Skip to content

Releases: dnncha/turbo-picard

turbo-picard 0.1.6

Choose a tag to compare

@dnncha dnncha released this 23 Jun 12:38

turbo-picard 0.1.6

This release promotes the algorithmic-overhaul branch with refreshed parity-gated benchmark evidence and the CollectGcBiasMetrics parity fix.

Highlights

  • Fixes CollectGcBiasMetrics detail-output parity for MEAN_BASE_QUALITY against Picard 3.4.0.
  • Refreshes public benchmark evidence from the 2026-06-23 median-of-5 suite run.
  • Corrects benchmark summary median calculation for even command counts.
  • Keeps the Picard-shaped command surface and fallback model unchanged.

Benchmark evidence

Fresh suite command:

TURBO_PICARD_CONDA_PREFIX=/opt/homebrew/Caskroom/miniforge/base/envs/.conda-turbo-picard \
python3 tools/bench_suite.py --repeats 5 --skip-build \
  --profile-output benchmarks/runs/full-20260623-130923/bench-suite-profile.json

Summary:

  • 32/32 benchmarked commands passed parity checks.
  • Top speedup: 75.77x on NormalizeFasta.
  • Floor speedup: 8.01x on RevertSam.
  • Median command speedup: 26.26x.
  • Geometric mean speedup: 26.67x.

Packaging

  • PyPI is published at 0.1.6 with the macOS Apple Silicon wheel plus source distribution.
  • PyPI wheel SHA-256: c183a9355f8462d71494864a2299fc3822a6129aa7ae98b313684c8b652da667.
  • PyPI source distribution SHA-256: c2c38ad03c7555b7d719e1715964282de647971031f651552a2c5700caeb2e85.
  • Bioconda recipes are staged for 0.1.6 with the immutable GitHub source archive SHA: ba6708a025e5e53e85a1637a5e28dc31d40f72c0d2fcc1fdb5e3e728f1981a5c.
  • The Bioconda channel package is still pending Bioconda PR review/merge.

v0.1.5

Choose a tag to compare

@dnncha dnncha released this 16 Jun 15:30

v0.1.5

Scientific correctness and release maintenance update.

Fixes

  • Fix WGS overlap bitmap lookups so mate-overlap exclusion checks the exact locus bit instead of any lower bit in the word.
  • Report real CollectGcBiasMetrics MEAN_BASE_QUALITY values in detail rows for both ALL and UNIQUE read sets.
  • Treat SAM SEQ=* and QUAL=* as missing values in SAM-text metrics collectors instead of counting the literal * sentinel as biological data or a base quality.
  • Recompute ValidateSamFile NM expectations correctly for = and X CIGAR operators.
  • Treat empty SAM read-group PL: values as missing platform values in SAM-text validation.

Release checks

  • cargo test --workspace
  • Benchmark evidence verifiers: log evidence, suite coverage, thresholds, README evidence, and site evidence.
  • Real-data release evidence verifier.
  • PyPI wheel and sdist built locally and passed twine check; wheel smoke-tested with turbo-picard --version, picard --version, and AccelerationStatus --help.

turbo-picard v0.1.4

Choose a tag to compare

@dnncha dnncha released this 13 Jun 19:36

Highlights

  • Refreshed release benchmark evidence from python3 tools/bench_suite.py --repeats 3 --skip-build on 2026-06-13.
  • Saved suite remains 32/32 PASS against Picard 3.4.0.
  • Current saved benchmark summary: 94.36x top speedup, 6.86x floor speedup, 26.72x median speedup, 24.94x geometric mean speedup.
  • Added targeted benchmark selection with tools/bench_suite.py --only ... for focused performance regression audits.
  • Updated README, docs, marketing site assets, and Bioconda/PyPI release metadata for 0.1.4.

