Conversation
Make the codecov project status a failing check (target auto, 0% threshold) instead of informational, so any drop in include/ + src/ coverage fails it. Requires the check to be marked required in branch protection. ci.yml already uploads coverage on every PR.
The join statement removes the race between thread_pool and main thread.
Several buffer free functions rendered their parameters as bulleted lists or not at all. MrDocs renders @param as a Name/Description table, and MrDocs added first-class function-object (niebloid) support, so the old workarounds are no longer needed. - buffer_slice: convert the @Par Parameters/@Li list to @param tags; document the deleted rvalue (const&&) overload. - when_any: add @param to the variadic overload. - front, buffer_copy, begin, end, buffer_size, buffer_empty: restore the original anonymous-struct niebloid form (drop the obsolete *_mrdocs_workaround_t type names), mark each with @functionobject, and move @param/@return onto each operator() so MrDocs renders an "Algorithm Function Object" page with a parameter table. ready already used @param and renders correctly; left unchanged.
… ranges The generic mutable overload took an lvalue reference `make_buffer(T&)`, so a `boost::span<char>` passed as a temporary could not bind to it and fell through to the const overload, returning a `const_buffer` (#147). `std::span<char>` only escaped this via a dedicated by-value overload. Change the generic mutable overloads to a forwarding reference `make_buffer(T&&)` so by-value mutable contiguous ranges bind there and return `mutable_buffer`. Remove the now-redundant explicit C-array, std::array, std::vector, std::basic_string, and std::span overloads, which the concept-based generic overloads fully subsume; prune the includes they required. Behavior is preserved for all non-empty inputs; for empty array/ const-string/span the returned (size-0) buffer's data pointer is now normalized to nullptr, matching vector/string. Add a prvalue std::span regression test plus a boost::span test guarded by __has_include so Capy still builds and passes without Boost present.
The run_async trampoline's unhandled_exception swallowed every escaping exception, so an unhandled task error vanished silently and the documented "exceptions are rethrown" behavior never happened. Make the trampoline call std::terminate instead: an exception reaches it only by escaping a handler (a handler that threw, or the default handler on an otherwise-unhandled task error), which is a genuine fault with no owner to receive it. Failing fast is loud and leak-free (the process ends), and needs no executor cooperation. To observe an error instead, pass an error handler (it receives the exception_ptr); to catch one, co_await the work inside a coroutine. Cooperative cancellation must not be treated as a fault: the default handler discards a stop_requested_exception (quitter's stop sentinel) so a stopped quitter completes silently rather than terminating. The result-only handler delegates to the same logic. Document the new behavior and exclude the terminate lines from coverage. Add a POSIX fork-based death test covering both terminate paths (no handler, and a rethrowing handler); it is guarded out on Windows.
read and write reported any error_code surfaced by the final read_some/write_some, even when that completion delivered the last bytes and filled the buffer. A generic observer (when_any/when_all) then saw a completed transfer as a failure. The most common trigger is end-of-stream coincident with a full read (read_some returns [eof, n] that fills the buffer). Report the contingency only when the buffer was not fully transferred: when n == buffer_size(buffers) the transfer succeeded and ec is success. Tighten the documented postcondition accordingly. Codify the same invariant in the ReadStream and WriteStream concept contracts: a read_some/write_some that fills its buffer reports success, so n == buffer_size(buffers) implies no error (a coincident condition such as end-of-stream surfaces on a subsequent call). Add a mock stream that reports a contingency in the same completion that transfers bytes (the test read/write streams report errors with zero bytes) and cover both the filled-is-success and short-transfer-is-reported cases.
Repo wg21-papers was reorganized, link bekame stale.
The macos-26 runner ships Xcode 26.5 / Apple Clang 21, so the CMake build no longer needs the visibility workaround there. macos-15 (Apple Clang 17) is still affected and keeps the flag. The workaround only ever mattered for CMake builds; b2 defaults to hidden visibility and is unaffected. Closes #194
This is an attempt to formally describe the contract of functions that take the dynamic byffer ant the initial size hint.
