Re: [Bug-tar] tar keeps reporting "file changed as we read it" on random files

[Mark O'Keefe]

Thanks for the comments. What about a directory? Part of the logic here was 
looking for and ignoring directory change when deleting files I think. Should 
it use mtime for directory only in that case? Is the mtime versus ctime 
security concern still relevant for a directory? Can you elaborate on the 
concern?

I've reverted to standard tar to find the case again.  When I was originally 
debugging this I was seeing ctime change by some small amount (can't remember 
the values) and there was nothing accessing the system other than McAfee on 
access scanner maybe. It was a serial job that copied files to a location then 
ran tar, so there would be nothing else interacting with the files.

Hopefully I can reproduce the issue again and then get better information on 
why it did this.

The option to ignore change is useless as tar still exits with an error. This 
will kill a build pipeline and the effort to determine that the error is for 
something you are ignoring is significant and sometimes even impossible. Maybe 
an option to have tar not exit with error for this case, not just ignore the 
error and exit with an error code...



Cheers,
Mark.

> On 3 Dec 2015, at 3:30 AM, Paul Eggert <address@hidden> wrote:
> 
>> On 12/01/2015 11:19 PM, Pavel Raiskup wrote:
>> FYI:https://lists.gnu.org/archive/html/bug-tar/2015-10/msg00021.html
> 
> In particular, see:
> 
> https://lists.gnu.org/archive/html/bug-tar/2015-10/msg00025.html
> 
> Most likely this is a bug in Mark's file system, not in tar.  Tar can't rely 
> on mtime here, as applications can modify mtime to whatever they want.
>