One of the things we do at All Around the World is apply the Pareto Principle: 80% of your results stem from 20% of your actions. We get hired to fix things for clients and don't have the luxury of sitting around for days to debate the perfect way of doing things before we start working. Today, I'll discuss our git workflow—created by following that principle—and how knowing only three simply git commands makes git easy for everyone.
I've introduced this workflow to several clients and all of them love it.
This repo contains a simplified subset of the git tools used by All Around
the World for our software development. It
makes it dead easy for teams using git (and in our case, github) to work
together.
There are only three new commands to remember:
git refresh (rebase your work on top of the current default branch)
git pushback (pushes your changes to origin)
git done (cleanly add your branch back to the default branch)
There are optional git-hub and git-lab commands available:
git hub $issue_number (optional. create new branch based on a github ticket)
git lab $issue_number (optional. create new branch based on a gitlab ticket)
The bin/git-hub command assumes you're using github and the bin/git-lab
command asumes you're using gitlab. The other commands
work fine without it.
The bin/git-lab and bin/git-hub commands require config files. The
bin/git-lab config file is documented via perldoc bin/git-lab. The
bin/git-hub config…
And that's it. Three simple commands to remember and git gets out of your way.
The Overview
The git command is extremely complicated. Invariably in a team environment, someone gets something "wrong" with git and the local git guru comes by and does mysterious things with git fsck, git reflog, and other commands that developers have never heard of. And while it's useful to have a git guru on hand, we find it's even easier to have a single, standard way of using git so that all developers are doing things the same way. Even developers new to git find our workflow very easy.
It's designed largely to avoid git histories which make the Gordian Knot look like a bowtie.
For us, the complicated history was the showstopper. We want a clean, linear history and an "easy-to-use" workflow. It's based on the following logic:
The master branch is the source of truth
All new development happens in a new branch off of master
When you're done, merge your code back into master and delete your branches
In other words, it's very clean, very simple, and focuses on the core development needs. There's nothing surprising about this workflow. How do releases work? That depends on your project and its needs. How do hotfixes work? That depends on your project and its needs. How do code reviews work? That depends on your project and its needs. This workflow is designed to be the core of what you do day in and day out. The bits you need to customize are up to you. I'm not dogmatic.
git-flow
As an aside, one popular workflow is git-flow. Yet git-flow has its detractors and it's easy to find many more. I found the convoluted history and strict requirements that things be "just so" to not be a good fit for our workflow. It might be the perfect fit for yours.
Our git Workflow
The git workflow tools are open source and they contain a simplified subset of the git tools used by All Around the World for our software development. It makes it dead easy for teams using git (and in our case, github) to work together.
There are only three new commands to remember:
git hub $issue Create new branch based on a github ticket
git refresh Rebase your work on top of current master
git merge-with-master Merge your branch back into master
Note: you don't even need the bin/git-hub command. It assumes you're using github, but other commands work fine without it. Further, the bin/git-hub code uses Perl 5.20 and needs some modules installed. However, git refresh and git merge-with-master are just bash scripts, so they'll work fine out of the box if you can use bash.
If you prefer bash, there's an experimental bin/git-hub-bash script. My bash isn't brilliant. Be warned.
Assumptions
The master branch is never worked on directly. Instead, new branches are created (usually for an individual github ticket), we hack, we regularly pull new changes into that branch, and after a pull request is created and the approved, we merge the code back into master.
The examples below assume the files in the bin/ directory are in your path. If they are not in your path, you have to type the commands explicitly:
I assume that branches are usually per ticket. You can manually create a branch, but I tend not to. Instead, if you are going to work on github issue 5738, with the title "Reduce reputation for failed crimes", you run the following command:
git hub 5738
And you're in a shiny new branch named reduce-reputation-for-failed-crimes-5738.
If that branch already exists, it's checked out.
If you're not using github, it's trivial to do this by hand:
git checkout -b name-of-your-branch
I just like the convenience of always having standard naming conventions. And if someone wants to contribute a bin/git-lab script to the project, I'll happily accept it (though a more general bin/git-start-ticket would probably be better).
Caveats
The new branch that is created is based on the branch you're on, so you probably want to run this from master
Branch detection is based on a heuristic, using the ticket number. If you have more than one branch with that ticket number, you will get a warning and the code will exit.
You'll need a config file for this. See perldoc bin/git-hub for this.
Assumes Perl 5 version 20
You will also need to install the following modules from the CPAN if you've not already done so.
autodie
Config::General
JSON
Text::Unidecode
Pithub
If you're not familiar with installing CPAN modules, check out cpanminus, or you can use the cpan command line tool that is standard with Perl.
Refreshing Your Branch
While working on your branch, you want to regularly pull in the latest master to keep your code up-to-date. Working on a change for a week with a fast-moving codebase can cause serious headaches when it's time to merge. Thus, you should regularly run the following in your branch (I try to run it at least once per day):
git refresh
Regardless of the branch you are on, this code:
Stashes changes (if any)
Checks out master
Does a fast-forward merge
Checks out your branch (if branch is not master)
Rebases onto master (if branch is not master)
Pops changes from stash, if any
In other words, it cleanly rebases your code on top of master, even if you have uncommitted changes.
Note: for everyone who keeps asking "why not git pull --rebase --autostash?"
The simple reason is backwards-compatibility. Sometimes developers have older versions of git installed and while that's usually perfectly compatible with newer versions, we don't force them to upgrade. Also, internally git-refresh does things like git fetch -p to remove references to remote branches which no longer exist. This regular "house cleaning" helps to keep git more performant. See the Pruning documentation for git-fetch for more information.
Note: if this command aborts (such as during a merge conflict), the stashed changes will remain stashed. Run git stash to see those changes and git stash pop to return them to your codebase when you've finished resolving the merge conflicts.
Merging into master
So you've finished your work and are ready to merge your branch back into master. Here's one way to do that very cleanly and safely:
And when you're done, it will also print out the instructions on how you can delete your local and remote copies of the branch, to help you be nice and not leave branches cluttering up the repository.
Merge Conflicts
With all of this rebasing, you're sometimes going to hit the dreaded ...
Auto-merging lib/Veure/Moose.pm
CONFLICT (content): Merge conflict in lib/Veure/Moose.pm
Failed to merge in the changes.
Patch failed at 0001 Issue 144 Make read-only the default for attributes
And then you're very unhappy. Worse, if you do this after a week of isolated hacking, you might get a huge amount of merge conflicts to work through. However, by running git refresh at least once a day, if not more often, those conflicts are minimized. And when you do get them? I have the following in my .bash_aliases file:
alias damnit='${EDITOR:-vim} $(git grep -l "<<<< HEAD")'
Then, when I get a conflict, I just type damnit and am automatically editing the files which caused the conflict. It saves me a huge amount of time.
If you have ack installed, you might find this better:
alias damnit="${EDITOR:-vim} $(ack -l --nopager '^(?:<|>|=){7}(?:\s|$)')"
Things To Do
If you're interested in helping, there are several tasks we've not gotten around to because we didn't need them. Others might.
Change git-hub to a bash script or remove the need to install dependencies
Have a config file for all the tools, not just git-hub. It would also specify the name of the main branch
Create a generic git ticket $identifier which knows about more than just github