[cvsnt] Betr.: CVSNT to CVS
a.krey at gmx.de
Mon Apr 13 20:36:27 BST 2009
On Thu, 09 Apr 2009 14:50:48 +0000, Tony Hoyle wrote:
> That's not how auditing works.. you need to audit everything. In some
> industries that's a legal requirement. That means right down to the
> diffs. You don't get to decide what stuff is worth storing.
If you don't want me to decide when I'm doing a 'git push' or 'cvs commit'
then you need to make regular snapshots of my sandbox, and also need to
monitor whether I ever pass around source files outside revision control.
(On the corp-git isssue: It would be thinkable to have a special version
of git that does all the auditing and passes it on to a central logging
server. Of course for the paranoid it needs to be made incompatible with
the regular git protocol and repository format.)
On the binary part: I did a pretty interesting experiment. I have a
lot of revisions of some software that is compiled into a set of
installable packages. Those contain the already-compressed binaries.
If I pack the set of packets of each release individually into a
tar file, they are hardly compressible. If I tar together all individual
tar files and zip the result, where is still hardly a saving.
If I commit all versions of the result directory into git, the resulting
repo is also the expected 400MB in size. However, after a 'git repack -d'
it shrinks to a mere 51 MB. Somehow git seems to be rather smart when it
comes to compress binary content revisions. I need to add that none of
the stored revisions of the files are identical; each build yielded different
files, however slightly (different build info in each package file,
and also in the contained and compressed binaries).
More information about the cvsnt