Wasted a lot of time because I didn't read the problem statement carefully enough. I thought I could move over the gap in the corner of the keypad, and it messed up all my attempts. Silly me thought I had found a bug in the puzzle because I was finding sequences shorter than the ones in the example. Sequences that were going over the gap, which are illegal.
Eventually, I went to reddit and found one discussion that mentioned "special rules" for avoiding the corner. That was all I needed.
I built abstractions for NumericalKeypad and DirectionalKeypad, along with
some helper classes for Coord and Vector to move across the keypads. The
logic got to have so many corner cases (pun intended) that I had to write some
tests for them.
The main method, #press_sequence, is kinda named backwards. You give it a
sequence that you want to enter on that keypad, and it returns all the movement
permutations that would result in that sequence. Naming is hard and I'm tired.
The Keypad#coalesce method is possibly misnamed, too.
Instead of 2 DirectionalKeypad, there are now 25.
I refactored Keypad so I can chain them instead, à la Decorator design
pattern. I also added a cache in each one to memoize repeated calculations.
It works just fine for Puzzle 1, with one NumericalKeypad and two
DirectionalKeypad, but it seems to get stuck in a loop the moment I add a
third DirectionalKeypad.
UPDATE: I've identified the loop as the #coalesce method. I didn't need to
return the strings of movements themselves, but just the length of the shortest
one. This way, I don't need the #coalesce method at all. The entire thing,
with 25 DirectionalKeypad instances, runs in about 62ms.
I wrote some tests to help me nail down the logic. You can run them with:
./bin/rspec specsOr see the full list of tests with:
./bin/rspec --format documentation specs
> You may need to install RSpec with:
>
> ```bash
> bundle install
> bundle binstubs --all
> ```