Intersection and difference of polygons leads to (almost) self-touching polygon

First of all, thank you very much for Shapely, it’s an awesome library!

Expected behavior and actual behavior.

I am intersecting and differencing polygons, expecting a valid, non-self-touching polygon. I often get almost-self touching polygons, which are a problem when I want to offset the contour of my intersection.

Steps to reproduce the problem.

Minimal example:

import numpy as np
import shapely.geometry
import shapely.wkt
import matplotlib.pyplot as plt


def plot_line(ob, clr='k', alpha=1.0):
    x, y = ob.boundary.xy
    ax.plot(x, y, color=clr, alpha=alpha)


container = shapely.geometry.Polygon(np.array([[20, 0],
                                               [35, 0],
                                               [35, -4],
                                               [20, -4],
                                               [20, 0]]))

pol = shapely.wkt.loads('POLYGON ((20.96 -2.168465542645916, 20.96 -1.189012617488881, 21.45875 -1.243766028311074, 22.62625 -1.364459833901519, 23.82 -1.480114679188134, 25.04 -1.590616286024315, 26.28625 -1.695512902750364, 27.5575 -1.794581333999776, 28.85375 -1.887359106259452, 30.175 -1.973623024162885, 31.52125 -2.053149892343578, 32.89125 -2.125596876361725, 34.28375 -2.18947836031152, 35.7 -2.244108675313356, 37.14 -2.290391324686642, 38.60124999999999 -2.328342391211475, 40.08375 -2.357831513961157, 41.58875 -2.37452684521826, 43.115 -2.381274616794401, 44.6625 -2.378874775715786, 46.23 -2.365052480903801, 47.8175 -2.336940056840027, 49.04 -2.308339084072406, 49.04 -3.307403246304138, 47.74722712561048 -3.337648286610996, 46.18464384097614 -3.365319470190588, 44.64476964447845 -3.378898159260675, 43.11990828613942 -3.381262892654494, 41.61527865723328 -3.37461070786003, 40.13672701151798 -3.358208774633484, 38.67965497406949 -3.329225451754934, 37.23932976613535 -3.291817836505181, 35.82084591271953 -3.246226731762603, 34.4280288046949 -3.192500315274461, 33.05991692034544 -3.129737645700606, 31.71239101679935 -3.058479114606998, 30.38724893438113 -2.980199151963237, 29.08723955504806 -2.895322026231939, 27.81243913152114 -2.804079485541084, 26.56253324382854 -2.706674399599603, 25.33767818412468 -2.603578590778966, 24.13915400788448 -2.49502215856048, 22.96699171228969 -2.381458808656548, 21.82119777103192 -2.263008931831816, 20.96 -2.168465542645916))')
pol = pol.intersection(container)
container = container.difference(pol)

rectangle1 = shapely.geometry.Polygon(np.array([[22, -1.4],
                                               [31, -1.4],
                                               [31, -2],
                                               [22, -2],
                                               [22, -1.4]]))
intersection1 = container.intersection(rectangle1)
container = container.difference(intersection1)

rectangle2 = shapely.geometry.Polygon(np.array([[29, -1.8],
                                                [31.5, -1.8],
                                                [31.5, -2.5],
                                                [29, -2.5],
                                                [29, -1.8]]))
intersection2 = container.intersection(rectangle2)

ax = plt.gca()
plot_line(ob=container, clr='r', alpha=0.5)
plot_line(ob=pol, alpha=0.5)
plot_line(ob=intersection1, alpha=0.5)
plot_line(ob=intersection2, clr='b')
plt.show()

With this code I get: Note the almost self-touching part of the polygon.

Illustration

On another intersection, I got: And by zooming (a lot):

Attempted solutions

The buffer(0) trick or this stackoverflow answer do not seem to solve my issue. I’ve been able to work around this issue until now but it has become a real problem. In this case, I need a clean polygon in order to obtain a linear_ring from an offset of my final intersection, but in general it would be very good to obtain clean intersections. Any idea where this issue comes from? Thanks for your help

Operating system

Windows 7 Python 3.6.7

Shapely version and provenance

Shapely-1.6.4.post1-cp36-cp36m-win_amd64.whl

[dr-jts]

Happy to help. I’m going to try and port this to JTS code, since it’s a fine example of why a precision model is needed. Improving JTS/GEOS support for this is the number one goal of development for this year. So maybe that will provide some motivation to do some Shapely development.

[Arkangus]

Thank you @sgillies and especially @dr-jts for the answers. I get a clean cut with your answer 😃