gitolite/doc/4-wildcard-repositories.mkd
2010-01-08 06:20:19 +05:30

7.3 KiB

repositories named with wildcards

IMPORTANT NOTE:

This branch contains features that are likely to be much more brittle than the "master" branch. Creating repositories based on wild cards, giving "ownership" to the specific user who created it, allowing him/her to hand out R and RW permissions to other users to collaborate, all these are possible. And any of these could have a bug in it.

"Brittle" also means some features in "master" may not work here. For example, you cannot specify gitconfig values for a wildcard repo; it only works for actual repos.

There may be other such missing features. Sometimes it's just not possible to make it work. Or it may be cumbersome enough that unless there are no workarounds I may not have the time to code it right away.


In this document:

  • wildcard repos
    • wildcard repos with creater name in them
    • wildcard repos without creater name in them
  • side-note: line-anchored regexes
    • contrast with refexes
  • handing out rights to wildcard-matached repos
  • reporting
  • other issues and discussion

This document is mostly "by example".


Wildcard repos

Which of these alternatives you choose depends on your needs, and the social aspects of your environment. The first one is a little more rigid, making it harder to make mistakes, and the second is more flexible and trusting.

Wildcard repos with creater name in them

Here's an example snippet:

@prof       =   u1
@TAs        =   u2 u3
@students   =   u4 u5 u6

repo    assignments/CREATER/a[0-9][0-9]
    C   =   @students
    RW+ =   CREATER
    RW  =   WRITERS @TAs
    R   =   READERS @prof

For now, ignore the special usernames READERS and WRITERS, and just create a new repo, as user "u4" (a student):

$ git clone git@server:assignments/u4/a12
Initialized empty Git repository in /home/sitaram/t/a12/.git/
Initialized empty Git repository in /home/gitolite/repositories/assignments/u4/a12.git/
warning: You appear to have cloned an empty repository.

Notice the two empty repo inits, and the order in which they occur ;-) Now make some changes and push, and after that, that specific repo (assignments/u4/a12) behaves as if the access control looked like this:

# effective config
repo    assignments/u4/a12
    RW+ =   u4
    RW  =   WRITERS @TAs
    R   =   READERS @prof

Wildcard repos without creater name in them

Here's how the same example would look if you did not want the CREATER's name to be part of the actual repo name.

repo    assignments/a[0-9][0-9]
    C   =   @students
    RW+ =   CREATER
    RW  =   WRITERS @TAs
    R   =   READERS @prof

We haven't changed anything except the repo name pattern. This means that the first student that creates, say, assignments/a12 becomes the owner. Mistakes (such as claiming a12 instead of a13) need to be rectified by an admin logging on to the back end, though it's not too difficult.

You could also repace the C line like this:

    C   =   @TAs

and have a TA create the repos in advance.

In either case, they could then use the setperms feature to specify which users are "READERS" and which are "WRITERS". See later for details.

Side-note: Line-anchored regexes

A regex like

repo assignments/S[0-9]+/A[0-9]+

would match assignments/S02/A37. It will not match assignments/S02/ABC, or assignments/S02/a37, obviously.

But you may be surprised to find that it does not match even assignments/S02/A37/B99. This is because internally, gitolite line-anchors the given regex; so that regex actually becomes ^assignments/S[0-9]+/A[0-9]+$ -- notice the line beginning and ending metacharacters.

Contrast with refexes

Just for interest, note that this is in contrast to the refexes for the normal "branch" permissions, as described in conf/example.conf and elsewhere. Those "refexes" are not anchored; a pattern like refs/heads/master actually matches foo/refs/heads/master01/bar as well, even if no one will actually push such a branch! You can anchor it if you really care, by using master$ instead of master, but anchoring is not the default for refexes.]

Handing out rights to wildcard-matached repos

In the examples above, we saw two special "user" names: READERS and WRITERS. The permissions they have are controlled by the config file, but who is part of this list is controlled by the person who created the repository.

The use case is that, although our toy example has only 3 students, in reality there will be a few dozen, but each assignment will be worked on only by a handful from among those. This allows the creater to take ad hoc sets of users from among the actual users in the system, and place them into one of two categories (whose permissions are, in this example, R and RW respectively). In theory you could do the same thing by creating lots of little "assignment-NN" groups in the config file but that may be a little too cumbersome for non-secret environments.

Create a small text file that contains the permissions you desire:

$ cat > myperms
R u5
RW u6
(hit ctrl-d here)

...and use the new "setperms" command to set permissions for your repo:

$ ssh git@server setperms assignments/u4/a12 < myperms
New perms are:
R u5
RW u6

'setperms' will helpfully print what the new permissions are but you can also use 'getperms' to check:

$ ssh git@server getperms assignments/u4/a12
R u5
RW u6

The following points are important:

  • note the syntax of the commands; it's not a "git" command,and there's no : like in a repo URL. The first space-separated word is R or RW, and the rest are simple usernames.

  • whoever you specify as "R" will match the special user READERS. "RW" will match WRITERS.

Reporting

Remember the cool stuff you see when you just do ssh git@server (grep for "myrights" in doc/3-faq-tips-etc.mkd if you forgot, or go here).

This still works, except the format is a little more compressed to accommodate a new column (at the start) for "C" permissions, which indicate that you are allowed to create repos matching that pattern.

Other issues and discussion

  • what if the repo name being pushed matches more than one pattern?

    I think it would be very hard to reason about access if we were to do something like combine all the access rights in all the matching patterns. No matter how you do it, and how carefully you document it, there'll be someone who is surprised by the result.

    And in security, that's a Bad Thing.

    So we don't combine permissions. At runtime, we die if we find more than one match. Let 'em go holler at the admin for creating multiple matching repo patterns :-)

    This can make some repos inaccessible if the patterns changed after they were created. The administrator should be careful not to do this. Most of the time, it won't be difficult; the fixed prefix will usually be different anyway so there won't be overlaps.


Enjoy, and please use with care. This is pretty powerful stuff. As they say: if you break it, you get to keep both pieces :)