Two empty geometries should have the same topology

[michaelkirk]

For example:

MultiPoint a = (MultiPoint) readerXY.read("MULTIPOINT EMPTY");
MultiPoint b = (MultiPoint) readerXY.read("MULTIPOINT EMPTY");
// This currently fails
assertTrue(a.equalsTopo(b));

Does the current behavior make sense?

[dr-jts]

I think the original idea was that EMPTY geometries contained no points, so could not be equal. But it's debatable whether that makes sense.

[dr-jts]

This was changed in GEOS for RelateNG so that empty geometries are equal (regardless of geometry type).

[michaelkirk]

The motivation for my question (which I should have included originally, sorry!) is that I'm comparing the outputs of some geometry transformations to known "correct" answers.

Occasionally the expected answer is an empty geometry.
So to check that I have achieved the correct answer, I currently have to say:

// (pseudo code)
if expected.empty() && actual.empty() {
   return true
} else {
   return expected.isTopoEquals(actual)
}

I'm somewhat emboldened by this change in RelateNG - it's what I expected.

[dr-jts]

That's a good use case for this semantic.

This actually isn't implemented in JTS yet. Perhaps I should make that change.

[michaelkirk]

For my own reference, in geos, it looks like we're talking about this code:

class EqualsTopoPredicate : public IMPredicate {

    std::string name() const override {
        return std::string("equals");
    }

    bool requireInteraction() const override {
        //-- allow EMPTY = EMPTY
        return false;
    };

    void init(int _dimA, int _dimB) override {
        IMPredicate::init(_dimA, _dimB);
        //-- don't require equal dims, because EMPTY = EMPTY for all dims
    }

    void init(const Envelope& envA, const Envelope& envB) override {
        //-- handle EMPTY = EMPTY cases
        setValueIf(true, envA.isNull() && envB.isNull());

        require(envA.equals(&envB));
    }
    ...
}

https://github.com/libgeos/geos/blob/29a4c7ea895da8bb721f1d9342b0a8ef3ad89b2f/include/geos/operation/relateng/RelatePredicate.h#L505

[dr-jts]

For my own reference, in geos, it looks like we're talking about this code:

https://github.com/libgeos/geos/blob/29a4c7ea895da8bb721f1d9342b0a8ef3ad89b2f/include/geos/operation/relateng/RelatePredicate.h#L505

Correct.

[dr-jts]

I changed RelateNG to have the EMPTY-equals semantics (#1089)

I also added an XML test TestRelateEmpty.xml for various EMPTY cases (since this was lacking, surprisingly). @michaelkirk are you making use of these in georust?

[michaelkirk]

Thank you for making this change.

I wrote a test harness to consume your xml tests here: https://github.com/georust/geo/tree/main/jts-test-runner

It was especially useful when I was porting the topology graph stuff for DE-9IM operations. I haven't yet found the time to port your OverlayNG work.

I'll pull in the new https://github.com/locationtech/jts/blob/master/modules/tests/src/test/resources/testxml/misc/TestRelateEmpty.xml momentarily.

[michaelkirk]

The new tests uncovered a couple errors in our own implementation, so thank you for those. All our tests are passing now.

https://github.com/georust/geo/pull/1227/files