[cvsnt] remove(delete) a revision from cvs

Gerhard Fiedler lists at connectionbrazil.com
Sun Jul 29 16:36:03 BST 2007


Glen Starrett wrote:

>>> There is a "cvs admin" command that will let an administrator destroy
>>> your SCM integrity by removing a revision, however it has not been
>>> tested in years and there are various reports that it can harm the
>>> repository.  
>> 
>> Do we have an authoritative list of commands (command/option
>> combinations, ...) that are considered tested and safe to use? If not,
>> is anybody interested in creating one? If so, what would be a good
>> procedure and format?
> 
> Arthur is talking about the issues with deleting versions of binary 
> files -- ...

Maybe he was, maybe not, but it wasn't explicit.

> ... there are problems removing the interim versions with binary deltas
> (and/or compression, I don't recall the specifics).

This is exactly what I'm talking about. It seems there is no place where I
(or any other repo admin) can read up on these specifics. I've had problems
with deleting interim revisions of files stored with non-compressed deltas.
But I don't know any more than this.

> The admin command itself falls into that gray area that is slowly being 
> replaced or deprecated, piece by piece.  

As you said later, 'admin -o' isn't likely to go away, it is dangerous, so
having it fully documented as to what is reasonably confirmed to be working
and what may not work is IMO very important.

> That is the only command that I know of that isn't fully supported (and
> it has the all-caps "THIS COMMAND..." warning on it as well.

This warning says in full: "THIS COMMAND IS FOR CVS ADMINISTRATORS ONLY.
DON'T USE IT UNLESS YOU KNOW EXACTLY WHAT YOU ARE DOING, AND WHAT THE
CONSEQUENCES ARE."

Where I'd like to get to is to be able to know exactly what I'm doing and
what the consequences are when I'm using 'admin -o'. Without the
consequences being documented, this warning is pretty much useless, IMO.

The recommended backups may or may not help -- if you find out a year later
that a certain revision is broken, the backup may already be gone.

> -o is dangerous and destroys the history of your repository (which is
> what it was designed to do).  

If it destroys only the part of the history that I'm telling it to destroy,
that's fine then -- obviously :)  I'm worried about the occasional reports
that it destroyed more than this. It seems nobody has any specifics, and
this is a situation that I'd like to remedy if possible.

Gerhard


More information about the cvsnt mailing list