winding-number

Winding number and point-in-polygon for arbitrary (non-convex, self-intersecting) polygons. Pure Python, zero dependencies.

What is the winding number?

The winding number of a closed polygon around a point is the integer count of how many times the polygon winds around that point. A CCW polygon around an interior point has winding number +1; a CW polygon gives -1; an exterior point gives 0.

This is the basis of the nonzero rule: a point is inside iff winding_number != 0.

Two fill rules

This library implements both standard fill rules:

Nonzero rule (point_in_polygon): inside iff winding_number != 0. Correct for non-convex and self-intersecting polygons; all regions with a non-zero winding count are considered inside.

Even-odd rule (point_in_polygon_even_odd): inside iff crossing_number is odd. Counts how many times a ray from the point crosses the polygon boundary; an odd count means inside. The two rules agree for simple (non-self-intersecting) polygons and diverge for self-intersecting ones.

Pentagram example

The canonical case where the rules disagree is a five-pointed star (pentagram) traced as the self-intersecting 5-vertex path (outer tips connected in skip-one order):

import math
from winding_number import winding_number, point_in_polygon, crossing_number, point_in_polygon_even_odd

n = 5
circle_pts = [
    (math.cos(math.radians(90 - 72 * i)), math.sin(math.radians(90 - 72 * i)))
    for i in range(n)
]
pentagram = [circle_pts[i % n] for i in [0, 2, 4, 1, 3]]

center = (0.0, 0.0)

winding_number(center, pentagram)               # -2  (wound twice, CW)
point_in_polygon(center, pentagram)             # True   (nonzero rule: inside)
crossing_number(center, pentagram)              # 2   (two crossings, even)
point_in_polygon_even_odd(center, pentagram)    # False  (even-odd rule: outside)

The central region is wound around twice, so the nonzero rule treats it as inside. The even-odd rule ignores winding direction and depth: two crossings is even, so outside.

Why zero-dependency vs. shapely or matplotlib?

The correct alternatives require heavy C extensions:

• shapely.geometry.Point.within(polygon) requires shapely, which depends on GEOS (a C/C++ library).
• matplotlib.path.Path.contains_point pulls in all of matplotlib (rendering, fonts, backends).

This library implements Sunday's winding-number algorithm in pure Python with zero runtime dependencies. No compilation, no GEOS, no plotting overhead.

Install

pip install winding-number

PyPI release pending. Install from source in the meantime:

git clone https://github.com/amaar-mc/winding-number
cd winding-number
pip install -e .

Quick start

from winding_number import winding_number, point_in_polygon, crossing_number, point_in_polygon_even_odd, signed_area, orientation, is_convex

# CCW unit square.
square = [(1, -1), (1, 1), (-1, 1), (-1, -1)]

winding_number((0, 0), square)      # 1  (inside, CCW)
winding_number((2, 0), square)      # 0  (outside)
point_in_polygon((0, 0), square)    # True
point_in_polygon((2, 0), square)    # False

signed_area(square)                 # 4.0  (positive = CCW)
orientation(square)                 # "CCW"
is_convex(square)                   # True

# Non-convex L-shape.
l_shape = [(0,0),(2,0),(2,1),(1,1),(1,2),(0,2)]
is_convex(l_shape)                  # False
point_in_polygon((1.0, 0.5), l_shape)  # True  (inside the bottom bar)
point_in_polygon((1.5, 1.5), l_shape) # False (the notch)

API

A point is tuple[float, float] and a polygon is Sequence[tuple[float, float]]. All functions raise ValueError if the polygon has fewer than 3 vertices.

On-edge convention

Nonzero rule: a point exactly on a non-horizontal polygon edge is counted as inside (winding_number returns +/-1). A point on a horizontal edge is counted as outside (winding number 0).

Even-odd rule: the crossing test uses a strict < for the crossing x-coordinate, so a point that lies exactly on a non-horizontal edge (where the ray x-coordinate equals the crossing x) is counted as outside. Points on horizontal edges are outside (horizontal edges are skipped). The two algorithms may therefore disagree at exact boundary points; they agree at all strictly interior and strictly exterior points of simple polygons.

Both conventions are deterministic and require no epsilon. See docs/architecture.md for the full derivation.

License

MIT. Copyright (c) 2026 Amaar Chughtai.