…ror_condition Override default_error_condition in error_cat so each capy error maps to its canonical portable condition: canceled -> std::errc::operation_canceled, timeout -> std::errc::timed_out, and eof/stream_truncated/not_found to their cond::* equivalents.
when_any selects a winner by success (!ec); a failing child neither wins nor cancels its siblings. This was reported as a bug (#265) but is by design. Document the behavior explicitly and the two patterns for treating an error as a win, and replace the dangling "see Racing Tasks" reference (which pointed at an unpublished doc/unlisted/ page). - doc/.../4f.composition.adoc: add "Errors Do Not Win" and "Treating an Error as a Win" subsections under when_any; drop the dead reference. - include/boost/capy/when_any.hpp: add a @note to the variadic when_any docstring pointing readers to the patterns.
Add an "Asynchronous Sleep" section to the concurrent composition page that frames delay as the non-blocking counterpart to sleep_for. Cross-link the custom-IoAwaitable guide to delay_awaitable as a production example of the stop-callback-must-post pattern, and list delay.hpp/timeout.hpp in the page reference table.
Audited the Antora book and header Javadoc against the implementation and corrected every factual and conceptual discrepancy, treating the code as canonical. Key corrections: - read_some/write_some error contract: n < buffer_size on error, and a full buffer always means success (conditions deferred to the next read); rewrote the chapter 9 rationale and fixed the 9n header misquote - BufferSource/BufferSink: span-based pull/prepare plus consume, matching the real concepts (was pointer+count with integer returns) - cancellation: this_coro::stop_token and error::canceled (removed the nonexistent get_stop_token and error::operation_aborted) - dynamic buffers: fixed-capacity adapters over external storage built via pointer or the dynamic_buffer factory (was grow-on-demand) - any_* wrappers: owning by value vs reference by pointer (no S& ctor) - IoAwaitable/IoRunnable concepts, Executor concept, composed-op return types (io_task), and assorted symbol/path fixes Verified all 489 code examples compile against the real API. Fixed real breakage: mutable_buffer(array) to make_buffer, co_await std::move on a named task, the mmap_source io_result return, the fabricated test::stream(fuse) constructor, and buffer_to_string on bufgrind slices. Build instructions link capy with no external dependencies, and the thread_pool examples call join() to drain outstanding work.
The OS Integration section listed only IOCP and io_uring and presented io_uring as the Linux mechanism. Corosio's default backend selection is IOCP > epoll > kqueue > select; io_uring is opt-in, so Linux defaults to epoll, not io_uring. Describe cancellation by backend class instead: completion-based backends (IOCP, io_uring when enabled) cancel in the kernel, while readiness-based backends (epoll, kqueue, select) remove the pending operation from the reactor before its syscall runs. State the reported error, std::errc::operation_canceled, and drop the fragile per-syscall names.
|
An automated preview of the documentation is available at https://331.capy.prtest3.cppalliance.org/index.html If more commits are pushed to the pull request, the docs will rebuild at the same URL. 2026-06-21 16:23:03 UTC |
Codecov Report✅ All modified and coverable lines are covered by tests. Additional details and impacted files@@ Coverage Diff @@
## master #331 +/- ##
==========================================
- Coverage 98.09% 98.07% -0.03%
==========================================
Files 164 164
Lines 8873 8774 -99
==========================================
- Hits 8704 8605 -99
Misses 169 169
Flags with carried forward coverage won't be shown. Click here to find out more.
... and 15 files with indirect coverage changes Continue to review full report in Codecov by Harness.
🚀 New features to boost your workflow:
|
|
GCOVR code coverage report https://331.capy.prtest3.cppalliance.org/gcovr/index.html Build time: 2026-06-21 16:38:43 UTC |
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
5 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
No description provided.