Migrate List-Compare from Subversion on One Machine to Git on Another

subversion-to-git-migration-of-cpan-module.pod

Migrate List-Compare from Subversion on One Machine to Git on Another

... and then up to github.com!

Initial Situation

On a Mac OS X iBook G4, I had a Subversion repository with a wide variety of projects: my CPAN modules, YAPC and other conference presentations, etc. I wanted to transfer the locus of development for certain projects, beginning with my CPAN distribution List-Compare, to a newer machine running Ubuntu 14.04 Linux. I also wanted to make that development process more publicly available. That meant migrating the projects from Subversion to git, then publishing those projects at https://github.com/jkeenan/.

In git, unlike CVS and Subversion, standard practice is to have one repository per project. That meant that I had to extract one project at a time from the Subversion repository on the Mac and re-store it in a git repository on the Ubuntu. This gist describes that process.

Work on the First Machine

On the basis of some Internet recommendations, I decided to create a dumpfile of the Mac's Subversion repository, then extract the commits for a given individual project into a dumpfile for that project. I would edit that project per Internet recommendations, getting a refined version of that dumpfile. To test the validity of that dumpfile, I would then create a new Subversion repository on the Mac and load it from the dumpfile. I would then do an svn checkout from that repository and verify the checkout was sound.

Commands:

The first repository to be transferred was named lc, standing for List-Compare.

• Dump the entire Subversion repository.

  svnadmin dump /path/to/repository > ~/repository-dumpfile

• Filter the dumpfile to create a project-specific dumpfile.

  svndumpfilter include --drop-empty-revs --renumber-revs lc \
    < repository-dumpfile \
    > lc-dumpfile

• Make a gzipped backup as a precaution.

  cp -v lc-dumpfile lc-dumpfile.raw
  gzip -v lc-dumpfile.raw
  gunzip -c lc-dumpfile.raw.gz > lc-dumpfile

• Edit the project-specific dumpfile

  vi lc-dumpfile

  1 Remove the first Node-path: block entirely.

  2 Trim lc from Node-path and Node-copyfrom-path.

  %s/Node-path: lc\//Node-path: /
  %s/Node-copyfrom-path: lc\//Node-copyfrom-path: /

  3 Save file.

  Net result should be zero output from:

  grep -n 'lc/' lc-dumpfile

• Validate dumpfile.

  Before shipping the dumpfile to the second machine, I took the precaution of making sure that I could create a new Subversion repository from its contents.

  cd ~
  mkdir svn_transfer_repositories
  cd svn_transfer_repositories
  svnadmin create lc
  svnadmin load --ignore-uuid lc < ~/lc-dumpfile

  Further precaution: create a new checkout from the new repository.

  cd ~
  mkdir transfer_checkouts
  cd transfer_checkouts
  svn checkout file:///Users/jimk/svn_transfer_repositories/lc/trunk lc
  cd lc
  svn log | less

  If svn log looks okay, we can proceed.

• Compress per-project dumpfile and send to second machine.

  cd ~
  gzip -v lc-dumpfile
  # scp to Linode server

Work on the Second Machine

I proceeded on the second machine as cautiously as I had on the first. I verified that I could create a new Subversion repository, populate it with the contents of the dumpfile and do an svn checkout from that repository. Only then would I proceed with a conversion from Subversion to git.

Commands:

• scp lc-dumpfile.gz from Linode server to ~/Downloads/

• Create a new Subversion repository and load it with the dumpfile.

  cd ~
  mkdir svnrepos
  cd svnrepos

  Anticipating creating a github.com repository named list-compare, on the second machine I began to use list-compare instead of lc as the name for a directory or repository.

  svnadmin create list-compare
  gunzip -c ~/Downloads/lc-dumpfile.gz | svnadmin load list-compare

• Create a location on second machine for svn checkouts

  cd
  mkdir svnwork

• Perform an svn checkout; verify its validity

  cd svnwork
  svn checkout file:///home/jkeenan/svnrepos/list-compare/trunk list-compare
  cd list-compare
  svn log |less

• Now migrate to git.

  Note that svn2git by default places files in your current directory. But I wanted a project-specific subdirectory, so I created one.

  cd ~/gitwork
  mkdir list-compare
  cd list-compare
  svn2git file:///home/jkeenan/svnrepos/list-compare

• Confirm validity of git repository

  git log

• Add a remote

  git remote add origin [email protected]:jkeenan/list-compare.git

Initiate Repository on github.com

I went to my account on github.com and used the GUI interface.

I clicked on Create new repository but did not create a README or .gitignore. I already had a README in the distribution and I planned to copy and revise a .gitignore from another project already on the Ubuntu.

I saved the new, empty repository on github.

Back on the Second Machine

List-Compare was now stored in its own git repository on my Ubuntu machine. I began by editing by adding a .gitignore and adding ^.git/ to MANIFEST.SKIP to what is now the master branch. I continued by updating the distribution's $VERSION number, all release and copyright dates, and adding an item to Changes.

Commands:

vi # edit files as needed
git add .gitignore MANIFEST.SKIP Changes [...]
git rm [...] # as needed

perl Makefile.PL
make
make manifest # then correct MANIFEST as needed
make test
make dist
mv -v List-Compare.0.39.tar.gz ~/localstorage/
# scp tarball to Linode archive
make clean

git push -u origin master

# upload to PAUSE

Results

• List-Compare now uses git for source control and is stored on a newer and more powerful laptop.

• List-Compare now has a development repository at https://github.com/jkeenan/list-compare.

• List-Compare has a new version which has been archived locally and on Linode and released to CPAN.