AFAICT, drek's method essentially just replaces one step of mine with a more efficient sub-algorithm. The stage in question generates a fully-connected tree, which I've done by culling starlanes, whereas drek is suggesting adding starlanes. There are minor differences in the results (I can get loops at this stage, he can't), but they're essentially the same, except that drek's should be faster.
Then again, maybe I'm wrong, and drek wan't to use just
the tree-maze generation as the final product... I don't think that'd be a very good starlanes
maze though... but it could be improved by adding severeal simultaneously growing trees starting at different locations.
If we want more other methods to generate starlanes, I'd suggest using something like what I used to generate this (from a few pages ago in the thread):
If that algorithm was paired with a pre-stage that added a fully connected tree, then it'd be guaranteed to be fully connected as well.
Also, the algorithm I'm uses does have a few magic numbers that can be replaced with user specified parameters (or parameters derived form user choices)... in particular, the maximum distance allowed between "adjacent" systems that aren't connected with a starlane before I say they're too far apart and need to be directly connected. This would have a significant impact on how "well connected" the final map is... and likely how many starlanes there are on an average star, though I haven't experimented much.
FYI, the only reason I'm not already using drek's suggestion in my implimentation was that I approached this particular soluation from a roundabout way... I originally was doing the culling / adding back in a single loop, but decided the results would be better with two separate loops, which in retrospect is essentially the same as making a fully-connected tree then adding some extras. The actual criteria I used to pick where to add more lanes than those of the initial tree is of my own creation (distance to adjacent systems) though...
That said, it'd probly be good to have a starlanes generation algorithm that doesn't start with any sort of fully connected tree...