Branching in Git is one of my favorite features. If you have used other version control systems, it's probably helpful to forget most of what you think about branches - in fact, it may be more helpful to think of them practically as contexts since that is how you will most often be using them. When you checkout different branches, you change contexts that you are working in and you can quickly context-switch back and forth between several different branches.
In a nutshell you can create a branch with
git branch (branchname), switch into that context with
git checkout (branchname), record commit snapshots while
in that context, then can switch back and forth easily. When you switch
branches, Git replaces your working directory with the snapshot of the
latest commit on that branch so you don't have to have multiple directories
for multiple branches. You merge branches together with
git merge. You can easily merge multiple times from the same
branch over time, or alternately you can choose to delete a branch
immediately after merging it.
The git branch command is a general branch management tool
for Git and can do several different things. We'll cover the basic ones
that you'll use most - listing branches, creating branches and deleting
branches. We will also cover basic git checkout here which
switches you between your branches.
Without arguments, git branch will list out the local
branches that you have. The branch that you are currently working on will
have a star next to it and if you have
coloring turned on,
will show the current branch in green.
$ git branch
* master
This means that we have a 'master' branch and we are currently on it.
When you run git init it will automatically create a 'master'
branch for you by default, however there is nothing special about the name -
you don't actually have to have a 'master' branch but since it's the default
that is created, most projects do.
So let's start by creating a new branch and switching to it. You can do
that by running git branch (branchname).
$ git branch testing
$ git branch
* master
testing
Now we can see that we have a new branch. When you create a branch this
way it creates the branch at your last commit so if you record some commits
at this point and then switch to 'testing', it will revert your working
directory context back to when you created the branch in the first place -
you can think of it like a bookmark for where you currently are. Let's see
this in action - we use git checkout (branch) to switch the
branch we're currently on.
$ ls README hello.rb $ echo 'test content' > test.txt $ echo 'more content' > more.txt $ git add *.txt $ git commit -m 'added two files' [master 8bd6d8b] added two files 2 files changed, 2 insertions(+), 0 deletions(-) create mode 100644 more.txt create mode 100644 test.txt $ ls README hello.rb more.txt test.txt $ git checkout testing Switched to branch 'testing' $ ls README hello.rb
So now we can see that when we switch to the 'testing' branch, our new files were removed. We could switch back to the 'master' branch and see them re-appear.
$ ls README hello.rb $ git checkout master Switched to branch 'master' $ ls README hello.rb more.txt test.txt
In most cases you will be wanting to switch to the branch immediately, so
you can do work in it and then merging into a branch that only contains
stable work (such as 'master') at a later point when the work in your new
context branch is stable. You can do this pretty easily with
git branch newbranch; git checkout newbranch, but Git gives
you a shortcut for this: git checkout -b newbranch.
$ git branch * master $ ls README hello.rb more.txt test.txt $ git checkout -b removals Switched to a new branch 'removals' $ git rm more.txt rm 'more.txt' $ git rm test.txt rm 'test.txt' $ ls README hello.rb $ git commit -am 'removed useless files' [removals 8f7c949] removed useless files 2 files changed, 0 insertions(+), 2 deletions(-) delete mode 100644 more.txt delete mode 100644 test.txt $ git checkout master Switched to branch 'master' $ ls README hello.rb more.txt test.txt
You can see there how we created a branch, removed some of our files while in the context of that branch, then switched back to our main branch and we see the files return. Branching safely isolates work that we do into contexts we can switch between.
If you start on work it is very useful to always start it in a branch (because it's fast and easy to do) and then merge it in and delete the branch when you're done. That way if what you're working on doesn't work out you can easily discard it and if you're forced to switch back to a more stable context your work in progress is easy to put aside and then come back to.
If we want to delete a branch (such as the 'testing' branch in the
previous example, since there is no unique work on it),
we can run git branch -d (branch) to remove it.
$ git branch * master testing $ git branch -d testing Deleted branch testing (was 78b2670). $ git branch * master
In a nutshell you use git branch to list your
current branches, create new branches and delete unnecessary or
already merged branches.
Once you have work isolated in a branch, you will eventually want to
incorporate it into your main branch. You can merge any branch into your
current branch with the git merge command. Let's take as a
simple example the 'removals' branch from above. If we create a branch
and remove files in it and commit our removals to that branch, it is
isolated from our main ('master', in this case) branch. To include those
deletions in your 'master' branch, you can just merge in the 'removals'
branch.
