2.5.1

Fixes task completion dispatch and improves Pusher-compatible publishing reliability, benchmark tooling, and test cleanup.

• Avoid dispatching unused task-finish events while preserving registered listeners and message handling.
• Respond before Pusher fanout to prevent an OpenSwoole coroutine deadlock.
• Add reproducible paired benchmarks, resource capture, trial validation, and safer process cleanup.
• Fix test counters for OpenSwoole 25 and 26 compatibility.

Benchmark results

Five valid trials per server measured median REST publishing throughput of 887.57 requests/sec for Conveyor and 649.22 for Laravel Reverb (1.37×) with 100 users, six Locust workers, and two-minute trials on a shared host; variation limits this result to the measured workload, rather than general server capacity.

Full benchmark results and environment · Methodology · Benchmark scripts

Raw trial artifacts are retained locally and are not bundled in this release.

Validation: 69 tests and 267 assertions, with PHPStan, Composer validation, shell checks, and independent review recorded in PR #3; hosted Tests passed on the release commit.