hgrecco / pint-pandas / 37543552727

90%

push

github

Fix eval recursion with mixed units (#300), continuing #304 (#320)

* adding test for issue 300

* make _get_common_dtype consistent return: instance of PintType

* clean up

* added a few exception to resolve #300

* Fix #300 without regressing concat, row access or dimensionless input

Continues #304. _get_common_dtype now returns a PintType() instance rather
than the class, which pandas treated as object dtype and which sent
DataFrame.eval with numexpr into infinite recursion.

Casting to that undefined-units dtype now falls back to object values
instead of building a pint[None] array:
- PintArray.astype(PintType()) returns an object ndarray of quantities
- PintArray._from_sequence(..., dtype=PintType()) returns a PintArray when
  all values are quantities of compatible units, else an object ndarray

The _from_sequence dimensionless change and the _to_array_of_quantity
change from #304 are reverted. numexpr is added to the test extras so the
eval test exercises the recursion path.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

---------

Co-authored-by: Laurent Mutricy <laurent.mutricy@ekium.eu>
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>

17 of 17 new or added lines in 1 file covered. (100.0%)

658 of 725 relevant lines covered (90.76%)

1.82 hits per line