$ git branch * master removals $ ls README hello.rb more.txt test.txt $ git merge removals Updating 8bd6d8b..8f7c949 Fast-forward more.txt | 1 - test.txt | 1 - 2 files changed, 0 insertions(+), 2 deletions(-) delete mode 100644 more.txt delete mode 100644 test.txt $ ls README hello.rb
Of course, this doesn't just work for simple file additions and deletions. Git will merge file modifications as well - in fact, it's very good at it. For example, let's see what happens when we edit a file in one branch and in another branch we rename it and then edit it and then merge these branches together. Chaos, you say? Let's see.
$ git branch
* master
$ cat hello.rb
class HelloWorld
def self.hello
puts "Hello World"
end
end
HelloWorld.hello
So first we're going to create a new branch named 'change_class' and switch to it so your class renaming changes are isolated. I'm going to change each instance of 'HelloWorld' to 'HiWorld'.
$ git checkout -b change_class M hello.rb Switched to a new branch 'change_class' $ vim hello.rb $ head -1 hello.rb class HiWorld $ git commit -am 'changed the class name' [change_class 3467b0a] changed the class name 1 files changed, 2 insertions(+), 4 deletions(-)
So now I've committed the class renaming changes to the 'change_class'
branch. If I now switch back to the 'master' branch my class name will
revert to what it was before I switched branches. Here I can change
something different (in this case the printed output) and at the same
time rename the file from hello.rb to ruby.rb.
$ git checkout master Switched to branch 'master' $ git mv hello.rb ruby.rb $ vim ruby.rb $ git diff diff --git a/ruby.rb b/ruby.rb index 2aabb6e..bf64b17 100644 --- a/ruby.rb +++ b/ruby.rb @@ -1,7 +1,7 @@ class HelloWorld def self.hello - puts "Hello World" + puts "Hello World from Ruby" end end $ git commit -am 'added from ruby' [master b7ae93b] added from ruby 1 files changed, 1 insertions(+), 1 deletions(-) rename hello.rb => ruby.rb (65%)
Now those changes are recorded in my 'master' branch. Notice that the class name is back to 'HelloWorld', not 'HiWorld'. Now I want to incorporate the 'HiWorld' change so I can just merge in my 'change_class' branch. However, I've changed the name of the file since I branched, what will Git do?
$ git branch
change_class
* master
$ git merge change_class
Renaming hello.rb => ruby.rb
Auto-merging ruby.rb
Merge made by recursive.
ruby.rb | 6 ++----
1 files changed, 2 insertions(+), 4 deletions(-)
$ cat ruby.rb
class HiWorld
def self.hello
puts "Hello World from Ruby"
end
end
HiWorld.hello
Well, it will just figure it out. Notice that I had no merge conflicts and the file that had been renamed now has the 'HiWorld' class name change that was done in the other branch. Pretty cool.
So, Git merges are magical, we never ever have to deal with merge conflicts again, right? Not quite. In situations where the same block of code is edited in different branches there is no way for a computer to figure it out, so it's up to us. Let's see another example of changing the same line in two branches.
$ git branch * master $ git checkout -b fix_readme Switched to a new branch 'fix_readme' $ vim README $ git commit -am 'fixed readme title' [fix_readme 3ac015d] fixed readme title 1 files changed, 1 insertions(+), 1 deletions(-)
Now we have committed a change to one line in our README file in a branch. Now let's change the same line in a different way back on our 'master' branch.
$ git checkout master Switched to branch 'master' $ vim README $ git commit -am 'fixed readme title differently' [master 3cbb6aa] fixed readme title differently 1 files changed, 1 insertions(+), 1 deletions(-)
Now is the fun part - we will merge the first branch into our master branch, causing a merge conflict.
$ git merge fix_readme Auto-merging README CONFLICT (content): Merge conflict in README Automatic merge failed; fix conflicts and then commit the result. $ cat README <<<<<<< HEAD Many Hello World Examples ======= Hello World Lang Examples >>>>>>> fix_readme This project has examples of hello world in nearly every programming language.
You can see that Git inserts standard merge conflict markers, much like Subversion, into files when it gets a merge conflict. Now it's up to us to resolve them. We will do it manually here, but check out git mergetool if you want Git to fire up a graphical mergetool (like kdiff3, emerge, p4merge, etc) instead.
