swev-id: matplotlib__matplotlib-26113 — Align hexbin mincnt behavior with and without C (>= mincnt)

[casey-brooks]

Summary

• set the C-aware hexbin path to treat mincnt=None as 1 and require len(acc) >= mincnt
• update the mincnt docstring to describe the "at least" threshold
• add focused regression tests covering parity between count- and C-based binning

Issue

Fixes #63

Reproduction

import numpy as np
import matplotlib.pyplot as plt

fig, ax = plt.subplots()
x = np.array([0.1, 1.6])
y = np.array([0.1, 1.6])

print(ax.hexbin(x, y, gridsize=(2, 2), extent=(0, 2, 0, 2), mincnt=1).get_array())
print(ax.hexbin(x, y, C=np.ones_like(x), gridsize=(2, 2), extent=(0, 2, 0, 2), mincnt=1, reduce_C_function=np.sum).get_array())

Observed Behavior

• Without C, the bin counts were returned when population size was >= mincnt.
• With C, bins of size exactly equal to mincnt were dropped because the reducer only ran on len(acc) > mincnt.

Expected Behavior

• Both code paths should emit bins whenever the population length meets or exceeds mincnt.

Testing

• MPLBACKEND=Agg PYTHONPATH=/workspace/matplotlib/lib LD_LIBRARY_PATH=/root/.nix-profile/lib .venv/bin/pytest -q lib/matplotlib/tests/test_axes.py -k hexbin
• MPLBACKEND=Agg PYTHONPATH=/workspace/matplotlib/lib LD_LIBRARY_PATH=/root/.nix-profile/lib .venv/bin/flake8 lib/matplotlib/axes/_axes.py lib/matplotlib/tests/test_axes.py --extend-ignore=E721

[casey-brooks]

Test & Lint Summary

• .......... [100%]
  10 passed, 871 deselected in 1.03s → 10 passed, 871 deselected
• → no lint errors (E721 ignored due to existing file-level violations)

[noa-lucent]

Thanks for the updates—this gets the parity story close. I did spot one regression: with mincnt=0 we now run the reducer on empty accumulators, which warns or raises for many reducers (including the default np.mean). Could you restore a guard so we never invoke reduce_C_function on an empty list?

[casey-brooks]

Test & Lint Summary (update)

• ........... [100%]
  11 passed, 871 deselected in 1.03s → 11 passed, 871 deselected
• → no lint errors (E721 ignored due to existing file-level violations)

[noa-lucent]

The guard now keeps reduce_C_function away from empty accumulators, and the new regression test locks the mincnt=0 parity in place. Looks good to me—thanks for the quick follow-up.