OOo and Subversion

From Apache OpenOffice Wiki
Revision as of 11:36, 14 October 2008 by B michaelsen (Talk | contribs)

Jump to: navigation, search migrates to Subversion

After a long discussion and preparation time it finally happens, migrates with milestone DEV300 m31 to a new SCM (Software Configuration Management) system. The new SCM system will be Subversion (SVN) for now, which might disappoint the one or other who had hoped for a distributed SCM like Git, Bazaar or Mercurial. Please be assured that we'll keep our mind open and we'll reevaluate the fast moving SCM tooling situation periodically. We were very careful to make certain that our move now to Subversion does not preclude a move to a DSCM in the future.
Documentation note.png why

Trunk only migration

Milestone DEV300 m31 is the last milestone which has been integrated via CVS and is the first milestone which is available in SVN. We did a trunk migration only. The trunk (DEV300) code line will be maintained via Subversion only starting with milestone DEV300 m32. Old code lines, including the OOo-3.0 release codeline(OOO300) will still be maintained via CVS.

Please let me repeat this: we did not migrate branches and tags to SVN. There are a number of advantages (and some disadvantages as well) to this approach. A discussion of the merits of different approaches to the actual repository conversion can be found here.

The most important reasons for choosing the "trunk only" approach were:

  • No quiet time for developers
  • The conversion of historical workspaces from CVS to SVN would be fragile at best due to some liberties we took in structuring our CVS repository. Having no historical branches and tags is preferable to having inaccurate branches and tags IMHO.

The "trunk only" approach enabled us to optimize the resulting SVN repository, it shrunk from 90 GBytes to about 6 GBytes overall size. The following rules have been applied during conversion

  • Only modules which have active content in DEV300 m31 are included in the conversion.
  • All files in "Attic" have been removed.
  • All but the last revisions of binary files have been dropped - after all you can't diff them anyway
  • All but the last revisions of "localize.sdf" files have been dropped - saves an incredible amount of repository space without loosing much

Required tools

  • subversion-1.5.1 or later; download...
  • ssh for committers, preferably OpenSSH 2.4 or later

We make heavy use of the new SVN merge tracking feature. So please do not use a SVN client older than subversion-1.5.1 if you work on a child workspace (CWS).

Windows users most certainly want to take a look at TortoiseSVN and lots of documentation available at

Subversion documentation

The excellent Subversion documentation can be found here: Please use the nightly build documentation for 1.5 and later, it covers new features like merge tracking.

Please read at least the "fundamental concepts" and "basic usage" part of the documentation!
CWS tooling is only a very thin layer around the SVN client hiding the lengthy URLs from sight plus some bookkeeping. Everything else is plain Subversion.

Server access methods

Read only SVN access

svn checkout svn://


svn checkout

Read only browser access

Point your browser to

Read/write access

Write access to the repository requires a ssh setup.

svn co svn+ssh://

Basic read/only SVN operations

Repository structure

The OOo SVN repository is structured like this:

../trunk                     <= the main development line, currently master workspace DEV300 is on trunk
../cws                       <= child workspaces live here, names need to be unique
../cws/foo                   <= for example CWS foo
../branches                  <= future master workspaces which are not on trunk
../branches/OOO310           <= for example MWS OOO310
../tags                      <= milestone tags, release tags
../tags/DEV300_m32           <= for example milestone DEV300 m32
../dist                      <= space for distribution specific stuff
../patches                   <= space for globally useful patches
../contrib                   <= do we need this?


Check out a milestone DEV300 m32 with:

svn checkout svn://

Switch from milestone DEV300_m32 to DEV300_m35 (safes potentially a lot of checkout time):

cd DEV300_m32
svn switch svn://
cd ..
mv DEV300_m32 DEV300_m35

Check out a child workspace foo:

svn checkout svn://

Update working copy of foo:

cd foo
svn update

Find out the base (milestone or cws) of working copy wc:

cd wc
svn info

List all available milestone and release tags:

svn list svn://

List all available child workspaces:

svn list svn://

View commit log on a single file:

svn log file.cxx

Include all paths affected by change sets in log command:

svn log --verbose file.cxx

Attribute blame:

svn annotate file.cxx

SSH Setup

SSH key generation

On Unix or Windows/cygwin a SSH identity (private/public key pair) is generated with:

ssh-keygen -t dsa

You'll be asked for pass phrase which protects your private key. Per default the public key is named $HOME/.ssh/

Note that SSH keys created between September 2006 and May 2008 on Debian or Debian derived systems like Ubuntu may be insecure. Please use recent SSH tools.

