Repository navigation
Publish to PyPI with Trusted Publishing; pin actions to SHAs - #348
Merged
Merged
Conversation
- All build jobs (generic sdist/wheel, macOS/Windows Cython wheels, manylinux cibuildwheel) run without secrets and upload artifacts - A single publish job, gated on every build succeeding, downloads all distributions and uploads via PyPI Trusted Publishing (OIDC) in the "pypi" environment; its only permission is id-token: write - No PYPI_USERNAME / PYPI_PASSWORD secrets are used anymore - Pin every action in both workflows to a full commit SHA - Default token permissions to none, checkout without persisted credentials, add job timeouts, twine check every build Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Merged
cdgriffith
added a commit
that referenced
this pull request
Nov 4, 2026
* Version fix * Fix issue #305: Merge dotted keys when non-dotted key with same prefix follows - Add test case for issue #305 documenting the bug - Fix __convert_and_store to merge dict values into existing Box when: - box_dots is enabled - key already exists as a Box (created from dotted keys) - during initialization (__created is False) - This ensures dotted keys like 'a.b: c' are preserved when followed by a non-dotted key like 'a: {d: e}' in YAML files * Preserve the Box subclass for nested boxes after unpickling Box.__init__ derives box_class from self.__class__, but Box.__new__ hardcoded the base Box. Unpickling calls __new__ + __setstate__ (not __init__), so a subclass instance ended up with box_class=Box, and nested boxes created on access after unpickling became base Box instead of the subclass. Use cls in __new__ to match __init__. * fix: preserve existing keys on dotted list writes BoxList.__setitem__ unconditionally replaced the element at the target index with a fresh empty container before recursing, so a second dotted write into an existing element (a[0].y after a[0].x) discarded the keys already stored there. Only replace the element when it is not already the right container type, matching the guard in Box.__setitem__. * fix: case-insensitive ConfigBox key lookup for __getitem__, __contains__, __delitem__, __setitem__ ConfigBox docstring advertises "loosey goosey" case-insensitive access but __getitem__ (dict-style), __contains__, __delitem__ and __setitem__ had no case-insensitive support. __getattr__ only retried with lowered input, missing MixedCase→lowercase resolution. Add _case_insensitive_key() helper and override the four dunder methods to resolve keys case-insensitively. __getattr__ now uses the same key-scan fallback instead of the broken .lower() retry. * Fix serialization of frozen_box with nested lists frozen_box stores lists as tuples of Boxes for immutability, but Box.to_dict() only recursed into Box and BoxList values -- not tuples -- so the nested Boxes were left unconverted. to_yaml() then failed with RepresenterError since ruamel can't represent a Box. Recurse into tuples in to_dict(), converting nested Box/BoxList members back to native types while preserving the tuple. Fixes #272. * Preserve box options in BoxList.__deepcopy__ `BoxList.__deepcopy__` built the copy as `self.__class__()` with no `box_options`, so `copy.deepcopy()` silently dropped every option (`frozen_box`, `box_dots`, `default_box`, `box_intact_types`, …). A frozen BoxList became mutable (and unhashable) after a deepcopy. `BoxList.__copy__` and `Box.__deepcopy__` both preserve options; `BoxList.__deepcopy__` was the outlier. Mirror `Box.__deepcopy__`: build unfrozen, register in memo before recursing (keeps cycle-safety), populate, then re-freeze via `__init__`. * Convert ruamel YAML scalars to builtin types on load * Test from_yaml returns builtin float/int/bool * Note #264 ScalarFloat fix in changelog * Fix nested boxes losing box_dots after pickle Pickle assigns items before __setstate__, so nested Box objects were recreated with default options and dotted lookups failed after load. * Add a merge option for box_merge_lists merge_update can already combine lists with extend and unique, but there was no way to merge lists of dictionaries the way nested dicts merge. Add a "merge" option that walks both lists by index, recursing into dicts at the same position so their keys combine and appending any extra incoming items. Closes #263. * Resolve CamelCase prefixes in dotted Box lookups * Test CamelCase paths, defaults and exact-key precedence * Document CamelCase path lookup fix and contributor * Flush pending default_box parents on assignment. default_box_create_on_get=False still needs b.foo.bar = 1 to persist. Lookups stay ephemeral. Writes attach the pending chain. * Protect frozen BoxLists against remaining in-place mutations * docs: correct optional dependency installation examples * Convert replacement values in BoxList assignments * Fix BoxList extending itself * Fix Box.popitem LIFO order * fix: report malformed BoxList bracket indices Guard unmatched index expressions in dotted reads, writes and deletes. Add regressions for nested paths, default/frozen modes and YAML loading. Independent review: passed (security and logic). AI-assisted contribution prepared with Hermes DEV. * Construct BoxList without testing the input truth value * docs: record JSON Lines string fix * fix: honor multiline JSON