What i’ll explain here is how to use the svndumpfilter program to extract a folder of an SVN repository in it’s own repository and make it disappear from the history of the original one.

What you’ll need

You will need direct access to the server and to warn all your users that they’ll need to checkout again everything you will move (if you don’t renumber the revisions otherwise all users will need to checkout again any part of the repository they use)

Knowing what to filter

You will need to include all the folders you want along with all the folders where something moved, for example if you want /some/dir but it was at some point in the history named /somedir and renamed later you need to include both in the filter.

The same apply if only some of the files moved from there in this case you need to include each file that moved.

You’ll end up with a list of files and directories representing all the history of the files that will be extracted.

Dump the repository

The first step is to dump the repository as svndumpfilter only act on dumps and not on directories themselves

svnadmin dump repo > repo.dump

Filter the history and load it in a new repository

To create a dump that contains only the change in our chosen directories the command is :

svndumpfilter include --drop-empty-revs --renumber-revs /some/dir /somedir < repo.dump > repo-only-some-dir.dump

The two options --drop-empty-revs and --renumber-revs are what you’ll need in 99% of the cases but anyway you’ll either want both or none :

  • Both will give you a clean new repository with new revisions starting at zero and no empty revisions. It’s clean but if you had anywhere else (in the source code comments, in your issue tracker or in some commit message) something referencing “r3628” it will now be incorrect.
  • None will give you a chance to have the same revision numbers (Only if you don’t need to create new directories otherwise all revision numbers will move, see next paragraph)

Then you’ll need to create a new repository

mkdir newrepo
svnadmin create newrepo

If some of the directories you included had a parent that wasn’t himself included (like our /some/dir directory) you need to create them back

svn checkout "file:////path/to/newrepo/" newrepocheckout
mkdir newrepocheckout/some
svn add newrepocheckout/some
svn commit newrepocheckout -m "Prepare to load filtered history"
rm -R newrepocheckout

Then finally load your filtered history :

svnadmin load newrepo < repo-only-some-dir.dump

Remove the files from the originating repository history

It’s simplier as you normally won’t have to create any directory, just load the dump in place of the old repo.

The same choice is offered to you regarding removing the now-empty revs from the history and renumbering the other revisions but in this case it’s more logical to keep the revisions as it will allow for all users that hadn’t checked out the removed directories to continue to work with their checkouts. Otherwise they’ll need a clean new checkout.

rm -R repo
svndumpfilter include /some/dir /somedir < repo.dump > repo-without-some-dir.dump
mkdir repo
svnadmin create repo
svnadmin load repo < repo-without-some-dir.dump