Git tips from the trenches

After a few years with git everyone has his own bag o' tricks - a collection of bash aliases, one liners and habits that make his daily work easier.

I've gathered a handful of these with varying complexity hoping that it can give a boost to you. I will not cover git or VCS basics at all, I'm assuming you're already a git-addict.

So fire up your favorite text editor and bear with me.

Check which branches are merged

After a while if you branch a lot you'll se your git branch -a output is polluted like hell (if you havent't cleaned up). It all the more true if you're in a team. So, from time to time you'll do the Big Spring Cleaning only to find it hard to remember which branch you can delete and which you shouldn't. Well, just check out your mainline branch (usually master) and:

$ git checkout master$ git branch --merged

to see all the branches that have already been merged to the current branch (master in this case).

You can do the opposite of course:

How about deleting those obsolete branches right away?

$ git branch --merged | xargs git branch -d

Alternative: use GitHub's Pull request UI if you've been a good sport and always used pull requests.

Find something in your entire git history

Sometimes you find yourself in the situation that you're looking for a specific line of code that you don't find with plain old grep - maybe someone deleted or changed it with a commit. You remember some parts of it but have no idea where and when you committed it. Fortunately git has your back on this. Let's fetch all commits ever then use git's internal grep subcommand to look for your string:

$ git rev-list --all | xargs git grep '<YOUR REGEX>'$ git rev-list --all | xargs git grep -F '<YOUR STRING>'# if you don't want to use regex

Fetch a file from another branch without changing your current branch

Local cherry-picking. Gotta love it. Imagine you're experimenting on your current branch and you suddenly realise you need a file from the oh-so-distant branch. What do you do? Yeah, you can stash, git checkout, etc. but there's an easier way to merge a single file in your current branch from another:

$ git checkout <OTHER_BRANCH> -- path/to/file

See which branches had the latest commits

Could also be useful for a spring cleaning - checking how 'old' those yet unmerged branches are. Let's find out which branch hadn't been committed to in the last decade. Git has a nice subcommand, 'for-each-ref' which can print information for each ref (duh) - the thing is that you can both customize the output format and sort!

$ git for-each-ref --sort=-committerdate --format='%(refname:short) %(committerdate:short)'

It will output branches and tags, too.

This deserves an alias, don't you think?

$ git config --global alias.springcleaning "for-each-ref --sort=-committerdate --format='%(refname:short) %(committerdate:short)'"

Making typos?

Git can autocorrect you.

$ git config --global help.autocorrect 1$ git dffi
WARNING: You called a Git command named 'dffi', which does not exist.
Continuing under the assumption that you meant 'diff'
in 0.1 seconds automatically...

Autocomplete, anyone?

If you download this file and modify your .bash_profile by adding:

source ~/.git-completion.bash

Git will now autocomplete your partial command if you press TAB. Neat.

Hate remnant whitespace?

Let git strip it for you. Use the mighty .gitattributes file in the root of your project and say in it:

Or say you don't want this for all files (*), only scala sources:

*.scala filter=stripWhitespace

But the filter is not defined yet, so chop-chop:

$ git config filter.stripWhitespace.clean strip_whitespace

(Actually there are two types of filters: clean and smudge. Clean runs right before pushing, smudge is run right after pulling)

We still have to define what strip_whitespace is, so create a script on your PATH and of course make it executable:

#!/usr/bin/env rubySTDIN.readlines.eachdo|line|putsline.rstripend

You could also do this as a precommit hook, of course.

Recovering lost data

The rule of thumb is that if you lost data but committed/pushed it somewhere, you're probably able to recover it. There are basically two ways:

reflog

Any change you make that affects a branch is recorded in the reflog. See:

$ git log -g
commit be5de4244c1ef863e454e3fb7765c7e0559a6938
Reflog: HEAD@{0}(Csaba Okrona <xxx@xx.xx>)
Reflog message: checkout: moving from master to master
Author: Robin Ward <xxx@xx.xx>
Date: Fri Nov 8 15:05:14 2013 -0500
 FIX: Pinned posts were not displaying at the top of categories.

If you see your lost commit(s) there, just do a simple:

$ git branch my_lost_data [SHA-1]

Where SHA-1 is the hash after the 'commit' part. Now merge your lost data into your current branch:

git-fsck

This gives you all the objects that aren't referenced by any other object (orphans). You can fetch the SHA-1 hash and do the same dance as above.

A nicer, one-line log

Get a color-coded, one-line-per-commit log showing branches and tags:

$ git log --oneline --decorate
355459b Fix more typos in server locale
b95e74b Merge pull request #1627 from awesomerobot/master
40aa62f adding highlight & fade to linked topic
15c29fd (tag: v0.9.7.3, tag: latest-release) Version bump to v0.9.7.3
c753a3c We shouldn't be matching on the `created_at` field. Causes tests to randomly fail.
dbd2332 Public user profile page shows if the user is suspended and why.

Highlight word changes in diff

Bored of the full-line highlights? This only highlights the changed words, nicely inline. Try:

A shorter, pro git status

Showing only the important things.

$ git status -sb## master...origin/master
?? _posts/2014-02-01-git-tips-from-the-trenches.md
?? images/git-beginner-share.png
?? images/git-beginner.jpg

Bored of setting up tracking branches by hand?

Make git do this by default:

$ git config --global push.default tracking

This sets up the link to the remote if it exists with the same branch name when you push.

Pull with rebase, not merge

To avoid those nasty merge commits all around.

Or do it automatically for any branch you'd like:

$ git config branch.master.rebase true

Or for all branches:

$ git config --global branch.autosetuprebase always

Find out which branch has a specific change

$ git branch --contains [SHA-1]

If you want to include remote tracking branches, add '-a'

Check which changes from a branch are already upstream

Show the last commit with matching message

Write notes for commits

These are only visible locally.

More cautious git blame

Before you play the blame game, make sure you check you're right with:

$ git blame -w # ignores white space$ git blame -M # ignores moving text$ git blame -C # ignores moving text into other files

Aliases make life easier

These go to the '[alias]' section of your .gitconfig

ds= diff --staged # git ds - diff your staged changes == review before committing.st= status -sb # smarter status - include tag and branch infofup= log --since '1 day ago' --oneline --author <YOUR_EMAIL> # I know what you did yesterday - great for follow-upsls= log --pretty=format:"%C(yellow)%h %C(blue)%ad%C(red)%d %C(reset)%s%C(green) [%cn]" --decorate --date=short # pretty one-line log with tags, branches and authorslsv= log --pretty=format:"%C(yellow)%h %C(blue)%ad%C(red)%d %C(reset)%s%C(green) [%cn]" --decorate --date=short --numstat # a verbose ls, shows changed files too# some resets without explanationr= resetr1= reset HEAD^r2= reset HEAD^^rh= reset --hardrh1= reset HEAD^ --hard# basic shortcutscp= cherry-pickst= status -scl= cloneci= commitco= checkoutbr= branch diff= diff --word-diffdc= diff --cached# stash shortcutssl= stash listsa= stash applyss= stash saverh2= reset HEAD^^ --hard

Did I miss something? Tell me in comments ;)