1. Identification of Variation Points Using Dynamic Analysis
Bas Cornelissen

2. Introduction Product lines Several versions for various customers
Each having its own set of features
Variation points
Configurable features

3. Introduction Identifying variation points in source code
Isolation of code responsible for specific features
Useful in merging product line members
Research proposal
Dynamic analysis
Comparison of traces generated by two versions
Detection and visualization of variation points

4. Method Code instrumentation Execution using similar scenarios
At method level
Aspect-oriented programming
Execution using similar scenarios
Aimed at invoking one particular feature

5. Method Detection algorithm Visualization Sliding windows Parameters
Checksum size
Minimum branch length
Maximum branch length
Visualization
Dot

6. Method Running example Pacman 20 classes, 1000 LOC Various versions
Original (reference)
Map extension
Additional game entities, e.g. holes

7. Preliminary results
Original vs. map version

8. Preliminary results Original vs. map version
Slightly different initialization
One fork…

9. Preliminary results Original vs. map version
Slightly different initialization
One fork…
…and a quick merge

10. Preliminary results
Original vs. hole version

11. Preliminary results Original vs. hole version
Slightly different initialization: a fork …

12. Preliminary results Original vs. hole version
Slightly different initialization: a fork … …and a quick merge

13. Preliminary results → Original vs. hole version
Slightly different initialization: a fork … …and a quick merge
Divergent behavior →

14. Conclusions Efficient algorithm Meaningful results
Parameterized
Branch lengths have upper bounds
Meaningful results
Architect gains quick insight into relevant source code
Creating similar scenarios may be hard in some cases
e.g., deterministic behavior is a must

15. Future work Parameter optimization Improved accuracy
Incorporation of stack depths
Incorporation of method arguments
Better visualization
Large software systems
Support for other abstraction levels

16. Discussion points How can the parameters be optimized?
Maximum/minimum branch length
How can we make the results more meaningful?