getInteriorPoint does not intersect narrow polygon

[dbaston]

An interior point generated from the following polygon is not found to intersect the polygon.

POLYGON ((2.0875049999999997 49.038467, 2.087497642793396 49.0384427212182, 2.087475 49.038368, 2.0875049999999997 49.038467))

As reported in https://trac.osgeo.org/postgis/ticket/4274

[dr-jts]

Well, yes... that's a pretty narrow polygon. 2.8E-16 width, in fact.

Have to think about what to do about this... This might be solvable by a better intersection computation (i.e. using extended precision).

Although, there is a public warning in the FAQ about this kind of thing: https://locationtech.github.io/jts/jts-faq.html#D7

Side note: I intend to improve the PointInArea computation using this code in the TestBuilder. The reason is to avoid calling Geometry.intersection(), which is expensive and non-robust. That would be a good time to look at this problem.

[dr-jts]

Really this is more of a "finite-precision" issue than an actual robustness failure. There are just some limitations to working in finite-precision floating point which cannot be worked around.

Perhaps the best that can be done at some point is to implement a tolerance/precision model for predicate calculations (which is the to-do list, and seems tractable).

[dr-jts]

Perhaps the simplest thing to do is to detect that the computed interior point does not intersect the polygon, and simply return a vertex of the polygon. This is not ideal, but may improve the outcome of some use cases (or then again, it may make them worse - hard to know without some real-world experience).

In fact I'm about to land an improved InteriorPointArea which has this behaviour (#369). So perhaps this will provide some feedback.

[dr-jts]

Possibly fixed by #369