Teaching basic lab skills
for research computing

Running a Code Retreat

I recently attended a scholarship retreat run by Google with about 40 other students. We spent one day doing a "code retreat" and I wanted to share my experience and what I learned about programming.

The code retreat was based on this template, and led by senior Google software engineer, Nathan Sandland. The task was the same for all the exercises -- pair programming Conway's Game of Life with unit tests. The first exercise was to simply program the game of life and then we did 4 more exercises that were variations on this, with different constraints. We were allotted about 45 minutes for each exercise, and we switched pairs for each exercise. After each exercise, we went around the room and debriefed people's thoughts on the exercise. People could program in any language as long as that language supported unit tests, and each pair needed to program in a common language that they both knew.

The exercises were:

  1. Program the Game of Life with no constraints other than time (don't forget the unit tests!)

  2. Extreme constraints: can't use if/else statements, only allowed 3 lines of code per function, can't pass primitives to functions or interfaces

  3. Ping pong pairing: Person A writes tests, Person B writes code to pass tests, then Person B writes tests, Person A writes code to pass the tests, etc.

  4. Evil mute A/B pairing: pairs are not allowed to talk. One person writes tests. The other person is a "lazy evil" programmer who writes the minimal code to pass the tests (but the code doesn't need to actually implement the specification).

  5. "Baby steps": Program in 5 minute rounds, alternating partners. Initialize an empty Git repository. Then the coder has 5 minutes to write a test and write code to pass the test. If the coder finishes, the work done during that round is committed to the repository. If the coder doesn't finish, the coder has to do a hard git reset to delete their work. (We were told that when this exercise is done with Google employees, they use 2.5 minute rounds -- I'm not sure how I could have managed with that time limit; I could barely finish in 5 minutes.)

The goal for each exercise is not to actually write the Game of Life, but to experience the process of coding under different conditions and constraints. In all 5 exercises, my partner and I never came close to finishing.

At the end of the day, we wrote down our observations as answers to the following questions:

Closing questions: On a piece of paper, write the answer to one of the following (to be shared with the group):

  • What, if anything, have you learned today?

  • What, if anything, surprised you today?

  • What, if anything, will you do differently?

During the discussions, we learned that software engineers write an average of 10 lines of code per day (!) I suspect that engineers spend a lot of time thinking and revising code, so only 10 lines of "forward progress" is made per day.

I thought the exercises were pretty interesting and sometimes entertaining (especially exercises 4 and 5). I had never done anything like it before. I learned about software engineering in the presence of constraints. For example, the crazy constraints in Exercise 2 forced me to do a lot of hacking. In Exercise 4 (the evil mute pairing), it was made clear that you can easily pass tests without the code actually implementing the featured tested.

Overall, I was forced to write a lot of tests quickly, which is something I'm normally pretty lazy about. It was also my first experience with pair programming. One of my first partners was a Java guy and it was interesting how differently he thought about programming. He wanted to write out a flow chart to design the program and do a lot of OOP stuff, whereas I just wanted to quickly prototype in Python. Switching partners frequently exposed me to different personalities and styles of programming.

Most of the students (about 40 of us) knew either Python or Java or both. I ended up programming in Python for all the exercises and using nose for my unit testing framework (nose was easier to use than unittest for quick dirty testing).

Overall, I thought the code retreat was fun and worth doing at least once. I'm not sure how these exercises would fit into the Software Carpentry curriculum, but they might be an interesting addition for an advanced workshop or maybe even a hacking day in a highly computational lab group. The barrier is that to get something out of these exercises, the students need to have written basic unit tests before (maybe that could be done in a pre-workshop prerequisite) and know a common programming language. Exercise 5 requires knowledge of Git. The students' laptops need to be setup with the programming language and unit testing framework (preferably for multiple languages) before the code retreat.