On the Impact of Feature Dependencies when Maintaining Preprocessor-based Software Product Lines
- 1. On the Impact of Feature Dependencies when Maintaining Preprocessor‐based So<ware Product Lines Márcio Ribeiro Felipe Queiroz Paulo Borba Társis Tolêdo Claus Brabrand Sérgio Soares {mmr3, Cq, phmb, twt, scbs}@cin.ufpe.br brabrand@itu.dk
- 4. Problem Developers can introduce compila?on and behavioral errors to features due to feature dependencies… 4
- 5. … leading to lower producKvity Late error detecKon Developers discover the problem only when compiling and execu?ng the problema?c product Difficult navigaKon Developers not aware of feature dependencies Even worse when using VSoC: features are hidden! 5
- 6. To minimize these problems: Emergent Interfaces 6
- 11. Agenda 11
- 12. Study secngs 43 Preprocessor‐based SoMware Product lines Java and C Different sizes and domains Script tool for compu?ng two metrics: MDi: number of methods with preprocessor direc?ves MDe: number of methods with feature dependencies 12
- 14. Effort esKmaKon: VSoC vs. Emergent Interfaces Number of Fragments Number of Features 14
- 15. Methods selecKon Randomly methods selec?on Only methods that contain the feature dependencies we focus on this work Which methods should we select? Many fragments: favoring emergent interfaces Few fragments: no differences 15
- 16. Groups Two groups: Why 2 as a threshold? Differences between both approaches appear from 2 1: both approaches have always the same effort es?ma?on 16
- 17. Methods selecKon to fit the groups Propor?onal selec?on according to each SPL Example: libxml2 Group 1: 953 methods (8 methods selected) Group 2: 125 methods (1 method selected) Majority (23 product lines) 1:1 17
- 18. General algorithm for effort esKmaKon While we do not reach 3 replica?ons For each product line Randomly select methods propor?onally according to the groups For each method Randomly select a variable From this variable, compute NoFa, NoFe, and LOC for both approaches 18
- 19. Results 19
- 20. QuesKon 1: how oMen methods with preprocessor direc?ves contain feature dependencies? 20
- 21. Frequency of feature dependencies Methods with Dependencies Methods with DirecKves 21
- 22. Our data reveal that… Methods with direc?ves that also contain the feature dependencies we studied: Max: 88% (mobile‐rss) Min: 12% (lampiro) So, the feature dependencies we considered may be common in the 43 SPLs we analyzed Only 6 SPLs have MDe/MDi < 50% 22
- 23. QuesKon 2: how feature dependencies impact maintenance effort when using VSoC and emergent interfaces? 23
- 25. Emergent Interfaces effort esKmaKon Effort es?ma?on in all SPLs Methods distribu?on by groups: 25
- 26. When increasing the number of fragments… … the percentage of methods where Emergent Interfaces achieve effort reduc?on also increases… 100% EsKmaKon of Effort reducKon 95% 90% 85% Replica?on 1 80% Replica?on 2 Replica?on 3 75% 70% 65% > 2 > 3 > 4 > 5 > 6 Number of Fragments 26
- 27. Threats to validity Metrics and effort es?ma?on Time beher measures effort… … but the metrics are s?ll useful Overhead to compute emergent interfaces Dependencies Interprocedural, chain of assignments… (not computed) So, the numbers we presented might be even higher… 27
- 28. Concluding remarks How oMen the feature dependencies we analyzed occur in prac?ce? Max: 88%; Min: 12%; Majority > 50% Emergent interfaces can achieve effort reduc?on For the majority of the studied methods (65%): same effort of VSoC Nega?ve impact of VSoC is not so common More significant effort reduc?ons: methods with many fragments (Group 1) 28