For a start, I fixed what I feel was the root problem with hallway routing by changing my algorithm to go from room center to room center and figure out the doors afterwards. It's still necessary to avoid all the other rooms, so it's not quite as simple as routing the hallways and then thinking about rooms, but the application of a few elementary set operations proved sufficient for this approach.
Things didn't go perfectly smoothly at first (pretty good though, really), so I wound up finishing work on the level generation visualizer. Being able to see what was going on proved both fascinating and helpful, and I'm making extensive use of it now. A* is fun to watch in action.
With the aid of the visualizer, I have been able to significantly improve the capability of the level generation engine. It's definitely still a work in progress; I need to work quite a bit more on getting the rooms closer together, particularly in the case of larger levels where things can really go wild at the moment. I haven't got it all robustified yet, so ugly cases still cause some crashes and endless loops, but I've seen it do enough that I'm convinced it can be made to produce an acceptable variety of levels. Eventually I'll get into things like divide-and-conquer to tackle more ambitious layouts, so the horizon still has plenty to offer.
Okay, enough talk, time to give up the goods. Here are a few fairly attractive levels from the latest engine:
|Hub-and-spoke layout with locks & keys|
|Chain of locks and keys. These can get plenty long.|
These are by no means the most complex levels possible now, but things do get a bit spread out looking as we go bigger:
|Longer chain of locks and keys, getting a bit spacey.|
This will obviously need a bit more work, and I've a number of ideas yet to try, so we'll see what comes of that.
I'll also be spending some time with a profiler to try and optimize out some slowness that's making my debug runs drag. It's mostly in the graphics layer (ironically), and I upgraded to a new version of the graphics library (SFML 2.0 RC), which helped some, but really it's got no excuse to be this slow and I'm going to have to hammer it into shape. There should be some good low-hanging fruit, though, so I'm not concerned; at any rate, it runs plenty fast in release mode.
I'm not sure how much longer I'll spend on level generation after that. Once it's running fast enough, can make larger levels that aren't way too spaced out, and I've got a decent library of structures ready to generate, I think it'll be time to proceed on to other things. I need to make the monster pathfinding work again and maybe start some other AI inroads, and then I'd like to change gears completely and start focusing on procedural narrative, to breathe life into these generated environments.
Thanks to everyone for staying interested! There's still plenty more to come.