Re: [Duplicity-talk] Preventing access time modification?

[Charles Duffy]

Peter Schuller wrote:
> One problem with changing it *back* is that you inherently introduce a
> race condition with other processes. If we allow for applications
> depending on atime in non-obvious ways, the potential consequences of
> atime moving *backwards* as opposed to unnecessarily forward could
> potentially be worse.
>   

So do like rsync does: Create the file under a hidden, temporary name; 
change the time back; and then, when you're done, move it to its final 
name as an atomic operation.