[Savannah-hackers-public] Re: mirroring cvs to git *on* savannah

[Jim Meyering]

Sylvain Beucler <address@hidden> wrote:
> If you make changes to the installation, please update ~/infra/cvs.txt
> (check ~/infra.README along the way :)).

Will do.
>
>> I expect to run a root cron job that su's to nobody.
>> It will maintain an rsync'd copy of each repository in
>> /var/tmp/c2g-mirror, as well as the resulting .git repo.
>> The repo sizes are negligible compared to available space (380GB):
>
> The available space is not the only factor to take into account -
> backups come to mind :) But indeed disk space should not be a problem
> here.

These directories may be marked as "no backup desired":

  /vservers/cvs/srv/git/emacs.git
  /vservers/cvs/srv/git/gnulib.git

>> The .git repository sizes:
>>
>>      13M  gnulib/.git
>>     203M  emacs/.git
...
> At a time, I wondered if the Git people had a recommended procedure
> for this kind of mirroring (eg had a ready-to-use tool). Is that the
> case, or are you using something custom?

A script called git-cvsimport does almost all of the work.
My scripts encapsulate that a little more.

> I don't like the fact that files are owned by nobody (other nobody
> processes may then mess with them). Maybe you could create special
> users that are part of the project group, as in the CVS rsync
> replication script (such as user libcrsync).

Good point.
I think it should all work fine with a single role-based owner.
I'll create/use a user named something like cvs2git.

Currently, I have everything (except the cron jobs -- doing it manually
for now) set up in the top level vserver.  Do you think it is worthwhile
to move the process into the cvs one?