A Uniform View of Backtracking

Backtracking search is a standard mechanism for solving constraint satisfaction problems (CSPs). Over the years a wide range of improvements of generic backtracking have been developed. These improvements have employed a seemingly wide range of insights, each accompanied by its own algorithmic techniques and data structures. In this paper we demonstrate that despite this seeming variety there is in fact a uniform way of viewing these improvements. In particular, we demonstrate that different backtracking algorithms can be categorized by the manner in which they discover, use, and store for future use, no-goods. This understanding can be used to provide a simplified presentation, a uniform implementation, and a simple theoretical framework for these algorithms. This unification also provides us with the appropriate conceptual apparatus to extend these algorithms in the non-binary case, and to identify new improvements.