I wanted to explore how to represent conveyor belts on a grid.

In Factorio^{[1]}, each grid tile with a conveyor belt has a *direction*. Adjacent tiles combine together to determine what type of conveyor belt is on the tile. The main operations are: build belt, destroy belt, rotate belt. The model is simple and can express straight belts, curved belts, and joins. It does not support splits, and Factorio uses a separate splitter object for that. Each **tile** has **5** states: empty, east, north, west, south. **Click a tile to cycle through the states**:

In Production Line^{[2]}, each grid tile with a conveyor belt may have multiple flows, allowing for both joins and splits. I implemented a data model that supports that functionality, but what I show here is not exactly the same as what the game uses. Each **tile** has **2** states: full, empty. Each **edge** between adjacent full tiles has **3** states: east/south, west/north, neither. **Click a tile or an edge to cycle through its states:**

Another model is for each tile to have flow information on all four edges, but that allows the possibility that adjacent tiles have inconsistent flow directions.

*Both games have a much nicer drag-move UI than what I have here.* The UI here was just for me to edit the state and see the results.

**Thoughts:** I was hoping to find that one of these models was much better than the other, but I didn't find that. The tile model is better by being simpler. The edge model is better by supporting 2-way and 3-way splits. The tile model has some nonsense configurations, such as two tiles pointing at each other. The edge model also has some nonsense configurations, such as a tile that doesn't point anywhere.

(Making-of: when I started this page I thought I would find a different model, involving connecting adjacent edges together. I built some diagrams and they made me realize my model wouldn't work. I slept on it, slept on it again, and came up with a simpler model.)

Implementation note: I used edge coordinates x,y,W and x,y,N as described in my article on grid parts. Each tile has data for itself (full or empty) and its *north* side and its *west* side. To get information about the east side, the tile at x,y asks the tile at x+1,y for its west edge. To get information about the south side, the tile at x,y asks the tile at x,y+1 about the north side. The edge flow is stored only once and used by both adjacent tiles.

Also see my page about grid edges to see a version of this without directions on each edge.