Git::Repository - Perl interface to Git repositories


version 1.314


    use Git::Repository;

    # start from an existing repository
    $r = Git::Repository->new( git_dir => $gitdir );

    # start from an existing working copy
    $r = Git::Repository->new( work_tree => $dir );

    # start from a repository reachable from the current directory
    $r = Git::Repository->new();

    # or init our own repository first
    Git::Repository->run( init => $dir, ... );
    $r = Git::Repository->new( work_tree => $dir );

    # or clone from a URL first
    Git::Repository->run( clone => $url, $dir, ... );
    $r = Git::Repository->new( work_tree => $dir );

    # provide an option hash for Git::Repository::Command
    # (see Git::Repository::Command for all available options)
    $r = Git::Repository->new( ..., \%options );

    # run commands
    # - get the full output (no errput) passing options for this command only
    $output = $r->run( @cmd, \%options );

    # - get the full output as a list of lines (no errput), with options
    @output = $r->run( @cmd, \%options );

    # - process the output with callbacks
    $output = $r->run( @cmd, sub {...} );
    @output = $r->run( @cmd, sub {...} );

    # - obtain a Git::Repository::Command object
    #   (see Git::Repository::Command for details)
    $cmd = $r->command( @cmd, \%options );

    # obtain version information
    my $version = $r->version();

    # compare current git version
    if ( $r->version_gt('1.6.5') ) {


Git::Repository is a Perl interface to Git, for scripted interactions with repositories. It's a low-level interface that allows calling any Git command, whether porcelain or plumbing, including bidirectional commands such as git commit-tree.

A Git::Repository object simply provides context to the git commands being run. It is possible to call the command() and run() methods against the class itself, and the context (typically current working directory) will be obtained from the options and environment.

As a low-level interface, it provides no sugar for particular Git commands. Specifically, it will not prepare environment variables that individual Git commands may need or use.

However, the GIT_DIR and GIT_WORK_TREE environment variables are special: if the command is run in the context of a Git::Repository object, they will be overridden by the object's git_dir and work_tree attributes, respectively. It is however still possible to override them if necessary, using the env option.

Git::Repository requires at least Git 1.5.0, and is expected to support any later version.

See Git::Repository::Tutorial for more code examples.



    Git::Repository->new( %args, $options );

Create a new Git::Repository object, based on an existing Git repository.

Parameters are:

git_dir => $gitdir

The location of the git repository (.git directory or equivalent).

For backward compatibility with versions 1.06 and before, repository is accepted in place of git_dir (but the newer name takes precedence).

work_tree => $dir

The location of the git working copy (for a non-bare repository).

If work_tree actually points to a subdirectory of the work tree, Git::Repository will automatically recompute the proper value.

For backward compatibility with versions 1.06 and before, working_copy is accepted in place of work_tree (but the newer name takes precedence).

If none of the parameter is given, Git::Repository will find the appropriate repository just like Git itself does. Otherwise, one of the parameters is usually enough, as Git::Repository can work out where the other directory (if any) is.

new() also accepts a reference to an option hash which will be used as the default by Git::Repository::Command when working with the corresponding Git::Repository instance.

So this:

    my $r = Git::Repository->new(
        # parameters
        work_tree => $dir,
        # options
        {   git => '/path/to/some/other/git',
            env => {
                GIT_COMMITTER_EMAIL => '',
                GIT_COMMITTER_NAME  => 'Philippe Bruhat (BooK)',

is equivalent to explicitly passing the option hash to each run() or command() call. The documentation for Git::Repository::Command lists all available options.

Note that Git::Repository and Git::Repository::Command take great care in finding the option hash wherever it may be in @_, and to merge multiple option hashes if more than one is provided.

It probably makes no sense to set the input option in new(), but Git::Repository won't stop you. Note that on some systems, some git commands may close standard input on startup, which will cause a SIGPIPE. Git::Repository::Command will raise an exception.

To create a Git repository and obtain a Git::Repository object pointing to it, simply do it in two steps:

    # run a clone or init command without an instance,
    # using options like cwd
    Git::Repository->run( ... );
    # obtain a Git::Repository instance
    # on the resulting repository
    $r = Git::Repository->new( ... );


Git::Repository supports the following methods:


    Git::Repository->command( @cmd );
    $r->command( @cmd );

Runs the git sub-command and options, and returns a Git::Repository::Command object pointing to the sub-process running the command.

As described in the Git::Repository::Command documentation, @cmd may also contain a hashref containing options for the command.


    Git::Repository->run( @cmd );
    $r->run( @cmd );

Runs the command and returns the output as a string in scalar context, or as a list of lines in list context. Also accepts a hashref of options.

Lines are automatically chomped.

In addition to the options hashref supported by Git::Repository::Command, the parameter list can also contain code references, that will be applied successively to each line of output. The line being processed is in $_, but the coderef must still return the result string (like map).

If the git command printed anything on stderr, it will be printed as warnings. For convenience, if the git sub-process exited with status 128 (fatal error), or 129 (usage message), run() will die(). The exit status values for which run() dies can be modified using the fatal option (see Git::Repository::Command for details).

The exit status of the command that was just run is accessible as usual using $? >> 8. See perlvar for details about $?.


Returns the repository path.


Returns the working copy path. Used as current working directory by Git::Repository::Command.


Return the option hash that was passed to Git::Repository->new().


Return the version of git, as given by git --version.

Version-comparison "operators"

Git evolves very fast, and new features are constantly added. To facilitate the creation of programs that can properly handle the wide variety of Git versions seen in the wild, a number of version comparison "operators" are available.