for strings * test: cover JSON Lines string round trips * Honor explicit pop defaults without creating missing values * Honor explicit ensure_ascii options in JSON output * Allow same-key updates with duplicate conversion checks * Accept source and keyword overrides in Box.update * Release converted attribute aliases after deleting keys * Fix pickling of frozen BoxList instances (#344) * Prep 7.5.0: add Python 3.15, drop 3.10 - Add Python 3.15 to test matrix, classifiers and Cython wheel builds - Add manylinux wheels for cp314 and cp315 (cp314 was missing) - Fix python_requires (was still >=3.9) to >=3.11 - Remove now-dead tomli python_version<3.11 requirements - Require Cython>=3.3.0 (first release with 3.15 support) - Bump actions/checkout, setup-python, upload-artifact to v7 - Bump version to 7.5.0 Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> * Consolidate 7.5.0 changelog and AUTHORS for merged PRs Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> * Fix stray changelog entry from #340 Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> * Format merged tests with black, tidy #305 test comments Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> * Bump version to 8.0.0 Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com> * Add _box_internal_attrs for internal-only attributes on Box subclasses (#269) Also make Box.copy() and __deepcopy__ return the same class instead of hardcoding Box; ConfigBox/SBox copy overrides are no longer needed. Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com> * Fix PR #326 edge cases: stale children, frozen boxes, pickling, dotted keys - never overwrite a key that gained a real value after the ephemeral child was handed out - flush only when data is actually stored, so failed writes/deletes/frozen boxes never mutate the parent - strip the pending-parent reference from pickled state - do not create literal dotted keys for box_dots lookups; register conversion_box alias - Box.update also creates missing parents Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com> * Move package metadata to pyproject.toml (PEP 621) setup.py is reduced to the optional Cython ext_modules; Cython is deliberately not a build requirement so sdist installs stay pure Python. Dropped the stray py_modules=['box'] and platforms='any' (meaningless for compiled wheels). Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com> * Replace requirements files and black with uv dependency groups and ruff config - [dependency-groups] test/dev in pyproject.toml (toon_format as a direct git reference), uv.lock added - [tool.ruff] (line length 120, py311, black compatible format; E/F/W/I lint), pre-commit uses ruff-pre-commit - coverage and cibuildwheel configuration moved into pyproject.toml - removed .black.toml, .coveragerc and requirements*.txt, dropped the no-op cythonize pre-commit hook Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com> * Use uv in GitHub workflows, update README and CONTRIBUTING for uv/ruff and pyproject builds - tests.yml/pythonpublish.yml use astral-sh/setup-uv, uv sync/build and uvx twine/cibuildwheel (same matrix, targets and secrets) - add a ruff lint/format-check job - document building the Cython version with --no-build-isolation now that metadata is in pyproject.toml Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com> * Apply ruff format/lint Mechanical changes from ruff format and ruff check --fix (import sorting, unused imports, quote style), plus noqa for the few intentional findings (== True assertions, one long test string). box/__init__.py is excluded from F401/I001 via per-file-ignores because its import order matters. config_box.pyi uses builtins.bool instead of a type-ignore that the formatter would have moved. Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com> * Update type stubs: Self for copy/__copy__/__deepcopy__, _box_internal_attrs Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com> * Finalize 8.0.0 changelog with a breaking changes section Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com> * Add bwanshoom to AUTHORS for the #269 suggestion Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com> * Review fixes for CI supply chain and build config - Use toon-format 1.2.0 from PyPI instead of an unpinned git URL (now hash-locked in uv.lock, and pinned for cibuildwheel/pip which read pyproject) - Split manylinux publish into a build job (runs tests, no secrets) and an upload-only job, so third party test deps never run alongside PyPI creds - Quote Cython>=3.3.0 in cibuildwheel before-build (was a shell redirect) - Pin uvx tools (ruff 0.17.3 to match lock/pre-commit, cibuildwheel 4.3.5, twine 6.2.0) so CI does not drift - Point uv at the setup-python interpreter (UV_PYTHON) in matrix jobs so each cell, including pypy, tests the interpreter it claims to Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> * Pin astral-sh/setup-uv to v10.3.3 commit (no floating v10 tag exists) Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> * CI: pass setup-python interpreter to uv sync/venv instead of exporting UV_PYTHON Exporting UV_PYTHON made uv build / uv pip target the bare interpreter rather than the project .venv, so --no-build-isolation builds could not find setuptools. