Re: [gentoo-dev] GLEP 27: Revisited (aka dynusers/creandus)

[Mike Kelly]

On Tue, 03 Oct 2006 08:09:08 -0400
Chris Gianelloni <wolf31o2@g.o> wrote:

> > I have a list of all packages in the tree that currently use either
> > of those functions[2]. If you maintain one of these packages, I'd
> > especially appreciate your feedback.
>
> You missed games-* (yes, all of them) via the games.eclass, but I'm
> sure there's a couple more eclasses that do user/group modification.

Oops, I forgot to account for eclasses. I'll redo my script and run it
again later to account for that.

> What about applications that aren't tied to a profile?  How do they
> work?

Well, user management is inherently tied to the userland which is being
used, which is in turn tied to the profiles (default-linux,
default-bsd, etc).

Settings which are userland-agnostic (like default uids, member groups,
GECOS comment fields), would be in the settings for the base/ profile.

> Doesn't this increase the size of the profiles pretty
> dramatically?

I don't think it makes that much of a size difference. Most of this
information is now duplicated over many enewuser/group lines in many
different ebuilds. With this system, ebuilds just need to have a EUSERS
and EGROUPS variable defined listing the users/groups needed.

> Does it need to be tons and tons of small files, or can
> we get away with a set of larger files with some sort of header?
>
> As you can see, this changes very little, but reduces the number of
> small files in the portage tree.  Is this necessary?  Who knows?  Will
> it makes syncs slightly faster? Not a clue.  I'm just throwing out an
> idea.

Hmm, parsing that would be a more difficult task for my scripts (which
are just basic bash code doing some greps). Also, it seems like the many
small files are easier to maintain. I don't know enough about rsync to
know how it would affect efficiency, though. It's something I'll try
and look into further.

> Anyway, this looks really good.  =]

Thanks!

-- 
Mike Kelly