If you want a dedicated identity for just OOo SVN usage, generate a specially named key pair

ssh-keygen -t dsa -f ~/.ssh/id_dsa_ooo_svn 

and instruct Subversion to use this identity (key pair) with the following configuration entry in $HOME/.subversion/config:

ssh = ssh -i /home/jr93709/.ssh/id_dsa_ooo_svn

Initial Migration of Keys

Please query IssueZilla for your issue with the attachment holding your key and add it as a dependency to issue 94002. If you are a new svn user, create the issue yourself.

We will then install those keys for DomainDevelopers for read/write access to the SVN repository.

Up to now only few code committers made use of the ssh key upload for the SVN repository. Please note that your key will not be migrated without your intervention.

Ongoing Maintenance

We will continue to use IssueZilla for managing and tracking key additions and changes. Please see Submitting Your Certificate (Public Key).

CWS tooling

The CWS tooling has been reworked to adapt to SVN. The basic CWS tool is now simply called cws and is invoked as cws <subcommand> in the style of the SVN client, there are no longer separate tools like cwsadd, cwsresync etc etc.

Note: the cws script will probably change over time as we learn how the usage patterns evolve. If you got a problem with the cws script, it might be worthwhile to rebase it to the latest version

$ cd <your_CWS>/ooo/solenv/bin
$ svn merge svn+ssh://

The most important change in the way CWSs are handled now is that there is no longer a need nor possibility to add modules. Well, there is still a sw and config_office but all modules of old are now just top level directories, at least as far as SVN is concerned.

CWS config file

Please keep all your entries in your $HOME/.cwsrc for now, you'll need them if you want to work with CWSs for OOo-3.0. Just add the following line to refer to the SVN server:


Example of a complete .cwsrc file:

# Network proxy for EIS connections. 
# Comment in if you need to access EIS via a proxy.
# Note: This is just for accessing the EIS database. For SVN have a look here:
# Example: PROXY=

# CWS Database server (SOAP). It's possible specify to several backup server,
# currently there are none.

# Your CVS login (for authentication with the CWS database).

# Your scrambled CVS password (for authentication with the CWS database).
# Take this one from .cvsrc

# OOo CVS tunnel
# Example:

# Path to the cvs binary (optional)
# Example: CVS_BINARY=/usr/bin/cvs

# SVN server for OOo-3.1 child workspaces
# r/w access via svn+ssh pprotocol
# r/o access via svn protocol

Working with the new cws tool

Help can be invoked with:

cws help

which yields

cws -- version:
usage: cws <subcommand> [options] [args]
Type 'cws help <subcommand>' for help on a specific subcommand.

Available subcommands:
        help (h,?)
        fetch (f)
        rebase (rb)
        analyze (an)
        query (q)
        task (t)
        integrate *** release engineers only ***
        eisclone *** release engineers only ***
help (h, ?): Describe the usage of this script or its subcommands
usage: help [subcommand]

The subcommands do more or less what their CVS predecessors did.

Create a CWS

Create a CWS with the cws create subcommand. Please note that this commands only creates a CWS branch on the OOo server and registers it with EIS. It changes nothing on your hard disk.

Example: create CWS foo on the latest published milestone of master DEV300

cws create DEV300 foo

Of course you can specify which milestone you want to use, here is the complete command reference:

cws -- version:
create: Create a new child workspace
usage: create [-m milestone] <master workspace> <child workspace>
       -m milestone:          Milestone to base the child workspace on. If ommitted the
                              last published milestone will be used.
       --milestone milestone: Same as -m milestone.

Your CWS branch is then available as


for check out.

Check out a CWS

$ cws fetch -c foo 

Currently this simply checks out your cws foo just as if you had issued the command

$ svn checkout svn+ssh://

A milestone can be fetched with

$ cws fetch -m DEV300_m32

this is equivalent to

$ svn checkout svn+ssh://

Over time I'll add a number of convenience functions to cws fetch, like configurable partial checkouts, use of prebuild solvers, switching CWSs etc etc.

Please note that if you have already a OOo SVN working copy lying around, you can save considerable checkout time with the svn switch command.

cd <wc>
svn switch svn+ssh://

will fetch the differences between your working copy in <wc> and the CWS foo

Rebasing a CWS

The successor of the cwsresync tool is called cws rebase. rebasing is now a two step operation:


How to migrate still active CVS based child workspaces destined for DEV300 to SVN?

