Code reading is one of the most frequent activities in software maintenance; before implementing changes, it is necessary to fully understand source code often written by other developers. Thus, readability is a crucial aspect of source code that may significantly influence program comprehension effort. In general, models used to estimate software readability take into account only structural aspects of source code, e.g., line length and a number of comments. However, source code is a particular form of text; therefore, a code readability model should not ignore the textual aspects of source code encapsulated in identifiers and comments. In this paper, we propose a set of textual features aimed at measuring code readability. We evaluated the proposed textual features on 600 code snippets manually evaluated (in terms of readability) by 5K+ people. The results demonstrate that the proposed features complement classic structural features when predicting code readability judgments. Consequently, a code readability model based on a richer set of features, including the ones proposed in this paper, achieves a significantly higher accuracy as compared to all of the state-of-the-art readability models.

1. Improving Code Readability
Models with Textual Features
Simone Scalabrino, Mario Linares-Vásquez, Denys Poshyvanyk, Rocco Oliveto

2. Software maintenance
accounts for
70%of the costs of a project

4. for (int i=1;i<=100;i++) {
int a=((528>>i%15-1)&1)*4;
int b=((-2128340926>>(i%15)*2)&3)*4;
System.out.println("FizzBuzz".substring(a,b)+(a==b?i:""));
}
What does it do?

5. Not really clear...

6. for (int i=1;i<=100;i++) {
String fizzBuzz = “”;
if (i % 3 == 0)
fizzBuzz += “Fizz”;
if (i % 5 == 0)
fizzBuzz += “Buzz”;
if (fizzBuzz.isEmpty())
fizzBuzz += i;
System.out.println(fizzBuzz);
}
What about this one?

7. Why is it easier
to understand?

8. Code readability

9. Code readability prediction

10. Source code

11. Source code
Source codeSource code
Features

12. Source code
Source codeSource code
Features
Training set Machine learner

13. Source code
Source codeSource code
Features
Readable Unreadable
Training set Machine learner

14. Code readability prediction

15. Code readability prediction

16. Code readability prediction

17. Code readability prediction
Structural features

18. Code readability prediction
Structural features
Visual features

19. Code readability prediction
Structural features
Visual features
Two datasets

20. Something is missing...

21. Code is text!

22. New features

23. Comments readability

24. Comments readability
Flesch-Kincaid index

25. Comments and
identifiers consistency

26. Comments and
identifiers consistency
Overlap between terms in comments
and terms in identifiers

27. Identifier terms
in dictionary

28. Identifier terms
in dictionary
Percentage of correct terms in identifiers

29. Narrow meaning
identifiers

30. Narrow meaning
identifiers
“Sum” more specific than “computation”

31. Number of meanings

32. Number of meanings
“Present” more ambiguous than “Attendee”

33. Textual coherence

34. Textual coherence
Consistency between pairs of syntactic blocks

35. Textual features

36. Source code
Source codeSource code
Features
Readable Unreadable
Training set Logistic regression

37. Case study

38. New datasetNew dataset
200
Java snippets
9
annotators

39. Do textual features complement
the others proposed
in the literature?

40. Overlap metrics

41. Textual Features vs Buse’s

42. Textual Features vs Posnett’s

43. Textual Features vs Dorn’s

44. Textual Features vs Dorn’s
Readability of 12%-21% of snippets can
be explained only using textual features

45. What is the accuracy of a
readability model based
on structural and
textual features?

46. 79%
81%
78%
80%
74%
Dataset by Buse and Weimer
ALL-F: All features
T-F: Textual features
D-F: Dorn’s features
P-F: Posnett’s features
BW-F: Buse’s features

47. 84%
79%
73%
80%
77%
Dataset by Dorn
ALL-F: All features
T-F: Textual features
D-F: Dorn’s features
P-F: Posnett’s features
BW-F: Buse’s features

48. 80%
71%
66%
76%
68%
New dataset
ALL-F: All features
T-F: Textual features
D-F: Dorn’s features
P-F: Posnett’s features
BW-F: Buse’s features

49. 80%
71%
66%
76%
68%
New dataset
ALL-F: All features
T-F: Textual features
D-F: Dorn’s features
P-F: Posnett’s features
BW-F: Buse’s features
A model which includes all features
achieves an higher accuracy on 2 datasets

50. In summary...

51. Code readability for
defect prediction

52. Thanks.