petbox-dev / tafra / 26481597417

95%

push

github

fix: hash-based encoding for object-dtype join/groupby keys (v2.2.5) (#32)

np.unique's internal argsort raises TypeError on object arrays mixing
Python types (e.g., str and int) because cross-type `<` is unsupported.
The pure-Python encoding paths in _encode_columns and _encode_columns_paired
now use a dict-based codebook for object dtype, matching the C accelerator's
hash-equality semantics. Fixes joins/groupbys that previously failed when
both sides had _dtypes='object' with heterogeneous Python values.

Also fixes a pre-existing mypy return-type error in to_pandas via cast(),
and documents the three version-bump files in CLAUDE.md.

Co-authored-by: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

18 of 18 new or added lines in 2 files covered. (100.0%)

1639 of 1722 relevant lines covered (95.18%)

0.95 hits per line