Edgewall Software

Multiple Repository Support for 0.12

This page details the development process that added support for accessing multiple repositories inside one Trac project.

The main part of the work has been completed and the sandbox branch has been merged into trunk.

The corresponding documentation can be found in 0.12/TracRepositoryAdmin.

This page still discusses a number of items that can be implemented in future Trac versions.

For now, there is support for the Mercurial (hg) and Subversion (both svn and direct-svnfs types of repositories).

See:

Checkpoints:

  • r6465: initial "release" of the branch, worksforme
  • r6469: a few bug fixes
  • r7590: that version was quite good
  • r7815: that version was also quite good ;-)
  • r7961: initial support for cached repositories
  • r8178: current version - should be "stable" (i.e. you can report bugs against that revision if you find some). Note that there's now support for cached repositories, and as this involved a schema change, an environment upgrade is needed.
  • r9084: real close to the end of the branch ;-)
  • r9125: branch landed on trunk; further stabilization work is now being done there.

The approach is quite simple: beyond repository connectors, there's now repository providers (IRepositoryProvider components) which are responsible for making available a list of named repositories. That name can be arbitrary (in particular, it may contain '/' to denote a pseudo-hierarchy) and will be used as a prefix for any repository path. Actually, a more complete description of the approach can be found in this mail: ​One trac env for X repositories ?.

If the only repository known is the default one (i.e. the one defined in TracIni, section [trac] repository_...), then Trac behaves exactly as before. But IRepositoryProvider components can make other repositories available. In particular, the default RepositoryManager itself provides a list of repositories listed in a new [repositories] section in TracIni. See r6447 for more details. Since r7969, trac-admin gained some commands to manage list of repositories and their aliases, storing them in a dedicated table in the Trac database.

Finally, repositories of type svn are now supported. This means there will be a cache created for each of those repositories. See MultipleRepositorySupport/Cache for details.

Related Tickets

closed status tickets:

#6654
Changeset view is slow when using sub repository path
#7116
fine-grained permissions are not applied to directories when browsing sources
#7706
ValueError: too many values to unpack
#7717
Multi-repo branch 'trac-admin resync' msg on one svn
#7743
Add/document commit-hook
#8203
Internal error when using multirepos, in search
#8369
ValueError: too many values to unpack
#8758
Timeline filtering in Multirepos
#8808
KeyError: <Option [account-manager] "account_changes_notify_addresses">
#10144
[Patch] Make it easier to switch off commit messages in timeline
#1781
Link from 'Browser Source' to Subversion Repository URL
#2086
better support for multiple repositories
#2482
document usage of post-revprop-change script
#3368
Repository subset support in commit hooks
#5640
SVN checkin comments float up past svn:authz barrier.
#5993
Confusing error message when repository_dir has changed.
#6016
trac-admin resync doesn't update revision message after editing a svn revision log
#6445
post-commit script example in trac-post-commit-hook is missing environment variable "TRAC_ENV"
#6826
Excessive calls of filter.pre_process_request for one request
#7723
Make repository cache work on multirepos branch
#7822
Manage repositories in DB instead of trac.ini
#8291
Create an admin panel for managing repositories
#8465
Subversion authorization for Browser Source
#8525
Multirepos Databse Upgrade from 0.11 Fails w/ PostgreSQL
#8536
Don't display Default Repository on main Browse page
#8572
AttributeError: 'sqlite3.Connection' object has no attribute 'get_db_cnx'
#8690
Missing changeset link
#8731
experiment with surrogate keys for the repository table
#9038
Latest multirepos revision chokes on revision log links
#9090
Browser "Download as zip" is not Multi-repos aware
#9112
Strange behavior with empty svn repositories
#5189
Feature request: add --no-notification option to trac-post-commit-hook
#6041
trac-post-commit-hook doen't work (nor error message)
#6370
Contrib post-commit for Win has issues with quotes in project name, failing
#5710
add link to last change to browse

new status tickets:

#2382
WebAdmin: resync from the web interface
#4048
Diff between repositories
#8700
patch for trac-svn-post-commit-hook.cmd

Installation (with mercurial)

