petbox-dev / tafra / 23695947661

95%

Build Type

push

github

Committed by web-flow

Commit Message

fix: composite key negative values, select generator exhaustion, cross join validation (#28)

* fix: composite key negative values, select generator exhaustion, cross join validation

- _build_composite_key: shift columns to non-negative range before computing
  cardinality — fixes key collisions when integer join keys are negative
- _resolve_join_cols: materialize self.select to frozenset — prevents
  generator exhaustion silently dropping columns
- _encode_columns_paired: add kind-match check before concatenating left/right
  columns — produces clear TypeError instead of silent object upcast
- CrossJoin: validate select columns exist, fix docstring (outer → cross)
- Move _STRING_KINDS to module-level constant (was per-call frozenset)
- Use astype(int64, copy=False) to avoid unnecessary allocation

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

* fix: move negative-key shift into _encode_columns_paired, materialize select

The shift-by-min in _build_composite_key was applied independently to
left and right sides, producing non-comparable keys when mins differed.
Moved the shift into _encode_columns_paired where both sides are visible,
using a combined min as the shared baseline. Only applies to signed
integer columns (kind 'i'), not datetime or other types.

Also added the same shift to _encode_columns (GroupBy path) so negative
integer group-by columns work correctly with _direct_labels_firstseen.

Added Join.__post_init__ to materialize self.on and self.select
from any Iterable into lists, preventing generator exhaustion on reuse.

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

* fix: address Copilot review — select validation, docstrings, tests

- Move select validation to _resolve_join_cols (consistent across all
  join types — InnerJoin, LeftJoin, CrossJoin all raise KeyError)
- Fix base.py cross_join docstring (how='outer' → how='cross')
- Fix _build_composite_key comment to list both callers
- Sort... (continued)

Run Details

170 of 178 new or added lines in 4 files covered. (95.51%)

1 existing line in 1 file now uncovered.

1602 of 1692 relevant lines covered (94.68%)

0.95 hits per line