RectanglePoints confuses me

When using CartesianIndex to define a rectangle, the resulting rectangle is not bounded by the pixels mentioned in the CartesianIndex.

Example:

julia>  RectanglePoints(CartesianIndex(1,2), CartesianIndex(3,4))
RectanglePoints(Point(1, 2), Point(3, 4))

But Point(1, 2) is the pixel with CartesianIndex(2,1).

The definition says:

 RectanglePoints(p1::CartesianIndex{2}, p2::CartesianIndex{2}) = RectanglePoints(Point(p1[1], p1[2]), Point(p2[1], p2[2]))

To me it would make a lot more sense to define it as

 RectanglePoints(p1::CartesianIndex{2}, p2::CartesianIndex{2}) = RectanglePoints(Point(p1[2], p1[1]), Point(p2[2], p2[1]))

Of course this would be breaking.

Am I just thinking about this the wrong way?

Start at the RectanglePoints constructor overload for CartesianIndex{2} and reproduce the example from the issue. Compare its coordinate ordering with the surrounding Point and CartesianIndex conventions and inspect relevant call sites before deciding whether the behavior should change. Done means an agreed, documented API decision with regression coverage, accounting for the breaking-change concern.