Does JavaScript Software Embrace Classes? (SANER 2015)

In this talk we report on a large and in-depth study to understand how class emulation is employed in JavaScript applications. We propose a strategy to statically detect class-based abstractions in the source code of JavaScript systems. We used this strategy in a dataset of 50 popular JavaScript applications available from GitHub. We found four types of JavaScript software: class-free (systems that do not make any usage of classes), class-aware (systems that use classes, but marginally), class-friendly (systems that make a relevant usage of classes), and class-oriented (systems that have most of their data structures implemented as classes). The systems in these categories represent, respectively, 26%, 36%, 30%, and 8% of the systems we studied.

Transcript

1. Does JavaScript Software
   Embrace Classes?
   L. Silva, M. Ramos,
   M. T. Valente
   UFMG, Brazil
   A. Bergel
   Univ. Chile
   N. Anquetil
   INRIA, France
   SANER, Montreal, March 2015
   

   View Slide

2. JavaScript is gaining popularity
   http://redmonk.com/dberkholz/2014/05/02/github-
   language-trends-and-the-fragmenting-landscape/
   2
   

   View Slide

3. 1995: Small Scripts
   3
   

   View Slide

4. 2015: Complex Systems
   4
   

   View Slide

5. Criticisms
   https://www.destroyallsoftware.com/talks/wat 5
   

   View Slide

6. Motivation
   How JavaScript software is organized?
   6
   

   View Slide

7. Motivation
   How JavaScript software is organized?
   More specifically,
   Do JS developers use class-like structures?
   7
   

   View Slide

8. JavaScript
   • Imperative
   • Objects maps strings to values
   – var person = { firstName: "Marco", lastName:"Valente" }
   • Functions are first-class objects
   – var fact = function factorial (n) { .... }
   8
   

   View Slide

9. Emulating Classes
   • new:
   –creates a new object
   –calls the constructor function
   –with this bound to the new object
   9
   

   View Slide

10. Emulating Inheritance
    • Each object has an implicit prototype property
    • To eval obj.p, the runtime searches for p in:
    –The current object obj
    –The prototype of obj
    –The prototype of the prototype of obj
    10
    

    View Slide

11. Prototype-based Inheritance
    11
    • Prototype chaining emulates inheritance
    

    View Slide

12. Heuristic to Detect Classes
    • A class is a tuple (C,A,M)
    –C is the class name
    –A= {a1
    , a2
    , ..., ap
    } are attributes
    –M= {m1
    , m2
    , ..., mq
    } are methods
    12
    

    View Slide

13. Heuristic to Detect Classes
    • A program P has a class (C,A,M) if:
    –P has a function called C
    • For each a in A,
    – C includes this.a = Exp;
    • For each m in M
    – C includes this.m = function {...}
    –P has a new C()
    13
    

    View Slide

14. JSClassFinder
    http://aserg.labsoft.dcc.ufmg.br/jsclasses/
    14
    

    View Slide

15. Evaluation
    

    View Slide

16. Research Questions
    1. Do developers use classes?
    2. Do developers use inheritance?
    3. What is the size of the classes?
    16
    

    View Slide

17. Dataset
    • 50 GitHub systems:
    –at least 1,000 starts
    –at least 150 commits
    –that are no forks of other projects
    17
    

    View Slide

18. Dataset
    • Data cleaning:
    – compacted files (min.js.*)
    – third-party libraries
    18
    

    View Slide

19. Dataset
    • Largest system (ace): 194 KLOC
    • Smallest system (masonry): 0.2 KLOC
    • Average size: 14 KLOC
    • Median is 2.5 KLOC
    19
    

    View Slide

20. RQ #1: Do developers use classes?
    • Class Usage Ratio (CUR):
    20
    

    View Slide

21. Key Finding: Four Types of Systems
    • Class-free systems (CUR = 0)
    – 13 systems, 26%
    • Class-aware systems (0 < CUR ≤ 0.25)
    – 18 systems, 36%
    • Class-friendly systems (0.25 < CUR ≤ 0.75)
    – 15 systems, 30%
    • Class-oriented systems (CUR > 0.75)
    – 4 systems, 8%
    21
    

    View Slide

22. Do large systems have more classes?
    22
    No correlation KLOC vs CUR
    

    View Slide

23. RQ #2: Do developers use inheritance?
    • Only 9 systems with subclasses (but very few)
    23
    

    View Slide

24. RQ #3: What is the size of JS classes?
    24
    Heavy-tailed distribution,
    with NOA, NOM < 10 for most classes
    

    View Slide

25. Few methods
    • In Java:
    – Classes have more methods than attributes
    • In JavaScript:
    – Classes with more attributes are common
    – Possible reason: lack of getters and setters
    25
    

    View Slide

26. In Summary
    • For many JavaScript developers:
    • Objects are not enough... they need classes
    • We found classes in 74% of the systems
    • But there is controversy :
    • “This is the first time a research paper has ever
    called me classless, and boy am I proud.”
    • (prototype-based) inheritance is rarely used
    26
    

    View Slide

27. Implications
    

    View Slide

28. ECMAScript 6
    • Syntactical support to classes
    • Conversion tool: “old JavaScript class styles” to ES 6
    28
    

    View Slide

29. Class-Oriented Tools and Techniques
    1. Reverse engineering (class diagrams)
    2. Bad smells (large class, feature envy etc)
    3. IDEs with class browsers (package explorer)
    4. Metrics (coupling, cohesion etc)
    29
    

    View Slide

30. Thanks!
    

    View Slide

31. Does JavaScript Software
    Embrace Classes?
    L. Silva, M. Ramos,
    M. T. Valente
    UFMG, Brazil
    A. Bergel
    Univ. Chile
    N. Anquetil
    INRIA, France
    SANER, Montreal, March 2015
    

    View Slide