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

Support for comparing tar archives (.tar, .tar.gz, .tar.bz2, .tar.xz) in addition to zip archives and plain directories
Unit test suite (pytest) covering the core comparison logic and the CLI
hatch test / pytest integration in hatch run lint:check

Migrated packaging from setup.py to pyproject.toml with hatch as build backend
Raised minimum supported Python version to 3.11 (developed and tested with 3.13)
Added a restrictive lint environment (black, bandit, ruff, flake8, pylint, mypy --strict) via hatch run lint:check
Added GitHub Actions workflows for linting across supported Python versions and for publishing to PyPI via Trusted Publishing
Relicensed from GPL-3.0-or-later to MIT
Unreadable files (broken symlinks, missing permissions) are now skipped with a warning on stderr instead of aborting the whole comparison
folderdiff CLI now prints a clean error message and exits with status 2 instead of a raw traceback when the comparison itself fails

Files with duplicate content could be silently dropped from the result when computing moved/removed/added file sets, hiding genuinely removed or added files whenever another file with identical content existed
--prefix had no effect when comparing two plain directories; it only ever worked for the zip-archive side
A corrupt zip archive's CRC-32 self-test result was computed but never checked, so a bad archive still crashed later with an unhandled exception
Fixed "profix" typo in the --prefix help text
Folder-mode relative paths used os.sep while zip entries always use /, so comparing a zip archive against a directory would report every file in a subfolder as changed on platforms where os.sep != "/"
A named pipe or other special file in the compared directory tree caused open() to block indefinitely, hanging the whole comparison; such entries are now skipped with a warning instead
Zip archive members were read fully into memory before hashing, which could exhaust memory on a maliciously crafted archive; hashing is now streamed in fixed-size chunks

initial release