The main tractability problem of the constraint-based polyhedra abstract domain can be derived from the costly (strong) join operation, that is, the convex hull computation. This paper presents a series of cheap weak join operations as a sound substitution for the convex hull operation of the constraint-based polyhedra domain. To achieve a trade-off between efficiency and precision, a heuristic strategy is proposed which dynamically combines both strong join and weak join during program analysis. Experimental results show that the weak join operation can significantly improve the efficiency, scalability and robustness of the constraint-based polyhedra domain.