[Ffmpeg-devel] SVN dump

[Ivan Kalvachev]

2007/4/15, Loren Merritt <lorenm at u.washington.edu>:
> On Sat, 14 Apr 2007, Andy Parkins wrote:
>
> > The most interesting point is the sizes of the repositories,
> > compressed subversion dump file     159M
> > subverion repository                124M
> > packed git repository                19M
> >
> > That's the entire subversion history in 19M.  That's a much more
> > manageable size for anyone who wanted the ability to fork.
>
> Some more points I found interesting:
>
> 805867412 ffmpeg-svn-dump
> 165832859 ffmpeg-svn-dump.gz
>    4601172 ffmpeg-svn-dump.lrz
>
>   91566080 ffmpeg-svn.tar
>   21777932 ffmpeg-svn.tar.gz
>   16936128 ffmpeg-svn.tar.lrz
>
>   17766400 ffmpeg-git.tar
>   16747134 ffmpeg-git.tar.gz
>   15984373 ffmpeg-git.tar.lrz
>
> I wonder what svn-dump does that makes it so much more compressible than
> the repository itself, and can git do the same?

Simple, svn-dump dumps the files and revisions in full form. No diffs
or anything.
The diffs could be considered form of compression (redundancy
removal), and compressing compressed material usually doesn't give
good results.

My guess is that lrz uses quite big search buffer and that allows
finding the revisions of the same files. gzip uses something like 64kb
(i think compressed) and bzip2 can have up to 1MB (source data). Well,
they use different algorithms but still.