[Techtalk] Using CVS to track a CVS tree?

[Almut Behrens]

On Mon, Apr 25, 2005 at 12:14:52PM +0000, Conor Daly wrote:
> This is an interesting one.  A user has a proprietary application which has an
> extensive configuration directory tree.  This tree contains CVS directories
> within it so it's the output of a 'cvs checkout <module>'.  If an issue
> arises with the application, he's instructed by the support team to send a
> tarball of the config area.  I presume they then do a 'cvsdiff' against
> their repository.  He wants to maintain his own repository of these configs
> (for quick restore purposes) but also wants to maintain compatibility with
> the support team.

What exactly does it mean to maintain compatibility with them?  Does he
need to keep the CVS directories in his local tree, referring to the
original repository?  Does he ever need to submit changes to their
repository himself, or would he only ever send them tarball snapshots? 
If so, do those have to contain the original CVS meta info?

In case he needs to keep those CVS dirs, it's probably easiest to use a
different revision management system on his side, in order to avoid
having them interfere.

OTOH, if he desperately wants to use CVS, he could - in theory - patch
the CVS sources to use a different name for the meta info directories
(e.g. MyCVS/), and then use his own special CVS version.  However, it
might turn out to be more involved than one would wish to find the right
places to edit: doing a recursive grep for "CVS" on the CVS source tree
yields 9269 (!) hits -- now you tell me which ones to edit... ;)
Well, it's probably not quite as bad.  With an educated guess, you'll
find that there are some definitions in src/cvs.h, which look like a
good starting point for experiments of this kind...   Anyhow, I'd
definitely do the experimenting only on a copy of the original data.

Cheers,
Almut