Changelog¶

All notable changes to this project are documented in the CHANGELOG.md file on GitHub.

Latest Releases¶

Added: The CLI supports sequential record-count splitting with --split N INPUT OUTPUT.
Added: load_archive and the CLI now auto-discover .jsonl and .ndjson archive members, including supported compressed variants. Explicit member patterns remain unchanged.
Breaking Change: dump_fork and dump_archive now limit open destination files to 64 by default. When the limit is reached, destinations are closed and reopened in append mode (at/ab). Custom opener functions must support append mode for destinations that may be reopened; pass max_open_files=None to retain the previous unbounded behavior.
Fixed: Auto-detect archives from remote URLs in the CLI without --member

Fixed: dump_archive rejects archive member paths that escape the staging directory.
Fixed: dump_archive supports archive filenames containing multiple dots.
Fixed: load detects compressed data consistently for paths, URLs, and binary file-like objects.
Fixed: File APIs consistently reject ambiguous bytes paths with a clear error.

Added: jsonl CLI with a json-style interface: stdin/stdout streaming, automatic compression, URLs, archive member auto-discovery and --broken record handling with well-defined exit codes.

Added: loads - Deserialize a JSON Lines formatted string into an object iterator.
Added: More tests to cover 100% of the codebase.
Added: CONTRIBUTING.md
Changed: Documentation - rewrite README and documentation for clarity and consistency
Changed: Documentation - Remove redundant sections and fix code examples
Fixed: warnings in CI/CD workflows

Breaking Change: Increase similarity with Python standard library json module behavior:
Now load and load_archive accept cls and **kwargs for custom decoding instead of json_loads and **json_loads_kwargs
Now dump, dump_fork, dumps and dump_archive accept cls and **kwargs for custom encoding instead of json_dumps and **json_dumps_kwargs

Added: More tests to cover 100% of the codebase.
Fixed: load Use context manager to close TextIOWrapper stream when loading from a URL.

Improved: load_archive from URL no longer loads the file into RAM, reducing memory consumption. Instead, the data is efficiently downloaded in chunks to a temporary file for loading.

Improved: Moved internal _openers variable to module level, eliminating per-call dictionary allocation.
Improved: loader by caching isinstance(line, bytes) check, evaluating only once instead of per iteration.
Improved: dump_archive worker using os.makedirs(exist_ok=True) to eliminate race conditions and reduce syscalls.
Improved: Streamlined dump function by combining isinstance(file, (str, os.PathLike)) check with os.fspath.
Improved: Removed redundant iter() call and improved _looks_like_url code structure for better readability.

Changed: Documentation - Use zensical instead mkdocs for documentation generation.

Improved documentation examples for json.dump and json.load keyword arguments.
Enabled additional ruff linting rules and passed all checks.
Enhanced docstrings with full sphinx-linter compliance.

Improved compression detection: load_archive and load now use magic-number detection as fallback.

For the complete history, see the full changelog on GitHub.