pandas 3.x compatibility: groupby().apply() drops grouping column

[conchoecia]

Summary

source/odp_functions.py:reciprocal_best_hits_blastp_or_diamond_blastp raises KeyError: 'qseqid' on pandas 3.x. The CI smoke test against tests/test_odp_basic/test_odp_basic.sh reproduces this; locally with pandas 2.x it works.

Root cause

The idiom

fdf = (fdf.groupby("qseqid")
         .apply(lambda group: group.loc[group["evalue"] == group['evalue'].min()])
         .reset_index(drop=True))

worked on older pandas because the grouping column was kept in apply's result. On pandas 3.x, groupby().apply() no longer includes the grouping column by default, so after reset_index(drop=True) the qseqid column is gone and the next line that does fdf["qseqid"] (line 390 in odp_functions.py) raises KeyError.

Workaround in place

CI pins pandas<3 in requirements-dev.txt so the smoke test passes against the API the codebase was written for. Users running odp from source on a fresh pandas 3.x install will hit the same error.

Fix scope

grep -n "groupby" source/*.py scripts/odp scripts/odp_nway_rbh scripts/odp_filechecker shows ~85 groupby sites. The problematic pattern is specifically groupby(col).apply(...).reset_index(drop=True). Hot spots:

• source/odp_functions.py:312, 321, 379, 384

Migration: either pass include_groups=False then reset_index() (without drop=True), or rewrite as a vectorized form, e.g.:

fdf = fdf.loc[fdf.groupby("qseqid")["evalue"].transform("min") == fdf["evalue"]]

Followup PR should audit all 85 sites and migrate to a form that runs cleanly on pandas 2.x and 3.x.

[conchoecia]

Workaround landed on main via #112 (commit 1dcf964) — requirements-dev.txt pins pandas<3 so the CI smoke test passes against the API the codebase was written for. Still need a followup PR to audit the ~85 groupby sites in source/ + scripts/ and migrate them to a form that runs cleanly on both pandas 2.x and 3.x.