Skip to content

CHANGELOG.md

Latest commit

 

History

History
57 lines (40 loc) · 3.96 KB

CHANGELOG.md

File metadata and controls

57 lines (40 loc) · 3.96 KB

Changelog

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

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

[Unreleased]

Added

  • Advisory pull-request benchmark regression check (Tier 1/2/3 on twitter and canada fixtures, advisory-only). The future blocking-flip is gated by the REQUIRE_NO_REGRESSION env var in .github/workflows/pr-benchmark.yml.

[0.1.4] - 2026-04-27

Fixed

  • Linux release builds no longer depend on custom std::basic_string allocator behavior that is rejected by the manylinux libstdc++ toolchain.

Changed

  • Bootstrap default-install recovery now pins bootstrap.Version to 0.1.4 after the v0.1.3 tag failed before artifact publication.

[0.1.3] - 2026-04-27

Fixed

  • Linux release artifacts now hide non-public dynamic symbols so the exported surface stays limited to the public pure_simdjson_* ABI.
  • macOS release smoke no longer reports parser diagnostic string cleanup as an untracked native allocator free.

Changed

  • Bootstrap default-install recovery now pins bootstrap.Version to 0.1.3 after the v0.1.2 tag failed before artifact publication.

[0.1.2] - 2026-04-25

Added

  • Benchmark harness coverage for Tier 1 full materialization, Tier 2 typed extraction, Tier 3 selective placeholder reads, cold/warm parser lifecycle, and the JSONTestSuite correctness oracle.
  • Committed benchmark evidence under testdata/benchmark-results/v0.1.1/, including phase7.bench.txt, coldwarm.bench.txt, and tier1-diagnostics.bench.txt.
  • Committed linux/amd64 benchmark baseline evidence under testdata/benchmark-results/v0.1.1-linux-amd64/ for future CI-based claim gates.
  • Committed testdata/benchmark-results/v0.1.2/ evidence and the claim-gated docs/benchmarks/results-v0.1.2.md benchmark snapshot.
  • Bootstrap/ABI release-state checks that reject stale bootstrap pins, Go/Rust ABI mismatches, and unknown ABI policy values before tag publication.
  • Repo-root LICENSE and NOTICE files covering the project MIT license and vendored simdjson attribution.

Changed — ABI

  • PURE_SIMDJSON_ABI_VERSION bumped from 0x00010000 to 0x00010001. Consumers linking against include/pure_simdjson.h must rebuild.
  • pure_simdjson_native_alloc_stats_t gained two fields: epoch (first, so callers can detect a pure_simdjson_native_alloc_stats_reset() race between snapshots) and untracked_free_count (last, incremented when the native allocator hook observes a free for a pointer it did not record — surfaces double-frees and stray frees that previously went silent). Field order is now [epoch, live_bytes, total_alloc_bytes, alloc_count, free_count, untracked_free_count], all uint64_t.

Changed

  • Releases now follow the org-standard tag-driven CI publish flow shared with pure-onnx and pure-tokenizers.
  • Bootstrap default-install alignment now pins bootstrap.Version to 0.1.2 for artifacts carrying ABI 0x00010001, while source checksums remain empty and runtime verification continues to use published SHA256SUMS.
  • Benchmark positioning now treats Tier 1 as the full-any worst-case workload on the current DOM ABI, while deferring ABI-level Tier 1 improvement and any new public patch-release decision to Phase 8 and Phase 9.
  • README benchmark-positioning now uses the linux/amd64 v0.1.2 evidence and keeps third-party comparator detail in docs/benchmarks/results-v0.1.2.md.

Documentation

  • Added a consumer-facing README.md with installation, quick start, supported platforms, and a benchmark snapshot linked to docs/benchmarks/results-v0.1.1.md.
  • Added docs/benchmarks.md to document tier definitions, omission rules, cold-start semantics, native allocator metrics, and rerun commands.

[0.1.0] - 2026-04-21

Added

  • Initial bootstrap distribution pipeline, five-platform release matrix, staged smoke gates, and release-preparation scaffolding for v0.1.0.