Lessons on Best Practices from Mozilla
One of the ways we facilitate improvement in development processes and the like at EventMobi is by having lunch and learns where someone will present on something interesting. Sometimes that's a cool technology they've used or a good development practice they've discovered or have had experience with. To that end I gave a presentation on some of the lessons I've learnt on best practices while working in the Mozilla open-source community.
Allthough many of these best practices may seem like no brainers to seasoned developers, I still hear way to many horror stories through the grapevine about software being built under seriously bad conditions. So, without further ado.
One of the things I think Mozilla does really well is the idea of code ownership. This essentially means identifying those who have a level of knowledge about a particular area or module of code and thrusting upon them the responsibility to oversee it. From a practical point of view this means answering questions about the module that others have and also reviewing all of the changes that are being made to the module in order to ensure that they are sane and fit into the larger architecture of the module as a whole.
The key part to this set up in my opinion is that it makes it clear what the requirements to become a code owner are and what their responsibilities are as a code owner. Too often I feel like, as with other things, if they aren't formalized they become subject to, well, subjectivity. And the details of code ownership and responsibility get lost in translation and hence, not enacted.
Bottom line, formalizing your code ownership policies and processes are a foundation for success. Without that it becomes unclear even who to ask to review code, is it the person in the blame? Possibly. Maybe not. Maybe that person didn't make correct changes, or made changes under the guidance of someone else. Maybe the code your changing has never even had a true 'owner'. No one knows the big picture for that piece of code and so no one knows enough about it to make informed decisions. That's a problem. If a code owner had been designated when that code was written that would never have been an issue.
We all know testing is a must for any sane development process. What I've learned through Mozilla is that having an insane amount of tests is okay. And honestly, preferable to just enough. As just enough is hard to gauge. The more tests I have in general, the more confident I feel in my code. Having a gauntlet of tests for your code makes it that much stronger.
Not only that, but it's important to be staying on top of tests. As a rule, not accepting code before it has a test and adding regression tests for bugs that are fixed. In exceptional cases code can merged without tests, but it should be tracked so that tests for it will be added later. Ten minutes spent writing a test now could save hours of developer time in the future tracking down bugs.
This is one of my personal favourites. And especially relevant I think in companies which are, in my experience, more driven to say yes—in order to hit that extra profit margin, to please that extra customer, to do whatever—as opposed to open-source projects who are able to say no because most of the time they have no deadline. They're driven by desire to build that next cool thing, to build it well, and to do it in a sane way.
From my experience working in Mozilla's and other open-source communities I've found it's important to say no when a feature isn't ready, when it's not good enough yet, when it needs an extra test, when it's not necessary, or when you just ain't got no time for that. I do think, however, that it's hard to say no sometimes while working under the constraints of a profit driven process. There is a healthy balance one can achieve though. We have to strive to achieve this zen like state.
Managing Technical Debt
One of the main ways I've seen this done is by creating tickets for everything. See something off about the code? Log a ticket. See something in need of refactoring? Log a ticket. See something that needs a test? Log a ticket. Any kind of piece of work that you think needs to get done, any kind of open question that needs to be answered about the state of the code, log a ticket for it.
Logging tickets for the problem gives it visibility and documents it somewhere. This enables a few things to happen. It enables communication about the state of your code base across your team and makes that information easily accessible as it's documented in your tracking system. It also puts the problems that are not necessarily bugs—your stinky, ugly, untested code, or otherwise—to be in your teams face all the time. It's not just getting swept under the rug and not being paid attention too. It forces your team to deal with it and be aware of it.
The key part of this strategy then becomes managing all these tickets and attempting to understand what they are telling you. One of the ways you can do this is by doing regular triages. This means going through the open tickets and getting an idea of what state your code is in and prioritizing how to go about fixing it. This is key as it turns the information that your team has been generating into something actionable and something that you can learn from.- Tagged in mozilla, open-source, cdot, and development processes.
comments powered by Disqus