$ vim README # here I'm fixing the conflict $ git diff diff --cc README index 9103e27,69cad1a..0000000 --- a/README +++ b/README @@@ -1,4 -1,4 +1,4 @@@ - Many Hello World Examples -Hello World Lang Examples ++Many Hello World Lang Examples This project has examples of hello world in
A cool tip in doing merge conflict resolution in Git is that if you
run git diff, it will show you both sides of the conflict
and how you've resolved it as I've shown here. Now it's time to mark
the file as resolved. In Git we do that with git add -
to tell Git the file has been resolved, you have to stage it.
$ git status -s UU README $ git add README $ git status -s M README $ git commit [master 8d585ea] Merge branch 'fix_readme'
And now we've successfully resolved our merge conflict and committed the result.
In a nutshell you use git merge to combine another
branch context into your current branch. It automatically figures out
how to best combine the different snapshots into a new snapshot with the
unique work of both.
So far we have been committing snapshots of your project and switching
between different isolated contexts, but what if we've forgotten how we've
got to where we are? Or what if we want to know how one branch differs
from another? Git provides a tool that shows you all the commit messages
that have lead up to the snapshot you are currently on, which is called
git log.
To understand the log command, you have to understand what information
is stored when you run the git commit command to store a
snapshot. In addition to the manifest of files and commit message and
information about the person who committed it, Git also stores the commit
that you based this snapshot on. That is, if you clone a project, what was
the snapshot that you modified to get to the snapshot that you saved? This
is helpful to give context to how the project got to where it is and allows
Git to figure out who changed what. If Git has the snapshot you save and
the one you based it on, then it can automatically figure out what you
changed. The commit that a new commit was based on is called the "parent".
To see a chronological list of the parents of any branch, you can run
git log when you are in that branch. For example, if we run
git log in the Hello World project that we have been working
on in this section, we'll see all the commit messages that we've done.
$ git log commit 8d585ea6faf99facd39b55d6f6a3b3f481ad0d3d Merge: 3cbb6aa 3ac015d Author: Scott Chacon <schacon@gmail.com> Date: Fri Jun 4 12:59:47 2010 +0200 Merge branch 'fix_readme' Conflicts: README commit 3cbb6aae5c0cbd711c098e113ae436801371c95e Author: Scott Chacon <schacon@gmail.com> Date: Fri Jun 4 12:58:53 2010 +0200 fixed readme title differently commit 3ac015da8ade34d4c7ebeffa2053fcac33fb495b Author: Scott Chacon <schacon@gmail.com> Date: Fri Jun 4 12:58:36 2010 +0200 fixed readme title commit 558151a95567ba4181bab5746bc8f34bd87143d6 Merge: b7ae93b 3467b0a Author: Scott Chacon <schacon@gmail.com> Date: Fri Jun 4 12:37:05 2010 +0200 Merge branch 'change_class' ...
To see a more compact version of the same history, we can use the
--oneline option.
$ git log --oneline 8d585ea Merge branch 'fix_readme' 3cbb6aa fixed readme title differently 3ac015d fixed readme title 558151a Merge branch 'change_class' b7ae93b added from ruby 3467b0a changed the class name 17f4acf first commit
What this is telling us is that this is the history of the development of this project. If the commit messages are descriptive, this can inform us as to what all changes have been applied or have influenced the current state of the snapshot and thus what is in it.
We can also use it to see when the history was branched and merged with
the very helpful --graph option. Here is the same command
but with the topology graph turned on:
$ git log --oneline --graph * 8d585ea Merge branch 'fix_readme' |\ | * 3ac015d fixed readme title * | 3cbb6aa fixed readme title differently |/ * 558151a Merge branch 'change_class' |\ | * 3467b0a changed the class name * | b7ae93b added from ruby |/ * 17f4acf first commit
Now we can more clearly see when effort diverged and then was merged
back together. This is very nice for seeing what has happened or what
changes are applied, but
it is also incredibly useful for managing your branches. Let's create a new
branch, do some work in it and then switch back and do some work in our
master branch, then see how the log command can help us figure
out what is happening on each.
First we'll create a new branch to add the Erlang programming language Hello World example - we want to do this in a branch so that we don't muddy up our stable branch with code that may not work for a while so we can cleanly switch in and out of it.
$ git checkout -b erlang Switched to a new branch 'erlang' $ vim erlang_hw.erl $ git add erlang_hw.erl $ git commit -m 'added erlang' [erlang ab5ab4c] added erlang 1 files changed, 5 insertions(+), 0 deletions(-) create mode 100644 erlang_hw.erl
Since we're having fun playing in functional programming languages we get caught up in it and also add a Haskell example program while still in the branch named 'erlang'.
