CVS to Git Migration for PLD Linux using GitHub

This page describes migration of PLD Linux CVS Packages repository to github.

Organization in github as been done: https://github.com/pld-linux, later referred as Organization.

PLD Developers wishing to join Organization, should create account in GitHub, and Organization Owners will add them to Team named Developers.

Each developer should set their <login>@pld-linux.org as e-mail alias in GitHub (Account Settings→Email Addresses)

Migration test was done using cvs2git, that is not incremental (but semi-incremental achieved with rsync.log).

• full git cvsimport (using local cvs files copy): ~470 minutes
• Secondary git cvsimport time (local cvs mirror): ~580 minutes
• git cvsimport size: 4.8G
• git root bare size: 3.2G
• imported users will be login@pld-linux.org in authors file, people can claim their identities by adding their @pld emails to their accounts in github

You can clone from test import at carme:~glen/git/pld-git/git-import (via ssh)

Scripts used to do the import: http://cvs.pld-linux.org/cgi-bin/cvsweb.cgi/git-migration/

• cvsps to understand cvsnt “Commit Identifier”. FIXED: cvsps patched
• fix problematic packages:
  1. libghttp, cvsps loops on libghttp-fixlocale.patch (two revisions 1.1)
  2. kernel - cvsps loops on kernel-ppc.config
  3. fltk2, classpath, joomla
  4. vim : spec.vim: cvs checkout: Skipping `$Log$' keyword due to excessive comment leader.
• latin2 commit messages. Test packages: mutt, audiocut. RESOLVED: see cvs2git.options
• cvs branches not imported Test package: audiocut. FIXED
• unnamed branches will be converted to tags: cvs-<BRANCHPOINT>. test packages: nagios-alert-jabber, nagios-plugins * <del>trim $Log:$ on importing?. DONE
• filter out imported archives: .tar, .gz, etc?
  • http://pld.pastebin.com/xF5yzZAv - .tar, .gz files
  • http://pld.pastebin.com/JT9RQJ2s - “extensions” from cvs server
  • maybe these files should be removed in cvs server? they are still present in “old” SOURCES.old if you really want some
• maybe we should set Comitter to value other then Author during conversion, so could later identify cvs and git commits.
• remove %changelog from old commits. DONE
• what to fill to .git/description? just package name?
• what to do with removed packages. suggestion: clone them to pld-linux-obsolete organization in github and drop in pld-linux
• ideas: register pld-projects for storage of non-packages from pld cvs/svn (geninitrd, rc-scripts, …)

We decided to try github hosting:

• supports organizations
• each dev is responsible for maintaining his own account (public key upload, password resets etc. handled by a web panel)
• has group-based ACL
• has commit hooks (can write our own and github will then enable them for all users)
• sent a request asking if they'd agree to host us for free, we have a green light
• github has polish interface translations, most devs should be happy? :D
• what to fill to Summary and URL fields (parse from .spec?)
• automate creation of new packages via GitHub API. DONE: pldpkg.py currently can add and create repos.

• builder script
  • rewrite from scratch or patch current?
    the patched version: https://github.com/draenog/PLDbuilder
  • look into fedpkg?
  • git clone on carme should use –reference to local git copy to save diskspace
• pld builder automation (building, auto tagging)
  minimal changes required; see: https://github.com/draenog/pld-builder.new
• restricting who can delete tags (auto-xxx tags should not be altered by humans)
• $log: keyword support for .spec, also $Revision: and $Date:
  • $Revision and $Date dropped
  • %changelog generated by git log in builder script (implemented)
  • ciabot, mailinglist notification, commit acl rules

Adoption of new commit message formatting rules

• http://tbaggery.com/2008/04/19/a-note-about-git-commit-messages.html

cvs up

git pull --rebase

cvs ci file.foo

git add file.foo; git commit; git push

cvs ci

git commit -a; git push

grep -r oauth_ SPECS

cvs ls packages

git log --pretty=format: --name-only --diff-filter=D

git log -- FILE

git show HASH:FILE

Other Readling:

• Using Fedora GIT

commands that work and have alternatives

./builder -g PACKAGE

fedpkg co PACKAGE; cd PACKAGE; fedpkg sources

./builder -ba PACKAGE

fedpkg local

./compile.sh PACKAGE

fedpkg compile --short-circuit

./repackage.sh PACKAGE

fedpkg install --short-circuit

./check-unused-files.py PACKAGE

fedpkg unused-patches

cvs diff

fedpkg diff

./builder -bs PACKAGE

fedpkg srpm

rm -f *; cvs up

fedpkg clean

• switch _builddir, _srpmdir and _rpmdir to packages/NAME when invoked from fedpkg?