stbrumme / microsat-cpp

A SAT solver in under 300 lines C++

MicroSAT/C++

Marijn Heule's MicroSAT ( https://github.com/marijnheule/microsat ) is a great compact SAT solver written in C.
It has a few limitations so I decided to convert it into a single-header C++ library.
Note: the main algorithm wasn't changed - therefore execution speed is pretty much identical.

Sample Code

  #include "microsat-cpp.h"

  MicroSAT s(numVars);                                           // set number of variables
  s.add(+var1); s.add(-var2);                                    // add units
  s.add({ -var1, var2 }); s.add({ -var1, -var3 });               // add clauses
  if (s.solve()) std::cout <<   "SATISFIABLE" << std::endl;      // run solver
  else           std::cout << "UNSATISFIABLE" << std::endl;
  std::cout << "variable 3 is " << std::boolalpha << s.query(3); // query variable (true or false)

The examples folder contains a fully working Sudoku and Hitori solver as well as a basic CNF file reader.

Features

MicroSAT was originally a standalone program which parses a CNF file and computes its satisfiability. On the other hand, my code is intended as a library being fed with units, clauses, etc. on-the-fly. The interface is as simple as possible, it has just 3 different public functions (plus constructor/destructor).

The syntax follows closely the CNF file format:

• variables are referred to by a number which must not be zero
• a positive unit is true, a negative unit is false
• clauses referencing negated variable use negative numbers
• unlike CNF, clauses are not terminated by a zero

The library should compile on most C++ compilers without any warnings. I tested it on a variety of GCC versions, CLang and Visual C++.

Improvements

1. fetch result

• get model / final state of a variable (query function)

2. memory handling

• allocated memory is properly freed after use
• slightly more compact (approx. 10% less memory while solving)
• throw exception "out of memory" if constructor's mem_max parameter was too small

3. overloaded add

• accepts units (single integer)
• accepts clauses (multiple integers), can be any STL container

4. expose only required functions

• just constructor, destructor, add, solve and query

5. minor bugfix

• m_false[0] needs to be initialized as zero

Removed

1. now no dependencies to any libraries: yes, there is no #include <> at all
2. no DIMACS CNF file handling in base library

• see cnfreader.h

Limitations

1. only very basic error handling
2. you need to know your approximate memory usage right away in the constructor