2 ms·
I plan for “done.” I remember taking a project management course, where the instructor kept using the phrase “What does ‘done’ look like?” I also worked with a
by ChrisMarshallNY 6y ago
I plan for “done.” I remember taking a project management course, where the instructor kept using the phrase “What does ‘done’ look like?” I also worked with a manager that declared "The #1 feature of a project is 'Ship.'”
Every project that I do; even experimental and “unending” projects are done as “ship-quality.” This is because I often revisit previous work, looking for snippets. If they are already at “ship” quality, then that’s one less problem.
I’m also a grizzled veteran of “the prototype becomes the product.” That’s something that many of us have experienced. It sucks, but it’s real. If the prototype is already “ship-quality,” then that’s one less problem.
I tend to tag ongoing projects, so the head may be dynamic, but cloning a tag will generate a stable release, or branch point. That’s a fairly typical Git workflow. I’m not a fan of too many branches. I used the Mainline Model (a Perforce pattern) for years, but I like no more than two branches, these days (dev and master, and sometimes, dev can be in a private repo, squashing to master). Often, I only have one branch. If I create a branch (like a retrospective fix to a release tag), I merge that branch back into master, or apply the same fix to master (not my preferred pattern). I will, occasionally, create experimental or exploratory branches, which are short-lived, and designed to be reintegrated with master. Again, not a unique workflow. This will be familiar to many folks.
So, lots of “done,” as mileposts on a continuum. In some cases (like when I am preparing course materials), I may do a lot of work in a dev branch, and squash over to master, in order to reduce the “noise” in master, but the “done bits” are always tags in master. I always reintegrate back to dev, so the branches remain in sync, and the tags propagate to dev.
One exception to the “continuum model,” is that fix tags may be in small, short-lived branches. If possible, I try to add the tag to master, at the reintegration point, but that isn’t always practical or safe, especially if I can’t actually reintegrate, but have to introduce the fix as new code, down the road.
Since I have started using the Swift Package Manager, this has been a good fit. I believe that many package managers rely on that pattern, so I don’t think the way that I work is especially creative or standout. Pretty much garden-variety configuration management.
Speaking of package managers, I often extract subprojects from my work in the master branch, as I progress, spinning them off as standalone package projects, with their own project lifecycle, and reintroduce them as dependencies. This gives me little pockets of “done,” that have encapsulated state and identity (and their own repo), along with focused testing. Each project I spin off, is one less thing to worry about, and another tool on the pegboard, for future projects.
So, in my case, “done” is really just a signpost along the road, but there can be many signposts.