Re: Re: Perfomance problem backupscript (more details and tests)

On Thu, 2005-10-06 at 13:05, [email protected] wrote:

> ctime is indeed the natural choice for a name change 
> indicating timestamp. The only problem is that it 
> always was so clearly designated to have a different
> meaning.

> > not adjusting it in rename is broken. 
> 
> The lack of a name change indicating timestamp is a
> birth defect of Unix.

No, unix originally required link() and unlink() operations
to create a new name and remove the old one. Each changed
the inode link count, and thus ctime().  I think the atomic
version of rename may have appeared in SCO's Xenix although
it did not break the filesystem semanatics - it just did both
operations with the inode locked.  Linux variations are the
ones that first ignored the side effects of shortcuts.

-- 
  Les Mikesell
   [email protected]