Let's assume your CVS based CWS is called 'foo42'.

CVS tasks

1) Find out the CVS branch tag and anchor tag of your CWS.

$ cd <added_module>
$ cvs log <some_file_which_already_existed_before_your_cws> | grep -i foo42
        CWS_SRC680_FOO42_ANCHOR: 1.36

The first tag (CWS_SRC680_FOO42_ANCHOR) is your anchor tag, the second tag (cws_src680_foo42) is your branch tag. Note them down.

2) Rebase your CVS based CWS to milestone DEV300 m31 with the old CWS tools

$ cwsresync -m m31 all
... solve conflicts ...
$ cwsresync -c all

3) Create a patch containing all modules. It is recommend to use the "rdiff" CVS command. A "rdiff" is quite handy for this purpose, because it allows the creation of one patch for all "added" modules. Use the "-u" switch for creating a so called "unified context patch".

$ cvs -d <your_tunnel_setup> rdiff -u -rCWS_SRC680_FOO42_ANCHOR -rcws_src680_foo42 `cwsquery modules` > foo42.patch

4) Check the patch for consistency. It should contain only stuff you did on your CWS. Throw out trivial changes.

5) Search the patch for lines which starts with

diff -u /dev/null <some_file>:<some_revision>

These files have been added to your CWS. Note the filenames down, you'll need them later.

6) Search the patch for lines which end with :removed

diff -u <some_file>:<some_revision> <some_file>:removed

These files have been removed in your CWS. Note the filenames down, you'll need them later.

SVN tasks

7) Check out the OOo trunk with

$ svn checkout svn+ssh:// my_working_copy

8) Configure and source the configured script as usual

$ cd my_working_copy/config_office
$ configure
$ cd ..
$ source  # YMMV

9) Create the CWS branch in SVN

$ cws create --migration -m m32 DEV300 foo42

This command creates the CWS branch in Subversion based on milestone DEV300 m32 and updates the "current milestone" information in EIS. Milestone DEV300 m32 is essentially the same as DEV300 m31 (just a few fixes for SVN and CWS tooling).

The --migration switch tells the cws create command that an existing CWS is migrated. Without the switch, EIS will complain that your CWS already exists.

10) Switch your working copy to your CWS branch

$ cd my_working_copy
$ svn switch svn+ssh://
$ svn info
Path: .
URL: svn+ssh://
Repository Root: svn+ssh://
Repository UUID: 8da58560-a4e7-4996-a0c2-a735b94b261c

Check the URL: line of the output of svn info. It should point to your CWS branch. Please be extra certain that you never commit anything on trunk. You can always check the branch on which you are working with svn info.

11) Apply the patch

$ cd my_working_copy
$ patch -p1 < foo42.diff

Since DEV300 m31 and DEV300 m32 are almost identical, your patch should apply cleanly. If you get rejects you'll need to fix them manually. The patch tool handles added and removed files quite nicely, the first ones are created by patch, the latter ones appear as zero byte sized files in the tree.

12) Handle added and removed files

This is pretty straight forward. You need to tell SVN which files have been added or removed by your patch.

$ svn add <list_of_added_files>
$ rm <list_of_removed_files>
$ svn remove <list_of_removed_files>

13) Commit the changes to your CWS branch

Review all changes with

$ svn status
$ svn diff

and if everything is OK, commit them to your branch

$ svn commit -m"#iXXXXX#: migrate CWS foo42 to SVN."

Remember, always check that your working copy has been switched to the CWS branch before committing.

Do I really need to resync first to DEV300 m31 in CVS? Can I use a later milestone than DEV300 m32 in SVN?

No and yes resp. If you are certain that your patch applies cleanly or you are prepared to handle the rejects, you can skip the resync part of the recipe or/and start of from a newer SVN based milestone. Working with DEV300 m31 (CVS) and DEV300 m32 (SVN) almost ensures that the patch applies cleanly which is why we recommend to work with them.

