The Perl Toolchain Summit needs more sponsors. If your company depends on Perl, please support this very important event.

USER TUTORIAL

As a Git user you may be interested in enabling some hooks for your local Git repositories. In particular, you may be interested in guaranteeing that the same policies that are being enforced by the remote repositories you push to are enforced earlier when you commit locally, so that you can avoid an onerous round trip to the common repository.

User Driver Script

Git::Hooks only need a single script to drive all hooks implemented by yourself or by the plugins you enable. If you do not need to create your own hooks, but want to use just the ones that come with Git::Hooks plugins, you can use a shared script like this for all your local repositories:

    #!/usr/bin/env perl
    use Git::Hooks;
    run_hook($0, @ARGV);

As a user, I save this script as ~/bin/githooks.pl under my $HOME directory. Just do not forget to make it executable!

If you invoke the driver script directly from the inside of a Git repository it should do nothing but exit normally:

    $ cd /my/git/repo
    $ ~/bin/githooks.pl
    $ echo $?

If you invoke it from the outside though, it should die:

    $ cd ..
    $ ~/bin/githooks.pl
    fatal: Not a git repository: . at /usr/share/perl5/Git.pm line 210.

Now you must create symbolic links under the .git/hooks directory of your repositories pointing to the common script. So, for example, if you want to enable some pre-commit and some commit-msg hooks, you would do this:

    $ cd .../.git/hooks
    $ ln -s ~/bin/githooks.pl pre-commit
    $ ln -s ~/bin/githooks.pl commit-msg
    $ ln -s ~/bin/githooks.pl pre-rebase

However, doing it manually for every repository is cumbersome and prone to mistakes and neglect. Fortunately, there is a better way. In order to make it easy to setup your hooks, it's useful to create a repository template for Git to use when you perform a git init or a git clone.

In Ubuntu, Git's standard repository template resides in /usr/share/git-core/templates. If you can't find it there, read the TEMPLATE DIRECTORY section of the git help init manual to see where is your Git's default template directory.

You may customize one for you like this:

    $ cp -a /usr/share/git-core/templates ~/.git-templates
    $ cd ~/.git-templates/hooks
    $ rm *
    $ for i in commit-msg post-commit pre-commit pre-rebase
    > do ln -s ~/bin/githooks.pl $i
    > done

These commands copy the default template directory to ~/.git-template (you may choose another directory), removes all sample hooks and creates symbolic links to the Git::Hooks driver script which we created above for four hooks: commit-msg, post-commit, pre-commit, and pre-rebase. These are all the hooks I'm interested in locally. If you're setting this up for a Git server you'll want to create links for other hooks, such as pre-receive or update.

You must tell Git to use your repository template instead of its default. The best way to do it is to configure it globally like this:

    $ git config --global init.templatedir ~/.git-templates

Now, whenever you git init or git clone a new repository, it will automatically be configured to use Git::Hooks.

User Configuration

By default Git::Hooks does nothing. At the very least, it must be configured to enable some plugins and configure them to your taste. You should read the plugins's documentation to understand them and decide which ones you would like to enable globally and which ones you would like to enable locally for particular repositories.

Here I show my personal preferences. You are encouraged to make your own variations.

This is what I have in my global Git configuration (~/.gitconfig):

    [githooks]
            plugin = CheckLog
            plugin = CheckRewrite
            abort-commit = 0
    [githooks "checklog"]
            title-max-width = 62
    [githooks "checkjira"]
            jiraurl  = https://jira.cpqd.com.br
            jirauser = gustavo
            jirapass = a-very-large-and-difficult-to-crack-password
            matchlog = (?s)^\\[([^]]+)\\]

The only plugins I want enabled for every repository are CheckLog and CheckRewrite. The latter is simple, as it doesn't require any configuration whatsoever. With it I feel more confident to perform git commit --amend and git rebase commands knowing that I'm going to be notified in case I'm doing anything dangerous.

The CheckLog is also useful to guarantee that I'm not deviating from the common Git policies regarding the commit messages. The only thing I change from the defaults is the title-max-width, because I think 50 characters is very constraining.

