Unexpected union of two MultiPolygons

[grimsa]

This is likely also caused by high-precision numbers, similar to #784 and #820.

JTS version used: 1.18.2 with OverlayNG enabled.

Can be reproduced by copy-pasting WKT to JTS TestBuilder:
A:
MULTIPOLYGON (((1.4020000000000001 0.6152725855529207, 2.017272585552921 0, 0 0, 0 2.0172725855529214, 0.6152725855529207 1.4020000000000001, 1.4020000000000001 1.4020000000000001, 1.4020000000000001 0.6152725855529207)), ((3.9999999999999996 0, 2.0172725856347586 0, 3.0086362927764605 0.9913637072235397, 3.9999999999999996 0)), ((0.9913637072235395 3.0086362927764605, 0 2.017272585634758, 0 3.9999999999999996, 0.9913637072235395 3.0086362927764605)))

B:
MULTIPOLYGON (((1.4020000000000001 0.6152725856097978, 1.4020000000000001 1.4020000000000001, 0.6152725855778813 1.4020000000000001, -0.0000000000000002 2.017272585552921, 0.9913637072235395 3.0086362927764605, 3.0086362927764605 0.9913637072235397, 2.0172725856347586 0.0000000000000001, 1.4020000000000001 0.6152725856097978)))

Result of A union B:
GEOMETRYCOLLECTION (POLYGON ((2.017272585552921 0, 0 0, 0 2.017272585552921, 0 2.0172725855529214, 0.0000000000000002 2.0172725855529214, 0.9913637072235395 3.0086362927764605, 3.0086362927764605 0.9913637072235397, 2.0172725856347586 0.0000000000000001, 1.4020000000000001 0.6152725856097978, 1.4020000000000001 0.6152725855529207, 2.017272585552921 0), (0.6152725855529207 1.4020000000000001, 0.6152725855778813 1.4020000000000001, 0.0000161232623158 2.0172564622906055, 0.6152725855529207 1.4020000000000001)), POLYGON ((2.0172725856347586 0, 3.0086362927764605 0.9913637072235397, 3.9999999999999996 0, 2.0172725856347586 0)), POLYGON ((0 2.017272585634758, 0 3.9999999999999996, 0.9913637072235395 3.0086362927764605, 0 2.017272585634758)), LINESTRING (0 2.017272585552921, -0.0000000000000002 2.017272585552921))

Expected result:

• Ideally, a single polygon
• Alternatively, a multipolygon (the 3 polygons without the almost 0-length linestring)

[dr-jts]

This is OverlayNG working as designed.

The input is not node-aligned - as can be seen using the TestBuilder in Reveal Topology mode:

The union result simply reflects this misalignment.

The reason the result is heterogeneous (i.e. including the additional LineString) is for legacy reasons, to match the semantics of the original overlay implementation. OverlayNG does provide a strict mode, which allows forcing results to be homogeneous. This is available in the TestBuilder:

Or, you can remove lower-dimension elements in a post-processing step.

[dr-jts]

Also note that the SnappingNoder can be used explicitly with OverlayNG (in code, and in the TestBuilder). Using snapping can eliminate the input misalignment.

With a suitable snapping distance (here 1E-7) the result is a single triangle polygon:

POLYGON ((0 0, 0 2.0172725855529214, 0 3.9999999999999996, 0.9913637072235395 3.0086362927764605, 3.0086362927764605 0.9913637072235397, 3.9999999999999996 0, 2.017272585552921 0, 0 0))

[grimsa]

Thank you for a quick investigation and a thorough explanation.

The Reveal Topology looks very useful, thank you. Took me a while to find it 😅
In case anyone stumbles upon this later, you can find it here:

It seems that ultimately we should use both SnappingNoder and the strict overlay mode in our application - that seems like a simpler solution than post-processing the result (what we currently do).

I looked into the code a bit, and did not find a way to configure this via system properties or Geometry#* operations (that we currently use).
In the other issue you mentioned OverlayNGSnappingFunctions code. It would seem that implementing some utility class/wrapper that provides a more convenient API on top of the OverlayNG operations is the intended approach when configuration knobs have to be tinkered with? I'll try that in near future.

Should I close the issue now that it's clear that it's not a bug?

[dr-jts]

Yes, implementing a (set of) wrapper functions providing an API for the OverlayNG settings you want is a good approach.

I'll close this now.