Hacking for Site5 Engineering is an awesome gig. I work with an amazing team of highly opinionated developers from a wide background. In early 2012, we adopted a semi-formal process for working on software. We’ve been very happy with how this has worked out, so I thought it was time to share.
Step 1: Discuss the change
Code written for Site5 projects usually falls under one of four categories:
- new features
- bug fixes
- code refactoring
- library upgrades
Before someone starts coding, they will typically discuss the changes they intend to make or the bug to be fixed. Developers use GitHub issues for most communications on a project.
Sometimes this discussion happens in real-time using our company chat room or Teamspeak. GitHub issues also make it easy to report bugs or request features to be discussed at a later date.
We tried Basecamp for these discussions, but found it was a bit difficult for developers to change context. We did find, however, that it was a great tool to discuss features with non-developers.
Wherever possible, we try to stick to one topic at a time and keep changes as small as possible. For example, features are added one at a time, bugs are fixed one at a time. We’ve found this makes it much easier for other developers to review code changes.
Step 2: Create a new git topic branch
Once a developer is sure of what they are working on, the next step is to create a new topic branch in git. We try to namespace these using one of the following:
This just makes it easier to determine what a branch is for using just its name.
Step 3: Write the code and commit changes
At this point, it is time to start coding. We are pretty much free to code things as we see fit, unless specifics were mentioned in the original discussion.
We’re strong proponents of BDD, so if a project has a test suite (and most of them do), we try to follow that process while working.
Where it makes sense, we will usually make one commit describing the changes that are being made.
If we need to test things on a staging environment, this is where it happens.
Step 4: Open a new Pull Request and solicit feedback
Once code has been committed, it is time to open a new pull request and solicit feedback from the rest of the team.
Other developers review the code to ensure that there are no issues with it, or to offer ideas for refactories. If there are any changes to be made, the original author will then make them. The new changes will be discussed for any further feedback, and the process continues.
At least one reviewer must informally sign off on a Pull Request in order for
it to be merged. This is usually just an emoji
Step 5: Rebase if needed
We try to avoid cluttering a project’s git history with a lot of extra
commits. If there are additional changes added from feedback in Step 4, we
git rebase -i master to squash them down into one commit. Of
course, this is only done where it makes sense – two commits are better than
one if they better describe the intended changes.
Step 6: Verify CI build and merge Pull Request
If the project has a test suite, we wait for our CI server (either Jenkins or Travis CI) to verify the build succeeds. After that, any reviewer can merge the Pull Request. As a rule, we don’t merge our own Pull Requests.
Step 7: Tie up loose ends
After a Pull Request is merged, we may need to tie up some loose ends such as deploying an updated Rails app, or replying to customers or staff to inform them a bug was fixed. This obviously depends on the situation, but it is an important step nonetheless.
Step 8: Beer (or Red Bull)
After all that work, it is time to enjoy a beer or a tasty Red Bull!