I had to bang my head against the trigonometry to get it all correct. Although I know the rules of trig, I always have a hard time applying them in the real world because I have a hard time imagining the shapes involved. Ended up drawing little pictures of what was doing on.
Its really great, btw .. I play with this toy with my son on a regular basis, and found myself just a few days ago wondering what is the maximum number of shapes we can build with the set .. okay, this got a bit screwed up by the addition of another mini-set gifted to my son by a friend, so there is room for improvement in my thinking now that I've learned what you have discovered. Nice job!
My thought process was something like this (over a number of days grabbing time when I could):
Emulate a human selecting pieces and clicking them together. Model pieces as straight lines of the right length. There are two important things: how long the line is and how much it modifies the angle at which the next piece will join. Thus all I need is to keep track of the position of the end of the last piece added and the angle.
Spend a while figuring out the trigonometry and getting that code working correctly. To generate the permutations I realized that the process could be split into three parts:
1. Always place the bridge. Since there's one just add it at the end.
2. Separate the straights and curves. Generate all the permutations of clockwise and anticlockwise curved pieces. To do that I used the trick of all 12-bit binary numbers between 0 and 2^12-1.
3. Then insert the straight pieces between the curved pieces in all places possible.
Feed each possible track layout into code that builds the track and check that the end pieces match up. Add code to eliminate end pieces that join impossibly (come from the wrong angle), add code to remove duplicates caused by rotation and reflection.
Once I got that working I then wrote code in Processing to draw a pretty picture.
2a. Continue with next iteration unless the turning number of the sequence of curves is a whole number.
You add 1/8 to the turning number for each clockwise curve and subtract 1/8 for each CCW (for numerical stability you would rather add of subtract 1 and check for multiples of 8).
That shouldn't be too hard. A little harder perhaps to then determine if a given permutation is a closed loop or not, but essentially it should be basic geometry (didn't look at the code). Even if that naive approach would not work, it would at least be a way to start. The remaining sub problems should be less scary to tackle.
You should be able to generate all possible loops relatively cheaply, since you're just moving characters around. You might need a separate character for the original circle/curve parts though, to make it possible to find the originals.
I never got around to implementing this - it was just an observation while building interesting tracks for my children, but I might have a go...