godotengine / godot

Godot Engine – Multi-platform 2D and 3D game engine
https://godotengine.org
MIT License
91.51k stars 21.26k forks source link

Geometry.is_point_in_polygon() fails when points are clockwise #69505

Open SteveSmith16384 opened 1 year ago

SteveSmith16384 commented 1 year ago

Godot version

3.5.1

System information

Linux Mint

Issue description

Geometry.is_point_in_polygon() returns an incorrect result in the example below where the polygon points are clockwise, but works if I invert the points.

Steps to reproduce

I'm running the following code:-


var polygon : PoolVector2Array 
polygon.append(Vector2(0, 20)) 
polygon.append(Vector2(100, 20)) 
polygon.append(Vector2(100, 0)) 
polygon.append(Vector2(0, 0))  
# Returns false:- 
var result = Geometry.is_point_in_polygon(Vector2(0,1), polygon) 

As far as I can see, "result" should be true since (0, 1) is inside this polygon (which is just a rectangle) but it returns false.

If I add:-


polygon.invert()
var result2 = Geometry.is_point_in_polygon(Vector2(0,1), polygon) 

It now returns true, which is correct.

Minimal reproduction project

n/a - See example code above

TheDuriel commented 1 year ago

This was solved on reddit. A careful look at the coordinates reveals that this is an anticlockwise polygon. It thus considers the point, which is on the exact edge (adding to the confusion), to be outside the inverted polygon.

SteveSmith16384 commented 1 year ago

Thanks. IMHO, the documentation should be updated though; AFAICS it makes no mention of clockwise or anticlockwise or that they are even a thing to know about.

kleonc commented 1 year ago

This was solved on reddit. A careful look at the coordinates reveals that this is an anticlockwise polygon. It thus considers the point, which is on the exact edge (adding to the confusion), to be outside the inverted polygon.

For reference: reddit thread.

@TheDuriel It's not solved. Even if it's intended to consider the edges to be a part of / not be a part of the polygon based on the winding order then it still seems to be bugged for the corners/vertices:

Test script: ```gdscript tool extends EditorScript func _run() -> void: var results_format_string := " %10s: %5s %5s" for rect in [ Rect2(0, 0, 5, 5), Rect2(0, 0, 10, 20), Rect2(0, 0, 7, 13), Rect2(0, 0, 111, 11), Rect2(0, 0, 45, 3), Rect2(231, 351, 5, 87), Rect2(0, 0, 100, 20), ]: print() print("Rect: %s" % [rect]) var polygon := PoolVector2Array([ rect.position, rect.position + rect.size * Vector2(1, 0), rect.position + rect.size * Vector2(1, 1), rect.position + rect.size * Vector2(0, 1), ]) var polygon_inv = polygon polygon_inv.invert() # print(" Inside points:") for y in range(rect.position.y + 1, rect.end.y): for x in range(rect.position.x + 1, rect.end.x): var point := Vector2(x, y) var r1 = Geometry.is_point_in_polygon(point, polygon) var r2 = Geometry.is_point_in_polygon(point, polygon_inv) if !r1 or !r2: # Assumption: these should always be considered inside. print(results_format_string % [point, r1, r2]) # print(" Edge points:") for y in range(rect.position.y + 1, rect.end.y): for x in [rect.position.x, rect.end.x]: var point := Vector2(x, y) var r1 = Geometry.is_point_in_polygon(point, polygon) var r2 = Geometry.is_point_in_polygon(point, polygon_inv) if r1 == r2: # Assumption: these should differ based on the winding order. print(results_format_string % [point, r1, r2]) for y in [rect.position.y, rect.end.y]: for x in range(rect.position.x + 1, rect.end.x): var point := Vector2(x, y) var r1 = Geometry.is_point_in_polygon(point, polygon) var r2 = Geometry.is_point_in_polygon(point, polygon_inv) if r1 == r2: # Assumption: these should differ based on the winding order. print(results_format_string % [point, r1, r2]) print(" Corner points:") for point in polygon: var r1 = Geometry.is_point_in_polygon(point, polygon) var r2 = Geometry.is_point_in_polygon(point, polygon_inv) # if r1 == r2: # Assumption: these should differ based on the winding order. # But let's print all of them for showcase: print(results_format_string % [point, r1, r2]) ```
Results (v3.5.1.stable.official [6fed1ffa3]): ``` Rect: (0, 0, 5, 5) Corner points: (0, 0): True True (5, 0): True True (5, 5): False False (0, 5): True True Rect: (0, 0, 10, 20) Corner points: (0, 0): True True (10, 0): True True (10, 20): False False (0, 20): True True Rect: (0, 0, 7, 13) Corner points: (0, 0): True True (7, 0): True True (7, 13): False False (0, 13): True True Rect: (0, 0, 111, 11) Corner points: (0, 0): True True (111, 0): True True (111, 11): True False (0, 11): True True Rect: (0, 0, 45, 3) Corner points: (0, 0): True True (45, 0): True True (45, 3): False False (0, 3): True True Rect: (231, 351, 5, 87) Corner points: (231, 351): True True (236, 351): True True (236, 438): False False (231, 438): True True Rect: (0, 0, 100, 20) Corner points: (0, 0): True True (100, 0): True True (100, 20): False False (0, 20): True False ```

Hence it's not just a matter of documenting it.


It's a 3.x-only issue. In master (v4.0.beta7.official [0bb1e89fb]) all points on the edges (including corners/vertices) seem to properly be considered as inside the polygon. It was changed as a part of #52110, and intentionally no backported to 3.x to avoid potential regressions.