SugarJar
[!IMPORTANT] As this was meant to replace arc/jf, which has now been open-sourced as Sapling, I highly recommend taking a look at that!
Sapling is a great tool and solves a variety of problems SugarJar will never be able to. However, it is a bigger workflow change, so existing SJ users may choose to stick with this. Similarly some workflows may not be suitable for Sapling. I still plan to maintain and develop SugarJar for the time being.
Welcome to SugarJar - a git/github helper. It needs one of the GitHub CLI's: either gh or the older hub.
SugarJar is inspired by arcanist, and its replacement at Facebook, JellyFish. Many of the features they provide for the Phabricator workflow this aims to bring to the GitHub workflow.
In particular there are a lot of helpers for using a squash-merge workflow that is poorly handled by the standard toolsets.
If you miss Mondrian or Phabricator - this is the tool for you!
If you don't, there's a ton of useful stuff for everyone!
Installation
Sugarjar is packaged in a variety of Linux distributions - see if it's on the
list here, and if so, use your package manager (or gem
) to install it:
If you are using a Linux distribution version that is end-of-life'd, click the above image, it'll take you to a page that lists unsupported distro versions as well (they'll have older SugarJar, but they'll probably still have some version).
Ubuntu users, Ubuntu versions prior to 24.x cannot be updated, so if you're on an older Ubuntu please use this PPA from our Ubuntu package maintainer.
For MacOS users, we recommend using Homebrew - SugarJar is now in Homebrew Core.
NOTE: If you previously used our custom Homebrew tap, you should remove and untap it:
homebrew uninstall sugarjar
homebrew untap jaymzh/sugarjar
Then you can install the core version (brew install sugarjar
).
Finally, if none of those work for you, you can clone this repo and run it directly from there.
Auto cleanup squash-merged branches
It is common for a PR to go back and forth with a variety of nits, lint fixes,
typos, etc. that can muddy history. So many projects will "squash and merge"
when they accept a pull request. However, that means git branch -d <branch>
doesn't work. Git will tell you the branch isn't fully merged. You can, of
course git branch -D <branch>
, but that does no safety checks at all, it
forces the deletion.
Enter sj bclean
- it determines if the contents of your branch has been merge
and safely deletes if so.
sj bclean
Will delete a branch, if it has been merged, even if it was squash-merged.
You can pass it a branch if you'd like (it defaults to the branch you're on):
sj bclean <branch>
.
But it gets better! You can use sj bcleanall
to remove all branches that have
been merged:
$ git branch
* argparse
master
feature
hubhost
$ git bcleanall
Skipping branch argparse - there are unmerged commits
Reaped branch feature
Reaped branch hubhost
Smarter clones and remotes
There's a pattern to every new repo we want to contribute to. First we fork, then we clone the fork, then we add a remote of the upstream repo. It's monotonous. SugarJar does this for you:
sj smartclone jaymzh/sugarjar
(also sj sclone
)
This will:
- Make a fork of the repo, if you don't already have one
- Clone your fork
- Add the original as an 'upstream' remote
Note that it takes short names for repos. No need to specify a full URL, just a $org/$repo.
Like git clone
, sj sclone
will accept an additional argument as the
destination directory to clone to. It will also pass any other unknown options
to git clone
under the hood.
Work with stacked branches more easily
It's important to break changes into reviewable chunks, but working with
stacked branches can be confusing. Enter binfo
- it gives you a view of your
current branch all the way up to master. In this example imagine we have a
branch structure like:
+- test2.1
/
master --- test --- test2 --- test3
This is what binfo
on test3 looks like:
$ sj binfo
* e451865 (HEAD -> test3) test3
* e545b41 (test2) test2
* c808eae (test1) test1
o 44cf9e2 (origin/master, origin/HEAD, master) Lint/gemspec cleanups
while binfo
on test2.1 looks like:
$ sj binfo
* 36d0136 (HEAD -> test2.1) test2.1
* e545b41 (test2) test2
* c808eae (test1) test1
o 44cf9e2 (origin/master, origin/HEAD, master) Lint/gemspec cleanups
Have a better lint/unittest experience!
Ever made a PR, only to find out later that it failed tests because of some small lint issue? Not anymore! SJ can be configured to run things before pushing. For example,in the SugarJar repo, we have it run Rubocop (ruby lint) and Markdownlint "on_push". If those fail, it lets you know and doesn't push.
