Typically there are three types of people who use GitHub to access source code: users, developers, and maintainers. Users simply want access to the latest code. Developers want to make some changes to the code. Maintainers are the ones who make sure the changes to the code make sense, and incorporate them into the code base. We will talk about each of these use cases in turn.
If you want to use software that is hosted on GitHub and get updates when they are released, but you don't think it's likely that you will ever want to fix bugs or add new features yourself, then probably you are a User (you can upgrade yourself to a Developer later if you need to).
Users will want to take the "official" version of the software, make a copy of
it on their own computer, and run the code from there. Using expyfun
software as an example, this is done on the command line like this:
$ git clone git://github.com/LABSN/expyfun.git $ cd expyfun $ python setup.py install
The git clone command will create a folder expyfun in the current
directory to store the source code, and the python setup.py install
command will install the necessary parts of expyfun in the special places
where python
looks when you do import expyfun and call expyfun functions in
your own scripts. The only thing you really need to decide first is where on
your computer to store the source code. When you want to update expyfun to a
newer version of the source code, just go back to that folder in your terminal
and type:
$ git pull $ python setup.py install
What is actually happening under the hood when you do these things? In addition
to making you a local copy of the source code, the initial git clone
command sets up a relationship between that folder on your computer and the
"origin" of the code. You can see this by typing:
$ git remote -v origin git://github.com/LABSN/expyfun.git (fetch) origin git://github.com/LABSN/expyfun.git (push)
This tells you that git knows about two "remote" addresses of
expyfun: one to fetch new changes from (if the source code gets updated
by someone else), and one to push new changes to (if you make changes that
you want to share). In this case the two addresses both point to the "official"
version of the expyfun software, and because you're a User, you won't
actually be allowed to push any changes you make: the remote copy that
your computer calls origin is picky about who it allows to make changes
(i.e., only the Maintainers can).
The initial git clone command also automatically gives your local copy
of the code a "branch" name. You probably won't need to use this functionality,
but the short explanation is that git allows you to add, delete, and
change files within the expyfun directory and then easily go back to where
you started from in case your changes didn't work. One way git enables
this is by having different "branches" of the same source code, and by default
it creates a branch called "master" when it first clones the remote repository
to your computer:
$ git branch * master
The asterisk means that master is the currently active branch. Later, if you
update expyfun to a newer version using git pull, what git
is really doing is git pull origin master, i.e., pulling changes that
exist in the origin copy of expyfun into your local master branch.
One of the most challenging concepts with git source code control
is the fact that its job is to seamlessly manage multiple different versions
of a code base. The idea is that git allows you to swap back and
forth between these different versions while working within the same directory
structure.
In other words, typically there is only one repository for a given project on
a developer's local machine. That repository can have multiple
branches
within it.
When a developer switches between branches, they are effectively switching
between different copies of that repository. The git protocol thus
modifies or updates files in-place, and you can keep track of different versions
of the files by keeping track of branch names, instead of storing temporary
copies in other folders, or appending _newVersion to the filename.
In addition to each developer having a local git repository, they usually become
associated with some number of remote repositories on
GitHub. The developer's local repository can "know"
about any number of remote repositories, and users get to name those remote
repositories whatever they want. Here is a diagram of the connections between
the remote repositories on GitHub and the local computers of each of three
collaborators on the expyfun project:
Developers (like rkmaddox in the diagram) typically first go to the
official repo of the project
and fork the repository so
that their changes can be sandboxed. The convention is to then set up their
local copy of the codebase with their own fork as the remote origin, and
connect to the official remote repo with the name upstream. So after forking
expyfun to his own GitHub account, user
rkmaddox would run:
$ git clone [email protected]:/rkmaddox/expyfun.git $ cd expyfun $ git remote add upstream git://github.com/LABSN/expyfun.git
Now this user has the standard origin/upstream configuration, as seen
below. Note the difference in the URIs between origin and upstream:
$ git remote -v origin [email protected]:/rkmaddox/expyfun.git (fetch) origin [email protected]:/rkmaddox/expyfun.git (push) upstream git://github.com/LABSN/expyfun.git (fetch) upstream git://github.com/LABSN/expyfun.git (push) $ git branch * master
URIs beginning with git:// are read-only connections, so rkmaddox can
pull down new changes from upstream, but won't be able to directly push his
local changes to upstream. Instead, he would have to push to his fork
(origin) first, and create a
pull request.
For example, to add some trivial file:
$ git branch fix_branch $ git branch fix_branch * master $ git checkout fix_branch $ git branch * fix_branch master $ touch foo.txt $ git add foo.txt $ git commit -am 'FIX: Add missing foo file' $ git push origin fix_branch
This creates a new branch called fix_branch on the local machine, checks out
that branch, adds a file, commits the change to the branch, and then pushes the
branch to a new remote branch (also called fix_branch) on the origin
repo (i.e., their fork of the official repo). rkmaddox could then navigate
to the website of the upstream repository
and they would find a nice Pull Request button available.
