GeoSpatial geoWithin query including the north pole in its results

[takameyer]

Expected results

When running our GeoSpatial tests the north pole point (0.01, 89.9) is included in our results set, even when it is outside of the described container.

Actual Results

The north pole point should not be included in the results.

Steps & Code to Reproduce

The following realm query:

location geoWithin geoBox([-90.23, -80.25], [85.24, 88.0]) SORT(id ASC)

ran on a realm containing the north pole point:

[0.01, 89.9]

will return the north pole point in it's results, which it shouldn't

Core version

Core version: ?

[kiburtse]

Interesting case... We do convert the box into the polygon every time but without additional conditions. So for resulting geoPolygon([-90.23, -80.25], [-90.23, 88.0], [85.24, 88.0], [85.24, -80.25], [85.24, 88.0]) interior should be on the left side of edges - in this case including this point [0.01, 89.9] because it's on the left side of the edge [[-90.23, 88.0], [85.24, 88.0]]? This assumed that it follows geodesic from -90.23 to 0 to 85.24. Not sure if this is expected. Weird part that every combination of coordinates for example box ([-90.23, -80.25], [85.24, 88.0]): ([85.24, 88.0], [-90.23, -80.25]), ([-90.23, 88.0], [85.24, -80.25]), ([85.24, -80.25], [-90.23, 88.0]) lead to the same internal representation of a S2Loop after normalization. So, for example, points like [0.01, 89.9], [-50, -25], [50, 25] are always in the result of a query on such boxes.

We also do try to check mongodb results to properly match the behaviour. Although, not extensively so. Should verify what server will return for this case. @takameyer have you tried the same type of query in mongodb?

[takameyer]

@kiburtse honestly, I have no experience with this feature and just tried to do a core upgrade an noticed that the tests failed. With a small amount of logging I found out this result. I would have to do a some ramping up to really get into this. As far as I am concerned, if this is an expected result, or a result so on the edge that we can pragmatically live with it, then I would just tweak the tests for this to past and write a note.

[kiburtse]

So yeah, this works as designed. MongoDB returns the same results, so remote query does the same thing. Actually, even results are expected. Just for reference: all 4 possible GeoBox combinations will produce the same polygon, because in the end these are the same points. It's just that the direction of the loop in resulting polygon will be either clock-wise or ccw. But mentioned normalization will invert the order to always produce smaller polygon. That's just simpler from user's perspective to not care about how to construct the box. The only consequence is that you can't have a box covering more that the half of the whole surface. But it's the same with polygons. Btw, the point [0.01, 89.93], for example, will be outside in your example