bsdtar POLA change in 10.0?

[Mark Felder]

On Thu, Mar 6, 2014, at 7:46, Marat N.Afanasyev wrote:
> Mark Felder wrote:
> >
> >
> > On Thu, Mar 6, 2014, at 7:12, Ronald Klop wrote:
> >> On Thu, 06 Mar 2014 14:00:22 +0100, Mark Felder <feld at freebsd.org> wrote:
> >>
> >>>
> >>>
> >>> On Wed, Mar 5, 2014, at 6:50, Marat N.Afanasyev wrote:
> >>>> I wonder why
> >>>>
> >>>> bsdtar --one-file-system
> >>>>
> >>>> suddenly started to skip archiving of mount-points? And no mention of
> >>>> such behaviour change either in UPDATING or errata :(
> >>>>
> >>>
> >>>       --one-file-system
> >>>               (c, r, and u modes) Do not cross mount points.
> >>>
> >>> Isn't that what it's supposed to do? Skip any mounted filesystems? Was
> >>> it broken before? Or do I simply not understand the behavior you're
> >>> seeing.
> >>
> >> The question is if the mountpoint (directory in the 'current' filesystem)
> >> itself should be tarred. After untarring you can than make the same
> >> mounts
> >> again, for example.
> >>
> >
> > So you're saying that if I have a disk mounted at /mnt/disk and my tar
> > with --one-file-system includes /mnt, the empty directory /mnt/disk will
> > not be in the tarball?
> 
> Yes, your /mnt/disk will not be in archive, while older version 
> certainly pick this directory
> 

After some minor sleuthing I wonder if this commit could be the culprit

https://github.com/libarchive/libarchive/commit/3439b41d9252372753dc8ca2c7b3e6c0dcfc2874

I don't have the free time to test further at the moment, but attempting
to reverse that commit on the libarchive in /usr/src/contrib and then
rebuilding bsdtar would be an interesting experiment.