Bounding box approach does not work well for distant points and antipodes

[CapacitorSet]

Consider the following:

• the shortest line between two points on a sphere is not a straight line on a map, not on a Mercator projection anyway.
• the difference between a straight line and a geodesic gets larger and larger as points get more distant in longitude.
• a bounding box does not accurately account for such a difference.

Suppose that one were to travel from Zurich to Vladivostok. A bounding box approach would get a significantly longer trip:

Furthermore, the bounding box can prove mildly inefficient for points which divide a geodesic into two roughly similar halves, like Zurich and the Bering Strait. The bounding box approach would search for a route that goes through the Pacific and Canada (i.e. to the west); however, there is a similar route that may prove more efficient, that goes through Asia (i.e. to the east). The difference between the two straight paths is insignificant (approx. 80 km).

[denysvitali]

Actually, you're right. A Bounding Box within two geometrical coordinates (as seen here) is wrong, because will cause the bounding box to include elements that are geometrically inside a box that acts like if the plane was projected. The solution here would be to change the cast to geometry to a cast to geography - that way the bounding box will be drawn correctly over the sphere.
I'll fix this issue in gtfs-server queries.

I'll look into the second issue (even though our project isn't capable of handling such big distances), and report back 👍

Thank you for opening this issue 🚀