Developing with branches
There was recently some discussion on the Koha mailing list about how “branches” fit into the development process. I replied with a description of how I use branches in my development process, and this post is an expansion of that.
In my discussion of Koha’s bug-reporting process I mentioned that a developer who wants to work on a bug will “accepts” a bug in Bugzilla before beginning work on it. There is a lot more to the developer’s workflow going on behind the scenes. I’ll describe what my process is.
When I first begin to work on fixing a bug, I create a “branch” in the git version control system. When I do so git creates a separate working copy of the main branch, “master.” This copy is of the very latest version of Koha available. I create a separate branch for each distinct change I want to make to Koha. That change might be defined as a bug fix for a particular bug, or a new feature. The changes to each branch might include modifications to many different files, but the changes should all be related to that one subject.
It’s vitally important that I keep my branches clean and separate from each other because the Release Manager is relying on me to make it easy for him to apply and test my patches. If I submit a patch that covers more than one “subject,” that makes it all the more difficult to test it properly. If I introduce a bug to one aspect of my update, the whole patch may have to be rejected. Keeping things as simple as possible makes it easier for everyone.
To see how this looks in my workflow, here’s a selection of my current working branches:
When I first sit down to work I create a branch with a prefix “ip-” for “in progress,” give it a short title and bug number if I have one, and add the date I started working on it. You can see I might have quite a view branches in progress at a time. This might be because I’m working on a big update which is time-consuming, or it might mean I’m stumped about what the solution is.
Each time I sit down to do more work on a bug I switch to that branch, download the latest updates which have been submitted and approved on the master branch, and tell git to “rebase” my branch. The rebase process merges my changes with the latest updates. If there are any problems with the merge git will warn me and ask me to make manual changes. Usually the process is automatic. I do this every time I sit down to work on an in-progress branch. This ensures that the changes I’m making are compatible with the latest version of Koha.
Once I have finished my work I check it and test it as carefully as I can. I make sure I’ve rebased against the very latest update in git, and I submit a patch. Submitting a patch takes the changes I’ve made and distills them down to a single text file which can be “applied” by the Release Manager or any other Koha developer to their own git installations. I also attach a copy of my patch to the bug report, if there is one, as I described in my bug reporting post. This is another way to get the patch out into the open where others can review it and test it.
After I’ve submitted a patch I rename the branch I was working on with “ps-” for “patch sent” and keep the branch until my patch is approved (fingers-crossed).
All this is on my computer in a virtual machine, so in my case I’m not being open as I might. If I had the resources to work on a real server, or if I kept my work on a service like GitHub I could share all these branches with everyone, from the time I first started working on it to the time my patch was submitted.
If some time has passed and I see that my patch from one of my “patch sent” branches hasn’t been approved I’ll switch to that branch and rebase: I’ll tell git to grab the most recent Koha commits and merge them with the changes in my patch. If not too much time has passed, or the files I changed haven’t been worked on by others the rebase will be successful and I’ll know I’m still on track. If not I might have to manually make changes to my original commit so that it fits in again with the work others have been doing. If manually merging was required, that tells me the Release Manager would have to do the same. I should consider resubmitting a revised version of my patch which can apply cleanly to the latest version.
Keeping all my “patch sent” branches intact until my patch has been approved is important to my workflow because it’s the easiest way for me to keep track of what has been approved and what hasn’t. I can see at a glance which patches of mine have not been approved. If it’s been a while, I might try to find out why not. If there was a problem with the patch I might be able to revise it to the RM’s satisfaction.
I can’t submit a patch and expect it to automatically work a month from now, a week from now, or even a day from now. I’m responsible for keeping up.
I’m not a git expert, and I’ve had lots of help and advice along the way to get me to this workflow. It works really well for me, and I hope it works as well from the point of view of our Release Manager and fellow developers.