An Empirical Study of Design Degradation: How Software Projects Get Worse Over Time

ESEM2015

Transcript

1. An Empirical Study of Design Degradation: How Software Projects Get Worse Over Time I"ekhar Ahmed, Umme Ayda Mannan, Rahul Gopinath, Carlos Jensen October 22, 2015
2. What this paper is about •  Large longitudinal study of code smells in open source projects •  Confirms and quanIfies what many assume to be true •  Corrects some misconcepIons •  Adds new details about how designs degrade July 15, 2016 1
3. What is a Code Smell 2 •  Neither syntax errors nor compiler warnings. •  Bad program design or bad coding pracIce. •  Duplicate code •  High CyclomaIc complexity •  Code smells are associated with •  Maintainability [Fowler 2002] •  Bugs [Oliva et al. 2013] •  Design debt [Izurieta & Bieman 2007] October 22,2015
4. Introduction 3 Project characteris-cs Code smell Longevity of project Decreased number of code smells Number of core developers Increased number of code smells Frequency of code changes Increased number of code smells Number of files in code base Increased number of code smells October 22,2015 Previous studies have shown relaIonships like:
5. Introduction 4 But how do code smells evolve? Do these findings hold over a large sample of projects? Project characteris-cs Code smell Longevity of project Decreased number of code smells Number of core developers Increased number of code smells Frequency of code changes Increased number of code smells Number of files in code base Increased number of code smells October 22,2015 Previous studies have shown relaIonships like:
6. Research questions •  How do code bases evolve over Ime wrt. code smells? •  Who addresses technical debt (code smells)? •  Is there a correlaIon between testedness and smellyness? •  Are we focusing on the smells coders struggle with? 5 October 22,2015
7. Methodology July 15, 2016 6
8. Methodology 7 October 22,2015 220 Projects Cap at 200 wks 33,070 commits 3,426 smells Analysis Randomly selected top 500 Java projects using Maven from Github •  10+ weeks of history •  500+ lines of code •  10+ files •  Could be successfully built
9. Methodology 8 October 22,2015 220 Projects Cap at 200 wks 33,070 commits 3,426 smells Analysis To remove the long tail, capped commit data at 200 weeks •  Threw out 5% of the commit data
10. Methodology 9 October 22,2015 220 Projects Cap at 200 wks 33,070 commits 3,426 smells Analysis For each commit to each project •  Calculate # of code smells at that point in Ime •  Calculate test coverage metrics (statement, branch & path) Within project metrics
11. Methodology 10 October 22,2015 220 Projects Cap at 200 wks 33,070 commits 3,426 smells Analysis Mapped smells to higher-level categories
12. Methodology 11 October 22,2015 220 Projects Cap at 200 wks 33,070 commits 3,426 smells Analysis Normalized results to allow inter-project comparisons •  0-1 mapping, 0=min count, 1=max count
13. Findings July 15, 2016 12
14. How code smells evolve 13 Week-wise average project smelliness October 22,2015
15. Some grow continuously 14 October 22,2015 Bloater smells
16. Others plateau… or dip 15 October 22,2015 OO abusers
17. How code smells evolve 16 Week-wise average project smelliness October 22,2015
18. Commits introducing or removing smells 17 October 22,2015
19. Who removes code smells? 18 October 22,2015 10% of developers responsible for removing most smells
20. Testing does not correlate with smell removal 19 October 22,2015
21. Research vs. Practice 20 October 22,2015 0 5 10 15 20 Data Clumps Data Class Cyclic Dependencies Blob OperaIon DuplicaIon Feature Envy SAP Breakers God Class Intensive Coupling Schizophrenic Class Blob Class Unstable Dependencies TradiIon Breaker Refused Parent Bequest Message Chains Shotgun Surgery Distorted Hierarchy UnnecessaryCoupling Rank From Literature From Projects More Important (Practitioner)
22. Conclusions •  Projects get smellier over Ime •  But not all smells created equal? •  Is this true for non open source projects as well? •  A small set of core contributors responsible for fixing smells, but not keeping up •  TesIng and refactoring do not go hand in hand •  Though coverage and some smells correlated •  Is it done by the same people though? •  PotenIal mismatch between interesIng smells and common smells •  May hinder interest and adopIon of code smell tools 21 October 22,2015
23. Thank you May 08,2014 22