setting up and using templates


This feature was added late September 2017, after 3.6.7. In terms of versioning, it will be part of 3.6.8, or you could just grab the latest from github.

overview

This feature describes a new method of specifying gitolite rules.

A "template" is a set of gitolite access rules with a name. A gitolite "expert" can setup a suitable set of templates, and then actual repos can use one or more of these templates, according to their need.

A simple example will illustrate. It's split into two parts.

First, the template definitions, where the rules comprising each template are defined:

# (obligatory warning: the order of the following lines matters!)

repo @is_public
    R                           =   @all

repo @is_suspended
    -                           =   @all

repo @has_personal_refs
    RW+     dev/USER/           =   teamleads team
    -       dev/                =   @all
    RW+     refs/tags/dev/USER/ =   teamleads team
    -       refs/tags/dev/      =   @all

repo @has_releases
    RW      refs/tags/v[0-9]    =   release_managers
    -       refs/tags/v[0-9]    =   @all

repo @base
    RW+                         =   teamleads
    RW                          =   team
    R                           =   READERS

Next, you have the template data section (i.e., the lines between the begin and end lines you see in the example below). This is where you declare actual repos (foo and bar below), giving each a set of one or more templates it will use, and map users to any roles that the template may require.

=begin template-data

repo foo = base is_public has_releases
    teamleads           =   sitaram
    team                =   dilbert wally
    READERS             =   ashok
    release_managers    =   alice

repo bar = base has_personal_refs
    teamleads           =   han
    team                =   luke chewie

=end

Notice that foo and bar use different sets of templates: foo is a public-readable repo that controls who can push versioned tags (releases), while bar is a basic repo which supports personal branches.

advantages

There are a few advantages with this approach:

  1. Maintaining access rules is much simpler. Just choose an appropriate set of "template names", assign people to roles, and you're done.

    There's no need to understand the intricacies of gitolite's ruleset. (The person who wrote the templates needs to, but not the person who is maintaining dozens of repos by merely using those templates).

    Heck you can probably roll a nice GUI around this. Finally!

  2. Reduces boilerplate. A good example is the "personal branches" one above -- why have each of those 4 lines in every repo if you can instead refer to the feature by name somehow

    A conf using this is often smaller, and definitely cleaner.

  3. Reduces possible errors. This should follow from the previous point. It is easy to make mistakes when changing something due to some new requirement. Did you remember to put in the two "deny" rules in the personal branches example above?

    What if management decides to suspend pushes to one particular repo, and you best choice was to add a catch-all "deny everyone" rule. Did you remember to put it at the top? (If the templates are written as above (including the order in which you see them), all you have to do is add the word is_suspended somewhere in the list of templates that apply to repo bar.)

  4. Makes gitolite compile much faster, especially if you have thousands (or tens of thousands) of repos.

how does it work?

a repo and its users

The wildcard repos feature already has a way to dissociate the actual user names from the rule set in gitolite.conf. For example, you can say

repo foo/..*
    C       =   @managers
    RW+     =   WRITERS
    R       =   READERS
    ...(etc.)...

This lets any "manager" create a repo whose name matches the pattern, then assign arbitrary users to WRITERS and other roles using the perms command. These role assignments are stored in a simple text file within the repository's bare directory (i.e., ~/repositories/$REPO.git), so they are specific to that repo, not common to all the repos matching that pattern (as they would be if you listed the users in gitolite.conf directly).

In other words, we've taken the actual users (say alice, bob, etc) out of the gitolite.conf file, and thus any changes to the users/roles no longer need to involve gitolite.conf.

a repo and its rules

In a "duh! Why didn't I think of this till now" moment, I realised I can do the same for the rules that apply to a repo -- take that association out of gitolite.conf. That is what the repo foo = [...list of templates...] lines are doing.

This list of templates is also stored in a plain text file just like the one that contains the user/role mappings, and in the same directory.

usage and syntax

First, you have to add all the new "roles" to the ROLES hash in ~/.gitolite.rc. If you edit that file, you'll see two pre-created roles READERS and WRITERS. Using the same syntax (including the trailing comma), add any other roles you would like to use. In our example up at the top, the role names are team, teamleads, and release_managers.

Rolenames must start with a letter, and be made up of only alphanumeric characters and the underscore -- basically the same rules as a shell variable.

Next, you define the templates, in the right order. This is the only order that matters (not the order in which the templates are used in any particular repo in the template-data section).

Thus, this is also the part that requires gitolite rules expertise, but it's hopefully a one-time or once-in-a-while thing. (Or you can ask on the mailing list!)

Finally, you define actual repos in gitolite.conf as shown in the example above (including the =begin template-data and =end lines). For each repo, you specify what templates it will use, and then you map actual users to the role names from those templates.

