[cvsnt] Repository Robustness

[Tony Hoyle]

Shawn Haigh wrote:
> I was just having a discussion with some colleagues about the scenario
> in which a repository would become corrupt. First of all, based on
> everyone's wide range of experience have you ever heard of this
> happening? Second, what would be the best way to keep a running backup
> of the repository.
> 
Assuming you're using a standard client/server configuration, it's 
pretty hard to corrupt a repository unless there's a corruption of the 
disk itself.  All file file level operations are atomic, so even if 
there's a power failure your repository will still be OK (you might have 
a partial commit, but I don't consider that 'corrupt' as you can recover 
it by doing an update/commit on the client).

Over file shares it's a lot more muddy - there have been reports of 
corruption using them, although rare.  I don't recommend that 
configuration for that reason (amongst others).

One scenario that could be called 'corrupt' is sharing sandboxes - 
checkout on an NT machine and checkin on a Unix machine.  You'll get the 
CR/LF pairs stored in the RCS file and the next checkout on NT will 
convert that to CR/LF/LF, etc.  The rule of thumb is don't share 
sandboxes across platforms (but if you must, never mix clients).

Tony