A Few Simple Rules


We held the post-mortem for Helium this morning. Lots of things went wrong, but many more went right—it was a fun, productive summer.

Lots of things contributed to the fun, but much of the productivity can be attributed to one thing: everyone learned to use modern tools, and adopted good working practices. Many projects stumble or fail because programmers aren't willing to do this—far too many, young as well as old, still insist on using Vim, Emacs, and other legacy tools. This doesn't just make them less productive; it also handicaps the rest of the team, since it prevents people learning from each other, sharing resources (such as useful keyboard macros), doubling up to debug hard problems, and so on.

So, with that, the post-mortem, and the Joel Test in hand, here are the rules by which this fall's projects will be run:

1. Everything important is versioned. Source code, web pages… anything and everything that you need to build, install, or debug the project goes into the Subversion repository.
2. Everyone uses Eclipse. And Ant. And JUnit. And all the other tools that are part of the environment. If you don't, you are slowing down your teammates, and making extra work for them.
3. The build is automated. And all tests are re-run by default every time the project is built (which will be every 30 minutes, round the clock).
4. Anything that isn't done immediately goes into the issue tracker. Why? Because even if you would remember it a month from now, teammates who weren't sitting with you when you slapped your forehead and said, "D'oh!" won't.
5. Everyone writes tests, preferably before they write the code itself. Take a look at Helium's progress chart; one of the reasons the team go that far, that fast, was that they tested early, and tested often. You owe it to your successors to do the same.
6. Everyone fixes bugs. Code is done when it's all typed in; code is done when it works, and "working" means "working for everyone, everywhere, every time". (Yes, this means that most software is never done…) When your tests turn up bugs, file 'em and fix 'em before implementing more features. Experience shows that this will make your work faster, since you won't constantly be doubling back to shore up the foundations.
7. Every large piece of work is sanity checked before code is written. Thinking of reorganizing some of the data model classes? About to write forty new classes to handle threading of email conversations? Grab a teammate and a whiteboard, and explain what you're about to do, or send email to the list, or fire up IM. Anything, just as long as someone else gets a chance to tell you when you should switch back to decaf…
8. Every meeting start with an agenda, and ends with minutes and an action list. Where possible, the agenda should be circulated before the meeting begins; the minutes and actions must be circulated soon after it's over (and added to the project web site).
9. Have fun. This is the most important rule of all. ("Floggings will continue until morale improves…") Your project isn't just about marks—it's about learning new things, making new friends, and building something you can be proud of. Most of the students who've done projects in the past have had a lot of fun—we hope you will too.