Testing with Hypothesis turned up some matches in Soffit that took a long time to run. Some optimization of constraint checking got the runtime down from 600ms to about 400ms, but now I'm kind of stuck.

Can you find a path in the following graph that sums to 100? When you follow a blue edge, you add the number, and when you follow a red edge, you must subtract the number. So, starting at 13 and following

I've continued work on Soffit, but more slowly in the past couple weeks. My focus had been getting better testing covering using Hypothesis. My first attempt was to generate graphs by their edges, like

Dr. Joris Dormans wrote a chapter in "Procedural Generation in Game Design" (Tanya X. Short and Tarn Adams, editors) about cyclic generation, and he included this short example of a graph grammar

For Soffit I implemented a few Constraint objects with the python-constraint library. These objects represent a particular restriction on the solution, and they can prune the search space either ahead

A lot of the existing packages for graph grammars seem to have GUIs for writing rules, and in some of them that seems to be your only option. Tonight I prototyped a way to export a graph grammar as a HTML

This is the first example I tried that really pushed my engine hard. It would be better broken up into multiple grammars, I think. I wanted to run Rule 30, a 1-d cellular automata, as a graph grammar in

Let's create a graph grammar with Soffit that generates a fixed-sized binary tree. How do we tell a graph grammar when to stop working? One option is to only run it for a fixed number of iterations, but

Suppose we want to create a graph grammar that makes rectangular grids. A first cut at it would be something like "expand outwards from any edge that hasn't already been used", something like

There's something in my soffit, and this pool is starting to drive me nuts.