They are named version_op where op is the equivalent of the Perl operators lt, gt, le, ge, eq, ne. They return a boolean value, obtained by comparing the version of the git binary and the version string passed as parameter.

The methods are:

version_lt( $version )
version_gt( $version )
version_le( $version )
version_ge( $version )
version_eq( $version )
version_ne( $version )

All those methods also accept an option hash, just like the others.

Note that in the git.git repository, v1.0.1 and v1.0.2 are lightweight tags that points respectively to v1.0.0a and v1.0.0b. As of Git::Repository 1.314, the comparison code internally converts v1.0.0a and v1.0.0b to their numerical equivalent before performing the comparison.

Prior to 1.4.0-rc1 (June 2006), compiling a development version of git would lead git --version to output 1.x-GIT (with x in 0 .. 3), which would make comparing versions that are very close a futile exercise.

Other issues exist when comparing development version numbers with one another. For example, is greater than both and, and 1.7.1 is less than both. Obviously, will compare as greater than (asciibetically), but in fact these two version numbers cannot be compared, as they are two siblings children of the commit tagged v1.7.1).

If one were to compute the set of all possible version numbers (as returned by git --version) for all git versions that can be compiled from each commit in the git.git repository, the result would not be a totally ordered set. Big deal.

Also, don't be too precise when requiring the minimum version of Git that supported a given feature. The precise commit in git.git at which a given feature was added doesn't mean as much as the release branch in which that commit was merged.


Git::Repository intentionally has only few methods. The idea is to provide a lightweight wrapper around git, to be used to create interesting tools based on Git.

However, people will want to add extra functionality to Git::Repository, the obvious example being a log() method that returns simple objects with useful attributes.

Taking the hypothetical Git::Repository::Plugin::Hello module which source code is listed in the previous reference, the methods it provides would be loaded and used as follows:

    use Git::Repository qw( Hello );

    my $r = Git::Repository->new();
    print $r->hello();
    print $r->hello_gitdir();

It's possible to load only a selection of methods from the plugin:

    use Git::Repository [ Hello => 'hello' ];

    my $r = Git::Repository->new();
    print $r->hello();

    # dies: Can't locate object method "hello_gitdir"
    print $r->hello_gitdir();

If your plugin lives in another namespace than Git::Repository::Plugin::, just prefix the fully qualified class name with a +. For example:

    use Git::Repository qw( +MyGit::Hello );

See Git::Repository::Plugin about how to create a new plugin.


Thanks to Todd Rinaldo, who wanted to add more methods to Git::Repository, which made me look for a solution that would preserve the minimalism of Git::Repository. The ::Plugin interface is what I came up with.


(This section was written in June 2010. The other Git wrappers have probably evolved since that time.)

A number of Perl git wrappers already exist. Why create a new one?

I have a lot of ideas of nice things to do with Git as a tool to manipulate blobs, trees, and tags, that may or may not represent revision history of a project. A lot of those commands can output huge amounts of data, which I need to be able to process in chunks. Some of these commands also expect to receive input.

What follows is a short list of "missing features" that I was looking for when I looked at the existing Git wrappers on CPAN. They are the "rational" reason for writing my own (the real reason being of course "I thought it would be fun, and I enjoyed doing it").

Even though it works well for me and others, Git::Repository has its own shortcomings: it is a low-level interface to Git commands, anything complex requires you to deal with input/output handles, it provides no high-level interface to generate actual Git commands or process the output of commands (but have a look at the plugins), etc. One the following modules may therefore be better suited for your needs, depending on what you're trying to achieve. was not on CPAN in 2010. It is packaged with Git, and installed with the system Perl libraries. Not being on CPAN made it harder to install in any Perl. It made it harder for a CPAN library to depend on it.

It doesn't allow calling git init or git clone.

The command_bidi_pipe function especially has problems:

The Git module from git.git was packaged as a CPAN distribution by MSOUTH in June 2013.


Git::Class depends on Moose, which seems an unnecessary dependency for a simple wrapper around Git. The startup penalty could become significant for command-line tools.

Although it supports git init and git clone (and has methods to call any Git command), it is mostly aimed at porcelain commands, and provides no way to control bidirectional commands (such as git commit-tree).


Git::Wrapper doesn't support streams or bidirectional commands.


(This description was added for completeness in May 2013.)

Git::Sub appeared in 2013, as a set of Git-specific System::Sub functions. It provide a nice set of git:: functions, and has some limitations (due to the way System::Sub itself works) which don't impact most Git commands.

Git::Sub doesn't support working with streams.


(This description was added for completeness in September 2014, upon request of the author of Git::Raw.)

Git::Raw provides bindings to libgit2, a pure C implementation of the Git core methods. Most of the functions provided by libgit2 are available. If you have complex workflows, or even if speed is of the essence, this may be a more attractive solution than shelling out to git.


Since version 1.17, Git::Repository delegates the actual command execution to System::Command, which has better support for Win32 since version 1.100.

Please report any bugs or feature requests to bug-git-repository at, or through the web interface at I will be notified, and then you'll automatically be notified of progress on your bug as I make changes.


You can find documentation for this module with the perldoc command.

    perldoc Git::Repository

You can also look for information at:


Please report any bugs or feature requests on the bugtracker website or by email to

When submitting a bug or request, please include a test-file or a patch to an existing test-file that illustrates the bug or desired feature.


Philippe Bruhat (BooK) <>


Copyright 2010-2014 Philippe Bruhat (BooK), all rights reserved.


This program is free software; you can redistribute it and/or modify it under the same terms as Perl itself.