A set of scripts to work locally on Subversion checkouts using Mercurial
This set of scripts allows to work locally on Subversion-managed projects using the Mercurial distributed version control system.
Why use Mercurial ? You can do local (disconnected) work, pull the latest changes from the SVN server, manage private branches, submit patches to project maintainers, etc. And of course you have fast local operations like “hg log”, “hg annotate”…
Three scripts are provided:
- hgimportsvn initializes an SVN checkout which is also a Mercurial repository.
- hgpullsvn pulls the latest changes from the SVN repository, and updates the Mercurial repository accordingly. It can be run multiple times.
- hgpushsvn pushes your local Mercurial commits back to the SVN repository.
Making a checkout of the Django trunk:
$ mkdir django && cd django # Make SVN checkout, initialize hg repository with first SVN revision $ hgimportsvn http://code.djangoproject.com/svn/django/trunk/ $ cd trunk # Pull all history from SVN, creating a new hg changeset for each SVN rev $ hgpullsvn
Then make your changes and use the “hg” commands to commit them locally. If you have commit privileges you can push back your changes to the SVN repository:
If you want to see what will be pushed back to SVN, use the “-n/–dry-run” flag. This is much like the “hg outgoing” command:
$ hgpushsvn --dry-run
Just type easy_install hgsvn. If easy_install is not available on your computer, download and uncompress the source tarball, then type python setup.py install.
Note: hgsvn makes use of the ElementTree library. It is bundled by default with Python 2.5, and the setup script should install it automatically for you if you are using Python 2.4. However, if you get some error messages, you might have to install it manually (at least one user reported he had to).
Unstable (development) version
Just run hg clone http://bitbucket.org/andialbrecht/hgsvn hgsvn and you’ll get the contents of the development repository.
hgpullsvn asks for SVN log entries in chunks, so that pulling history does not put the remote server on its knees.
hgpullsvn can be interrupted at any time, and run again later: you can pull history incrementally.
hgsvn reflects commit times (using the local timezone) and commit author names. Commit messages can contain Unicode characters. File copies and renames as reflected as well, provided they occur inside the branch.
These scripts encourage the use of named branches. All updates using hgpullsvn are made in the branch named from the last component of the SVN URL (e.g., if the SVN URL is svn://server/myproj/branches/feature-ZZZ, hgpullsvn will create and use the named branch ‘feature-ZZZ’).
You can thus do local development using your own named branches. When you want to fetch latest history from the SVN repository, simply use hgpullsvn which will update to the original (pristine) branch, leaving your local work intact (you can then merge by yourself if your want).
This also means that hg di -r name-of-pristine-branch will immediately give you a patch against the pristine branch, which you can submit to the project maintainers.
(Note: in a non-trivial setup where you work on several features or bugfixes, you will clone the pristine repository for each separate piece of work, which will still give you the benefit of named branches for quickly generating patches).
Detecting parent repository
If the SVN URL has been created by copying from another SVN URL (this is the standard method for branch creation), hgimportsvn tries to find an hgsvn repository corresponding to the parent SVN URL. It then creates the new repository by cloning this repository at the revision immediately before the creation of the SVN branch.
In other words, let’s say you are operating from myworkdir/. In myworkdir/trunk, you already have an hgsvn repository synced from svn://server/myproj/trunk. You then hgimport svn://server/myproj/branches/new-feature. It will find that the ‘new-feature’ branch has been created by copying from ‘trunk’ at rev. 1138. It will thus create the ‘new-feature’ hg repository by cloning from the ‘trunk’ repository at the revision immediately preceding rev. 1138: for example rev. 1135, identified by the local tag ‘svn.1135’.
This means you will have an hgsvn repository containing two named branches: ‘trunk’ for all the changesets in the trunk before rev. 1138, and ‘new-feature’ for all the changesets in the SVN branch (therefore, after rev. 1138). This way, you can easily track how the branch diverges from the trunk, but also do merges, etc.
SVN externals are purposefully ignored and won’t be added to your Mercurial repository.
- hgpushsvn fully integrated in this version. This command pushes local Mercurial commits back to the SVN repository.
- Add a -n/–dry-run flag to hgpullsvn and hgpushsvn to list incoming and outgoing changes without committing them to the local or remote repository.
- Add verbosity levels to all command line scripts. The default verbosity level limits the output of hgpullsvn and hgpushsvn to messages that have a certain relevance to the user. Additional messages will be shown with the -v/–verbose flag. The –debug flag enables the output of debugging messages including shell commands.
- Log files from SVN repositories with path-based authentication caused hgpullsvn and hgimportsvn to fail with an XML parsing error. Restricted paths are now silently ignored (issue5, reported by Andreas Sliwka).
- Updated the Mercurial-based lock file mechanism introduced in the previous release to work with Mercurial >= 1.2.0. The exception class was moved in Mercurial 1.2.0 (issue4).
- hgpullsvn and hgpushsvn can now be called from any sub-directory within the working copy. Both scripts now take care to change their working directory to the root of the working copy (issue3).
- hgimportsvn exits with a error message when the target directory is already a Mercurial repository controlled by hgsvn.
- Detect conflicts when running SVN update. To avoid conflicts when pulling new SVN revisions, hgpullsvn checks if the hg repository has uncommitted changes before actually pulling new revisions from SVN (issue6, reported and initial patch by Robert).