A current installation procedure which works is the following:

  1. Check out the mercurial plugin: 
    svn co http://svn.edgewall.com/repos/trac/plugins/multirepos/mercurial-plugin
  2. Build the mercurial plugin. From within the checked-out directory, build the egg file (it will end up in the dist subdirectory): 
    python setup.py bdist_egg
  3. Get Trac 0.12 or later. Check out the multirepos branch: svn co http://svn.edgewall.com/repos/trac/sandbox/multirepos
  4. Install Trac; see TracInstall. Install the multirepos version of trac (from within the multirepos directory, as root): python setup.py install
  5. Create your trac environment directory.
  1. Copy the mercurial plugin egg file to the <trac-env>/plugins directory, or use easy_install <plugin.egg> for global installation.
  1. Make sure the following is somewhere in <trac-env>/conf/trac.ini: 
    [components]
tracext.hg.* = enabled
    or use the 'Admin' panel in the browser to enable the plugin.
  2. Add all your repositories in the Repositories admin panel. Alternatively, you can add the repositories to the [repositories] section of trac.ini, but note that this has some restrictions on the repository names (i.e. lowercase, no '.'): 
    [repositories]
my_repo_a.dir = <path_to_a_repository>
my_repo_b.dir = <path_to_another_repository>
my_repo_c.dir = <path_to_yet_another_repository>
my_repo_c.type = svnfs

[trac]
repository_dir = <path_to_default_repository>
repository_type = hg
    Note that when the repository .type is not specified, the value of [trac] repository_type is used ([versioncontrol] default_repository_type since Trac 1.2).
    In the above example, this means that my_repo_a and my_repo_b are Mercurial (hg) repositories.
  3. Synchronize all your repositories, either with a wildcard: 
    trac-admin <trac-env> repository resync '*'
    or, if you want, explicitly: 
    trac-admin <trac-env> repository resync "(default)"
trac-admin <trac-env> repository resync my_repo_a
trac-admin <trac-env> repository resync my_repo_b
trac-admin <trac-env> repository resync my_repo_c
  4. Please note that at the time of writing, no initial resynchronization or any hooks are necessary for Mercurial repositories - see #9485 for more information. However, if you wish, you can install post-commit hooks for all your repositories, and have them call trac-admin <trac-env> changeset added for every new changeset: trac-admin <trac-env> changeset added my_repo_c $REV However, why you would want to do this in a normal 0.12 installation is beyond me.
  5. Cross your fingers, and hope it works :)
    It really should or it's a bug ;-)
    Mercurial works fine without hooks and resynchronization.

To make links to the non-default repository, just add a slash and the alias name after the revision identifier, for example: "[changeset:cffd9e480266/my_repos_b a changeset]". The default repository can be one of the explicitly named repositories.

Note that Step 9 & 10 are not needed for Mercurial integration, because TracMercurial will access Mercurial repository structure when any user view the version control system via Trac.

TODO

Merge

Now with #7116 nearly done, it's time to plan the merge.

  • What remains to be done is mainly documentation, as when people using trunk will get the multirepository changes, they'll need to find out what to do, in 0.12/TracRepositoryAdmin:
    • mainly what they need to do for adapting their repository hooks
    • what they can do with their default repository (setting up an alias for default, or better, "migrating" their unnamed default repository to a named repository and add an alias to the empty repo pointing to the new name)
    • trac-admin (repository and changeset commands)
      • hidden and url attributes
    • admin panel intro

A first version is now available, and should hopefully be sufficient for the merge. It could still be improved before releasing 0.12.

Not strictly needed before the merge but would be nice:

  • #7743 for mercurial - should add a contrib file there, containing a hook function calling the Trac API appropriately, and have an example of a command hook calling trac-admin directly.
    • A separate script is not strictly necessary (at least on Linux). The following section in hgrc does the trick: 
      [hooks] 
commit = trac-admin /path/to/env changeset added repo_name $HG_NODE 
changegroup = hg log -r $HG_NODE: --template "{node} " | xargs trac-admin /path/to/env changeset added repo_name
      This should be documented in TracMercurial, though. OTOH, for Windows we will need a Python hook anyway, so we could as well create a generic one.
    • Mercurial Repositories

Please note that at the time of writing, no initial resynchronization or any hooks are necessary for Mercurial repositories - see #9485 for more information.

