io_uring 0.1.1

A pure-Dart, syscall-level interface to Linux io_uring.

0.1.1

• UringLoopPool / UringLoopLease: exclusive, reference-counted loop leases with idle-loop reuse, reaper lifecycle, close coordination, and reaper error propagation.
• UringLoop.acquire(entries: ..., spin: ...) convenience for short-lived callers using the shared pool.
• Optional low-latency busy-poll reaper (spin: true): trades a CPU core for much lower async QD1 wake-up latency.
• Async completion path uses synchronous completers and the direct List<int> message fast path.
• Benchmarks cover spin-mode async I/O and pool acquire/release; chart uses an honest linear scale.
• Dartdoc coverage raised above the pub.dev 20% threshold (previously 17.2%), restoring the documentation score for v0.1.1.

0.1.0

• Initial implementation: ring setup/teardown, SQ/CQ submission, NOP/READ/ WRITE/FSYNC, OPENAT/CLOSE/STATX/READV/WRITEV/FTRUNCATE/FALLOCATE.
• Registered buffers, registered files, eventfd registration and kernel probe.
• Network operations: SOCKET/BIND/LISTEN/ACCEPT/CONNECT/SEND/RECV/SHUTDOWN, with a loopback TCP echo test and example.
• SPLICE/TEE with pipe helpers and tests.
• Links, relative/absolute timeouts, timeout removal, link timeouts, async cancellation, poll add/remove, provided buffers and multishot receive.
• IoUringBuilder with SQPOLL/IOPOLL/CQ-size/taskrun flags and SQPOLL wakeup.
• Registered-file updates (registerFilesUpdate / FilesUpdateOp) and provided-buffer rings (registerBufRing / UringBufRing).
• Future-based UringLoop: owner submits directly, a reaper isolate blocks in io_uring_enter and completes Futures.
• C vs Dart benchmark suite: raw-syscall C ring and benchmark_test AOT Dart suite, with a comparison runner (benchmark/compare.dart).
• bench-svg integration: canonical JSON + benchmark/bench-svg.toml render assets/benchmarks.svg, embedded in the README.
• CI benchmark job after tests: runs C/Dart benchmarks, emits canonical JSON, renders the SVG with unk-404/benchmark_svg@master (config-defined series, no positional input), uploads artifacts, and commits the refreshed chart back to the branch.
• Hot-path performance work: 8×64-bit SQE zeroing, cached SQE Struct views, x86_64 barrier-fence elision (TSO), removed redundant CQ-head stores, and reapCqes raw batch reaping without per-completion object allocation.
• Native buffers, raw-fd helpers (openat/close/eventfd/read), and a small memory-barrier abstraction.
• x86_64 support; aarch64 ring publication uses libatomic fences with a pthread fallback and is exercised by CI.
• Benchmark harness (benchmark/run.dart) with NOP throughput, QD1 NOP latency, and QD1 file-read metrics; CI builds and runs it on both architectures.