You can configure SugarJar to tell it how to run both lints and unittests for a given repo and if one or both should be run prior to pushing.
The details on the config file format is below, but we provide three commands:
git lint
Run all linters.
git unit
Run all unittests.
git smartpush # or spush
Run configured push-time actions (nothing, lint, unit, both), and do not push if any of them fail.
Better push defaults
In addition to running pre-push tests for you smartpush
also picks smart
defaults for push. So if you sj spush
with no arguments, it uses the
origin
remote and the same branch name you're on as the remote branch.
Cleaning up your own history
Perhaps you contribute to a project that prefers to use merge commits, so you like to clean up your own history. This is often difficult to get right - a combination of rebases, amends and force pushes. We provide two commands here to help.
The first is pretty straight forward and is basically just an alias: sj
amend
. It will amend whatever you want to the most recent commit (just an
alias for git commit --amend
). It has a partner qamend
(or amendq
if you
prefer) that will do so without prompting to update your commit message.
So now you've rebased or amended, pushing becomes challenging. You can git push
--force
, but everyone knows that's incredibly dangerous. Is there a better
way? There is! Git provides git push --force-with-lease
- it checks to make
sure you're up-to-date with the remote before forcing the push. But man that
command is a mouthful! Enter sj fpush
. It has all the smarts of sj
smartpush
(runs configured pre-push actions), but adds --force-with-lease
to
the command!
Better feature branches
When you want to start a new feature, you want to start developing against
latest. That's why sj feature
defaults to creating a branch against what we
call "most master". That is, upstream/master
if it exists, otherwise
origin/master
if that exists, otherwise master
. You can pass in an
additional argument to base it off of something else.
$ git branch
master
test1
test2
* test2.1
test3
$ sj feature test-branch
Created feature branch test-branch based on origin/master
$ sj feature dependent-feature test-branch
Created feature branch dependent-feature based on test-branch
Additionally you can specify a feature_prefix
in your config which will cause
feature
to create branches prefixed with your feature_prefix
and will also
cause co
to checkout branches with that prefix. This is useful when organizations
use branch-based workflows and branches need to be prefixed with e.g. $USER/
.
For example, if your prefix was user/
, then sj feature foo
would create
user/foo
, and sj co foo
would switch to user/foo
.
Smartlog
Smartlog will show you a tree diagram of your branches! Simply run sj
smartlog
or sj sl
for short.
Pulling in suggestions from the web
When someone 'suggests' a change in the GitHub WebUI, once you choose to commit
them, your origin and local branches are no longer in-sync. The
pullsuggestions
command will attempt to merge in any remote commits to your
local branch. This command will show a diff and ask for confirmation before
attempting the merge and - if allowed to continue - will use a fast-forward
merge.
And more!
See sj help
for more commands!
Using SugarJar as a git wrapper
SugarJar, by default, will pass any command it doesn't know straight to hub
(which passes commands it doesn't know to git
). If you have configured
SugarJar to use gh
instead of hub
, then it will pass commands straight to
git
since gh
doesn't act as a git
wrapper.
As such you can alias it to git
and just have a super-git.
$ alias git=sj
$ git config -l | grep color
color.diff=auto
color.status=auto
color.branch=auto
color.branch.current=yellow reverse
color.branch.local=yellow
color.branch.remote=green
$ git br
* dependent-feature 44cf9e2 Lint/gemspec cleanups
master 44cf9e2 Lint/gemspec cleanups
test-branch 44cf9e2 Lint/gemspec cleanups
test1 c808eae [ahead 1] test1
test2 e545b41 test2
test2.1 c1831b3 test2.1
test3 e451865 test3
It's for this reason that SugarJar doesn't have conflicting command names. You
can turn off fallthru by setting fallthru: false
in your config.
The only command we "override" is version
, in which case we not only print
our version, but also call hub version
which prints its version and calls
git version
too!
