redroot/gogen

Gogen Puzzle Solver

Gogen Puzzle Solver

Caveat - I haven't really used anything other than Ruby beyond simple tutorials before this.

Attempts at solving a gogen puzzle solver in various languages

cd ruby & ruby gogen.rb
cd go & go run gogen.go
cd elixir & elixir gogen.exs
cd clojure & lein repl # then (-main) I have no idea what I'm doing here

Coming Soon: Clojure

Thoughts

Ruby

• I was keen to use set operations and there are available on arrays out of the box which was a pleasant surprise

Go

• Set isn't part of the standard library so used a great third party representation. I definitely missed the unified API that Ruby provides for array/sets and maps.
• Typed definition especially involving maps took a little while to grok, plenty of the shortcuts in Ruby unavailable but I appreciated it, if it compiles it tended to run
• Lack of cond || default was a little frustrating.

Elixir

• Fun to use list comprehension style syntax to work on all_positions and all_letters methods
• Piping syntax made nested reduces for data_from_input method messy, decided to use fixed grid size and modulus/rounding to get correct positions
• I initially used a %Position{x,y} struct, but building maps of remaining positions was super slow compared to straight maps, so dropped that in favour of straight maps
• Building letter adjacencies was interesting, reduce a list of strings to a map of Letters of a List of the adjacent ones. Without the ability to set a default values per array matching the reduce approach from the Ruby attempt was messy, but when I realised I could intialise the map outside
• The solve function was another interesting issue using the parameters I'd used in previous solutions, namely: letters_to_find, map of unfound letters and potential positions, letters_found a map of found letters and known positions, and adjacencies, a map of letter adjacencies from the word list. The inability to modify variables outside of the loop, i.e. updating the maps - meant that I ended up nesting reduce statements, and implementing my own break/continue pattern using the accumulator tuple, see here.
• One question that came to me while I was doing above if I could do the same with only one map of known letters - possible but I doubt I could use guards since determining if there was still work to do would involve map_size(Enum.filter(letter_pos_set, fn {k,v} -> length(v) > 1 end)) which I wouldn't be able to write in a guard. Unless I passed the count of remaining letters through as a parameter on each pass. Thats what I did in the elixir-alt branch - simplifying the data structures making the reduce accumlator simpler too and only involving a control parameter within the loop once! See the commit here
• But then I realised I didn't even need to control variable now I was doing the length check in the nested reducer! So I managed to make it even simpler, which was the final edition.

Clojure

• lein repl is super useful, but a pain if there are compilation errors as you have to restart the whole repl to get things to load properly.
• Errors messaging relating to syntax aren't the most helpful out of the box
• Simple pure functions are a joy to write, many different ways of doing it, however complicated control flows took a little while longer and doesn't seem as neat, such as print a few blank lines either side of the grid - later I found out you can have multiple statements in a function body with negates the wrapping method I found first. Feels like its best for people who know exactly what they are doing from the get go rather than stepping through an idea.