Changelog

0.9.0

Breaking changes

• CsvDirFile: Stitched body lines stream in sorted file path order (aligned with read_dir). Canonical header sequence: expected_headers if set; else strict_headers uses the first sorted discovered file; else lexicographically smallest joined header. Match / error / skip behavior follows that path order — callers relying on the previous “emit first matching file out of order” behavior must adapt.

Additions and fixes

• read_dir / read_dir_chunks / CsvChunksDir: Non-positive chunksize raises ValueError.
• get_csv_paths: Recursive discovery returns an empty list when path does not exist (same as non-recursive).
• CsvDir: strict_headers pins the schema for the iteration without mutating expected_headers.

Packaging, tooling, and docs

• Require Python 3.10+; PEP 621 / SPDX metadata; typed package.
• CI: ruff, mypy, pytest on 3.10–3.13; mkdocs build --strict on docs.
• Remove unused iterators.py and chunks_iterators.py.
• read_dir overloads for clearer return types.
• Documentation on Read the Docs: MkDocs Material + mkdocstrings (guides, options reference, set vs sequence headers for dict readers vs CsvDirFile, pandas example).

0.8.0

• Name and path iterator options
• CsvDirFile for pandas compatibility
• Keyword-only extension, delimiter, and chunksize parameters

Earlier releases

See PyPI release history for versions 0.7.0 and below.