[josm-dev] Geometric calculations and projections

[Robert Scott]

On Tuesday 08 December 2009, Matthias Julius wrote:
> I guess local cartesian coordinates would be the best for things like
> orthogonalize since that's the way buildings are built.  But, in
> practice, using Mercator should be close enough for real world
> building sizes.

Close enough maybe, but I don't think doing the calculation in 3-space cartesian coordinates is significantly more complex. In fact, I'd argue that it's simpler.

And remember people aren't just editing buildings. Think of rectangular landuse areas that may have side lengths kilometers long.

> In other words: if the current projection makes circles look like eggs
> then I would expect this to be true for both the imagery and the
> circles created by AlignInCircleAction.

Yes, I understand that, but a user tracing imagery will probably be expecting the application to behave like a drawing application, with an 'extrude' operation extruding perpendicular to the object as seen on screen, not going off in some random direction that it thinks is clever. Some users don't understand/want to care about projections, and just want to draw things 'as-seen'. Take for example a straight edged object that's non-square, but happens to be distorted vaguely inversely to the map distortion, so that it appears nearly square on screen. A user wishing to trace it would get puzzled and frustrated with an extrusion going off in the seemingly wrong direction.

This may be a far fetched case for extrusion, but I think it's a more obvious problem for rotation. A user using the rotation operation often doesn't actually want to literally rotate the literal object as they would in real life, rather people sometimes use it just to visually manipulate the data they're working with. In which case they would want to rotate in screen space, not real space.

I don't know if I'm explaining this very well, but my point is I think maintaining the option to do operations in screen space is valuable, even if it is not the default.


robert.