Now that the algorithm does not need a new test_board every time, we no longer need to treat the Node object as a Flyweight. We can move the board into the Node object and remove it from the parameter list of all the functions that operate on it.

The code in tree_to_solution_string mixes the logic of solving the puzzle with the management of a linked list. Splitting your attention between these two levels can make it hard to track down errors. To continue teasing these two aspects apart, we need to make heavier use of the iterator. We’re in the middle of it now, and the code might actually feel like it is more of a mess than when we started. That is common, natural, and nothing to be afraid of.

Well, unless we get directed on to a different task tomorrow. Lets finish this up tonight.

In a previous article, I had to shorten a bunch of lines that had a row and column value used as indexes to the board array. This repeated pattern is a call-to-action.

We want to encapsulate the logic for referring to a particular place on the board, and for advancing through the board. This is the responsibility of the Iterator pattern.

Spoiler Alert: we don’t get all the way there in this article.

This refactoring is my bread and butter. Functions tend to grow. Eventually, you need to split them. Find a section of the method that has its own self-containerd reason- for existence, and make that its own function.

I have in the back of my head that I want to extract a class that abstracts the boards from this code. I’ve been resisting the urge thus far, as keeping the board as a 2D array of cells is really conducive to sharing with other implementations of this code. However, the following refactoring should work to support either direction: pull out the code that constructs a board from the string. This has the added benefit of making it possible to write a unit test that calls tree_to_solution_string without having to parse all of the strings.