This paper deals with generating paths for cutting irregular parts nested on a metal sheet. The objective is to minimize the total non-cutting time for the cutter head starting at a known location, cutting all the required elements and returning to the known location. In contrast to most literature on this topic, a part does not have to be cut at once. If this reduces the total non-cutting time, it is possible to cut a number of elements on a given part, then cut other parts and then return later to finish cutting the given part. The problem is modeled as a generalized traveling salesperson problem with special precedence constraints. An initial feasible solution is generated and improved by local moves embedded in a tabu search framework. The proposed algorithm shows promising results in comparison with a commercial software package on a limited set of test cases.