Abstraction and Dynamic Analysis for the Reconstruction of Scenario Diagrams

1. Abstraction and Dynamic Analysis for the Reconstruction of Scenario Diagrams Bas Cornelissen Leon Moonen Arie van Deursen

2. Introduction • Scenario • event from the user’s perspective • Scenario diagram • UML sequence diagram at the scenario level • Long-term goal • e.g., conformance checking, design pattern detection, architecture reconstruction (?)

3. Challenges • Static analysis • source code • independent of input • potentially covers all object interactions • Dynamic analysis • execution traces • scenario- and input-driven • polymorphism: detection of late binding • merely applicable in “complete”, executable systems

4. Challenges • Abstraction • typically, initial diagrams: • are too large • show too many details • contain uninteresting interactions • this calls for abstraction mechanisms • omit irrelevant messages and/or objects • allow user to choose the level of detail

5. Method • Reconstruction of scenario diagrams using JUnit and AspectJ • Scenario determination • Producing an execution trace • Abstraction • Visualization

6. Running example • Pacman • 25 Java classes • GUI • Large traces • Polymorphism

7. 1. Scenario determination • JUnit testcase • basically a use case • contains one or more scenarios public class PacmanTest extends TestCase { public PacmanTest(String arg0) { super(arg0); } public void testTopLevelAlphaOmega() throws InterruptedException { Pacman p = new Pacman(); p.start(); System.out.println("SCENARIO START"); p.up(); System.out.println("SCENARIO END"); p.quit(); p.exit(); } }

8. 2. Producing a trace • Aspect for tracing methods and constructors • exclusion of certain classes and/or methods • distinction between run-time objects • at call sites, log: • method/constructor ID • caller + object ID • callee + object ID • method/constructor signature • actual parameters METH 533: Pacman [1] :: Engine [2] :: public synchronized void Engine.movePlayer(int, int) :: 0, -1

9. 3. Abstraction • Compression • summarize recurring messages and patterns • requires pattern recognition • post-mortem • Pruning • hide uninteresting interactions (and objects) • define minimum and maximum stack depths • both at the tracing and the visualization stage

10. 4. Visualization • GNU plotutils package • pic2plot • library containing pic macros for seq. diagrams • Perl script to create pic programs from traces object(Pacman1,"Pacman1"); object(Engine2,"Engine2"); object(Game3,"Game3"); ... step(); active(Pacman1); message(Pacman1,Engine2,"movePlayer(int, int)"); active(Engine2); message(Engine2,Engine2,"inPlayingState()"); message(Engine2,Game3,"initialized()"); ...

11. Example (1) • Compression

12. Example (2) • Pruning

13. Example (3)

14. Example (4)

15. Conclusions (1) • JUnit testcases • enable scenario diagram reconstruction for both complete systems and incomplete systems/modules • AspectJ • flexible: allows for (not) tracing specific components and/or messages • distinguishes between run-time objects

16. Conclusions (2) • Effective abstraction mechanisms • human-readable, high-level model is easily obtained • though omission of messages at deep stacklevels may be infeasible • Better visualization is needed • interactive • e.g., zooming, expanding

17. Questions Fire at will!