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",
Three scripts are provided:
* ``hgimportsvn`` initializes an SVN checkout which is also a Mercurial
* ``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
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
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
.. _Django: http://www.djangoproject.com
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``. Type ``python setup.py install --help``
for additional options.
*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
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.
``hgpullsvn`` tags each new Mercurial changeset with a local tag named
'svn.123' where 123 is the number of the corresponding SVN revision.
Local tags were chosen because they don't pollute the hg log with
superfluous entries, and also because SVN revision numbers are only
meaningful for a specific branch: there is no use propagating them
you can assign this tags manualy to provide hgvsn scripts operation
on repository from any revision
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
(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
Named branches map and Hystory tree
default hg-branch naming can be overriden by option --branch=<hg-branch name="">
hystory of svn-to-hg branches mapping saves into a map-file .hgsvn/branches.map
this contents of strings with pairs:
it can be edited manual, or new pairs can append by option
appending and updating this file controls by option -mapbranch=<options>
hgpull\pushsvn automaticaly recognises current hg-branch and svn-branches, and
using this map can pull\push hystory tree. now pull\push stops at merge points
by default, to enable this operation use option:
--merge-branch=(push\pull | try<push\pull> | skip ...
hgpullsvn: can automaticaly open named branch as svn-branch mirror, when ones
hgpushsvn: start of new svn-branch mirrors current hg`s one can be do now follows:
update svn+hg to desired branch base rev.
tag\switch from there svn to apropriated svn-branch name
update hg to some rev in desired branch
hgpullsvn -l --branch-add=<new svn-branch="">=<hg-branch>
and so current hg-branch pushes into new svn one.
it is nesesary for hgpushsvn to have svn-synched base of desired branch,
and a valid branch map to know where to push it
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
+ svnhead bookmark - provides bookmark topmost synced rev with bookmark "svnhead",
and tophost synced rev of every branch as "svnhead/<hg_branch_name>"
! py3.4 compatibity provided on error messages of hgclient
* py3.4 compatibity provided, and success test by hgsvnpush
! svn:locale:url - svn url now decodes and can be nationalised
+ ``hgpushsvn --auto-start-branch now creates branch in svnrepo, if it have maped
! ``hgpushsvn - "through revision", and some another bugs fixed
! pull - fixed svn sync revision loose for -l option
+ setuptools 20 migrated
+ start Python3.4 compatibility
! hgpush - few fixes of branch-map detection, svn copy/add pathes, merge proof on mixed-source wc,
cleanup wc for avoid mixed sources,
fixed message for commit that crushes svn 1.9
+ ``hgpullsvn --noversioned_change=<skip|abort(default)>`` - allows ignore when try delete files that
! fix crach for above situation for files that not versioned and absent
! hgpush - anonimous branches has tested in situation with multiple heads in current branch.
option -l now use current hg-rev as point that directs hystory when choose between
multiple anonimous branches, or to drop aside heads.
! hgpush - issue #107 David Johansen : hgsvn fails when file deleted in moved directory
accepted and fixed
+ hgpush - now automaticaly detects started svn-branch on maped hg-branch, that
not yet synced, and try to sync one with hg-branch.
now to start synch new hg-branch need only 'svn tag' origin svn-revision to maped svn-branch.
! push - hg v3.4 fixed bug for log --follow option
! hgpush - now -l option can ommits WC changes if svn status = not changed
! hgpush - fix anonim branches confuses, now they not interfere,
on trypush and skip modes pushed are first branch
+ ``hgpush --autoimport`` - now can auto start push on empty svn repo
that not syncronised yet
+ pull : merge - now automated prompts on hg mergeing
! test fixed unittests
! hgpush - fix broken hg deletes detection
! hglib - fix hg operations with files with national characters
! ``svn`` - critical fix of determining replaced and deleted svn status
! ``hgpullsvn -l`` - now normaly switch to hg branch to clean state
* ``hglib`` - migrate to this lib for hg commands invoking. it must provide faster
execution due hgserver standup once and ready for all script session
! ``hgpullsvn`` fixed loose svn changes when starts with svn is not clean due to
update just merge changes. now more stict svn status check after update
+ ``hgpullsvn`` with -f and -l options there no break by hg durty status if
svn status is clean
! ``hgpullsvn`` now svn status tree-conflicted take care, ignoring of one causes
to loose fileadded\deleted changes
! ``hgpullsvn`` remove files from hg-control try to "hg forget" if "hg remove"
can`t delete ones
! ``hgpushsvn`` - invoke snv cleanup after resolve if one reports error
"Run 'svn cleanup'"
+ ``hgpushsvn`` - now can start push branch that have no syn-revs, but have
synced branch origin revision, and branch-maping defined
! ``hgpushsvn`` issue89 - revision with "Nothing to do" tags by last svn.rev
after last push
! `` --svn-accept`` = "working" by default now, this helps resolve conflicts
to merge and switch svn
* ``--svn-accept`` default is "working"
! strategy of evaluate sync start point: now if it was not defined peg revision,
it try to found last synched rev wich from to push in current svn branch;
if --hg-follow defined, then it try to switch svn to current hg branch, and
find last synched revision
if peg revision defined, then it switches svn to it, and start push\pull into it
from current hg revision
* ``hgpullsvn --hg-follow`` - option renamed from --hg-branch.
