Hello fellow Indico, Invenio, and Inspire developers:
As you may know, our CVS repository is hosted on an aging machine that
we are progressively moving away from. The forthcoming Invenio 0.99.0
release will provide us with a very good opportunity to migrate the
CVS repository onto a more recent hardware.
Moreover, we can take advantage of this move in order to possibly
upgrade our SCM system too. We started to discuss such an upgrade in
2004 already, but we decided to stick to CVS at that time, the CVS
being CERN-wide SCM. I think the time has come to reevaluate the
options once again.
Here is a short overview of our principal options:
Centralized SCM paradigm:
1. CVS
- moving to CVS hosted on another box is an obvious candidate
- nothing changes for developers; we only need to obtain firewall
opening for the new machine
- but we would be stuck again with all CVS problems that are often
as basic as proper file renaming
2. Subversion (svn)
- natural evolution of CVS centralized SCM paradigm
- adopted by many big projects since years
- small adaptation/learning curve needed
- CERN might be deploying a central SVN service in 2008, so we may
eventually use this central service later (this can be seen both
as a plus and as a minus)
- but merging capabilities not really much better than with CVS
- but tagging capabilities can even be worse than with CVS
Distributed SCM paradigm (may fit better our more and more distributed
development environment):
3. Git
- adopted by many big projects (besides Linux kernel)
- featureful and powerful
- but also more complex; possibly steeper learning curve from CVS
- but not running very well on MS Windows (which can be seen both
as a plus and as a minus)
4. Mercurial (hg)
- cleaner and simpler git-like alternative
- possibly smoother transition from CVS
- written in Python, so easy to run everywhere (e.g. it was no prob
to create a recent RPM for the aging SLC4 OS in a minute)
- easy native transport setup over HTTP
- but less powerful than Git
- but less mainstream adoption than Git (but e.g. Mozilla)
5. Bazaar (bzr)
- like Mercurial, written in Python too
- recently became an official GNU project; e.g. GNU Emacs seems to
be going to adopt it to replace its CVS
- used to be much slower than Mercurial, but it's catching up
There are other interesting choices such as 6) Darcs that, as you may
recall, I liked in 2004, but nowadays there are perhaps more
"mainstream" alternatives in 3-5.
Some issues to have in mind:
a) other personal projects: distributed SCM can be used easily for
other local hacks too, so it may be nice to standardize upon one;
can retain a notion of "central" repo
b) transfer channel: HTTP/HTTPS read/write transport channels do not
require official firewall port opening and local users and stuff,
so it's easier to (i) hop on various machines and (ii) to be
"different" from the main CERN SCM tool (currently CVS, soon
probably SVN)
c) integration with IDEs: to take Emacs an example, the native
version control ("C-x v *") supports all these alternatives well,
with Git, Mercurial, and Bazaar natively only from Emacs 23
d) OS dependency: if we host the repo, it should be easy to install
very recent SCM versions on an aging OS such as SLC4
For more musings and pros/cons of various systems, you can check out
e.g. Elijah's recent blog posts:
<http://blogs.gnome.org/newren/2008/03/01/happenings-in-the-vcs-world/>
<http://blogs.gnome.org/newren/2007/11/17/adoption-of-various-vcses/>
Please think of this email as a kick-off to stem some discussion and
compare our preferences...
Best regards
--
Tibor Simko ** CERN Document Server ** <http://cds.cern.ch/>