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> * Fix #321 review issues: convert only numbers and bools, handle anchored booleans and keys The submitted conversion turned anchored booleans (ruamel ScalarBoolean, not an int subclass in the bool sense) into int 1, and converted ruamel string types to str, which dropped literal block style and preserved quotes on to_yaml. Strings are now left untouched, ScalarBoolean maps to bool, and mapping keys are converted too. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> * Skip the PyYAML backend test when PyYAML is not installed (cibuildwheel env) Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> * Rework #314: case-insensitive ConfigBox keys where the latest write wins The PR as submitted crashed on non-string keys, silently merged keys that differ only by case into the first spelling, made update and merge_update disagree, and scanned every key on each miss. * Only str keys are matched ignoring case, other keys match exactly * A ConfigBox never holds two keys differing only by case: the latest write wins, keeping its value and its spelling (construction, [] and attribute assignment, update, merge_update and |) * Lookups, in, del, pop, get and box_dots paths (first segment, nested ConfigBoxes match the rest) all resolve case-insensitively * A lower-case index in the box config makes misses O(1). It is rebuilt when the dict size changes behind ConfigBox's back * A stale default_box ephemeral child is not attached when the key was since assigned under a different spelling * Frozen ConfigBoxes still refuse every write Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> * 8.0.0 features: create_on_get off by default, subclass defaults, YAML key order, camel keep keys, int keys in dotted paths * #262 default_box_create_on_get now defaults to False, so attribute probing (IPython, Spyder, hasattr) never adds keys. pop() of a missing key and hasattr + delattr (argparse namespaces) keep working with it. * #296 / #223 / #210 _box_defaults class attribute: subclasses change option defaults without overriding __init__. Options are now read from kwargs in one place (_build_box_config), so an explicit option still wins. DDBox uses it. * #104 to_yaml keeps insertion order for PyYAML and ruamel safe / unsafe too, sort_keys=True sorts. * #279 camel_killer_box_keep_keys keeps CamelCase keys and adds snake_case attribute aliases through conversion_box. * #123 box_dots path segments made of digits also match int keys. Only for path segments, a str key of the same value wins, plain keys are not mapped. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> * Convert nested containers during BoxList in-place addition * Publish to PyPI with Trusted Publishing and pin actions to SHAs (#348) - All build jobs (generic sdist/wheel, macOS/Windows Cython wheels, manylinux cibuildwheel) run without secrets and upload artifacts - A single publish job, gated on every build succeeding, downloads all distributions and uploads via PyPI Trusted Publishing (OIDC) in the "pypi" environment; its only permission is id-token: write - No PYPI_USERNAME / PYPI_PASSWORD secrets are used anymore - Pin every action in both workflows to a full commit SHA - Default token permissions to none, checkout without persisted credentials, add job timeouts, twine check every build Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com> * Reach 100% line and branch coverage, enforce it in CI Tests for the remaining reachable paths (error branches, optional YAML / TOML / TOON backends, ConfigBox index upkeep, default_box_attr call signatures, IPython probes, reflected operators). Code that could not run was removed instead: * _parse_box_dots error handlers at call sites that already guarantee a "." or "[" in the key, and its unreachable `setting` split * the _dunder_ check in __get_default (__getattr__ refuses those names first) and the __getstate__ AttributeError case (Box defines __getstate__) * tomli / toml read fallbacks (tomllib is always present on Python 3.11+) and try/except blocks around TOML writes that only caught decode errors * box_from_string's BoxList fallback for TOML (a TOML document is always a table) * DDBox.__repr__, identical to the inherited SBox one Import time environment fallbacks and the Python 3.11 slice branch are marked `pragma: no cover`. A new pure Python CI job fails below 100%. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> * Add 80 edge case tests and fix the 5 bugs they found * `in` / get() for box_dots paths through lists ('a[0].x'), and no longer treating substrings of string values as keys * IndexError on a box_dots key ending with "." with default_box * _get_property_func assumed every class attribute was a property; the msgpack / toon converters are now protected keys like the others * BoxList.__deepcopy__ of a frozen list unfroze the copied Boxes * BoxList pickling appended items before restoring box_options, so items were rebuilt with default options (and unfrozen). __reduce_ex__ now keeps the items in the state, Box 7 pickles still load through __setstate__ A sixth finding, BoxList += not converting items, stays a strict xfail: PR #347 fixes it. The setdefault box_dots fallback is gone, `in` covers it now. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> * Update development dependencies to their latest versions * pyproject dependency group minimums raised to the current releases, and the pytest-cov<6 cap lifted (pytest-cov 7.1.0); uv.lock upgraded with uv 0.13.7 * Cython 3.3.1 for cibuildwheel and the publish workflow, twine 7.0.0 * pre-commit: pre-commit-hooks v6.0.0 (drops the removed fix-encoding-pragma hook), mirrors-mypy v2.4.1 with current type dependencies * Fix the 4 errors the per-file pre-commit mypy run reports (StringIO / file variable type, list operator signatures, Callable in a type comment) ruff 0.17.3 and cibuildwheel 4.3.5 were already the latest. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> * Keep options for dict | Box and dict + Box, class factories for default_box_attr, ConfigBox "off" * __radd__ / __ror__ build the result from a plain dict with the Box's class and options (unfrozen, top level namespace), like __sub__ already did * default_box_attr: list maps to BoxList like [] does, other classes and functions are called like defaultdict factories. key / box_instance are passed by keyword (positional only parameters positionally), any other required parameter is an error * ConfigBox.bool: "off" is false, as in configparser Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> * Add comments on Box keys and BoxList items, kept through YAML (#184) Comments live outside the data, in a per container store ({key: {"before": [lines], "eol": text}} plus document start / end), so to_dict, to_json, equality and hashing never see them. * from_yaml with ruamel's round trip loader extracts them (box.comments re-assigns ruamel's "comment above a key" tokens, which it parses onto the previous key), to_yaml rebuilds CommentedMap / CommentedSeq objects and writes them back. Without comments, or with PyYAML / the safe dumper, the output is exactly as before. * get_comment / set_comment / del_comment on Box and BoxList, positions "eol" / "before", key None for the document ("start" / "end"), box_dots paths, negative list indexes, frozen boxes refuse changes. * Kept in sync: deletes, pop, popitem, clear drop them; ConfigBox latest write wins keeps the comment on the new spelling; camel_killer keys; BoxList insert / pop / remove / del / reverse / sort move them with their items (slice writes drop them); copy, deepcopy, pickle and Box(other_box) carry them. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> * Comments: treat every YAML line break as a line break, reject control characters A comment containing \r, \x85, \u2028 or \u2029 was written with that character inside a "#" line, and YAML parsers read the rest of the line as document content, so set_comment("a", "x\rinjected: 1") added a key on reload. All YAML line breaks now split comment lines (end of line comments reject them) and other control characters are refused. Found by the automated security review. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> --------- Co-authored-by: Jeroen van Bemmel <jvb127@gmail.com> Co-authored-by: kigland <shuaizhicheng336@gmail.com> Co-authored-by: Vincent Gao <gaobing1230@gmail.com> Co-authored-by: apoorvdarshan <ad13dtu@gmail.com> Co-authored-by: chuenchen309 <48723787+chuenchen309@users.noreply.github.com> Co-authored-by: tan_unique <143566827+vinson0522@users.noreply.github.com> Co-authored-by: SeaStar Deng <37767638+DSeaStar@users.noreply.github.com> Co-authored-by: Chris (ChrisJr404) <11917633+ChrisJr404@users.noreply.github.com> Co-authored-by: Guo Jiarui <799423879@qq.com> Co-authored-by: Gyanu Mayank <gyanumayank0@gmail.com> Co-authored-by: Vitaliy <160263432+vitalivo@users.noreply.github.com> Co-authored-by: Alex <100605452+Likio3000@users.noreply.github.com> Co-authored-by: x0Lazarus <113273587+x0Lazarus@users.noreply.github.com> Co-authored-by: machines-6657 <machines-6657@users.noreply.github.com> Co-authored-by: Arindam Sikder <88596876+arindamsikder@users.noreply.github.com> Co-authored-by: fhgffy <102001626+fhgffy@users.noreply.github.com> Co-authored-by: shguild <shifensuperme@gmail.com> Co-authored-by: Huanyi Xie <74809554+xiehuanyi@users.noreply.github.com> Co-authored-by: Muneeb Sheikh <muneikh@gmail.com> Co-authored-by: AryanRG <aryanrganjoo@gmail.com> Co-authored-by: Gary <182863204+gary1110086@users.noreply.github.com> Co-authored-by: Abhinav Maurya <owner@wegenz.in> Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com> Co-authored-by: lin <hanxianfan123@qq.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Brings the release workflow in line with PyPI Trusted Publishing, so no PyPI password lives in GitHub. Previously, four jobs each had
PYPI_USERNAME/PYPI_PASSWORD.Changes
twine checkand uploads its files as an artifact.publishjob waits for every build to succeed, downloads all distributions, and uploads them withpypa/gh-action-pypi-publishfrom thepypienvironment. Its only permission isid-token: write. A failed build can no longer leave a release half-uploaded.pythonpublish.ymlandtests.ymlis pinned to a full commit SHA, with the release tag in a comment.permissions: {}, checkout without persisted credentials, and job timeouts.Before the 8.0.0 release (maintainer)
cdgriffith, repoBox, workflowpythonpublish.yml, environmentpypi.PYPI_USERNAME/PYPI_PASSWORDrepo secrets.Lands on
develop, so it ships with the Version 8.0.0 PR (#346).🤖 Generated with Claude Code