A few additional points:

  1. Not all role names need to be mapped to users (for example, we did not assign any READERS to repo bar, even though the base template specifies that role).

  2. Within the gitolite.conf file, the placement of the template-data section does not matter. (It's not even parsed by the conf compiler, which completely skips it. It's processed by a new program that is run internally, and directly manipulates the gl-repo-groups and gl-perms files).

  3. You can even have multiple template-data sections, with normal gitolite.conf rules, group definitions, config and option lines, etc., in between. (That's why there's a begin and an end!)

    If you use include files, I strongly suggest -- in the interest of sanity -- that you do not let a template-data section cross over a file boundary (i.e., define the begin in one file, and the end in another). It will work, if you understand what order the files are picked up, but I'd still avoid such tricks if I were you!

  4. If you want to insert some rules for a repo that is defined in a template-data section, you need to be careful where you place it.

    Rules defined by templates are deemed to occur exactly where the template definition is. So, speaking of repo foo, pretend that the line repo @is_public was replaced by repo foo, and similarly for repo @has_releases and repo @base.

    While we're on the subject, you can also pretend the role names on the right hand side of the rules are replaced by the actual user names you supplied in the template-data section.

    In the example above, say you wanted to insert a new rule for repo foo, which says that no one can rewind master, not even teamleads.

    Clearly, the rules you need are:

    repo foo
        RW  master      =   sitaram
        # notice we had to expand 'teamleads' from foo's definition in the
        # template-data section
        -   master      =   @all
    

    But where do you place them? The answer is, at least before the base template is defined. Otherwise, the RW+ in the base template will kick in, and this restriction will fail to take effect.

    Having said that, I would rather add a new template to deal with this, placing it just before repo @base):

    repo @limits_master
        RW  master      =   teamleads
        -   master      =   @all
    

    and then add limits_master to the list of templates that foo uses.

    This has the advantages of being able to reuse that logic for other repos, but even more important, you're avoiding repeating the actual teamleads name(s) in more than one place! (Potentially a huge future inconsistency if someone forgot to update both places when the teamleads change!)

  5. You can also do multiple repos in one shot, as well as repo groups:

    # before the '=begin' line
    @repogroup1 = r1 r2 r3
    ...
    ...
    ...

    =begin template-data repo foo bar @repogroup1 = base is_public has_releases ... ... ... =end

bypassing gitolite.conf for huge sites

Some sites have all their access control information in a web-based system, and generate gitolite.conf as needed. If they have tens of thousands of repos, this "generated" gitolite.conf becomes humongous, and slows down compiles. Worse, the more repos you have, the more churn you have in terms of changes to users accesses, so you do more compiles per hour than a smaller site, which only makes things worse!

With this feature, you can bypass gitolite.conf and directly create/update those text files to change the users and rule-sets for a given repo. It doesn't even have to touch gitolite or gitolite.conf (assuming the templates and roles are already defined in gitolite.conf and ~/.gitolite.rc of course!)

generating the text files externally

The actual text files involved are very simple. Remember these files go into ~/repositories/$REPO.git (or more accurately, $(gitolite query-rc GL_REPO_BASE)/$REPO.git).

For the example above, here's the file gl-repo-groups in repo foo:

$ cat ~/repositories/foo.git/gl-repo-groups
base is_public has_releases

As you can see, this text is just what is after the = sign in the repo line in the template data section of gitolite.conf.

and the file gl-perms is:

$ cat ~/repositories/foo.git/gl-perms
teamleads           =   sitaram
team                =   dilbert wally
READERS             =   ashok
release_managers    =   alice

Again, this text is exactly the same as in the gitolite.conf!

creating new repos

Gitolite has no mechanism to create repos out of thin air, so if you don't want to go via gitolite.conf, one way to do this is to add the following lines to the conf file (one-time):

repo [a-zA-Z0-9].*
    C   =   gitolite-admin

and then, at the server, run this:

GL_USER=gitolite-admin gitolite create foo/bar

That creates the repo, and you can now populate its gl-perms and gl-repo-groups` files.

thanks to...

...pingou on irc, and the Fedora project, for having 42,000 repos in a conf file over 560,000 lines long. Which made me think about this real hard for days, including two false starts (one of which I published and have just now reverted, and one which was so kludgey I refuse to acknowledge it exists -- thank God I did not publish that!)

miscellanea

This feature is not the same as wild repos; repos here are created by the gitolite admin or a server-side backend, not by a gitolite user. (However, this feature piggy-backs on a lot of the code for wild repos, adding just a wee bit -- the "duh" comment earlier in this document -- to complete it).