Re: SV: Different kind of access methods

[Larry Jones]

Richard Cobbe writes [quoting me]:
> > 
> > Doing :local: access to an nfs/smb/samba mount repository frequently
> > doesn't work reliably at all, which is why it is discouraged. 
> 
> Why is this?  In particular, what sort of problems can one encounter when
> using this sort of system?  (Is it perhaps related to the lockfiles CVS
> uses to prevent multiple people from modifying the repository at the same
> time?)

There is a potential locking problem with NFS (although no one has ever
reported it actually occurring in practice), but it only causes denial
of service -- a client can create a lock without knowing it so the lock
will never be cleared -- not loss of data.  The more serious problem is
that most implementations are buggy.  The bugs tend to be
interoperability bugs -- as long as all the machines are running the
same code, everything works fine, but when you try to mix different
systems things start to go wrong and both vendors insist that the
problem must be with the other system.

If the bugs cause immediately noticeable errors, then it's annoying
(perhaps to the point of unusability), but at least you don't lose any
data.  Unfortunately, it's not uncommon for the bugs to instead cause a
chunk of data in the middle of a repostory file to be replaced with NUL
bytes.  Because of the way the RCS file format works, it's entirely
possible that you won't notice the corruption for quite some time and
there's a good possibility that when you do notice it, you'll no longer
have any backup copies old enough to recover the lost data.

-Larry Jones

Hey Doc, for 10 bucks I'll make sure you see those kids in the
waiting room again real soon! -- Calvin