Skip to content

Changelog

All notable changes to this project will be documented in this file. The format is based on Keep a Changelog, and this project adheres to Semantic Versioning.

Unreleased

Added

  • Codecov badge to the README so coverage is visible at a glance from the repository landing page and PyPI listing.

Changed

  • Default CLI output no longer shows the log-level prefix (INFO, DEBUG, etc.) in front of every line. The level prefix is now only rendered when --debug is set, where the distinction between DEBUG and INFO actually matters. Plain runs print bare message text.

0.3.1 - 2026-05-05

0.3.0 - 2026-05-05

Added

  • fill keyword argument and --fill CLI option to substitute a placeholder string for cells that have no matching row in the input.
  • keep_duckdb keyword argument and -k / --keep-duckdb CLI option that persists the staged input as a DuckDB database file at <input>.duckdb, so users can re-query the data with ad-hoc SQL after the pivot finishes.

Changed

  • click is now pinned to >=8,<9 because crosstab.cli.OptionEatAll reaches into Click's private parser API (parser._long_opt, _short_opt, our_parser.process) to implement argparse-style variadic options. The pin makes a future Click 9 release require explicit verification rather than silently breaking the CLI.

Removed

  • The "README" sheet that previously appeared as the first sheet of every output workbook. It was leaking the runner's local username and absolute filesystem paths into shared workbooks, and the Crosstab and optional Source Data sheets are sufficient for downstream use.
  • keep_sqlite keyword argument and the -k / --keep-sqlite CLI flag. These have been deprecated since 0.2.0 and emitted a DeprecationWarning; they are now removed entirely. Replace with keep_duckdb.
  • The _current_user helper, getpass/os/datetime imports, and the tox passenv workaround that existed solely to power the now-removed README sheet's metadata fields.

0.2.1 - 2026-05-04

Fixed

  • Docker image build. The previous Alpine-based image (python:3.13-alpine) failed to install DuckDB because DuckDB only ships prebuilt wheels for manylinux (glibc), not musl, and the Alpine builder has no C++ toolchain to compile from source. Switched both stages of the Dockerfile to python:3.13-slim (Debian) so the prebuilt DuckDB wheel is used.

0.2.0 - 2026-05-04

Changed

  • Replaced the argparse CLI with a rich-click app in crosstab.cli and routed the crosstab console-script entry point at it. Help output and log output use Rich; validation errors come through Click with a non-zero exit code. The multi-value flags -r, -c, and -v keep the original space-separated UX (e.g. -r loc sample) via a small custom OptionEatAll Click option subclass.
  • Replaced the SQLite-backed engine with a single-pass DuckDB pivot. CSV input is now read directly via read_csv(..., all_varchar=True), duplicates are detected with one GROUP BY ... HAVING COUNT(*) > 1 query, and the result is materialized in a single aggregate query rather than O(R × C) per-cell SELECTs. On a 100,000-cell pivot this is roughly two orders of magnitude faster.
  • Replaced openpyxl with XlsxWriter for output. XlsxWriter is write-only and significantly faster and lower-memory for large workbooks; openpyxl remains a development dependency for reading workbooks in tests.
  • Centralized SQL identifier quoting in _quote_ident(). Headers containing spaces, parentheses, embedded double quotes, unicode, leading digits, or SQL reserved words (e.g. Depth Interval (ft), select, Param "raw") now pass through end to end without being rewritten or breaking the engine.
  • Row keys and column keys are now sorted before output, so re-running the same input produces a byte-identical workbook (modulo the README timestamp).

Deprecated

  • The keep_sqlite keyword argument and the -k / --keep-sqlite CLI flag are deprecated and ignored — the engine no longer uses SQLite. Passing keep_sqlite=True emits a DeprecationWarning. The argument will be removed in a future release.

Added

  • pytest-benchmark is a development dependency, with one baseline benchmark covering a 100,000-cell pivot.
  • Test coverage for _quote_ident (parametrized over awkward identifier shapes) and an end-to-end round-trip test that exercises every problematic header category at once.

0.1.0 - 2026-05-04

Changed

  • Switched build backend from setuptools to hatchling and adopted the src/crosstab/ package layout.
  • Refreshed development toolchain to match geocoug/execsql and geocoug/pg-upsert: ruff 0.15.x, pre-commit hooks (gitleaks, uv-lock, mdformat, markdownlint, typos, ruff, validate-pyproject), tox-uv, and branch coverage with a 100% floor (defensive unreachable branches are marked # pragma: no cover).
  • Expanded the test suite from 8 to 26 tests, covering all argument validation paths, the __repr__, the keep_src=False path, duplicate row/column detection, and the full CLI surface (--version, --debug, --log, error exit codes).
  • Replaced MkDocs configuration with zensical, with mkdocstrings-python for auto-generated API reference.
  • Replaced Makefile with a justfile for development tasks.
  • Rewrote the GitHub Actions workflow: separate lint job, 3-OS × 5 Python-version test matrix, OIDC PyPI publishing, and auto-generated GitHub releases on tag push.
  • __version__ is now read from package metadata via importlib.metadata.version("crosstab") rather than hard-coded in the source file. Versioning is driven by bump-my-version, which keeps pyproject.toml and this CHANGELOG.md in sync.

0.0.15 and earlier

See the project's git history for changes prior to the 0.1.0 toolchain revamp.