Heuristics for Discovering Architectural Violations (WCRE 2013)

Software architecture conformance is a key software quality control activity that aims to reveal the progressive gap normally observed between concrete and planned software architectures. In this paper, we present ArchLint, a lightweight approach for architecture conformance based on a combination of static and historical source code analysis. For this purpose, ArchLint relies on four heuristics for detecting both absences and divergences in source code based architectures. We applied ArchLint in an industrial-strength system and as a result we detected 119 architectural violations, with an overall precision of 46.7% and a recall of 96.2%, for divergences. We also evaluated ArchLint with four open-source systems, used in an independent study on reflexion models. In this second study, ArchLint achieved precision results ranging from 57.1% to 89.4%.

Transcript

1. Heuristics for Discovering Architectural Violations Cristiano Maffort, Marco Tulio Valente, Mariza Bigonha, Nicolas Anquetil, Andre Hora 1 WCRE, Koblenz, Germany, October 2013
2. Software Architectures 2 Idealized
3. Software Architectures 3 Idealized Concrete
4. Architecture Conformance  Goal: expose architectural violations  Implementation decisions not following the planned architecture 4
5. Architecture Conformance  Goal: expose architectural violations  Implementation decisions not following the planned architecture  Two techniques:  Domain-specific languages  Model-based techniques 5
6. Domain-Specific Languages 6 .QL DCL
7. Domain-Specific Languages 7 .QL DCL Problem: detailed definitions of constraints
8. Model-based Techniques 8 Reflexion Model
9. Model-based Techniques 9 Reflexion Model Problem: successive model refinements
10. Our Goal 10  Lightweight approach for conformance:  Without constraints  Without model refinements  Minimal input information:  Just a high-level component model (HLM)
11. ArchLint  Our inspiration: 11 Bug Finding Tools (Lint, FindBugs, PMD, FxCop, etc) Warnings
12. ArchLint  Our inspiration:  Our approach: 12 Bug Finding Tools (Lint, FindBugs, PMD, FxCop, etc) Warnings
13. ArchLint  Heuristics to detect absences and divergences  Absences: dependencies prescribed by the HLM  But that are not present in the concrete model  Divergences: dependencies not prescribed by the HLM  But that are present in the concrete model  Heuristics:  False positives are expected 13
14. Heuristics 14
15. Heuristics for Detecting Absences 15 Evidence #1: 4 classes dependend on TargetClass
16. Heuristics for Detecting Absences 16 Evidence #1: 4 classes dependend on TargetClass Evidence #2: 3 classes initially created without this dependency, but inserted it later
17. Heuristics for Detecting Absences 17 Evidence #1: 4 classes dependend on TargetClass Evidence #2: 3 classes initially created without this dependency, but inserted it later Evidence #1 && Evidence #2  architectural warning C2 missing a dependency with TargetClass
18. Heuristics #1 for Divergences 18 Evidence #1: cp2 is a heavy-user of TargetModule
19. Heuristics #1 for Divergences 19 Evidence #1: cp2 is a heavy-user of TargetModule Evidence #2: c2 is the only class in cp1 that depends on TargetModule
20. Heuristics #1 for Divergences 20 Evidence #1: cp2 is a heavy-user of TargetModule Evidence #2: c2 is the only class in cp1 that depends on TargetModule Evidence #3: dependency to TargetModule were removed from Cp1
21. Heuristics #1 for Divergences 21 Evidence #1: cp2 is a heavy-user of TargetModule Evidence #2: c2 is the only class in cp1 that depends on TargetModule Evidence #3: dependency to TargetModule were removed from Cp1 Evidence #1 && Evidence #2 && Evidence #3  architectural warning C2 should not depend on TargetModule
22. Evaluation #1 22
23. Evaluation 23  SGA System:  EJB-based information system  4,923 revisions  March, 2009 to September, 2011  1,852 classes and interfaces  127 KLOC
24. Thresholds  Absences:  Asca = 0.9 -- 90% of the classes have dependency  Ains = 0.35 -- 35% inserted the dependency  Divergences  Dsca = 0.05  Ddel = 0.65  Ddir = 0.05 24
25. Precision and Recall 25  Classification made by the system’s architect  For absences, architect classified each warning  For divergences, architect classified existing dependencies
26. Precision and Recall 26  Classification made by the system’s architect  For absences, architect classified each warning  For divergences, architect classified existing dependencies
27. Precision and Recall 27  Classification made by the system’s architect  For absences, architect classified each warning  For divergences, architect classified existing dependencies
28. Evaluation #2 28
29. RM Dataset  RMs independently proposed by Bittencourt et al.  Ant  ArgoUML  Lucene  SweetHome3D  False positives/negatives regarding such models  Same thresholds from the 1st evaluation 29
30.  Precision Results for Divergences 30
31.  Precision  Recall Results for Divergences 31
32. Conclusion  ArchLint:  Lighweighted approach for architetcure conformance  Minimal input  Static and historical analysis  Architetcural warnings  Future work: heuristics based on data mining techniques (association rules) 32
33. Thanks! Heuristics for Discovering Architectural Violations Cristiano Maffort, Marco Tulio Valente, Mariza Bigonha, Nicolas Anquetil, Andre Hora 33 WCRE, Koblenz, Germany, October 2013