CSC/ECE 517 Fall 2012/ch2a 2w9 ms
Pair Programming
This chapter will explain basic concepts of pair programming which is one of the popular Agile Software Development technique. We would also explain its benefits, advantages and disadvantages. In the end we will conclude the topic by mentioning ongoing research in this field.
Introduction
Pair programming is a style of programming in which two programmers work side-by-side at one computer, continuously collaborating on the same design, algorithm, code or test. Use of this practice has been demonstrated to improve productivity and quality of software products.<ref name = ref-paper />.
In Pair programming practice, two software engineers work on one task at one computer. One of the engineer acts as a driver who has control of the keyboard and mouse and (s)he writes the code for an implementation. The other engineer acts as a navigator (also called as an observer) and watches and reviews the driver's implementation to identify defects and participates in on-demand brainstorming. The roles of driver and navigator are periodically rotated between the two software engineers. The observer also considers the strategic direction of the work, coming up with ideas for improvements and likely future problems to address. This frees the driver to focus all of his/her attention on the "tactical" aspects of completing the current task, using the observer as a safety net and guide.
Extreme Programming (XP), is an emerging software development methodology. Pair programming is one of the corollary practices of Extreme Programming. XP attributes great success to the use of “pair programming". XP advocates pair programming with such fervor that even prototyping done solo is scrapped and re-written with a partner.
Principles of Pair Programming
There are some principles of pair programming in the context of Fulghum's Poem <ref name = principle-pair />.
In pair programming, two programmers are required to produce one piece of software (design, algorithm, code etc.) within a specified time limit. The two programmers are working with one mind responsible for every aspect of this piece of software. Both the driver and a navigator are equal participants in the process and both of the partners own everything and hence they should not blame each other for any defect in the artifact.
Play fair
With pair programming, one person has control of the keyboard or is recording design ideas, while the other is continuously reviewing the work. Even when one programmer is significantly more experienced than the other, it is important to take turns “driving,” so that the observer does not become disjoint or feel unimportant.
Don't hit your partner
Make sure your partner stays focused on task. The driver is awaiting continuous contribution and input from the navigator and hence both the partners should be focused on their part of work.
Clean up your mess
Pair programming makes many obvious unnoticed defects go away because of another person watching over the shoulders. Additionally, these defects can be removed without the natural animosity that might develop in a formal inspection meeting.
Don't take things too seriously
Pair programming demands ego-less programming approach from both the partners. It is essential for effective programming effort. Excess ego can manifest itself in two ways, both damaging the collaborative relationship. Attitude problems can prevent the programmer from considering others ideas. Secondly, excess ego can cause a programmer to be defensive when receiving criticism or to view this criticism as mistrust.
Take a break from working together
It is acceptable to work alone 10-50% of the time. Many programmers prefer to do experimental prototyping and logical thinking alone. Simple, well defined coding is more efficiently done by a solitary programmer and then reviewed with a partner.
Be aware of the power of two brains
Human beings can remember only to a certain extent and hence one must consult others to increase their knowledge. When two people are working together, each one has their own set of skills and expertise, which should be properly used while producing a piece of software. The unique skills of each individual will allow them to engage in interactions which pool their resources to accomplish their tasks.
Partner picking principles
- Partners should have complementary skills.
- Partners should be of similar skill level.
- Both the partners should be flexible and open to new ideas.
- Both partners should have good communication skills.
- Partners should have compatible schedule.
- Both programmers should be knowledgeable.
How to achieve Pair Programming
There are some ways to achieve pair programming for an effective development process <ref name = ways-pair-prg />.
- Prepare together for the coding you are going to do so that no misconceptions will be developed during the coding time.
- Start with a reasonably well-defined task before you sit down. The task should be something you are confident that you can complete in an hour or two.
- Agree on a solution: Decide on a general strategy to tackle the task that you decided upon. You may find it helpful to outline what you plan to do before you begin to code.
- Agree on one tiny goal at a time: something you can complete within a few minutes. Stating the problem in words to another person helps focus your mind and helps engage your partner's mind. It also ensures that you both know what you are working on right now.
