwise, schedule the handle for execution as usual, and return true<\/code>.<\/p>\n

Adding to our gradually-improving understanding of the compiler code generation of co_await<\/code>:<\/p>\n\n\n\n\n\n\n
 <\/td>\n\n
calculate x<\/code><\/div>\n
obtain awaiter<\/code><\/div>\n<\/td>\n<\/tr>\n
co_await<\/code><\/td>\n\n
(We’re not ready to talk about this step yet.)<\/div>\n
save state for resumption<\/div>\n
if (awaiter.await_suspend(handle))<\/code> \u21d0<\/span><\/div>\n
{<\/code><\/div>\n
return to caller<\/div>\n<\/td>\n<\/tr>\n
\n
[Invoking the handle resumes execution here]<\/div>\n
}<\/code><\/div>\n
restore state after resumption<\/div>\n
result = awaiter.await_resume();<\/code><\/div>\n<\/td>\n<\/tr>\n
 <\/td>\nexecution continues<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n

Let’s add fancy await_<\/code>suspend<\/code> support to our resume_<\/code>in_<\/code>any_<\/code>apartment<\/code> function:<\/p>\n

template<typename Async,\r\n         typename = std::enable_if_t<\r\n             std::is_convertible_v<\r\n                 Async,\r\n                 winrt::Windows::Foundation::IAsyncInfo>>>\r\n[[nodiscard]] auto resume_in_any_apartment(Async async)\r\n{\r\n  struct awaiter : std::experimental::suspend_always\r\n  {\r\n    bool<\/span> await_suspend(\r\n        std::experimental::coroutine_handle<> handle)\r\n    {\r\n      if (async.Status() != Windows::Foundation::AsyncStatus::Started) {\r\n        return false;\r\n      }<\/span>\r\n      async.Completed([handler](auto&&amp...) { handler(); });\r\n      return true;<\/span>\r\n    }\r\n\r\n    auto await_resume()\r\n    {\r\n        return async.GetResults();\r\n    }\r\n    Async async;\r\n  };\r\n  return awaiter{ {}, std::move(async) };\r\n}\r\n<\/pre>\n
If at the time of suspension, the asynchronous activity is not in the Started<\/code> state, then that means that it completed (successfully, with an error, or with cancellation). Therefore, there’s no point waiting for it to complete. We can report it as already-completed and continue execution directly.\u00b9<\/p>\n
This type of short-circuit is commonly seen when the await_<\/code>suspend<\/code> function tries to schedule the continuation, and the framework says, “Dude, it’s already done!” For example, you might be performing an asynchronous read: If the Read\u00adFile<\/code> function returns TRUE<\/code>, then the operation completed synchronously, and you can go straight to the resumption code.<\/p>\n
There’s one last piece of the compiler code generation that is marked “We’re not ready to talk about this step yet.” We’re just about ready to talk about that step. Next time.<\/p>\n
\u00b9 There is a small race window if the asynchronous activity completes just after we check whether it has completed. Therefore, this change does not eliminate the stack accumulation completely, but it greatly reduces its likelihood.<\/p>\n
 <\/p>\n","protected":false},"excerpt":{"rendered":"
Avoiding needless stack buildup.<\/p>\n","protected":false},"author":1069,"featured_media":111744,"comment_status":"open","ping_status":"closed","sticky":false,"template":"","format":"standard","meta":{"_acf_changed":false,"footnotes":""},"categories":[1],"tags":[25],"class_list":["post-103210","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-oldnewthing","tag-code"],"acf":[],"blog_post_summary":"
Avoiding needless stack buildup.<\/p>\n","_links":{"self":[{"href":"https:\/\/devblogs.microsoft.com\/oldnewthing\/wp-json\/wp\/v2\/posts\/103210","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/devblogs.microsoft.com\/oldnewthing\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/devblogs.microsoft.com\/oldnewthing\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/devblogs.microsoft.com\/oldnewthing\/wp-json\/wp\/v2\/users\/1069"}],"replies":[{"embeddable":true,"href":"https:\/\/devblogs.microsoft.com\/oldnewthing\/wp-json\/wp\/v2\/comments?post=103210"}],"version-history":[{"count":0,"href":"https:\/\/devblogs.microsoft.com\/oldnewthing\/wp-json\/wp\/v2\/posts\/103210\/revisions"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/devblogs.microsoft.com\/oldnewthing\/wp-json\/wp\/v2\/media\/111744"}],"wp:attachment":[{"href":"https:\/\/devblogs.microsoft.com\/oldnewthing\/wp-json\/wp\/v2\/media?parent=103210"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/devblogs.microsoft.com\/oldnewthing\/wp-json\/wp\/v2\/categories?post=103210"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/devblogs.microsoft.com\/oldnewthing\/wp-json\/wp\/v2\/tags?post=103210"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}