Fix autoscaling for Axes.hist histtype='step' density=True (swev-id: matplotlib__matplotlib-24177)

[casey-brooks]

Fixes #35

Summary

• address density autoscale regression for step and stepfilled histograms
• replace the fill-based Polygon construction with explicit PathPatch instances that preserve histogram peaks in data limits
• add regression tests covering density autoscale and scale invariance for step/stepfilled histograms

Reproduction

import numpy as np
import matplotlib.pyplot as plt

rng = np.random.default_rng(20250101)
scale = 1.2
x = rng.standard_normal(100000) * scale
fig, ax = plt.subplots()
hist, bins, patches = ax.hist(x, bins=100, density=True, histtype="step")
print(f"hist max: {hist.max():.4f}")
print(f"ylim upper: {ax.get_ylim()[1]:.4f}")

Observed output before the fix:

hist max: 0.3386
ylim upper: 0.2818

Observed failure details (before fix)

$ MPLBACKEND=Agg python -m pytest lib/matplotlib/tests/test_axes/test_hist.py
E   assert 0.30000000000000004 >= 0.33859214437970275
E   assert 0.30000000000000004 >= 0.33859214437970275
========================= 2 failed, 1 passed in 0.20s =========================

Fix

• build histogram outlines (“step”/“stepfilled”) with Paths that carry explicit MOVETO/LINETO (and CLOSEPOLY for stepfilled) codes and render them via a PathPatch subclass
• disable simplification on the step Paths and trigger autoscale updates after attaching the patches so data limits honor the peak density
• keep sticky edges semantics and facecolor/edgecolor behavior while a new regression test module exercises the autoscale bug and scale invariance

Testing

• MPLBACKEND=Agg python -m pytest lib/matplotlib/tests/test_axes/test_hist.py
• MPLBACKEND=Agg python -m pytest lib/matplotlib/tests/test_axes.py -k hist_step
• flake8 lib/matplotlib/tests/test_axes/test_hist.py lib/matplotlib/axes/_axes.py

[noa-lucent]

Replacing the step histogram patches with a PathPatch helper breaks the documented Polygon return type. Please keep returning Polygon instances (or a subclass) while applying the autoscale fix.

[noa-lucent]

[major] The and branches of have always returned instances, and we still advertise that via the a few lines below. Switching the return type to a subclass breaks that published contract (e.g. existing code that does will now fail). Could we keep returning a and still set the explicit path codes/ flag (e.g. by creating a and then mutating its path), or alternatively derive from so that the type relationship remains intact?}

[casey-brooks]

Addressed by replacing the PathPatch helper with direct Polygon construction while keeping simplification disabled for autoscale. Explicit step fill requests now reuse the edge color so the geometry tests still pass.

[casey-brooks]

Test & lint summary:

• flake8 lib/matplotlib/tests/test_axes/test_hist.py lib/matplotlib/axes/_axes.py
• MPLBACKEND=Agg python -m pytest lib/matplotlib/tests/test_axes/test_hist.py # 3 passed
• MPLBACKEND=Agg python -m pytest lib/matplotlib/tests/test_axes.py -k hist_step # 9 passed, 823 deselected

[noa-lucent]

Thanks for restoring Polygon returns and for the thorough regression coverage. The autoscale fix now looks solid to me.