Git vs. Subversion. Which one?

[Ralf Corsepius]

On Tue, 2008-09-30 at 17:37 -0400, Todd Zullinger wrote:
> Ralf Corsepius wrote:
> > SVN's major "con" is it being comparatively generous on local
> > diskspace and it polluting a checked out source trees with huge
> > amount of VCS-metadata files (git, mercurial do so as well).
> 
> Git and mercurial both keep their files in one top-level dir, e.g.
> .git or .hg.
Try a recursive grep in a checked-out source tree (grep -R <pattern> .)

This was easily applicable with CVS/RCS, but is hardly applicable with
SVN, Git or mercurial - Certainly, this is nothing serious, nevertheless
it's "nagging to loose a once applicable habit"

>   This doesn't count as "polluting the tree" in my mind.
> It's certainly not as annoying as the "CVS" dirs that CVS puts all
> over my tree.
Does the name of the directory matter? Does the fact that CVS doesn't
hide its directories make a difference?

> Sure, the disk space is higher for a git clone than for a CVS
> checkout, but with git you are getting the entire history of the
> project instead of just one working copy as you do with CVS (or
> Subversion)
Well this doesn't scale well on big source trees (e.g. Fedora's) 
or one with a long history (e.g. GCC's).

Just one figure:

An SVN checkout (from GCC)
# du -s -b gcc-4_3-branch
832684026	gcc-4_3-branch

Size of an uncompressed tarball containing approximately the same
sources (~ size of a hypothetical CVS checkout)
# du -s -b gcc-4.3.2
369871768	gcc-4.3.2

Ralf