Tag Archives: practices

Pair programming – is it worth it?

I was recently tasked with creating a “quality improvement” strategy for one of our projects. One of the controversial propositions turned out to be pair programming. How are we going to justify this cost to the client? That question caught me off guard. I didn’t even think that might be an issue. I didn’t even think about mentioning it to the client, too. For me it’s just a natural part of my work, like using CVS. I never had to justify doing that to anybody. In fact, it used to be quite the opposite.

Stage 1: I hate it

For me personally, pair programming was very challenging. It was the least favourite part of my work. I am a perfectionist (luckily, recovering one) and ever since I can remeber, I hated people looking over my shoulder when I was working on something. Even when I was in primary school and my mum wanted to check how my essay is progressing, I got aggressive. That’s the main reason why I always insisted on not having anybody sit behind my back. I hated the idea of somebody staring at my screen. What if they saw me doing something very dumb?

Over time, I got pretty good at using various avoidance techniques, altering aggression with jokes and plain refusal. Of course, I could got away with this behavior at home, but it wasn’t professional to say the least. Pair programming was not only allowed, but strongly encouraged by my team-lead when I worked at VSoft. I’ve also heard a lot about its benefits. Like it or not, since everybody around me was doing it from time to time (and I couldn’t really say no, every time they wanted to pair on some task), over time I simply got used to it.

Stage 2: I learn a lot

My first “real” pair programming experience was slightly over two years ago and I can’t believe how much has changed. I don’t mind other people looking at my code. I love it. Getting feedback is awesome. I like getting new ideas to improve my code during reviews. I got comfortable with criticism, use it to learn about different approaches and coding preferences. At the same time, I learned to defend my point of view, discuss the motivation behind the decisions I made and be ok with changing my mind, when reviewer points flaws in my thinking.

I can’t recall specific words, I don’t remember all of those small and big ideas I learned through pairing with other people. Some of them applied only to the situation at hand. But there were a few that I benefit from today, such as testing untestable code (thank you Lukasz for being such a pain in the neck about it, now it’s second nature:)), getting familiar with all the crazy shortcuts, learning about new tools, etc. I’m also much better in giving feedback to sb’s code and learned a lot by thinking hard how their code might be improved or why something “didn’t feel quite right”.

Overall, once I got over this irrational “don’t you dare look at my code” attitude, my rate of learning went up significantly. Besides, I think that my productivity was higher, when I worked in a pair, because I was fully focused on a task at had. There was no email checking or talking to colleauges. I did my best not to waste somebody else’s time. Unfortunately, I don’t have any hard data to support that claim.

Stage 3: It’s not a silver bullet

When people ask whether pair-programming is good or not, I think the only honest answer is (of course) “it depends”. There are some people I can’t pair with. When we tried working together it was simply counterproductive, because we just kept arguing and nothing got done. Working with some others was challenging, but I learned a lot. Finally, when working with some people I felt that magical synergy, we were more effective together than each of us working separately, because our working styles and our typical focus areas were complementary.

So it depends to large extent on whom we work with. It’s good to get out of our comfort zone, but I think that pushing through, just for the sake of activity when it’s clearly not working, doesn’t make any sense.

Then it depends on the task at hand. When the task is a no-brainer or we’re very familiar with the area of codebase we’re working on and we’ve done similar tasks before, then pairing might be a waste of time. For me it was most beneficial when I worked on something very new or especially complicated. I could have asked for a review afterwards, but pairing seemed to be more efficient way than stopping every now and then to discuss my idea with a colleague.

Stage 4: It’s hard to work otherwise

There are two situations when pairing seems to be great, irrelevant of other factors. The first is introducing a new person to the project. I can’t understand why people don’t see how wasteful it is to give a vaguely described task to the newbie and instructing them to “ask when they need anything”. Even if they are experienced developers, it takes some time to understand business requirements and be able to understand the impact of every decision on other areas. We lose opportunity to make sure early on that they follow our interal standards and practices. Not to mention the frustration they experience, when they happily announce their task is finished, only to be told that they have to re-do it because of some stupid catch they were unaware of (happened to me recently a few times, extremely frustrating experience).

The other situation is making significant changes or working on some complex piece of logic. It’s beneficial to discuss our design ideas, pay close attention to implementation, make sure we don’t make stupid mistakes, that can be avoided by using second pair of eyes.

For me the main benefits of pair programming are exchange of knowledge between the devs (both domain specific and regarding tools and skills) and avoiding painful surprises during reviews or testing (Oh, how could you miss that edge case? How could you not know about this email I sent to XYZ half a year ago? Why don’t you know about our super-secret document, where all this is described? How can you not know that some external app relies on this file being in that location?).

At this moment, I can’t think about any more efficient and sustainable practice for continuous quality improvement and spreading knowledge among team members.

Stage 5: Back to square one

So how do you justify to your client not doing it? Is there any more efficient way for achieving the goals I’ve just mentioned (improving quality, mentoring, sharing knowledge, avoiding bugs)? Is pairing something one should discuss with the client? Should client be allowed to make such decisions for us?

For me the only honest answer is that I can’t justify not doing it. The other ways for konwledge sharing and keeping high standards that I know, are not as effective (e.g. workshops, presentations). Code reviews should be enough for ordinary tasks in an experienced team, but even for them, I think the two situations mentioned in Stage 4 are still relevant. When it comes to clients I believe that it’s my job to be the best professional I can be and client shouldn’t decide on standards in our industry.

I think this is true not only for software. For example, I don’t go to the dentist and tell him he shouldn’t have an assistant, because I have to pay more for the service. I trust that they do their job well and that this is the best industry practice. I also suppose that if they worked alone, the overall cost would be much higher, they would do the same task longer with more mistakes and caused me more discomfort. Better service for less money? Even if not true in every case, I think this is more common than most people realize.