* ``hgpushsvn --hg-follow`` - on clean working dir, this option force svn
switch to last synched rev of current hg-branch
+ ``hgpullsvn "--subrepo=<[break]|skip>"`` - if 'skip' - pulling ommit changes
occures in subrepositiories
+ add support pull\push merge revisions, and automation of push\pull dependent branches
--merge-branches = (push/pull | trypush/trypull | skip | break) - defines how to process
merge-points : push/pull - synchronize dependent branches
skip - process them as ordinary revisions
break - stops on it
trypush/trypull - synches branches if can, else process like skip
merge branches automation bases on branch-maping <svn base="" path=""> to <hg-branch name="">. this map
accumulates in file .hgsvn/branches.map, and it can be edited manualy. accumulation controled by
--mapbranch=<[n][o][a]> - set branches.map update mode: allow add, overwrite mapings to one
--branch-add='<svnpath>=<hgname>' - add/overwrite maping
+ ``hgpullsvn --hg-branch`` - pulling can switch svn to current hg branch and pull one from last sync point.
in contrast standart behaviour requires to pull from current svn revision
* ``hgpushsvn --svnignore-use`` - append node to .hgsvn/.svnignore - list of files wich is ignored
from push\pull process. ordinary there should be .hgtags
+ ``--svn-accept=<x>`` - allow override default ``svn merge --accept`` option
+ ``--terminal-width=x`` - declares used maximum log line length
* ``--debug`` - this option now automaticaly setup --terminal-width=250
* Improved handling of files containing @ sign in filename (issue52).
* ``hgimportsvn --branch=myname`` imports the SVN repository to a
named Mercurial branch other than 'default' (patch by Matthias
* ``hgpullsvn --svn-retry`` retries to retrieve information on flaky
connections (patch by Stefanus Du Toit).
* Restrict internal SVN update when running ``hgpushsvn`` to last
pushed revision (issue79, by ankon).
* Improved syncing of empty directories (issue77, by ankon, reported
* ``hgpushsvn --keep-author`` keeps author given in hg log when
committing to SVN (by wwwjfy).
* Mercurial 1.7 compatibility (issue91, by easye).
* Convert a local SVN checkout into a hgsvn controlled Mercurial
repository with ``hgimportsvn --local-only``. No network access is
needed when setting this flag (aka airplane mode). The Mercurial
history then starts with the current revision of the SVN
checkout. Patch by Matt Fowles.
* Commit messages when pushing back to SVN can be edited before
committing using the -e/--edit command line flag. Issue #29, patch
* It's now possible to use ``hgsvn`` with the mq extensions. Both
``hgpushsvn`` and ``hgpullsvn`` abort with an error message if mq
patches are applied when running those commands. Issue #43, patch by
* Removal of temporary file when pushing new revisions to SVN on
Windows fixed. Issue 8 reported by Daniel Dabrowski.
* Changed 'hg log' command line arguments that interfered with default
options in .hgrc file. Issue 12 reported by Simon Percivall, initial
patch by Joel Rosdahl and issue 16 reported by Wladimir Palant.
* ``hgpushsvn`` is now compatible with Python 2.4.
* If a SVN changeset contains empty changeset comments, ``hgpullsvn``
has failed. Empty changeset messages are now handled correctly.
* SVN repositories at revision 0 couldn't be imported. Now it's
possible to import empty SVN repositories with ``hgimportsvn``.
Issue #13 reported by tiktuk.
* ``hgpushsvn`` failed when a non-empty directory was removed from
version control. Issue #15 reported by Keith Yang.
* Proper encoding of commit messages on Windows systems. Issue #19
reported and patch by Chunlin Yao.
* Change svn:executable property on mode changes when pushing back to
SVN repository. Issue #24 reported by sterin.
* Improved parsing of SVN messages. Issues #27, #14 patch contributed
* ``hgpushsvn`` whiped uncommitted changes in working directory. Issue
#32 reported by foxcub.
* ``hgpullsvn`` in dry-run-mode displayed latest fetched revision
too. Issue #35 reported by Dmitriy Morozov.
* New sub-directories are now created properly. Issue #46 reported by
* Several fixes by IanH: issue #64, #94, #95.
* ``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
* 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
* hgsvn can now be bundled as standalone executables using py2exe. Patch by
* 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
* 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
* 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
* 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
* 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.