$ vim haskell.hs $ git add haskell.hs $ git commit -m 'added haskell' [erlang 1834130] added haskell 1 files changed, 4 insertions(+), 0 deletions(-) create mode 100644 haskell.hs
Finally, we decide that we want to change the class name of our Ruby program back to the way it was. So, we can go back to the master branch and change that and we decide to just commit it directly in the master branch instead of creating another branch.
$ git checkout master Switched to branch 'master' $ ls README ruby.rb $ vim ruby.rb $ git commit -am 'reverted to old class name' [master 594f90b] reverted to old class name 1 files changed, 2 insertions(+), 2 deletions(-)
So, now say we don't work on the project for a while, we have other
things to do. When we come back we want to know what the 'erlang' branch
is all about and where we've left off on the master branch. Just by looking
at the branch name, we can't know that we made Haskell changes in there, but
using git log we easily can. If you give Git a branch name,
it will show you just the commits that are "reachable" in the history of
that branch, that is the commits that influenced the final snapshot.
$ git log --oneline erlang
1834130 added haskell
ab5ab4c added erlang
8d585ea Merge branch 'fix_readme'
3cbb6aa fixed readme title differently
3ac015d fixed readme title
558151a Merge branch 'change_class'
b7ae93b added from ruby
3467b0a changed the class name
17f4acf first commit
This way, it's pretty easy to see that we have Haskell code included in the branch (as I've highlighted). What is even cooler is that we can easily tell Git that we only are interested in the commits that are reachable in one branch that are not reachable in another, in other words which commits are unique to a branch in comparison to another.
In this case if we are interested in merging in the 'erlang' branch we
want to see what commits are going to effect our snapshot when we do
that merge. The way we tell Git that is by putting a ^ in
front of the branch that we don't want to see. For instance, if we want
to see the commits that are in the 'erlang' branch that are not in the
'master' branch, we can do erlang ^master, or vice versa.
$ git log --oneline erlang ^master 1834130 added haskell ab5ab4c added erlang $ git log --oneline master ^erlang 594f90b reverted to old class name
This gives us a nice, simple branch management tool. It allows us to easily see what commits are unique to which branches so we know what we're missing and what we would be merging in if we were to do a merge.
In a nutshell you use git log to list out the commit
history or list of changes people have made that have lead to the snapshot
at the tip of the branch. This allows you to see how the project in that
context got to the state that it is currently in.
If you get to a point that is important and you want to forever remember
that specific commit snapshot, you can tag it with git tag.
The tag command will basically put a permanent bookmark at
a specific commit so you can use it to compare to other commits in the
future. This is often done when you cut a release or ship something.
Let's say we want to release our Hello World project as version "1.0".
We can tag the last commit (HEAD) as "v1.0" by running
git tag -a v1.0. The -a means "make an annotated
tag", which allows you to add a tag message to it, which is what you almost
always want to do. Running this without the -a works too, but
it doesn't record when it was tagged, who tagged it, or let you add a tag
message. I would recommend always creating annotated tags.
$ git tag -a v1.0
When you run the git tag -a command, Git will open your editor
and have you write a tag message, just like you would write a commit
message.
Now, notice when we run git log --decorate, we can see our
tag there.
$ git log --oneline --decorate --graph
* 594f90b (HEAD, tag: v1.0, master) reverted to old class name
* 8d585ea Merge branch 'fix_readme'
|\
| * 3ac015d (fix_readme) fixed readme title
* | 3cbb6aa fixed readme title differently
|/
* 558151a Merge branch 'change_class'
|\
| * 3467b0a changed the class name
* | b7ae93b added from ruby
|/
* 17f4acf first commit
If we do more commits, the tag will stay right at that commit, so we have that specific snapshot tagged forever and can always compare future snapshots to it.
We don't have to tag the commit that we're on, however. If we forgot to
tag a commit that we released, we can retroactively tag it by running the
same command, but with the commit SHA at the end. For example, say we had
released commit 558151a (several commits back) but forgot to
tag it at the time. We can just tag it now:
$ git tag -a v0.9 558151a
$ git log --oneline --decorate --graph
* 594f90b (HEAD, tag: v1.0, master) reverted to old class name
* 8d585ea Merge branch 'fix_readme'
|\
| * 3ac015d (fix_readme) fixed readme title
* | 3cbb6aa fixed readme title differently
|/
* 558151a (tag: v0.9) Merge branch 'change_class'
|\
| * 3467b0a changed the class name
* | b7ae93b added from ruby
|/
* 17f4acf first commit