Comment by unclebobmartin

Ah, yes. The famous Sudoku solver controversy.

In 2006 Ron Jeffries wrote four blogs about solving Sudoku in Ruby with TDD. His blogging effort ended before he completed the solution. I think he got interested in something else and left the whole thing hanging.

That same year Peter Norvig wrote a Sudoku solver in Python using a constraint based approach. You can see their respective documents here. https://ronjeffries.com/categories/sudoku/… https://norvig.com/sudoku.html

The anti-TDD lobby, at the time, hailed the two documents as proof that TDD didn't work. Ha Ha, Nya Nya Boo Boo.

I was aware of this silliness, but never bothered to study it. I had better things to do. Until March of 2020. Then I thought I'd use Sudoku as a case study for Episode 62 in http://cleancoders.com.

I had not read either of the previous documents and decided to maintain that ignorance while writing the solver in Clojure using TDD. It turned out to be a rather trivial problem to solve. You can see my solution in http://github.com/unclebob/sudoku

I don't know why Ron stopped blogging his Sudoku solver in 2006; but he picked it up again in 2024 and has written much more about it.

The trick I used to solve Sudoku with TDD was to consider the degenerate cases. Sudoku is usually a 3x3x3x3 grid. Let's call this a rank-3 problem. I started with a rank 1 problem which is trivial to solve. Then I moved on to a rank 2 problem which was relatively simple to solve; but was also very close to a general solution. After that I could solve rank N problems.

The TDD strategy of starting with the most degenerate case (rank 1) and then gradually adding complexity may not have been well known in 2006. TDD was pretty new back then. If you explore Ron's first four blogs you can see that he briefly considered rank 2 but opted to go straight into the rank 3 case. The sheer number of variables for each test (81) may have played a role in his loss of interest. In my case (rank 2) I had far fewer variables to deal with.