I disable the githooks.abort-commit option so that pre-commit and commit-msg hooks don't abort the commit in case of errors. That's because I find it easier to amend the commit than to remember to recover my carefully crafted commit message from the .git/COMMIT_EDITMSG file afterwards.

The section githooks "checkjira" contains some global configuration for the CheckJira plugin, which I enable only for some repositories. Since the CheckJira plugin has to connect to our JIRA server, it needs the server URL and some credentials to authenticate. The matchlog regex makes JIRA issue keys be looked for only inside a pair of brackets at the beginning of the commit messages's title line.

I enable other plugins for specific repositories, since they depend on the context in which they are developed.

At CPqD we use JIRA and Gerrit. So, for my work-related repositories I have this in their .git/config:

    [githooks]
            plugin = CheckJira
            plugin = GerritChangeId
    [githooks "checkjira"]
            project = CDS

GerritChangeId doesn't require any configuration. It simply inserts a Change-Id line in the messages of all commits. These are required by Gerrit.

I use CheckJira to remind me to cite a JIRA issue in every commit message. The project value makes it accept only issues of the CDS JIRA project for this particular repository.

Disabling plugins temporarily

If you prefer the default behaviour of having your pre-commit and commit-msg abort on errors, it's sometimes useful to disable a plugin temporarily in order to do a commit that otherwise would be rejected. For instance, if you enable CheckLog's spelling checks and it rejects a commit because you used a cute-but-not-quite-right word in its message you can disable it for the duration of the commit by defining the environment variable CheckLog as 0 like this:

    CheckLog=0 git commit

You can disable any plugin in the same manner. Just define as zero (0) an environment variable homonymous to the plugin (you can use the plugin module full name or just its last component, as in the example above) for the duration of the commit and the plugin will be disabled.

ADMIN TUTORIAL

As the administrator of a Git server you may be interested in enabling some hooks for your Git repositories to enforce project policies through source code verification or access rights.

Server Driver Script

The same driver script described above for user repositories can be used for server repositories:

    #!/usr/bin/env perl
    use Git::Hooks;
    run_hook($0, @ARGV);

As a Git administrator, I save it as /usr/local/bin/githooks.pl in my Git server. You may save it elsewhere in the machine your hooks will run. Just do not forget to make it executable!

As a Git administrator, you would be interested in the back-end hooks. So, you should create some symbolic links under the .git/hooks directories of your repositories pointing to the drive script:

    $ cd .../.git/hooks
    $ ln -s /usr/local/bin/githooks.pl pre-receive
    $ ln -s /usr/local/bin/githooks.pl update
    $ ln -s /usr/local/bin/githooks.pl ref-update

Also, read the section about "Automating the creation of links" to know to have such links automatically created for you when you initialize or clone a repository.

Server Configuration

In your Git server you should insert global configuration in the ~/.gitconfig file at the HOME of the user running Git. This is an example using some of the available plugins:

    [githooks]
            plugin = CheckJira
            plugin = CheckLog
            admin = gustavo
    [githooks "checkjira"]
            jiraurl  = https://jira.cpqd.com.br
            jirauser = gustavo
            jirapass = a-very-large-and-difficult-to-crack-password
            matchlog = (?s)^\\[([^]]+)\\]
    [githooks "checklog"]
            title-max-width = 62

In the server the CheckJira and CheckLog plugins are enabled for every repository. The <githooks.checkjira> section specifies the URL and credentials of the JIRA server as well as where in the commit message the JIRA references are to be looked for.

The githooks.checklog section specifies a nonstandard value for the title-max-width option.

As the administrator, I've configured myself (githooks.admin = gustavo) to be exempt from any checks so that I can brag about my superpowers to my fellow users. Seriously, though, sometimes it's necessary to be able to bypass some checks and this is a way to allow some user to do it.

In particular repositories you can make local configurations to complement or supersede the global configuration. This is an example .git/config file:

    [githooks]
            disable = CheckJira
            plugin = CheckAcls
            groups = integrators = tiago juliana
    [githooks "checkacls"]
            acl = @integrators CRUD ^refs/(heads|tags)/
            acl = ^.           CRUD ^refs/heads/user/{USER}/'
            acl = ^.           U    ^refs/heads

In this repository the CheckJira plugin is disabled, even though it is enabled globally.

