petbox-dev / tafra / 23523601345

92%

push

github

fix: StringDType / <U interop and left join null handling (v2.2.1) (#25)

* fix: StringDType / <U interop in joins, unions, and update_dtypes

- Join _validate_dtypes: compare base types via _reduce_dtype instead of
  raw numpy dtypes, so StringDType vs <U and <U8 vs <U12 are accepted
- Union.apply: same base-type comparison for column dtype validation
- update_dtypes_inplace: preserve raw numpy dtypes for casting instead of
  round-tripping through _format_dtype string labels (fixes silent no-op
  when both <U and StringDType map to 'str')
- Add test_mixed_string_dtypes covering all interop scenarios
- Bump version to 2.2.1

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

* fix: left join preserves string/float dtypes on null-fill

- String columns use StringDType(na_object=None) instead of object
- Float columns use NaN instead of object
- Int/bool columns still fall back to object (can't represent None)
- Both C-path and fallback path updated
- Tests updated for new null-fill behavior

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

* fix: emit warning when left join casts column to object dtype

Warns once per column when an int/bool column has unmatched rows and
must fall back to object dtype. String and float columns are handled
natively (StringDType(na_object=None) and NaN respectively) and do
not warn.

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

* test: add coverage for update_dtypes string conversion and object warning

- test_update_dtypes_string_conversion: verifies <U/StringDType handling,
  int->float, float->int conversions
- test_left_join_object_fallback_warning: asserts warning fires for int
  columns, does NOT fire for string/float columns

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

* chore: remove stale config files

- .readthedocs.yml: replaced by GitHub Pages / MkDocs
- MANIFEST.in: unnecessary with setuptools>=64 + pyproject.toml
- d... (continued)

64 of 67 new or added lines in 4 files covered. (95.52%)

1398 of 1514 relevant lines covered (92.34%)

0.92 hits per line