Do Maintainers Utilize Deployed Design Patterns Effectively

1
Do Maintainers Utilize Deployed Design Patterns
Effectively?

• ACM International Conference on Software
  Engineering (ICSE)
• T.H. Ng, S.C. Cheung, W.K. Chan, Y.T. Yu
• Presented by Jason R. Beck and Enrique G. Ortiz

2
Outline

• Research Questions
• Participant Descriptions
• Hotel Example
• Question Formalization and Null Hypothesis
• Target Programs
• Required Changes
• Experiment / Subjects / Procedure
• Results
• Pros / Cons

3
Research Questions

• Will maintainers use design patterns?
• How likely are maintainers to perform specific
  tasks (i.e. T1, T2, and T3)?
• Is the maintainer more likely to use specific
  tasks over others?
• Is the code produced of higher quality using the
  design patterns?

4
T1 Concrete Participants

• Participant that is subclass of another
  participant

5
T2 Abstract Participants

• Participant that is superclass of another
  participant

6
T3 Client Partcipants

• Participant that is not a subclass or superclass,
  but statically binds to abstract participant
  (superclass)

7
Hotel Example

• Focus on Perfective Maintenance (adding new
  capabilities to program)

8
T1 Adding new class as a concrete participant

• New Local Setting Hong Kong
• Must implement new class that implements locale
  settings to inherit Locale (abstract) class

9
T2 Modifying the existing interfaces of a
participant

• Display currently selected country in the user
  interface
• Add a new method to the Locale class hierarchy

10
T3 Introducing a new client

• Add support for console-based user interface
• Introduce new client codes for Locale inheritance
  hierarchy

11
Question Formalization

• Q1 What proportion of maintainers utilize the
  relevant design patterns in completing the
  required anticipated change?
• What proportion of them perform T1?
• What proportion of them perform T2?
• What proportion of them perform T3?
• Are the above proportions significantly different?

12
Question Formalization

• Q2 Comparing the codes delivered by maintainers
  who utilize the relevant design patterns in
  completing the required anticipated change to
  those delivered by maintainers who do not utilize
  patterns, are they equally faulty?
• Comparing those who perform T1 with those who do
  not, are their codes equally faulty?
• Comparing those who perform T2 with those who do
  not, are their codes equally faulty?
• Comparing those who perform T3 with those who do
  not, are their codes equally faulty?

13
Null Hypothesis

• HA0
• Maintainers have same tendency to perform each
  task
• HB0
• Maintainers who perform tasks and those who do
  not provide same code quality
• HC0
• No strong correlation between the number of
  faults in the delivered codes and whether or not
  a task is performed to complete the anticipated
  change

14
Target Programs

• JHotDraw
• Open-source drawing editor
• Java based
• GUI
• MCM
• Multiple-user calendar manager
• Edit individual appointments
• Console-based
• HMS
• Hotel management system prototype
• Edit room usage (check-in and check-out)
• Console-based

15
Target Program Analysis

• Different applications
• Different sizes (few hundred LOC to ten-thousands
  LOC)
• Defined anticipated changes
• Created design patterns

16
Required Changes

• Perfective changes (most frequent maintenance)

17
Required Changes

• Must spend at least three hours performing
  changes (ensures effort)

18
Experiment Subjects

• 215 Undergraduate Java course students.
• Grouped and individually assigned to tasks
• 27 students ? Image modification
• 28 students ? Language modification
• 100 students ? Group appointments and Undo
• 60 students ? GUI-based HMS and Search

19
Experimental Procedure

• Briefed the students in a two hour information
  session.
• Covered
• Requirements and specification and demo of
  correctly operating program.
• Presented the source code.
• Documentation (API and UML)
• Also covered general object-oriented concepts
• Given a month to work on the assignment

20
Results

• Subjects more likely to perform tasks T1-T3 than
  not use them at all.
• Maintainers were more likely to perform task T1
  (adding a concrete participant).
• Followed by T3 (introduce a new client) and then
  T2 (modify existing interface).

21
Do maintainers utilize deployed design patterns?

• Yes, they do use them more often than not.

22
Do maintainers utilize deployed design patterns?
23
Do maintainers who utilize deployed design
patterns deliver better code?

• Functional correctness measured in number of
  faults on the submitted programs.
• Means compared between those that used one of T1
  T3 in each program and those who did not.
• Fewer faults for those who used one of T1 T3
  results statistically significant at 5 level.

24
Threats to validity

• Internal (Authors opinion)
• Plagiarism
• Wide variety of student expertise (all approaches
  feasible even if no design pattern used)
• External
• Only used a few target applications.
• Design patterns used were not exhaustive
• Only used finite number of perfective changes

25
Related Work

• Others have researched maintenance, but not also
  how design patterns factor in.
• Other experiments had result that conflicted
  between each other regarding design pattern ease
  of maintenance.
• Experiment to see if additional design patterns
  in code is beneficial regardless or work
  experience. (By the authors)

26
Conclusion

• When design patterns are initially implemented in
  anticipation of maintenance work, they are more
  often used than not by maintainers.
• Conclude that T1 is most always recognized and
  utilized, but work is needed to raise the
  maintainers awareness to perform T2 and T3.

27
Pros

• Good goals. Maintainers should make use of
  existing design.
• Statistical significance of the fault results
  helped strengthen the argument.
• Good variety of maintenance tasks and
  applications. Covered a good variety of design
  pattern tasks.

28
Cons

• Are student representative of maintenance
  programmers?
• Undergraduates in a Java class. (likely very new
  to programming)
• Covered general object-oriented concepts when
  giving them the assignment.
• Showed the students how the final results should
  look, showed them the requirements, and original
  source code. (probably not the hand holding a
  real maintainer gets).
• A month to complete these tasks.

29
Cons

• Results showed those which used patterns in the
  tasks had fewer faults.
• Students who recognize design patterns are
  probably more skilled in the language.
• The statistical significance is not between using
  patterns and not using patterns but between
  students who are skilled to use patterns and
  those who are not.

30
Cons

• None of this confronted in the threats to
  validity, instead they thought plagiarism was
  more of a threat.

31
Areas to Improve / Comments

• Take out the student factor and perform an
  experiment on professionals who perform
  maintenance on a regular basis.
• Gain insight on decisions made by professionals.
• Creates a less bias fault statistic.
• Choosing not to utilize the pattern only makes
  things worse.
• Mismatched documentation
• Ruining the designers intent
• Likely to cause problems for future maintainers

32
Questions?