Polygon clipping fix

[artemp]

No description provided.

[flippmoke]

@artemp or @ericfischer could you produce an image of the results here compared to #100 (comment)

I would love to visually see the differences again.

[e-n-f]

I'll try to build both versions to compare the output, unless @artemp gets there before I do.

[e-n-f]

As far as I can tell, both implementations are now producing visually identical output:

[flippmoke]

After thinking about this more two situations are concerns here for me, first can this turn the following into two exterior rings after clipping?

Next, would this resolve to a single exterior ring and single hole.

These are two common failure cases for many clipping algorithms and things I would be concerned that Boost may not resolve correctly.

[artemp]

@flippmoke - thanks for providing pics ^ - it's always easier to see the issue. Lets establish what is a correct expected result for both cases. Could you post some more info, pls!

[flippmoke]

The expected solution for the first one is two polygons each with no interior rings, one small triangle and one polygon with a larger square with a triangle cut out of it.

The expected solution for the second one is one polygon with one interior ring with a hole cut out of it. The interior ring is a triangle that touches the square exterior ring at one point.

[artemp]

• boost-1.64 (earliest I have installed)

One polygon with interior ring touches exterior at the tangent (which is valid according to OGC/ISO standards)

POLYGON((2560 3072,1024 3072,1024 1024,2560 1024,2560 3072),(2560 2048,2048 1536,1536 2048,2048 2560,2560 2048))
 Is valid:1
Is simple:1

two separate polygons

POLYGON((2048 3072,1024 3072,1024 1024,2048 1024,2048 1536,1536 2048,2048 2560,2048 3072))
 Is valid:1
Is simple:1
POLYGON((2048 2560,1792 2048,2048 1536,2048 2560))
 Is valid:1
Is simple:1

I'm yet to incorporate these tests into vector tiles but here is the start. Results are as expected ^
/cc @flippmoke @ericfischer @springmeyer

[e-n-f]

I think either of these PRs would be an improvement over the current situation. Maybe fuzzing will reveal a case that one of them can handle that the other can't.

[flippmoke]

LGTM, lets ship it.