The InverseKinematics() method is using step_tol for checking both convergence (compare with delta_q_norm) and constraint satisfication (compare with error_norm), where constraint_tol is not used in anywhere, but its comment in Kinematics.h says

// Constraint tolerance (default = 1.0e-12). If error_norm is smaller than this value the algorithm terminates successfully, i.e. all constraints are satisfied.

Should it use constraint_tol to compare with error_norm instead of step_tol for inverse kinematics calculation?