cosmic-crunch 2.1.2

Adds distribution-validation tooling; CI and publish workflow hardening.

cosmic-crunch 2.1.1

Review-driven fix batch: --year_regex/--date_regex now survive multiprocessing spawn/forkserver (previously silently ignored with --processes > 1 on macOS/Windows), request timeouts on every HTTP call, alternation-safe filter patterns, non-zero exits when convert finds nothing or errors, txt->nc rewrite anchored to exact segments, and support for no-L2 archive layouts. Plus the Read the Docs site shipped in-tag and a README "Use from Python" section with the xarray group read-back.

Full details: CHANGELOG.md

v2.1.0

Added

• Optional lossless zlib compression of netCDF4 output: --compress / --complevel flags (on get and convert) and compress/complevel params on the conversion functions. Off by default (HDF5 per-variable overhead makes it pay off only for long profiles; crossover ≈ 1000 levels).
• .zenodo.json + author ORCID in CITATION.cff for a citable Zenodo DOI.

See CHANGELOG.md.

v2.0.2

Docs & packaging release — no code changes.

• New "Problems this solves" section in the README mapping common queries (bulk-downloading COSMIC-1 radio-occultation data, converting Level-2 ASCII profiles to netCDF4)
• Enriched packaging metadata: trove classifiers, expanded keywords, Source and Bug Tracker project URLs

See CHANGELOG.md for details.

v2.0.1

[2.0.1] — 2026-07-08

Docs/packaging only — no code changes.

Changed

• README now leads installation with pip install cosmic-crunch (the package is
  published on PyPI), keeping the editable install as the contributor path. The
  v2.0.0 PyPI page had been built before this fix and still showed
  pip install -e .; this release republishes with the corrected instructions.
• Repo/project URLs point at the renamed cosmic-crunch GitHub slug.

v2.0.0

[2.0.0] — 2026-07-08

The 2020 JPL site restructure broke v1 (it silently downloaded nothing), and v1
carried latent security and correctness defects. v2 is a modernization: same
purpose, packaged, tested, and fixed.

Added

• Packaged as the cosmic_crunch distribution with a single cosmic-crunch
  console entry point (get and convert subcommands) via pyproject.toml
  (hatchling); requires Python 3.10+.
• --base-url flag and COSMIC_CRUNCH_BASE_URL environment variable to override
  the crawl root.
• --instrument flag (default cosmic) to select other instrument trees
  (champ, gracea, gracefo1, …).
• Atomic, resumable downloads: stream to <name>.part then rename; skip files
  already present with a matching size.
• Offline pytest suite and a GitHub Actions CI workflow (ruff + pytest on
  Python 3.10–3.13 + build/twine check).
• CHANGELOG.md and CITATION.cff.

Changed

• New crawl root https://genesis.jpl.nasa.gov/ftp/glevels/ (the v1
  /ftp/pub/genesis/glevels root is dead).
• Fail loud: a crawl that finds zero data files now raises and exits
  non-zero instead of silently "succeeding" with nothing downloaded.
• Bounded retry: the v1 infinite silent retry is replaced with a bounded
  retry (default 3 attempts, exponential backoff, logged, re-raises on failure).
• Relicensed from AGPL-3.0 to MIT.
• Logging is configured only under the CLI, not at import time (importing the
  package no longer creates a stray log file).
• Deduplicated the copy-pasted parallelize/flatten/retry helpers into
  cosmic_crunch/_parallel.py; single-sourced __version__.

Fixed

• Security: removed eval() of downloaded header content (arbitrary code
  execution); header values are parsed with ast.literal_eval + raw-string
  fallback, preserving brace-set element order.
• Security: path traversal in downloads — a hostile or tampered directory
  listing could smuggle ../ through a filename and write outside the save
  directory. Downloaded filenames are now reduced to a bare leaf and confined
  to the save root (irreducible unsafe names are rejected before any fetch).
• The v1 FILES_TO_GET = -1 default silently dropped the last crawled file on
  every non-test run ([:-1]); it now defaults to downloading everything.
• crawl_convert processed only the first path argument (a return inside the
  per-path loop, plus a paths shadowing typo) — it now converts all paths.
• Column-name loop-leak: rows of DataTypes with differing field counts were
  parsed with the wrong column names — each type now keeps its own columns.
• Single-file conversion creates its output directory; directory creation uses
  os.makedirs(exist_ok=True).

Migration

• The invocation changed: python get_files.py … / python convert_files.py …
  become cosmic-crunch get … / cosmic-crunch convert …. This is the breaking
  change behind the major version bump. There are no deprecation shims (the tool
  was never published and has no known downstream users).