- Rely on your partner, support your partner. When you're the driver, complete the current tiny goal as quickly as you can, ignoring larger issues. When you're the observer, read the code that the driver is writing as he or she writes it. Your job is code review. You should pay total attention, aiming to let nothing get by you. Think about possible bugs, larger issues, and ways to simplify or improve the design.
- Talk a lot! Say what you are about to do, ask for an implementation idea, ask for a better way to solve the problem at hand, bring up alternative ideas, point out possible inputs that the code doesn't cover, suggest clearer names for variables and subroutines, suggest ways to implement the code in smaller steps, tell the driver that little bit of API knowledge that they need right at the moment they need it, etc. Listen a lot, too, of course. When people are pairing well, they are talking back and forth almost non-stop.
- Sync up frequently. As you are working together, you will find yourself getting out of sync: becoming unsure of what your partner is doing, or becoming unclear about the current task. This is normal. When it happens, sync up again. The key to good pairing is to sync up very frequently— within seconds or a minute of noticing that you're out of sync.
- Take a moment to celebrate as you complete tasks and overcome problems.
- Switch roles often — at least every half hour. This keeps both the participants fully engaged, and both are in tune with the low-level details and the big picture.
When to use Pair Programming
Pair programming can be beneficial in certain scenarios. Some of them are mentioned here <ref name = benefit-scenarios />.
- Pair programming is very appropriate when a new team member is introduced into a group as a form of training.
- It may also be useful in complex re-factoring exercises, where one programmer thinks and the other does the coding.
- It may be used when programming efforts require significant object oriented designs, or development work on algorithms or procedures that have performance considerations.
Pair Programming Benefits/Advantages
- Better communication skills are developed among the individuals.
- Better quality software is developed because of improved knowledge exchange and different perspectives from different people.
- The cycle time of software development is improved.
- Programmers are more satisfied in case of pair programming.
- Pair rotation makes staff training easier and reduces product risk. It enhances team building.
Pair Programming Disadvantages
- The development progress can get hampered due to conflicts between the programmers and debate breakout among the group.
- Everyone has their own ideas and sometimes it becomes difficult to collaborate different ideas.
- It is possible that only one programmer does more than half of the work in the project.
Research
Academia
A group of researchers from the University of California, Santa Cruz did a study of "The Effects of Pair-Programming on Performance in an Introductory Programming Course" <ref name = ref-pp-academia-ucsc/>. Computer programming was essentially taught and practiced as a solitary activity. But, they wanted to see if the upcoming trend of Xtreme programming (XP) and especially Pair Programming could help in learning programming.
For this purpose during the 2000-2001 academic year, data was gathered from students enrolled in two sections of an introductory programming course at the University of California – Santa Cruz. These courses were designed for Computer Science, Management and other Engineering majors. One of the two sections required students to complete programming assignments in pairs (N=172), while the other required students to write programs independently(N=141). The programming assignments, lectures, and quizzes were comparable, and the final exam was identical in both sections. The analysis of the data gathered lead to some interesting finds. These are summarized below.
To compare whether programming scores differed as a function of pair-programming experience, analysis of variance (ANOVA) was conducted. Among all students who completed the course, students in the pairing class scored significantly higher on the programming assignments (M=86%) than those in the non-pairing class (M=67%). Overall, the scores from the entire pair-programming section (M=86%) were significantly higher than the even scores of the top half of the non-pairing class (M=77%). So we can reject the theory that the scores for the pair programming group merely reflects the work of the brighter student in the group. The mean exam score in the non-pairing class (M= 75%) was slightly higher than the mean exam score in the pairing class (M=73%). This small difference, however, was not statistically significant. Another major statistic was that the percentage of students who finished the final was dramatically higher in the pairing section (92% vs. 76%). So pairing increased the likelihood that students complete introductory programming class.