Configuration
Sugarjar will read in both a system-level config file
(/etc/sugarjar/config.yaml
) and a user-level config file
~/.config/sugarjar/config.yaml
, if they exist. Anything in the user config
will override the system config, and command-line options override both. The
yaml file is a straight key-value pair of options without their '--'. For
example:
log_level: debug
github_user: jaymzh
In addition, the environment variable SUGARJAR_LOGLEVEL
can be defined to set
a log level. This is primarily used as a way to turn debug on earlier in order to
troubleshoot configuration parsing.
Repository Configuration
Sugarjar looks for a .sugarjar.yaml
in the root of the repository to tell it
how to handle repo-specific things. Currently there options are:
lint
- A list of scripts to run onsj lint
. These should be linters like rubocop or pyflake. Linters will be run from the root of the repo.lint_list_cmd
- A command to run which will print out linters to run, one per line. Takes precedence overlint
. The command (and the resulting linters) will be run from the root of the repo.unit
- A list of scripts to run onsj unit
. These should be unittest runners like rspec or pyunit. Test will be run from the root of the repo.unit_list_cmd
- A command to run which will print out the unit tests to run, one more line. Takes precedence overunit
. The command (and the resulting unit tests) will be run from the root of the repo.on_push
- A list of types (lint
,unit
) of checks to run before pushing. It is highly recommended this is onlylint
. The goal here is to allow for the user to get quick stylistic feedback before pushing their branch to avoid the push-fix-push-fix loop.commit_template
- A path to a commit template to set in thecommit.template
git config for this repo. Should be either a fully-qualified path, or a path relative to the repo root.include_from
- This will read an additional repoconfig file and merge it into the one being read. The value should be relative to the root of the repo. This will not error if the file does not exist, it is intended for organizations to allow users to optionally extend a default repo config.overwrite_from
- Same asinclude_from
, but completely overwrites the base configuration if the file is found.
Example configuration:
lint:
- scripts/lint
unit:
- scripts/unit
on_push:
- lint
commit_template: .commit-template.txt
Commit Templates
While GitHub provides a way to specify a pull-request template by putting the
right file into a repo, there is no way to tell git to automatically pick up a
commit template by dropping a file in the repo. Users must do something like:
git config commit.template <file>
. Making each developer do this is error
prone, so this setting will automatically set this up for each developer.
Enterprise GitHub
Like hub
, SugarJar supports GitHub Enterprise. In fact, we provide extra
features just for it.
We recommend the global or user config specify the github_host
. However, most
users will also have a few repos from upstream so always specifying a
github_host
is sub-optimal.
So, when you overwrite the github_host
on the command line, we go ahead and
set the hub.host
git config in that single repo so that it'll "just work"
from there on out.
In other words, assuming your global SJ config has github_host:
github.sample.com
, and the you clone sugarjar with:
sj clone jaymzh/sugarjar --github-host githuh.com
We will add the hub.host
to the sugarjar
clone so that future hub
or sj
commands work without needing to specify..
Choosing a GitHub CLI
SugarJar will use gh
if it is available or otherwise fall back to hub
. You
can override this by specifying --github-cli
on the command line or setting
github_cli
to either gh
or hub
(it defaults to auto
) in your
configuration.
FAQ
Why the name SugarJar?
It's mostly a backronym. Like jellyfish, I wanted two letters that were on home row on different sides of the keyboard to make it easy to type. I looked at the possible options that where there and not taken and tried to find one I could make an appropriate name out of. Since this utility adds lots of sugar to git and github, it seemed appropriate.
Why did you originally use hub
instead of the newer gh
CLI?
When I originally wrote SugarJar, gh
was in early development, and hub
had
many more features. In addition, I wrote SugarJar to be a wrapper for git/hub,
and hub
allows this but gh
does not.
When gh
matured, we added experimental gh
support in 0.0.11, and switched the
default to prefer gh
in 1.0.0.
I'd like to package SugarJar for my favorite distro/OS, is that OK?
Of course! But I'd appreciate you emailing me to give me a heads up. Doing so will allow me to make sure it shows up in the Repology badge above.
What platforms does it work on?
Since it's Ruby, it should work across all platforms, however, it's developed and primarily tested on Linux as well as regularly used on Mac. I've not tested it on Windows, but I'll happily accept patches for Windows compatibility.
How do I get tab-completion?
If the package for your OS/distro didn't set it up manually, you should find
that sugarjar_completion.bash
is included in the package, and you can simply
source that in your dotfiles, assuming you are using bash.