manahl / PyBloqs / f5371eee-6366-4f4a-a251-cf08a843ef06

89%

master: 89%

Pull #177

circleci

fix: pandas 3 compatibility in table_formatters

The 8 test failures on build_3_11 and build_3_12 come from three pandas 3
changes. Reproduced on clean master against pandas 3.0.5.

1. Series._append and DataFrame._append were private and are gone in pandas 3.
   Replaced with pd.concat behind two helpers next to the existing
   _pandas_df_map_compat.

2. Assigning "" into a numeric Series or DataFrame no longer silently upcasts
   to object, it raises TypeError: Invalid value '' for dtype 'float64'. Widen
   to object explicitly first. The widening is guarded on the blanking mask
   being non-empty, because pandas 2 only upcast when something was actually
   assigned, and casting unconditionally turned an all-numeric totals row or
   column into object and changed the result.

3. Concatenating an empty object Series onto a numeric row widens it to object,
   which Series._append did not do, so the concat is skipped when there are no
   non-numeric columns to blank.

With these, FmtAppendTotalsRow, FmtAppendTotalsColumn and FmtExpandMultiIndex
return identical values and dtypes on pandas 2.3.3 and 3.0.5.

One test change: test_FmtAppendTotalsRow_modify_dataframe built its expected
Series from a list of strings, which infers the new "str" dtype on pandas 3,
while the appended row is object on both. Pinned dtype=object, which is what
the function returns on either version.

Signed-Off By Sandeep Singh Rai sandeeprai_dsp@hotmail.com

16 of 16 new or added lines in 1 file covered. (100.0%)

2522 of 2845 relevant lines covered (88.65%)

3.55 hits per line