Evolution of version control in opensource - fossa2010

Evolution of Version
Control in Open Source
Lessons learned along the path to distributed version control

Chris Aniszczyk (Red Hat)
Principal Software Engineer
zx@redhat.com
http://aniszczyk.org

About Me
I've been using and hacking open source for ~12 years
- contribute{d} to Gentoo Linux, Fedora Linux, Eclipse

Eclipse Board of Directors, Committer Representative

Member in the Eclipse {Architecture,Planning} Council

I like to run! (just finished Chicago marathon in 3:20)

Co-author of RCP Book (www.eclipsercp.org)

Evolution of Version Control in Open Source | © 2010 by Chris Aniszczyk

Agenda
History of Version Control (VCS)
The Rise of Distributed Version Control (DVCS)
Lessons Learned at Eclipse moving to a DVCS
Conclusion
Q&A

Picture 5


Version Control
Version Control Systems manage change

“The only constant is change” (Heraclitus)


Why Version Control?
VCS became essential to software development because:

They allow teams to collaborate
They manage change and allow for inspection
They track ownership
They track evolution of changes
They allow for branching
They allow for continuous integration


Version Control: The Ancients
1972 – Source Code Control System (SCCS)
Born out of Bell Labs, based on interleaved deltas
No open source implementations as far as I know

1982 – Revision Control System (RCS)
Released as an alternative to SCCS
Operates on single files
Open source implementation hosted at GNU


Version Control: The Centralized
One centralized server with the revision information

Clients checkout a working copy locally

Most operations happen on the server

Linear revision history


Version Control: The Centralized
1990 – Concurrent Versions System (CVS)
Initially released as some scripts on top of RCS
Made branching possible for most people
Revisions by commits are per file :(
No atomic commit :(
Not really maintained anymore...

2000 – Subversion (SVN)
Released as an improvement to CVS
Atomic commits via transactions
Open source implementation hosted at Apache


“Hey, get back to work!”
… “My code's merging” - remember those days you
spent merging in changes in CVS/SVN?

Picture 5


Version Control: The Distributed
Every client has a copy of the full repository locally

All repository operations are local (except sharing)

Intelligent network operations when sharing content

A very non linear revision history

Large online communities to share changes


Version Control: The Distributed
2001 – GNU arch
First open source DVCS
Deprecated; not maintained anymore

--- In 2005, Bitkeeper was no longer open source ---

2005 – Git
Created as the SCM for the Linux kernel by Linus

2005 – Mercurial (Hg)
Cross-platform DVCS

2007 – Bazaar (BZR)
Sponsored by Canonical

Agenda
- How does a DVCS work?
- The benefits of a DVCS
Conclusion
Q&A

Picture 5


How does it work?
A DVCS generally operates at the level of a changeset

Logically, a repository is made up from an initial empty
state, followed by many changesets

Changesets are identified by a SHA-1 hash value

e.g., 0878a8189e6a3ae1ded86d9e9c7cbe3f


It's all about the changesets

Changesets contain pointers to the previous changeset

previous: 48b2179994d494485b79504e8b5a6b23ce24a026
--- a/README.txt
+++ b/README.txt
@@ -1 +1 @@
-SVN is great
+Git is great

previous: 6ff60e964245816221414736d7e5fe6972246ead
--- a/README.txt
+++ b/README.txt
@@ -1 +1 @@
-Git is great
+SVN is great


Branches
The current version of your repository is simply a pointer
to the end of the tree

The default "trunk" in Git is called "master"

The tip of the current branch is called "HEAD"

Any branch can be referred to by its hash id

Creating branches in a DVCS is fast, you simply point to
a different element in the tree on disk already


Merging
DVCS are all about merging

Merges are just the weaving together of two (or more)
local branches into one

However, unlike CVCS, you don't have to specify
anything about where you're merging from and to; the
trees automatically know what their split point was in the
past, and can work it out from there.

Merging is much easier in a DVCS like Git


Pulling and Pushing
We've not talked about the distributed nature of DVCS

Changes flow between repositories by push and pull

Since a DVCS tree is merely a pointer to a branch...

There's three cases to consider for comparing two trees:
• Your tip is an ancestor of my tip
• My tip is an ancestor of your tip
• Neither of our tips are direct ancestors; however, we
both share a common ancestor


Cloning and Remotes (git)
git clone git://egit.eclipse.org/egit.git

Where you can push or pull to is configured on a per
(local) repository basis

git remote add github http://github.com/zx/myegit.git

origin is the default remote; you can have many remotes


Software Trends and Revolution
Most major open source projects use some form of DVCS

Git, Hg, Bazaar

Linux
MySQL
OpenJDK
Android
JQuery
Gnome
Fedora
Bugzilla and so on...

But why?
Using Git in Eclipse | © 2010 by C. Aniszczyk and M. Sohn

Benefits of Distributed Version Control
Can collaborate without a central authority

Disconnected operations

Easy branching and merging

Define your own workflow

Powerful community sharing tools

Easier path to contributor to committer


Collaboration
Developers can easily collaborate directly without
needing a central authority or dealing with server
administration costs


Disconnected operations rule!
Developers can still be productive and not worry
about a central server going down... remember the
days of complaining that CVS was down and you
couldn't work?

Also, there's a lighter server
load for administrators!


Branches everywhere
Creating and destroying branches are simple
operations so it's easy to experiment
with new ideas

Very easy to isolate changes


Define your own workflow
Define your own workflow to meet your team needs.
Different workflows can be adopted as your team
grows without changing VCS toolsets!


