As the traditional approaches to generative planning have failed to produce efficient, general planning systems, many researchers have turned their attention to planning as constraint satisfaction. The fundamental insight behind this interest is that by limiting the length of the plan, the planning problem is in NP instead of being in PSPACE. In this paper we show that despite some advances in this area, there is no satisfactory method for solving planning problems as constraint satisfaction problems. Current methods are either not general enough, or they yield satisfiability problems too large for satisfiability engines to handle. We analyze the current state of the art, hoping that future research efforts will benefit from the lessons learned.