[pacman-dev] Road to 5.1

[Andrew Gregory]

On 01/10/18 at 09:58am, Eli Schwartz wrote:
> On 01/10/2018 09:29 AM, Andrew Gregory wrote:
> > On 01/09/18 at 09:00am, Dave Reisner wrote:
> >> On Mon, Jan 08, 2018 at 03:26:49PM +1000, Allan McRae wrote:
> >>> Hi all,
> >>>
> >>> We have not had a release in a long, long time.  Lets fix that!
> >>>
> >>> There is a substantial number of bugs tagged for a 5.1 release [1].
> >>> These were added when we were all younger and more optimistic about the
> >>> world...  I have added a 5.2 target, so we can start bumping some to a
> >>> future release.
> >>>
> >>> Please reply here with anything you think is essential to the next
> >>> release - even better if you provide a patch!
> >>>
> >>> Cheers,
> >>> Allan
> >>>
> >>>
> >>> [1] https://bugs.archlinux.org/roadmap/proj3
> >>
> >> Can we fix extraction based on lookups of user/group so that the
> >> nonsense that Arch has been subjected to via systemd-sysusers can be
> >> fixed? This should be a simple matter of calling
> >> archive_read_disk_set_standard_lookup(archive) prior to extraction. I
> >> think agregory has a branch that he's been working on to do this.
> > 
> > I do have a branch to use user/group names, but the change is not
> > simple.  It breaks alternate rootdir usage hard.  I haven't noticed
> > the systemd-sysusers nonsense myself and only caught bits of IRC
> > discussions, can you elaborate on exactly what the problem is?
> 
> users/groups work absolutely fine... just as long as we hardcode uids
> for any files that need to be packaged with a specific user ownership.
> 
> And systemd-sysusers isn't really different from useradd in that respect
> (we still hardcode uids in sysusers files).
> 
> But while trying to move to sysusers, we've hit "interesting" issues
> like this: https://bugs.archlinux.org/task/56818
> It did not used to be a problem, until we started relying on systemd's
> own basic.conf to create the "lp" user rather than hardcoding an id in
> filesystem's /etc/group
> 
> But if we define our own sysuser with a hardcoded uid, then it will
> clash with basic.conf...

So Arch used a static GID, then accidentally switched to a dynamic
GID without the necessary install script update?  Seems easy enough to
fix by adding the relevant chown calls like every other package with
dynamic UIDs/GIDs.