The TeTra approach is novel, both in its application of transformation to the problem of
automated test data generation and in the way in which it becomes necessary to change the traditional view of
transformation in order to achieve this:
- Transformation merely as a means to an end
Program transformation has previously been regarded as an end in itself, rather than merely
as a means to an end. Hitherto, the goal of transformation research has been to transform poor programs
into better ones. Thus research has tended to assume that the original program will be discarded once
the transformed program has been constructed.
By contrast, in the TeTra approach, it is the transformed program which is discarded and the original
which is retained. Testability transformation
requires the transformed program solely for the purpose of test data generation. Definition 1
guarantees that such test data will be adequate for the original.
Once the test data is generated, the transformed program is no longer required.
- Novel Notions of Equivalence Preserving Transformation
Program transformation has traditionally been concerned with the preservation of
functional equivalence. In the TeTra approach, the transformed program must provide
a wholly different guarantee; that test data generated from it is adequate for the original program.
This means that the transformations applied need not preserve any notion of
traditional equivalence, opening the possibility of a novel set of program transformation rules and
algorithms.
Software test data generation is widely recognised as a hard problem, yet industry standards rightly require thorough testing
which meets well-defined and understood test adequacy criteria [3,12].
Evolutionary testing is a promising approach to attack
the problem of automated generation of adequate test data, yet the structural problems described
earlier inhibit
its wider application.
The TeTra project presents an opportunity to make a significant step forward in
automated test data generation, thereby moving forward research and practice in an area of major concern.
The proposed combination of transformation and evolutionary testing could not come
at a better time.
The DaimlerChrysler evolutionary testing system and the FermaT transformation
system represent the current state of the art. The FermaT system was made publically available under GPL
only very recently (in November 2001), yet the proposers have had two years experience working with it as part of
the GUSTT project.
Mark Harman, Department of Information Systems and Computing, Brunel University, Uxbridge, Middlesex, UB8 3PH.