Fixed another bug in to_mesh_iter!

This was a more minor bug - but a bug that was making it produce
non-manifold geometry.

README.md

@@ -2,8 +2,7 @@
 
 ## Highest priority:
 
-- Clean up `ramhorn_branch` because it's fugly.  Also, fix
-  non-manifold stuff at higher recursions.
+- Clean up `ramhorn_branch` because it's fugly.
 - See `automata_scratch/examples.py` and implement some of the tougher
   examples.
   - `spiral_nested_2` & `spiral_nested_3` (how to compose
@@ -15,8 +14,9 @@
 - Elegance & succinctness (my recent closure work may help with this):
   - Why must I repeat myself so much in these definitions?
   - What patterns can I factor out?  I do some things regularly, like:
-    the clockwise boundaries, the zigzag connections, the iterating over
-    a `Vec<Vertex>` to transform each element and make another vector.
+    the clockwise boundaries, the zigzag connections
+  - Procedural macro to shorten this `Tag::Parent`, `Tag::Body`
+    nonsense - and perhaps force to groups of 3?
 - Docs on modules
 - Grep for all TODOs in code, really.
 - Look at performance.  Can I save on copies of geometry by using

src/rule.rs

@@ -203,9 +203,16 @@ impl<S> Rule<S> {
                 geom = geom.connect(vec![(geom2, child.vmap.clone())]).0;
                 // TODO: Fix clone?
 
-                // and backtrack:
+                // If we end recursion on one child, we must end it
+                // similarly on every sibling (i.e. get its geometry &
+                // final geometry, and merge it in) - so we increment
+                // s.next and let the loop re-run.
+                s.next += 1;
+                if s.next >= s.rules.len() {
+                    // Backtrack only at the last child:
                     stack.pop();
                     n -= 1;
+                }
                 continue;
             }