- Prefix commit messages with [svn r123] (where 123 is the corresponding SVN revision number), rather than just [svn]. Also, trim leading whitespace in the original commit message, to avoid blank changeset summaries when the message begins with a carriage return.
- Introduce a .hgsvn private directory at the top of the working copy. This will us to store various things in a common location without polluting the base directory.
- Introduce a lock file (named .hgsvn/lock) to disallow running two hgpullsvn instances in parallel on the same repository. The locking mechanism is imported from mercurial’s own mercurial.lock. If the mercurial package is not available, a dummy lock is used instead. Initial patch by Ori Peleg.
- Add a –no-hgignore option to hgimportsvn, for situations where the source SVN repository already contains a versioned .hgignore file. Patch by Ori Peleg.
- hgsvn can now be bundled as standalone executables using py2exe. Patch by Paul Moore.
- More descriptive error message when either hg or svn cannot be executed (e.g. not installed). Patch by Joonas Paalasmaa.
- Very long commit messages (> 16000 characters) are provided to Mercurial through a temporary file rather than the command line, which could fail on some platforms. The corresponding commit messages were generated by svnmerge. Reported by Ralf Schmitt.
- Filenames starting with a hyphen were considered by hg and SVN as command-line options. Report and patch by Mirko Friedenhagen.
- If the last hg changeset mapped to more than one SVN revision, trying to update again with hgpullsvn failed.
- A replaced directory can have added and removed files without them being mentioned in the SVN log; we must detect those files ourselves.
- More robust atomicity check for (hg commit, hg tag) sequence. Reported by Florent Guillaume.
- Fix a bug when comparing local filesystem paths on Windows. We now invoke os.path.normcase on both paths before comparing. Reported by Pavol Murin.
- In the initial import, parse the svn:ignore property and add suggestions to the .hgignore file accordingly. These suggestions are commented by default because they are based on the latest version of the svn:ignore property and could make us miss some files along the SVN history, if enabled blindingly.
- Critical fix for Mercurial 0.9.5, which is stricter with named branches. This bug made hgimportsvn fail when cloning from an auto-detected parent repository.
- Honor the SVN_ASP_DOT_NET_HACK environment variable when deciding the name of private SVN folders (.svn or _svn). Thanks to Anton Daneika for the report and the original patch.
- Change setuptools options to solve bdist_rpm bug under Fedora and other Linux distributions. Patch by Tim Wegener.
- Be able to pull dead (removed) SVN branches by introducing a -p (–svn-peg) option to specify the SVN “peg revision”. The option must be used with both hgimportsvn and hgpullsvn. Patch by Cameron Hutchison.
- Allow copying directories with non-ASCII names (reported by Andre Klitzing).
- Make rmtree reliable under Windows. Thanks to Mark (mwatts42) for finding both the bug and the solution.
- Fix a problem where there is a symbolic link in the SVN repository that points to a directory. Patch by Cameron Hutchison.
- svn log can output invalid XML when a commit message contains control characters. Reported by Tim Wegener.
- License upgraded to GNU GPL v3 (or later).
- Performance improvement with svn log command in hgpullsvn (suggested by Mads Kiilerich and Daniel Berlin).
- Less obscure error message when svn info fails while returning a successful return code.
- Two simplistic man pages added.
- Windows compatibility fix by Bill Baxter.
- hgimportsvn failed when used on a whole repository.
- Fix crash on empty commit message (also reported by Neil Martinsen-Burrell and Walter Landry).
- Handle file and directory renames properly (reported by Bill Baxter).
- SVN allows copying from a deleted file by having mixed revisions inside the working copy at commit time, but Mercurial doesn’t accept it (reported by Neil Martinsen-Burrell).
- Automatically generate .hgignore file. Not only does it produce cleaner output for commands like hg status, but it speeds things up as well.
- hgpullsvn is more robust in the face of errors and user interruptions.
- Try to be Windows-compatible by not using the commands module.
- Remove dependency on the pysvn library; we use the XML output option of SVN commands instead.
- Fix a bug in parent repository detection.
- Detect the wicked case where the SVN branch has been overwritten with contents of another branch (witnessed with Nose trunk and 0.10-dev branch). We can’t properly handle this situation, so fail with an explicit message.
- svn info on base repository URL does not always succeed, use the specific project URL instead (reported by Larry Hastings).
- pysvn doesn’t really ignore externals, so use the command line for svn update instead (otherwise we get failures for obsolete URLs)
- .svn directories were not always ignored.
- On large repositories, adding more than 32765 files at once failed because of too many arguments on the command line.
- On slow SVN servers, the chunked log fetching algorithm ended up asking for 0 log entries.
Release history Release notifications | RSS feed
Download the file for your platform. If you're not sure which to choose, learn more about installing packages.
|Filename, size||File type||Python version||Upload date||Hashes|
|Filename, size hgsvn-0.1.7-py2.4.egg (107.4 kB)||File type Egg||Python version 2.4||Upload date||Hashes View|
|Filename, size hgsvn-0.1.7-py2.5.egg (115.0 kB)||File type Egg||Python version 2.5||Upload date||Hashes View|
|Filename, size hgsvn-0.1.7-py2.6.egg (115.0 kB)||File type Egg||Python version 2.6||Upload date||Hashes View|
|Filename, size hgsvn-0.1.7.tar.gz (48.6 kB)||File type Source||Python version None||Upload date||Hashes View|