Skip to content

CHANGELOG.md

Latest commit

 

History

History
142 lines (112 loc) · 4.9 KB

CHANGELOG.md

File metadata and controls

142 lines (112 loc) · 4.9 KB

Changelog

All notable changes to nimsync will be documented in this file.

The format is based on Keep a Changelog, and this project adheres to Semantic Versioning.

[1.1.1] - 2026-02-22

Changed

  • Moved 10 experimental modules (actors, streams, group, etc.) to src/nimsync/experimental/
  • Backward-compatible shims kept at old paths — existing imports unchanged
  • Moved BENCHMARKS.md to docs/
  • nimble bench now points to tests/performance/run_all_benchmarks.sh

Fixed

  • Version inconsistency: VERSION.nim and nimsync.nimble now both reflect v1.1.0
  • Committed ELF binaries removed from git history (17 test/example binaries)
  • Stale benchmark scripts in scripts/ removed (referenced deprecated benchmarks/ dir)
  • Duplicate docs/LICENSE removed
  • Accidentally committed benchmark CSV results removed

Removed

  • Deprecated root benchmarks/ folder (contents superseded by tests/performance/)
  • src/nimasync_simple.nim — orphan file with typo in name, never exported

[1.1.0] - 2025-11-02

Added

  • MPSC (Multi-Producer Single-Consumer) channel mode
  • ChannelMode.MPSC enum value for multi-producer support
  • Wait-free MPSC algorithm using atomic fetchAdd (based on dbittman + JCTools)
  • Comprehensive MPSC benchmark suite (tests/performance/benchmark_mpsc.nim, 316 lines)
  • MPSC unit tests (tests/unit/test_mpsc_channel.nim, 259 lines)

Performance (Verified Benchmarks)

  • SPSC micro-benchmark: 558M ops/sec peak, 31ns P99 latency
  • SPSC realistic threaded: ~35M ops/sec
  • MPSC 2 producers: 15M ops/sec micro, ~15M realistic
  • MPSC 4 producers: 8.5M ops/sec
  • MPSC 8 producers: 5.3M ops/sec (memory-bandwidth limited)
  • Key finding: SPSC is 3.5× faster than MPSC in realistic threaded workloads

Changed

  • Updated README with MPSC usage examples and honest performance data
  • Updated Nimble badge to v1.1.0
  • Expanded performance documentation with micro vs realistic comparison

Implementation Details

  • Wait-free producer coordination via atomic CAS
  • Cache-line padding prevents false sharing
  • ORC-safe memory management, zero GC pauses
  • Single consumer uses relaxed atomic operations

[1.0.0] - 2025-11-02

Added

  • Production-ready SPSC channels with verified performance
  • 7-benchmark validation suite (Tokio, Go, LMAX Disruptor methodologies)
  • Latency profiling: 20ns p50, 31ns p99, 50ns p99.9
  • Published to official Nimble registry
  • GitHub issue templates for contributors

Performance (Verified)

  • Peak micro-benchmark: 558M ops/sec
  • Average across suite: 551M ops/sec
  • Realistic threaded: ~35M ops/sec
  • P99 latency: 31ns
  • 0% contention under stress (500K ops)
  • Async overhead: 512K ops/sec (Chronos wrappers)

Documentation

  • BENCHMARKS.md with methodology
  • Performance validation in tests/performance/
  • Roadmap updated to show v1.0.0 complete
  • Honest disclaimers about micro vs realistic performance

Fixed

  • Added test binaries to .gitignore

[0.2.1] - 2025-11-01

Fixed

  • Async channels now actually work - Previous implementation was completely broken
    • Fixed send()/recv() to use actual SPSC implementation (was referencing non-existent fields)
    • Removed duplicate Channel/ChannelMode type definitions
    • Added exponential backoff (1ms → 100ms) to reduce CPU usage
    • Fixed Chronos deprecation warnings
  • Deduplicated channels.nim - Removed 100+ lines of duplicate code
    • Now properly imports from channel_spsc instead of reimplementing everything

Added

  • Comprehensive async channel tests (tests/unit/test_async_channel.nim)
  • Added async tests to CI workflow

Changed

  • Updated KNOWN_ISSUES.md to reflect async improvements

[0.2.0] - 2025-11-01

Reality Check Release

Downgraded from v1.0.0 to v0.2.0 to accurately reflect what's production-ready.

Production-Ready

  • SPSC Channels: 212M+ ops/sec verified
    • Tests passing: tests/unit/test_channel.nim
    • Benchmark verified: tests/performance/benchmark_spsc.nim

Removed from Public API

  • TaskGroup (has bugs, removed from exports)
  • Cancellation (untested, removed from exports)
  • All other features remain internal/experimental

Changed

  • Version: 1.0.0 → 0.2.0
  • README.md: Replaced with honest version
  • Build: "production" → "experimental"

Added

  • STATUS.md: Feature audit
  • CI workflow: Test + benchmark validation
  • CHANGELOG.md: This file
  • KNOWN_ISSUES.md: Documented bugs and limitations

Known Issues

  • Async send/recv use polling (exponential backoff)
  • TaskGroup: nested async macro bugs (not exported)
  • Actors: won't compile, needs MPSC
  • Streams: compiles but untested
  • Scheduler: fake (just metrics)
  • NUMA: broken node detection

See KNOWN_ISSUES.md for complete list.

[1.0.0] - RETRACTED

Premature release. Use v0.2.0 instead.

Retracted because:

  • Claimed features that don't work
  • TaskGroup exported but broken
  • Examples showed inaccessible API
  • No CI validation