Major

  • split repository names at "/" in the repository browser (i.e. show only the a entry at toplevel when we have repositories a/one and a/two, show one and two when expanding a)
  • currently, a "default" repository (the one defined in the [trac] section) needs to be defined, otherwise the versioncontrol modules are disabled r7972 - repository_dir entry needs simply to be present (can be empty)
  • fix AJAX query in annotate
  • IRepositoryProvider must have a way to signal new content, in order to rebuild the repository name cache - trac-admin ... repository notify <repos> changeset_added <rev>, see #7723. See sandbox/multirepos/contrib/trac-svn-hook.
  • support the .url repository metadata (#1781)
  • small issue: visual glitch in Firefox 2 for the new Rev column, where the changeset icon doesn't align properly to the right (but works fine for every other web browser, including Firefox 3) FF 2 is now dead, isn't it?
  • "normal" feature: it's currently not possible to compare across repositories (#4048) (the situation is detected, but raises a "not implemented" error). It would be quite useful to be able to do so, and it's doable without much work (only need to compare folders in a backend neutral way, we already do our own diff at the file level)
  • extend to cached repositories, see MultipleRepositorySupport/Cache and #7723
    In a first step, the existing cache could be adapted to simply handle more than one repository.
    Bonus points if that cache is smart enough to handle multiple scoped repositories without data duplication.
    A second (much) more ambitious step would be to generalize the cache so that it can support other (d)vcs, but this is somewhat out of scope of the MultiRepos task itself.

Minor Points

  • the browser quickjump menu wants to go to the repository youngest (svn), not the youngest cached so far. There might be more places like that, we need to deal consistently with the youngest cached. fixed in r8396
  • browser expansion doesn't work for the repository index. I'm pretty sure it used to work at some point. fixed in r8076.
  • More IRepositoryProvider components, though this is becoming less pertinent now that we manage the repositories in the db and the repository table:
    • HgParentDirRepositoryProvider, for listing all the working copies present in a given folder
      • work-in-progress on ​http://hg.trbs.net repo here is not up-to-date, but code is and works on latests multirepos tree, waiting for the opportunity to submit the provider back into mercurial-plugin
    • HgForestRepositoryProvider, for providing forest repositories and all sub-repositories managed by each forest
    • SvnParentDirRepositoryProvider, for listing all the Subversion repositories present in a given folder
  • Add an admin panel to manage repositories (#8291). Complete as of [8257].

Other Open Points

taken from #6665:

A few comments / questions

  1. How can I get the name in the Repository Index to keep the case in the name? It is converted to lower case.

ConfigParser limitation (#7378).

  • rblank: Note that this is less of a concern now that repositories can be added from the web interface.
  1. Since you use the init tag as the name to show in the Repository Index, it seems I cannot have a period in the name of the repository as shown.

Also related to the way the the .ini keys are read. It would be possible to look first for the ending .dir part and then get the name.

  • rblank: Again, not an issue anymore with the web interface.
  1. I get a default in the Repository Index, as well as that same repository listed in the previous single repository manner. I think that when the multiple repositories are listed, and one of them is the same as the trac repository_dir, it does not need to be listed twice or three times.

Yes, known issue: (verify)

  • one possibility would be to not list aliases (even aliases for the [trac] repository_dir). I think that's acceptable, but there should be an easy way to learn about the defined aliases, maybe they can be listed at the bottom of the toplevel page.
  • OTOH, aliases might be defined simply because they are more convenient names than the "real" ones, so in the end, more flexibility is needed. Add a .browseable flag, a la Samba?
    • there's a .hidden flag now (and a corresponding hidden attribute on Repository that can be set via the admin panel or via: 
      trac-admin ... repository set hidden true

Other than that, so far this is going to meet my needs quite nicely. Thanks!

Thanks :-)

Last modified 7 years ago Last modified on Aug 6, 2017, 12:34:24 AM
Note: See TracWiki for help on using the wiki.