Using plain SVN

Showing the differences between a milestone on the master and a cws

$ svn diff

Showing the differences in a module or directory between a milestone on the master and a cws

$ svn diff

This shows the differences in module sw between the milestone DEV300_m32 (a tagged version in svn-speech) and the most current checkin on cws swrefactormarks2. Since modules are only directories in svn (not in any way special as they where with CVS), this can be used to compare any directory too. For example:

$ svn diff

A specific revision can be compared to the master like this:

$ svn diff

This compares the cws at revision 4711 to the milestone m32

Tips and tricks

Ignoring output trees

Platform output trees (unxlngi6, unxsols4, wntmsci12 etc) will show up as unversioned directories marked with a "?" in the output of svn status and other SVN commands. This is inconvenient and reduces the usefulness of these commands. In principle it's possible to deal with this in two ways:

  • attach and commit a svn:ignore property to every top level directory listing all possible platforms
  • add all the platforms which are usually build to the so called global ignore list

Actually the global ignore list is more of a kind of "personal" ignore list which is applied to all svn commandos which access your personal subversion configuration.

Since maintaing svn:ignore properties in about 200 directories for about 20 platforms is pretty clumsy, I suggest that we use the global ignore list for this purpose. Here is my personal ignore list, YMMV.

$ cat $HOME/.subversion/config

global-ignores = common* unxsoli4* unxsols4* unxlngi6* unxlngx6* unxubti8* wntmsci12* unxlngi6* unxmacxi* 

Detailed instructions on how to use ignore lists can be found here.

Ignoring contents of .svn when using grep

To avoid being driven insane by getting grep hits on the .svn dirs contents you can use

$ export GREP_OPTIONS="--exclude=\*.svn\*"

Speeding up svn+ssh: access by caching ssh connections

You might have notice that nearly all SVN operations are somewhat slower via svn+ssh: access than via plain (and read-only) svn: access. This is due to the ssh handshake, some SVN operations even suffer several times from the handshake overhead.

There is an easy and convenient way to get rid of this overhead by using a ssh connection cache.

The following description is taken more or less verbose from the the GCC wiki:

SSH connection caching works with OpenSSH 4.0 on all Unix look alikes, but not on Windows/cygwin. You can check your OpenSSH version with

$ ssh -V
OpenSSH_4.6p1, OpenSSL 0.9.8e 23 Feb 2007

Create a ControlPath entry for in $HOME/.ssh/config

$ cat $HOME/.ssh/config
     ControlPath /tmp/ssh_ooo_control

Set the permissions 0600 on $HOME/.ssh/config, otherwise SSH refuses to work.

$ chmod 0600 $HOME/.ssh/config

Set up up a permanent connection to the ooo SVN server, for example right after login:

$ ssh -fMN

Subsequent svn commands will use the already established connection, resulting in noticeably less lag for short commands. It's also an alternative to using ssh-agent if you have protected your private key with a pass phrase.

Create a OOo repository mirror

The repository can be replicated with the svnsync tool. No special server side setup is necessary (read-only access is sufficient, please use either the svn: or http: method), but you need to make certain that the target repository can't be modified by other means than svnsync.

  • First create an empty target repository:
$ svnadmin create /absolute/path/to/rep
  • Implement the pre-revprop-change and start-commit hooks
$ cat /absolute/path/to/rep/hooks/pre-revprop-change


if [ "$USER" = "syncuser" ]; then exit 0; fi

echo "Only the syncuser user may change revision properties" >&2
exit 1
$ cat /absolute/path/to/rep/hooks/start-commit


if [ "$USER" = "syncuser" ]; then exit 0; fi

echo "Only the syncuser user may commit new revisions" >&2
exit 1
  • Initialize the target repository:
$ svnsync init --username syncuser file:///absolute/path/to/rep svn://
  • And finally synchronize the target repository with the source repository
$ svnsync synchronize file:///absolute/path/to/rep

The full details for replicating SVN repositories can be found [here].

Reporting Problems with the Repository

The list is used to coordinate the initial migration and ongoing work on the repository and related tools. Please raise problems there first. Issues which need the attention of repository administrators should be filed in IssueZilla against the component www/ SVN.

Read/Write access with TortoiseSVN

Please have a look at this page.

[More SVN Tips and tricks will appear here]


Documentation note.png why
Why not just wait until one of the DSCM systems emerges as a clear front runner? Well, first, this might never happen because each of the mentioned DSCM system emphasizes certain strength, leaving no one clear winner in all categories (ease of use, performance, integration possibilities and platform support). Second, and this is way more important, the CWS/CVS system is crumbling under the heavy weight of 8 years worth of OOo coding.
Personal tools