Verification

  • python3 tools/bioconda_release_preflight.py
  • python3 -m unittest discover tools
  • cargo check --workspace

Bioconda source archive SHA-256 for https://github.com/dnncha/turbo-picard/archive/refs/tags/v0.1.4.tar.gz: 8878603e00945f575aa7f6846f0658554229403b9165024fc59b6e7b3015fe4e.

turbo-picard v0.1.3

Choose a tag to compare

@dnncha dnncha released this 12 Jun 20:32

turbo-picard v0.1.3

Public-release hardening release for the current turbo-picard evidence set.

Highlights

  • Adds doctor and explain utility commands to make runtime/fallback state and native command coverage easier to audit.
  • Tightens CollectWgsMetrics parity against Picard 3.4.0, including default BQ histogram behavior and theoretical sensitivity sampling behavior.
  • Keeps the full benchmark suite at 32/32 passing commands. Latest local benchmark run showed median speedup 29.61x and geometric mean speedup 28.13x against Picard, compared with saved public values of 26.82x and 26.74x.
  • Removes internal notes, host-specific benchmark runners, and generated evidence work products from the public tree while preserving compact public evidence metadata.
  • Refreshes locked transitive crate versions so cargo test --workspace --locked resolves cleanly against the current crates.io index.

Bioconda

This release is linked to the pending Bioconda submission:

The public repo includes updated Bioconda recipe metadata for v0.1.3 and source archive SHA-256:

5aa23e085628114c168ddf9e7df94b5762769c3e6aed2b571eae136d380b0b28

Verification

  • cargo test --workspace --locked
  • python3 -m pytest tools
  • release verifier chain including benchmark evidence, parity docs, real-data evidence, site/readme links, release versions, and Bioconda recipe checks
  • python3 tools/bioconda_release_preflight.py

turbo-picard v0.1.2

Choose a tag to compare

@dnncha dnncha released this 11 Jun 10:29

Updates Bioconda packaging to the current release state and includes the improvements shipped after v0.1.1.

Notable changes since v0.1.1:

  • improved CMM pipeline handling, options, tests, and docs
  • added SamToFastq unpaired FASTQ output support
  • added RevertSam fast paths and BGZF thread handling
  • refreshed parity comparators, benchmark evidence, PyPI packaging, and release checks

turbo-picard v0.1.1

Choose a tag to compare

@dnncha dnncha released this 04 Jun 12:06

Patch release focused on RevertSam performance and release evidence.\n\nDOI: https://doi.org/10.5281/zenodo.20541928\n\nChanges:\n- RevertSam can now stream unsorted output, and queryname output when records are already queryname-monotonic. If query names are not monotonic, it falls back to the existing sort path.\n- Saved benchmark evidence refreshed on 2026-06-04: 32/32 parity checks passing, 8.55x floor speedup, 26.74x geometric mean speedup, 84.46x top speedup.\n- Version, citation metadata, site assets, and Bioconda recipe templates bumped to 0.1.1.\n\nVerification run before release:\n- cargo test -p turbo-picard-cli revertsam -- --nocapture\n- python3 tools/verify_benchmark_log_evidence.py\n- python3 tools/verify_benchmark_suite_coverage.py\n- python3 tools/verify_benchmark_thresholds.py\n- python3 tools/verify_readme_benchmark_evidence.py\n- python3 tools/verify_site_benchmark_evidence.py\n- python3 tools/verify_release_versions.py\n- python3 tools/verify_real_data_evidence.py --release-ready

turbo-picard v0.1.0

Choose a tag to compare

@dnncha dnncha released this 01 Jun 09:47

First public packaging release for turbo-picard.

Highlights:

  • Native Rust command surfaces for selected Picard-compatible SAM/BAM, metrics, FASTA, VCF, and interval workflows.
  • Separate optional turbo-picard-picard-shim package recipe for users who deliberately want a picard compatibility entrypoint.
  • CITATION.cff, release evidence, benchmark disclosures, and Bioconda-ready recipes.

Bioconda submission: bioconda/bioconda-recipes#65922

Source archive SHA-256 used for Bioconda: 95923bebbc7f6ab59e73c436b31d84c8da547939c1e6c63be984747acfbc387c