{"id":108373,"date":"2023-06-26T07:00:00","date_gmt":"2023-06-26T14:00:00","guid":{"rendered":"https:\/\/devblogs.microsoft.com\/oldnewthing\/?p=108373"},"modified":"2023-06-19T06:20:49","modified_gmt":"2023-06-19T13:20:49","slug":"20230626-00","status":"publish","type":"post","link":"https:\/\/devblogs.microsoft.com\/oldnewthing\/20230626-00\/?p=108373","title":{"rendered":"How to wait for multiple C++ coroutines to complete before propagating failure, initial plunge"},"content":{"rendered":"

I’ll start by repeating a very handy cheat sheet from a debugging case study of memory corruption from a coroutine that already finished<\/a>.<\/p>\n\n\n\n\n\n\n\n\n\n\n\n
Language<\/th>\nMethod<\/th>\nResult<\/th>\nIf any fail<\/th>\n<\/tr>\n
C++<\/td>\nConcurrency::when_all<\/td>\nvector<T><\/code><\/td>\nfail immediately<\/td>\n<\/tr>\n
C++<\/td>\nwinrt::when_all<\/td>\nvoid<\/code><\/td>\nfail immediately<\/td>\n<\/tr>\n
C#<\/td>\nTask.WhenAll<\/td>\nT[]<\/code><\/td>\nwait for others<\/td>\n<\/tr>\n
JavaScript<\/td>\nPromise.all<\/td>\nArray<\/code><\/td>\nfail immediately<\/td>\n<\/tr>\n
JavaScript<\/td>\nPromise.allSettled<\/td>\nArray<\/code><\/td>\nwait for others<\/td>\n<\/tr>\n
Python<\/td>\nasyncio.gather<\/td>\nList<\/td>\nfail immediately by default<\/td>\n<\/tr>\n
Rust<\/td>\njoin!<\/td>\ntuple<\/td>\nwait for others<\/td>\n<\/tr>\n
Rust<\/td>\ntry_join!<\/td>\ntuple<\/td>\nfail immediately<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n

Python’s asynchio.gather<\/code> lets you choose whether a failed coroutine causes gather<\/code> to fail immediately or to wait for others before failing. The default is to fail immediately.<\/p>\n

The problem we saw was that the C++\/WinRT when_all<\/code> fails immediately, but the code wanted it to wait for the others before failing. This is a potentially useful general pattern, so let’s try to write our own when_all_completed<\/code> function.<\/p>\n

It’ll take a few tries to get there.<\/p>\n

The idea behind the function is simple. Here’s the pseudocode:<\/p>\n

template<typename... T>\r\nIAsyncAction when_all_complete(T... asyncs)\r\n{\r\n    std::exception_ptr eptr;\r\n\r\n    \/* Repeat for each element \"async\" of asyncs... *\/\r\n    try {\r\n        co_await async;\r\n    } catch (...) {\r\n        if (!eptr) {\r\n            eptr = std::current_exception();\r\n        }\r\n    }\r\n    ...\r\n\r\n    if (eptr) std::rethrow_exception(eptr);\r\n}\r\n<\/pre>\n
The idea is that we co_await<\/code> each of the passed-in coroutines, but do so inside a try<\/code>\/catch<\/code> block. If an exception occurs, then we save it, assuming we don’t have an exception already: The first exception thrown is the one that is reported. (Naturally, you can remove the if (!eptr)<\/code> if you want to report the last exception thrown.)<\/p>\n
The “repeat for…” part can be solved with  expansion statements<\/a>:<\/p>\n
template<typename... T>\r\nIAsyncAction when_all_complete(T... asyncs)\r\n{\r\n    std::exception_ptr eptr;\r\n\r\n    for... (auto& async : async) {\r\n        try {\r\n            co_await async;\r\n        } catch (...) {\r\n            if (!eptr) {\r\n                eptr = std::current_exception();\r\n            }\r\n        }\r\n    }\r\n\r\n    if (eptr) std::rethrow_exception(eptr);\r\n}\r\n<\/pre>\n
But before we clap the dust off our hands, note that expansion statements failed to be completed in time for C++20 and were postponed to C++23. And even if it were available in C++20, it will take time for projects to migrate off of C++17, so we’ll have to find a solution that at least works on C++17.<\/p>\n
We’ll start our explorations next time.<\/p>\n
Warning: There will be a lot of failure. If you’re looking for an answer to be handed to you, you’ll have to skip ahead to the end of the series, whenever that ends up happening.<\/p>\n
Bonus chatter<\/b>: Also, there’s a frustrating edge case in the above code, but I don’t want to try to fix it yet.<\/p>\n","protected":false},"excerpt":{"rendered":"
Remembering the exception while finishing the other work.<\/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-108373","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-oldnewthing","tag-code"],"acf":[],"blog_post_summary":"
Remembering the exception while finishing the other work.<\/p>\n","_links":{"self":[{"href":"https:\/\/devblogs.microsoft.com\/oldnewthing\/wp-json\/wp\/v2\/posts\/108373","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=108373"}],"version-history":[{"count":0,"href":"https:\/\/devblogs.microsoft.com\/oldnewthing\/wp-json\/wp\/v2\/posts\/108373\/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=108373"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/devblogs.microsoft.com\/oldnewthing\/wp-json\/wp\/v2\/categories?post=108373"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/devblogs.microsoft.com\/oldnewthing\/wp-json\/wp\/v2\/tags?post=108373"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}