Understanding Concerns in Software: Insights Gained from Two Case Studies Meghan Revelle, Tiffany Broadbent, and David Coppit

Size: px

Start display at page:

Download "Understanding Concerns in Software: Insights Gained from Two Case Studies Meghan Revelle, Tiffany Broadbent, and David Coppit"

Buddy Martin
5 years ago
Views:

1 Understanding Concerns in Software: Insights Gained from Two Case Studies Meghan Revelle, Tiffany Broadbent, and David Coppit Department of Computer Science The College of William and Mary

2 Separation of Concerns in Software Encapsulation, separation, and abstraction of concerns key ways to fight complexity [Dijkstra76, Parnas72] Programming language support: subroutines, abstract data types, classes, aspects Reduce scattering of concerns across modules, and tangling of concerns within a module Reduced complexity yields less faults, lower maintenance costs, higher productivity 2

3 No Consensus on Concerns Definitions vary: General: "any consideration about the implementation of a program" [Robillard] Specific: features [Lai & Murphy] IEEE: "concerns are those interests which pertain to the system s development, its operation or any other aspects that are critical or otherwise important to one or more stakeholders" 3

4 Research Questions What is a concern? What types of concerns are there? How subjective are concerns to programmers? Do programmers think at similar levels of abstraction? How are concerns manifested in code? At what granularity? Are there factors that impact identification of concerns in code? 4

5 Approach Two case studies, where two programmers identified concerns in an existing system GNU sort: C, imperative, batch, feature-oriented Minesweeper: Java, object-oriented, interactive Compared the set of concerns found by each Compared the code for the concerns Classified concerns into types Studied factors for agreement 5

6 6

7 Metrics Overlap: Concern code programmer agreement Tried character and line accuracies Spread: The distribution of a concern in code 7

8 Concern Abstraction Mapping related concepts between programmers Shapes Shapes Triangles Squares Triangles Circles Squares Circles 8

9 Case Study #1: GNU sort Sorts lines from STDIN or files, merges sorted files, checks sortedness 2145 lines of C Batch Feature-oriented Revelle identified 50 concerns Previous data for 83 concerns [Carver & Griswold 1999] 9

10 Case Study #1: Concern Comparison 23 concerns in common, with >80% average character overlap Differences due to: Different levels of abstraction (more detail) System-specific concerns Sort Mode Modes Check Mode Merge Mode After abstraction mapping: Ours C&G's - 7 system-oriented concerns 10

11 11

12 Case Study #2: Minesweeper Game to find mines without exploding them 2776 lines of Java in six classes Interactive GUI-oriented Revelle identified 30 concerns Broadbent identified 26 concerns 12

13 Case Study #2: Concern Comparison 13 concerns in common, with about 50% character overlap Differences due to: Different levels of abstraction Not as feature-oriented as sort After abstraction mapping: Revelle's - 1 Broadbent's

14 14

15 15

16 Abstraction and Concern Code Overlap Level of Abstraction GUI 15.00% Game State 34.86% Game Difficulty 30.00% Debug 40.12% Graphics 35.98% Minefield 47.65% LED 45.80% Images 52.93% Cell State 79.32% Flag Cell 57.38% Error Handling 81.25% Menu 88.87% Timer 78.87% 0.00% 20.00% 40.00% 60.00% 80.00% % Character Concern Overlap 16

17 Spread and Concern Code Overlap Number of Classes GUI 15.00% Game Difficulty Game State 34.86% Graphics 35.98% Debug 40.12% Images 52.93% LED: 45.80% Minefield 47.65% Flag Cell 57.38% Error Handling 81.25% Timer 78.87% Cell State 79.32% Menu 88.87% % 20.00% 40.00% 60.00% 80.00% % Character Concern Overlap 17

18 Observations Measuring overlap accurately is tricky Character-level only within constant strings Tokens elsewhere Can be helped with tagging guidelines Context matters if(currentcell.getstate()!= Cell.STATE_FLAGGED) The meaning of a concern is important Programmers identify similar concerns, but work at different levels of abstraction 18

19 Types of Concerns Feature GUI, sorting, merging Domain-Independent Unit of Functionality Debugging, error handling Input/Output Internal Program Characteristic Caching, buffers Language Characteristic Exception handling, constructors 19

20 Concern Identification Guidelines 1. Review the file and look up any unfamiliar constructs of the language. 2. Identify the main pieces of the program (features); they are concerns. 3. Constants, user-defined types, class attributes and imported classes are good indicators of concerns. 4. Entire functions usually relate to a concern or support a concern (except for main). 5. When you create a concern, decide what it encompasses. 6. Look for domain independent concerns such as debugging and error handling. 20

21 Concern Tagging Guidelines 7. Different levels of concerns can be tagged in the same code fragment. 8. Even though a code fragment is tagged with one concern, it can be tagged with another concern. 9. Use the search feature to find code that is possibly related to the current concern of interest, but take the time to figure out the context of the code before tagging it. 10. If a function is tagged with a concern, the calls to it should also be tagged. 11. If the whole body of an if or switch statement is tagged, tag the if or switch as well as the beginning and ending braces. 12. If the whole body of a loop is tagged, tag the loop conditionals as well as the beginning and ending braces. 13. Make sure to tag both the declaration and use of variables associated with a concern. 14. When a variable is an argument or parameter to a function, tag only the argument or parameter and associated type. 15. Tag the whole expression when it affects a concern variable. When a concern variable is used on the right side of an assignment statement, tag only the use of that variable. 16. White space, new lines and comments should be included when tagging concerns. 17. Most to all of the code in a file should be tagged. 21

22 Evaluation Only two data points for small systems Need more study of contributing factors Studied scatter (average distance between instances of a concern in the file) Taxonomy only a start Patterns Hard to control all factors Example: Spread vs. size 22

23 Conclusion Two case studies investigate how programmers think about concerns, and identify concern code Identified 5 types of concerns Developed (but did not evaluate) guidelines Spread and abstraction are key factors that affect concern identification Programmers do identify same concepts But at different levels of abstraction Concerns manifested at character level 23

Software Plans for Separation of Concerns

Software Plans for Separation of Concerns David Coppit and Benjamin Cox The College of William and Mary 9/11/13 1 Elevation 9/11/13 2 Floor Plan 9/11/13 3 Foundation Plan 9/11/13 4 Software 9/11/13 5 Example: