gitredux - Let's remake the git CLI

This project takes inspiration from gitless and legit, and is influenced by this blog post and this fantastic diatribe. However, I feel all of these tools are either underdeveloped or too opinionated. I want a tool that gives me all the same control as git but without the headache of its impossible to remember commands.

Project Goals

1. Stay as close to the original jargon of git as possible, rather than create a dialect or "new" abstractions.

• Some exceptions must be made for the horribly confusing unintuitive names.

• Have a consistant, predictable command names
• Have more useful and predictable default behaviors
• Make recovering from mistakes easier

Equivalent Command List

Stage operations

There are a lot of really terribly named commands here: add, reset, checkout. (checkout? What is this, SVN?) They are probably named so because they are seen from the point-of-view of the stage, rather than the working directory. But since the stage is a new concept to beginners, thinking from the working directory point-of-view is how most new users (or even most normal developers) think.

So here is where I make perhaps the biggest naming changes. Updating the stage copy of a file is now stage as a verb. Reseting the stage copy is now called unstage. get reset now resets the file in the working directory, rather than the file in the stage.

get	git equivalent
get add untracked_file	git add untracked_file
get rm tracked_file	git rm -r tracked_file Then prompt y/n to delete untracked files in deleted directories.
get stage	git add -u :/
get stage tracked_file	git add tracked_file
get unstage	git reset HEAD
get unstage staged_file	git reset HEAD staged_file
get reset	git checkout -f HEAD
get reset file	git checkout file

Making commits/branches/tags

Here I mostly just enhanced the porcelain commands. At some point, I would like to consolidate the syntax for creating and deleting things, where things can be branches, tags, remotes, etc.

get	git equivalent
get commit	Prompts for commit message inline, rather than opening a text editor
get commit message	git commit -m message
get uncommit	git reset --soft HEAD~1
get branch branch	stashes working tree, creates or switches branch, and checks out branch
get rmbranch branch	git branch -d branch TODO: rename?
get tag tag	git tag tag
get untag tag	Deletes local tag and shows Y/N prompt to delete remote tag.
get branches	Print all local and remote branches sorted by last commit date, and more.

For get branches it's easiest just to show an example:

user@host:~/project get branches
  2 days ago    c1632be  master -> origin/master
* 11 days ago   1e8eab0  develop -> origin/develop  (merged)
  5 months ago  d58b929  feature/historical -> ?    (merged)
  5 months ago  3b22622  save -> ?
  2 days ago    c1632be  origin/HEAD
  2 days ago    c1632be  origin/master
  11 days ago   1e8eab0  origin/develop
  11 days ago   1e8eab0  origin/will
  4 weeks ago   f58098a  origin/paul

Viewing changes

By default, all diff commands compare the working directory to something else. By default, that something else is HEAD, not the stage as it is in git, because I think the most common query is "what have I changed since my last commit?" not "what is different between the stage and my working directory?". STAGE refers to the git staging area. I choose to always use the keyword stage rather than index, because they're the same thing and having two names for the same thing is unnecessarily confusing. And calling it the cache in the --cached option just makes it even worse. I avoid the stage/index/cached confusion by choosing to give it a name (STAGE) and treating it like a special reference akin to HEAD.

get	git equivalent
get status	git status
get review	git diff --cached
get diff	compare working tree with HEAD (git diff HEAD)
get diff STAGE	compare working tree with stage (git diff)
get diff ref	compare working tree with ref (git diff ref)
get diff STAGE ref	compare stage with ref (git diff --cached ref)
get diff refA refB	compare refA with refB (git diff refA refB)

Working with remotes

Here, I want practicality and "it just works (TM)". The clone command installs submodules by default. I am (slowly) adding support for popular hosting services. Right now its just Github but I will probably add Bitbucket.

The biggest grievance when working with remotes is the "git pull" command. Which as anyone will tell you, just does "git fetch" followed by "git merge". But merging can be downright dangerous, especially if you have unsaved changed, leading to unsuccessful "git pull" attempts. I am going to avoid "pull" altogether, but to speed things up, my "fetch" operation goes ahead and fast-forwards branches.

get	git equivalent
get clone PATH	git clone --recurse-submodules [-b branch] PATH Paths like "username/repo" will be expanded to a Github URL. Paths like "repo" will be expanded to a Github url if your username is stored in git config.* Paths ending with #branch (the URL style used by NPM and Bower) will clone that branch.
get fetch	Fetches all remotes and fast-forwards local branches when possible
get fetch branches	Fetches and fast-forwards the specified branches
get push	pushes to upstream. If upstream not set, prompt user to name a remote branch. If multiple remotes exist, prompt for which remote to use.
get remote	Interactive prompt to add a remote.

* To save your Github username where get can see it: git config --global github.user your_user_name

Miscellaneous

get	git equivalent
get log	git log --graph --pretty=format:'%h - %d %s (%cr) <%an>' --abbrev-commit -10
get ignore file	Adds file to the current .gitignore file
get submodule	Finds all git repos underneath the main repo and makes them submodules
get squash n	Squash the last n commits into one commit. Prompts for a new commit message. Moves HEAD without actually deleting the old commits.

Shortcuts

Because sometimes you just want to type one letter.

Short Aliases	Full
get ?	get status
get +	get add
get -	get rm
get =	get stage
get !	get commit
get @	get branch
get #	get tag
get ^	get push

Why?

I started this project because git log --no-pager gives an error. Apparently I wanted git --no-pager log. This was the last straw. So I decided to "fix" the git CLI.

EDIT 1:

Here is an even better example of the inanity of git CLI. To get the SHA reference of HEAD, which of these would you use?

git show-ref HEAD --abbrev --hash

or

git rev-parse --short HEAD

Questions:

• Why are they different results?
• Why does show-ref use --abbrev but rev-parse use --short?
• Why are the options after HEAD in show-ref but before HEAD in rev-parse?

I leave answering these questions as an exercise to the reader.

EDIT 2:

Another example of git's terrible option parsing. This command lists all the local branches that have been merged:

$ git branch --list --merged
* develop
  feature/latest

So what do you think this command does?

$ git branch --merged --list 
fatal: malformed object name --list

OH MY GOD COULD YOU BE ANY MORE FRAGILE GIT?

TODO

• Now that I've added tab completion, I think "stage" and "status" are too similar. I've solved that for now by adding short aliases 'get =' and 'get ?'. Not intuitive, but I didn't want to rename status or stage. (Remember, I'm trying to keep as much lingo unchanged.) It's the same number of keystrokes as if using tab completion and one letter. I know, I know, still feels inconvenient somehow.
• Also, I have three four commands that start with "r" which is a problem for tab completion. Seriously considering renaming 'rmbranch' to 'unbranch' to match 'untag' and 'unstage'.

Changelog

• I renamed "update" to "fetch" to avoid completion conflicts with "unstage". Also, it really is just fetch + fast-forward. One of my goals is NOT to use different terminology than git, because that makes StackOverflow less useful.