Search-based automatic software test data generation for structural testing depends on the instrumentation of the test goal to construct a many-valued function which is then optimised. The method encounters difficulty when the search is in a region in which the function is not able to discriminate between different candidate test cases because it returns a constant value. A typical example of this problem arises in the instrumentation of branch predicates that depend on the value of a boolean-valued (flag) variable. Existing transformation techniques can solve many cases of the problem but there are situations for which transformation techniques are inadequate.

This paper presents a technique for directing the search when the function that instruments the test goal is not able to discriminate candidate test inputs. The new technique depends on introducing program data-state diversity as an additional search goal. The search is guided by a new evaluation (cost) function made up of two parts, one depends on the conventional instrumentation of the test goal, the other depends on the diversity of the data-states produced during execution of the program under test. The method is demonstrated for a number of example programs for which existing methods are inadequate.