The CheckAcls plugin is enabled and configured in the githooks.checkacls section with three ACLs.

The first ACL allows the two users belonging to the integrators group to create, rewrite, update, and delete any branch or tag.

The second ACL allows any user to create, rewrite, update, and delete any branch with a name beginning with user/USER, where USER is the username with which she authenticated herself with Git. This is useful to allow developers to backup their own local branches in the server while they aren't good enough to be shared.

The third ACL allows any user to update any branch, which means, to push to any branch and have it be fast-forwarded.

Distributed configuration

By default you only get a single global and one local configuration file for each repository in the server. Sometimes it's useful to factor out some configuration in specific files. If you have, say, three development teams holding their repositories in a single server but each one of them wants different CheckAcls configuration you may separate these configurations in three files and include one of them in each repository using Git's include section. For example, team A's repositories could have this in their .git/config files:

    [include]
            path = /usr/local/etc/githooks/teamA.acls

Using include files you can manage complex configurations more easily.

GERRIT TUTORIAL

Gerrit is a Git server but since it uses JGit instead of standard Git, it doesn't support the standard Git hooks. It supports its own hooks instead.

Git::Hooks supports just three of the many Gerrit hooks so far: ref-update, patchset-created, and draft-published. The first one is much like the standard hooks pre-receive and update in that it can reject pushes when the commits being pushed don't comply. However, since Gerrit's revision process takes place before the commits are integrated, it's more useful to enable just the other two.

First, you have to create the same driver script as described for the server. There is a catch though. Gerrit undefines the HOME environment variable when it runs, so that when we invoke the git command inside one of our hooks it won't be able to read the global configuration file ~/.gitconfig. So, you should define HOME in the script like this (obviously, using the correct value in the assignment):

    #!/usr/bin/env perl
    BEGIN { $ENV{HOME} = '/home/gerrit' }
    use Git::Hooks;
    run_hook($0, @ARGV);

Then we must create the symlinks from the hook names to the driver script. However, in Gerrit there's a single hooks directory per server, instead of one per repository. Normally, when you install Gerrit, the hooks directory isn't created. It should be created below the Gerrit's site directory. Create it and the two symlinks like so:

    $ cd .../gerrit-site
    $ mkdir hooks
    $ cd hooks
    $ ln -s /usr/local/bin/githooks.pl patchset-created
    $ ln -s /usr/local/bin/githooks.pl draft-published

The patchset-created hook is invoked when you push a patchset to Gerrit for revision, but Git::Hooks only enable it for non-draft patchsets, because draft patchsets can only be reviewed by their onwners and invited reviewers. The draft-published hook is invoked when you publish a draft-patchset. Both hooks run asynchronously so that they can't reject the push. Instead, they review the patchset as a normal reviewer would, casting a positive or negative vote, depending on the result of the checks made by the enabled plugins.

All the (standard) Git::Hooks plugins that attach to the pre-receive and update hooks also attach themselves to both the patchset-created and the draft-published hooks, so that you can use the same configuration we did above.

You have to do a little extra configuration in the githooks.gerrit section:

    [githooks "gerrit"]
            url  = https://gerrit.cpqd.com.br
            username = gerrit
            password = a-very-large-and-difficult-to-crack-password
            review-label = Verification
            vote-ok  = +1
            vote-nok = -1

The three options url, username, and password tells where to connect to Gerrit and with which user's credentials. This is the user that will appear to be making comments and reviewing the patchsets.

Then you have to tell Git::Hooks how it should vote. First, you tell which label it should vote in. By default it votes in the Code-Review label. In the example above I've changed it to vote in the Verification label instead.

The last two options, vote-ok and vote-nok, tells which number it should use to vote positively and negatively. +1 and -1 are the defaults, but you may tell it to use different numbers if you like.

Gerrit has a notion of a hierarchy of repositories (called 'projects' in Gerrit). Gerrit's own configuration uses this hierarchy so that child repositories inherit their ancestor's configuration. Git's own configuration mechanism has no such notion, but you can fake it using the same include mechanism discussed above. But you have to do it manually, though.