So according to the study, it appears plausible that as a result of pair-programming, students that might otherwise have dropped the course, completed the course. Also, it also appears that the programs of even the better students benefited from pair-programming. The data also suggested that students who work in pairs produce better programs. Furthermore, they perform comparably on exams (when not adjusted for varying attrition rates), and possibly significantly better on a final exam, to students required to program individually.
Industry
Research done by Andrew Begel and Nachiappan Nagappan on Pair Programming at Microsoft by was published in the paper entitled "Pair Programming: What's in it for Me?" <ref name = ref-pp-industry-ms/>. Their main focus was to understand how pair programming methodologies are used, what kinds of problems and benefits they are perceived to have, the types of partners people would like to work, and a general consensus on PP's usefulness in the software engineering professional community. They did this by asking a sample of developers at Microsoft directly about their current development practices.
Their findings were interesting. They found that 22% of the participants had practiced pair programming, but only 3.5% did it in their current project. Most practitioners are more experienced than the average Microsoft employee practicing other Agile development methodologies. Around two-thirds like pair programming and believe it is a workable practice, but less than half would agree to that of their team's use of pair programming. Three-eighths of the respondents believe that pair programming takes more time than programming alone, but two-thirds believe that the quality of the resulting software is better.
Also part of the survey was trying to rank the perceived benefits and problems of pair programming. The top benefit reported was fewer bugs in the source code and spotting of bugs earlier. The second most cited benefit indicated that pair programming helps to spread code understanding between the members of the pair. So, "There is never only one person in the team who knows all the code." In third place was higher quality code. Fourth was the ability to learn from a partner. The use of pair programming is a good way to “quickly ramp-up new members.” The fifth benefit was the perception that software being built had a “better architecture and implementation” mainly due to “adherence to good design and standards.”
The participants noted few problems with pair programming too. The number one problem reported with pair programming was cost. Two people are being paid to the do the work of one. The second problem is scheduling time to work in pairs. Two partners require equivalent schedules and suffered double scheduling conflicts. The third most cited problem is a clash of personalities. Finding compatible partners is a difficult process according to many. The fourth problem was trouble resolving disagreements and the fifth problem was that engineers were worried that they will be paired with a partner who is not as smart or skilled as they are.
The study also published a list of qualities that programmers desired in their partners ordered according to preference. The top attribute of a good pair programming partner is that the person has complementary skills to your own. The sentiment in most responses was that my partner “usually looks at things from a different angle.” and he has a “different background which provides a different perspective.” The second attribute is flexibility. An ideal partner is “open minded,” and “open to new ideas.” The third attribute that people looked for was communication skills. Every study of pair programming has shown that it is a communications-intensive process. Tied for fourth is that the person is smart and personable.
This study presented an interesting view of the programmers from the industry.
Overall, there is considerable research going on in the field of pair programming and most of which is in academia.
Conclusion
Pair programming is an agile programming practice which is gaining popularity in both academia and industry. Considerable research is also going on in this area. Major perceived benefits of pair programming are increased quality of code, fewer bugs, better awareness of code among team members etc. It still has problems like lesser productivity, conflicts among partners etc. Many of these problems could be solved by proper selection of team members and applying pair programming to appropriate type of projects. This a useful software engineering practice which teams should consider to increase the quality of their software products.
References
<references> <ref name = ref-paper> Laurie A. Williams, Robert R. Kessler. All I Really Need to Know about Pair Programming I Learned In Kindergarten </ref> <ref name = ways-pair-prg> http://www.wikihow.com/Pair-Program </ref> <ref name =ref-pp-academia-ucsc>Charlie McDowell, Heather Bullock, Julian Fernald, Linda Werner. The Effects of Pair-Programming on Performance in an Introductory Programming Course</ref> <ref name = ref-pp-industry-ms>Andrew Begel, Nachiappan Nagappan. Pair Programming: What’s in it for Me?</ref> <ref name = principle-pair> http://anh.cs.luc.edu/170/Kindergarten.html </ref> <ref name = benefit-scenarios> http://ctotodevelopers.blogspot.com/2008/04/when-to-use-pair-programming.html </ref> </references>