The mixed-revisions bug was fun… Also cannot clean history or make shitty branches everywhere, it was one of my worst experience. Nowadays Jujutsu is my favorite.
It caters more for a linear workflow, though. So modern large teams won’t find joy with SVN
For what it’s worth, I work at a FAANG company and we don’t use branches at all. Instead, we use feature flags. Source control history is linear with no merges.
All code changes have to go though code review before they can be committed to the main repo. Pull requests are usually not too large (we aim for ~300 lines max), contain a single commit, aren’t long-lived (often merged the same day they’re submitted unless they’re very controversial), can be stacked to handle dependencies between them (“stacked diffs”), and a whole stack can be landed together. When merged, everything is committed directly to the main branch, which all developers are working off of.
I know that both Google and Meta take this approach, and probably other companies too.
Usually, feature branches mean that all the work to implement a particular feature is done on that branch. That could be dozens of commits and weeks of work from several developers. The code isn’t merged until the feature is complete. It’s more common in the industry compared to trunk-based development.
My previous employer had:
Feature branches for each new feature.
A dev/trunk branch, where features branches were merged once they were done.
A beta branch, branched from dev once per week.
A live/prod branch, branched from beta four times per year.
This structure is very common in enterprise apps. Customers that need stability (don’t want things to change a lot, for example if they have their own training material for their staff) use the live branch, while customers that want the newest features use the beta branch.
Bug fixes were annoying since you’d have to first do them in the live branch then port them to the beta and dev branches (or vice versa).
On the other hand, feature flags mean that all the code goes directly into the trunk branch, but it’s turned off until it’s ready. A basic implementation of feature flags would be a static class with a bunch of booleans that get checked throughout the code, but they’re usually dynamic so a misbehaving feature can be turned off without having to redeploy the code.
Some codebases use both feature branches and feature flags.
Ah okay, places I’ve worked have tried to keep tasks as small as possible so you don’t work on your feature branch more than a day. If it takes over a day, should’ve been an epic (and therefore multiple feature branches). Seen different approaches to the whole release thing too. Weekly deployments, 3x per year, or in my current company: deploy as soon as someone has tested it
The bottom picture should be SVN. I miss incremental revision numbers.
The mixed-revisions bug was fun… Also cannot clean history or make shitty branches everywhere, it was one of my worst experience. Nowadays Jujutsu is my favorite.
SVN is still great if there is a need for strict access controls and central control matters a lot. Auditing is also a bit easier with SVN.
It caters more for a linear workflow, though. So modern large teams won’t find joy with SVN.
For what it’s worth, I work at a FAANG company and we don’t use branches at all. Instead, we use feature flags. Source control history is linear with no merges.
All code changes have to go though code review before they can be committed to the main repo. Pull requests are usually not too large (we aim for ~300 lines max), contain a single commit, aren’t long-lived (often merged the same day they’re submitted unless they’re very controversial), can be stacked to handle dependencies between them (“stacked diffs”), and a whole stack can be landed together. When merged, everything is committed directly to the main branch, which all developers are working off of.
I know that both Google and Meta take this approach, and probably other companies too.
What’s the difference between that and feature branches? Sounds like you still have PRs that get merged to main from somewhere - forked repos I guess?
Usually, feature branches mean that all the work to implement a particular feature is done on that branch. That could be dozens of commits and weeks of work from several developers. The code isn’t merged until the feature is complete. It’s more common in the industry compared to trunk-based development.
My previous employer had:
This structure is very common in enterprise apps. Customers that need stability (don’t want things to change a lot, for example if they have their own training material for their staff) use the live branch, while customers that want the newest features use the beta branch.
Bug fixes were annoying since you’d have to first do them in the live branch then port them to the beta and dev branches (or vice versa).
On the other hand, feature flags mean that all the code goes directly into the trunk branch, but it’s turned off until it’s ready. A basic implementation of feature flags would be a static class with a bunch of booleans that get checked throughout the code, but they’re usually dynamic so a misbehaving feature can be turned off without having to redeploy the code.
Some codebases use both feature branches and feature flags.
Ah okay, places I’ve worked have tried to keep tasks as small as possible so you don’t work on your feature branch more than a day. If it takes over a day, should’ve been an epic (and therefore multiple feature branches). Seen different approaches to the whole release thing too. Weekly deployments, 3x per year, or in my current company: deploy as soon as someone has tested it
This makes me happy. 🙂
Trunk based dev is GOAT.
git rev-list HEAD | wc -l