A simple Sudoku solver that I've done to experiment with SAT/SMT solvers. It currently only runs for boards up to 36x36 (k = 6). For anything bigger, MiniSAT chokes anyway.

THIS WAS ONLY TESTED ON CYGWIN!!!

Install Haskell Platform and do

  $ make clean all -j4
  $ make check

To profile, export PROFILE=1 and rebuild.

• plot timing graphs (time vs. #constraints, #constraints vs. #vars)
• see how rule order influences MiniSAT timing:
  • randomize
  • spatial (rules for ij go together)
• find all solutions (plot graph number-of-sols vs. percentage)
• experiment with different encodings:
  • Sudoku as a SAT Problem (http://anytime.cs.umass.edu/aimath06/proceedings/P34.pdf)
  • A SAT-based Sudoku Solver (https://www.lri.fr/~conchon/mpri/weber.pdf)
  • Optimized CNF Encoding for Sudoku Puzzles (http://www.cs.cmu.edu/~hjain/papers/sudoku-as-SAT.pdf)
• see how Haskell people do interfaces to MiniSAT or other external solvers (and how they cope with big CNFs)
• experiment with SMT solvers
• find all solutions

• pass several lists to CNF instead of one (this should speed things up because we won't copy data when concatting rules)
• store temp files to separate folder
• Haskell coding style
• write Haddocks
• run for inputs up to 100x100 (k=10); ideas:
  • understand where is all the memory going to
  • generate all rules for each cell in one function
  • use streams