Re: [Bug-tar] Apply ignore-failed-read to "file changed as we read it" case

[Stefanos Harhalakis]

On Wed, Jan 6, 2016 at 6:58 AM, Pavel Raiskup <address@hidden> wrote:
> Anyways, it would be fine to have rather a better option for it, something
> similar to "--warning", but the effect would be
> dont-switch-to-nonzero-status.

I agree. If it's an option to implement a new option, then that would
be great. Alternatively, see below.

> Other than that;  If the file have shrunk/grew, it _is_ serious issue from
> the consistency POV!  And returning non-zero status is IMO really worth.
> I can imagine that in really *rare* situation you can expect that the
> archived file may be restored a safe way.  And to me, --ignore-failed-read
> ignores way too much (orthogonal) situations.

I'm the author of vbackup, a tool for easing system backups and one of
the methods used is tar. The tool attempts to detect whether tar has
failed or not. I can tell you from my experience that it is
practically impossible to setup automated backups of a running system
without such an error, because there are always files that change and
for which you may have no interest. I understand that this may result
in the files not being backed up properly, but there needs to be an
option that ignores single file errors. Otherwise it's impossible to
auto-detect more critical failures (e.g. out of space).

This bug report came as a request from exactly that: A user of vbackup
reporting tar backups failing because files were changing underneath.

My workaround at this point is to ignore exit code 1 from tar. If you
think that this is a better option then can you consider documenting
exit codes 1 and 2 in the man page?  I.e. to document PAXEXIT_DIFFERS
and PAXEXIT_FAILURE, so that a user of tar can chose to ignore
PAXEXIT_DIFFERS.

Thanks,
Stefanos