bboxHexEstimate: Calculated area is way too big

I noticed that maxPolygonToCellsSize is giving me numbers way too big for my input geometries. I could track it down to bboxHexEstimate. There the calculated area is way to big.

Here is an example geometry:
Polygon ((7.54402069689999966 47.13895667900000319, 7.54419258430000017 47.13895667900000319, 7.54419258430000017 47.15093149700000197, 7.54402069689999966 47.15093149700000197, 7.54402069689999966 47.13895667900000319))

It is a street. So it is quite long but narrow.

bboxHexEstimate calculates a distance (bbox diagonal) of 1.33km. This sounds about right. But the calculated area then is 103.87km2 which is way too big. We call polygonToCells with an index resolution of 15 and this causes memory problems in our application.

I do not understand your formula to calculate the bbox area:
double a = d * d / fmin(3.0, fabs((p1.lng - p2.lng) / (p1.lat - p2.lat)));

But it seems to work really bad for bboxes where one side is much much longer than the other.

Do you have any idea how to fix this?

Start by reading bboxHexEstimate and its caller, maxPolygonToCellsSize, then trace how polygonToCells uses the estimate. Reproduce the provided narrow, elongated polygon and inspect the diagonal, aspect-ratio, and area calculations. Done means the estimate reflects the bounding box dimensions without causing excessive memory use at resolution 15, with regression coverage if the existing test structure is identified.