DVCS and Building Community
Developers can easily discover and fork projects. On
top of that, it's simple for developers to share their
changes

“Distributed version control is all about empowering
your community, and the people who might join your
community” - Mark Shuttleworth


Agenda
- Version control at Eclipse
- Code review at Eclipse
- Challenges in moving to a DVCS
Conclusion
Q&A

Picture 5


Version Control at Eclipse
Eclipse defined a roadmap to move to Git in 2009
CVS is deprecated; SVN will be deprecated in the future

EGit is an Eclipse Team provider for Git
http://www.eclipse.org/egit/

JGit is a lightweight Java library implementing Git
http://www.eclipse.org/jgit/

The goal is to build an Eclipse community around Git

So why did Eclipse.org choose Git?


#1: Git-related projects at Eclipse.org
… both the core Git library (JGit) and tooling (EGit)
are actively developed at Eclipse.org by a diverse set
of committers and contributors with a common goal

Picture 5


History of JGit and EGit
2005 Linus Torvalds starts Git

2006 Shawn Pearce starts JGit

2009 Eclipse decides roadmap for Git migration
JGit/EGit move to eclipse.org
SAP joins JGit/EGit

3/2010 Released 0.7 (first release at Eclipse)
Diff/Merge Algorithms, Automatic IP Logs

6/2010 Released 0.8 (Helios)
Usability Improvements, Git Repositories View, Tagging

9/2010 Released 0.9 (Helios SR1)
Merge, Synchronize View, .gitignore

Planned: 12/2010 0.10 (Helios SR2) 3/2011 0.11 6/2011 1.0 (Indigo)


#2: Git is fast
… Git is fast and scales well

Picture 5
*whyisgitbetterthanx.com


#3: Git is mature and popular
… Git is widely used and is the most popular
distributed version control system

Picture 5


#4: Git community tools
… the Eclipse community is interested in taking
advantage of such Git tools like Gerrit Code Review
(used by the Android community) and GitHub

Picture 5


Roles at Eclipse and Code Review
Committer
Formally elected
Can commit own changes without review

Contributor
Small changes
reviewed by committers
Bigger changes
also formal IP review by legal team
in separate protected Bugzilla (IPZilla)

Review Tool
patches attached to bug in Bugzilla
comments in Bugzilla

Code Review via Bugzilla


Eclipse – Review Process
Contributors
• create patch using CVS, SVN, Git (since 2009)
• attach patch to bug in Bugzilla

Committers
• do code and IP review
• comment, vote in Bugzilla
• create CQ for changes needing IP review
• commit accepted changes

IP Team
• does IP review bigger changes from contributors


Eclipse – Review Process
Review not done for all changes

Each Eclipse.org project does it differently

Review tedious for contributors
(and also for committers mentoring contributors)


Gerrit Code Review

Gerrit is a Code Review system based on JGit
http://code.google.com/p/gerrit/

Also serves as a git server
adding access control and workflow

Used by
• Android https://review.source.android.com/
• JGit, EGit http://egit.eclipse.org/r/
• Google, SAP, …

Eclipse wants to use it …


History: Google and code review tools
Mondrian (Guido van Rossum)
• based on Perforce, Google infrastructure
• Google proprietary

Rietvield (Guido van Rossum)
• based on Subversion
• Open Source hosted on GoogleApp Engine

Gerrit (Shawn Pearce)
• started as a fork of Rietvield
• based on JGit and GWT
• Open Source (Android)
• Apache 2 license


One Branch One Feature
Master
branch
contains
only
reviewed
and
approved
changes
• master
moves
from
good
to
be7er
state
a)er
each

(approved)
change

Each
feature
branch
is
based
on
master
branch
• stable
star6ng
point

A
change
can
really
be
abandoned
because
• no
other
approved
change
can
depend
on
a
not
yet

approved
change
• Gerrit
will
automa6cally
reject
a
successor
change
of
an

abandoned
change


Gerrit – Lifecycle of a Change
•
create local topic
topic
branch
master
•
commit change
1 •
push it for review
a •
do review
•
automated
verification


•
create local topic
topic branch
master
•
commit change
•
push it for review
1 •
do review
•
automated
a
verification

topic master

c
•
refine based on
3 review
•
push new patchsets
b
2 until review votes ok

1 a


•
create local topic master
topic branch
master •
commit change
•
push it for review d
topic
1 •
do review
a
•
automated
verification c
3

b
topic master 2

c
•
refine based on a
1
3 review
•
push new patchsets
b until review votes ok
2
•
Submit may lead to
server-side merge
1 a •
or merge / rebase before
push

Gerrit Workflow


Eclipse.org: Challenges moving to a DVCS
Convincing management and peers was tough
- At first, everyone is resistant to change

The learning curve of DVCS systems is high
- Initially, the Eclipse tooling was “alpha”
- People refuse to drop to the CLI

Legacy is a pain in the ass!
- 200+ projects at Eclipse used CVS/SVN
- The existing VCS tooling was high quality

Picture 5


No free lunch!
… trust me, the only way to learn DVCS is to start using
it... there is a learning curve, you need to rewire your
brain!

Picture 5


Git Resources
http://git-scm.com/documentation is your friend

Watch Linus' talk at Google
http://www.youtube.com/watch?v=4XpnKHJAok8

Read the Pro Git book - http://progit.org/book/


Conclusion
The future of version control is distributed!

Moving to a DVCS takes time

Gerrit enables a nice code review workflow

Open source has embraced the way of DVCS


Q&A

Picture 5


Evolution of version control in opensource - fossa2010

More Related Content

What's hot

Viewers also liked

Similar to Evolution of version control in opensource - fossa2010

More from fOSSa - Free Open Source Software Academia Conference

Recently uploaded

Evolution of version control in opensource - fossa2010