LargestEmptyCircle crash on malformed input

[pramsey]

From https://trac.osgeo.org/postgis/ticket/5626#comment:3

Test case for LargestEmptyCircleTest.cpp

// https://trac.osgeo.org/postgis/ticket/5626
template<>
template<>
void object::test<20>()
{
    checkCircle(
        "POINT(-11 40)", // input
        "POLYGON((-71.1 42.3,-71.1 42.4,-71.2 42.3,-71.1 42.3))", // boundary
        1, // build tolerance
        5.5, 4.5, 2.12 ); // expected x, y, radius
}

This doesn't appear to be structural, other combinations of point geometry outside of the polygon boundary don't yield a crash. On the other hand, the crash doesn't seem to be a tightly bounded numerical issue, as this rounded up version of the initial bug report still crashes.

[pramsey]

Moving the point input around also changes nothing, the problem seems to be in the facet distance search. It's not the 3-edge polygon, other 3-edge polygons work, and if you round the triangle coordinates even more, so it's a one-unit edge, the crash goes away too.

[dr-jts]

Works in JTS.

[dr-jts]

It looks like the issue is the horizontal botton edge of the triangle ( LINESTRING(-71.2 42.3,-71.1 42.3)) along with the right angle. Lengthening the edge while keeping it horizontal and keeping the right angle still crashes. Moving other vertices, or lengthening the bottom edge to the right doesn't cause a crash.

[robe2]

@pramsey @dr-jts are you sure this isn't a recent issue. I saw the error locally when I was fixing the garden tests, but even with my fixes I didn't see berrie and berrie64 (the two raspberry pis we have testing PostGIS and GEOS). But then I realized I had them testing against 3.12.0 (instead of following the branches).

I'm going to check the GEOS old versions I have lieing around to check, but given JTS can't trigger it, I'm thinking its an issue introduced in a recent micro release.

[dr-jts]

I'm going to check the GEOS old versions I have lieing around to check, but given JTS can't trigger it, I'm thinking its an issue introduced in a recent micro release.

Sounds good, I'll check too.

[pramsey]

You are correct, 3.12.0 does not crash.

[dr-jts]

It might be due to #939. But that may not help - it's probably just the change that reveals the problem.

[robe2]

You are correct, 3.12.0 does not crash.

I just tested on 3.12.2dev and no crash (so looks like this is a new issue in 3.13 cycle)