PR summary

On polar plots, radial gridlines (#20388) and the outer spine (#26972) could collapse to a near-empty arc for certain set_theta_zero_location / set_theta_offset / set_theta_direction combinations.

The root cause is in Path.arc's angle-unwrap step. A span of 360*n plus a tiny floating-point rounding error was reduced modulo 360 to a near-zero arc instead of a full circle:

eta2 = theta2 - 360 * np.floor((theta2 - theta1) / 360)   # 360 + 1e-9 -> 1e-9

Fix

Detect spans that are within floating-point tolerance of a whole number of turns and draw a complete circle; otherwise unwrap theta2 to the shortest arc within 360 degrees exactly as before:

n_turns = (theta2 - theta1) / 360
if (theta2 != theta1 and round(n_turns) >= 1
        and abs(n_turns - round(n_turns)) < 1e-9):
    eta2 = theta1 + 360            # complete circle
else:
    eta2 = theta2 - 360 * np.floor(n_turns)   # shortest arc

The change is byte-identical for all non-degenerate inputs (e.g. 90°, 359.99°, 360°, 410°→50°, 540°→180°, 720°→full circle) and only affects the 360*n + ε collapse cases, so no call sites need to change.

Closes #20388
Closes #26972

Tests

• test_arc_full_circle_snap / test_arc_unwrap_partial_turn in test_path.py cover the snap-vs-unwrap behaviour at the unit level.
• test_path.py and end-to-end polar gridline / outer-spine regression tests in test_polar.py.

PR checklist

• Code is PEP8 compliant.
• New and changed code is tested.
• Plot-related changes verified in CI (no image baselines change; behaviour is byte-identical outside the bug case).

AI Disclosure

AI was used to help proofread the grammar of this PR description and to understand some parts of the existing codebase during development. The code changes and design decisions were developed independently.