With LDD Being Discontinued I've elected to migrate to LDraw and would like some help

[Roland Melkert]

I understand why snapping is preferred over collision detection.  With collision detection, you've got about 1000 triangles per part with only so much precision to their vertices.  Plus, let's not forget that it's not really collision detection - collision detection is for finding when and how two objects touched each other, not about finding when the static friction between two non-overlapping objects is greater than some particular applied force yet less than what a child could exert.

I don't think this kind of snapping is possible using just the raw LDraw data as even with bfc you are not always sure about what is inside or outside etc.

But what "degenerate" cases would you watch out for if you were to try to implement "collision detection"?

Not all fitting connections are modeled using the exact same shapes / tube orientation etc. In LDCad I mainly simplified most things to pen / hole variants and most of the rest is done using simple (group) named orientation hotspots.