Maintainers would then typically comment on the pull request and ask for
some changes. For example, maybe the user forgot to also add the necessary
bar file. The user could then do:
$ git branch * fix_branch master $ touch bar.txt $ git add bar.txt $ git commit -am 'FIX: Add missing bar file' $ git push origin fix_branch
After this set of commands, the pull request (PR) is automatically updated to reflect this new addition. The cycle of commenting on and updating the continues until the Maintainers are satisfied with the changes. They will then merge the pull request to incorporate the proposed changes into the upstream GitHub repo.
Once their branch gets merged into the master branch of
the upstream repo, the developer can do the following to get
up to date on their local machine:
$ git checkout master $ git fetch upstream $ git pull upstream/master $ git branch -d fix_branch $ git branch * master
This synchronizes their local master branch with the master branch of
the upstream remote repo, and deletes their local fix_branch (which is
no longer needed, since its changes have been merged into the upstream master).
Occasionally as a developer you might run into a bug that you are uncertain when it was introduced. Instead of going through all the commits manually that have occurred since the code last worked, you can use the git bisect command.
First, you should write a simple script that reliably reproduces the bug, so you can easily test whether it is present in older versions of the code. Next, you will need to find a commit where the bug is not present (where it is working). Search through the git log from the time you know things were working, find a commit hash that you believe to be bug-free (shown here as abc1234) and confirm that this commit does indeed work by checking out that commit and running your test that reproduces bug:
$ git checkout abc1234 $ # now run your test
If you still see the bug, keep testing older and older commits until you find one that is bug-free. Next, note down a commit hash for a point in history that you know the bug exists. You can use the most recent commit hash, or an older hash if in the previous step you happened to discover one where the bug existed. Here the “known bad” commit hash is represented as zyx9876.
Now you are ready to use the git bisect command. To start, enter the following:
$ git bisect start $ git bisect good abc1234 $ git bisect bad zyx9876
This tells git bisect where the “known good” commit was and where the “known bad” commit is. It then looks at all the commits between these two, picks one half way between, and checks out that commit. You will see something like this:
Bisecting: 5 revisions left to test after this
Now run your test to see if this middle commit has the bug. If it does, then the bug was introduced before this middle commit; if the bug is not present then the problem was introduced after this middle commit. If the bug is gone, run:
$ git bisect good
and the middle commit will become the new “known good”. If the bug is present, run:
$ git bisect bad
and the middle commit will become the new “known bad”. Git will then select a new middle commit for you to test. Continue this process until there are no more commits to bisect:
Bisecting: 0 revisions left to test after this (roughly 1 step)
Run your test one last time, and let Git know:
$ git bisect good # or git bisect bad, if the bug is present in that commit
At this point you know which commit introduced the bug: it is either the last commit you checked (if it failed your test) or the second-to-last commit you checked (if the final one passed your test). Make a note of the commit hash in the issue tracker. Now you can return to the most recent commit (aka, HEAD) by letting git bisect know that you are done:
$ git bisect reset
You can also use git bisect reset to abort/stop the bisecting before you've finished, if need be.
This section was done in part from the following sources webchick and help articles on git.
Maintainers start out with a similar set up as Developers. However, they might
want to be able to push directly to the upstream repo as well as pushing to
their fork. Having a repo set up with git:// access instead of
[email protected] or https:// access will not allow pushing. So
starting from scratch, a maintainer Eric89GXL might fork the upstream repo
and then do:
$ git clone [email protected]:/Eric89GXL/expyfun.git $ cd expyfun $ git remote add upstream [email protected]:/LABSN/expyfun.git $ git remote add ross git://github.com/rkmaddox/expyfun.git
Now the maintainer's local repository has push/pull access to their own personal
development fork and the upstream repo, and has read-only access to
rkmaddox's fork:
$ git remote -v origin [email protected]:/Eric89GXL/expyfun.git (fetch) origin [email protected]:/Eric89GXL/expyfun.git (push) ross git://github.com/rkmaddox/expyfun.git (fetch) ross git://github.com/rkmaddox/expyfun.git (push) upstream [email protected]:/LABSN/expyfun.git (fetch) upstream [email protected]:/LABSN/expyfun.git (push)
Let's say rkmaddox has opened a PR on Github, and the maintainer wants
to test out the code. This can be done this way:
$ git fetch ross $ git checkout -b ross_branch ross/fix_branch
The first command allows the local repository to know about the changes (if
any) that have occurred on rkmaddox's fork of the project
(github.com/rkmaddox/expyfun.git).
In this case, a new branch named fix_branch has been added.
The second command is more complex. git checkout -b $NAME is a command
that first creates a branch named $NAME, then checks it out. The
additional argument ross/fix_branch tells git to make the
branch track changes from the remote branch fix_branch in the remote
repository known as ross, which you may recall points to
github.com/rkmaddox/expyfun.git. The full command can thus be interpreted
in human-readable form as "create and check out a branch named
ross_branch that tracks the changes in the branch
fix_branch from the remote repo named ross". The maintainer can
now inspect and test rkmaddox's code locally rather than just viewing the
changes on the GitHub pull request page.
Once the code is merged on GitHub, the maintainer can update their local copy in a similar way as the developer did earlier:
$ git checkout master $ git pull upstream/master
Still looking for more help? Check out all the help articles on Github and the help articles on git.