No libc shared lib number bump ?

Alexander Leidinger Alexander at Leidinger.net
Fri Nov 9 09:22:14 PST 2007


Quoting "David O'Brien" <obrien at freebsd.org> (Fri, 9 Nov 2007 08:28:43 -0800):

> On Fri, Nov 09, 2007 at 04:43:01PM +0100, Alexander Leidinger wrote:
> > Quoting Daniel Eischen <deischen at freebsd.org> (Fri, 9 Nov 2007 09:54:46 -0500 (EST)):
> > > On Fri, 9 Nov 2007, Alexander Leidinger wrote:
> > > > I'm curious, why do we need to reset it back to .0?
> > > 
> > > We don't have to.  It would just make things clearer to have all
> > > versioned symbol libraries with the same version number since
> > > they shouldn't ever have to be bumped again.  Solaris has all
> > > their libraries at .1.  We've already used .1, but .0 has never
> > > been used.  obrien suggested it, and it seems to make sense
> > > to me.
> > 
> > So it's just "cosmetics"...
> 
> It also clearly denotes the lib is symbolized.  Years from now, (if not
> today) its hard to remember which .so numbers relate to which FreeBSD
> releases.  So I would call it clarity instead of cosmetics.

What would you need the version number of... let's say FreeBSD4.x for,
now (= what's the scenario where you need this information)? If you
just want to know if it is a lib which is versioned or not, it's easy:
if it is the same version number as "currently" (at this time) in use,
then it is versioned (as the version number is not supposed to change
anymore).

> > What we gain in not doing is, is that users of those libs don't have to
> > recompile all ports. Compared to the number of FreeBSD installations in
> > total the number of affected users are small, but those are the users
> > which help us debug -current (and ideally "all" (sort of)
> > src-committers). I think those people have more interesting things to
> > do than to recompile everything.
> 
> When things like large Xorg or GNOME or KDE changes hit the Ports
> Collection, one already has to practically recompile everything.. except
> for figlet and jive.

You talk about the use as a desktop system. I have 11 of my 12 jails on
my 3 -current machines where no package depends upon any Xorg port
(WITHOUT_X11=yes) is installed (server jails, one as a MTA, one as a
IMAP server, one as a nntp server, one as a samba server, ...). The 12th
jail actually contains my desktop. So for my laptop and my desktop
jail your assumption is true, for the remaining 13 of 15 instances of
-current I have locally, it is not true. Normally I generate packages
on those systems, so for the 2 instances where I use gnome, the amount
of work could be reduced, but for the server jails there's not that
much in common. This is just my setup and doesn't cover the use of
other people. Both of us don't know what the real use of the other
people is.

If you coordinate that the version change happens at the same time when
GNOME and KDE are updated _at the same time_, then you cover the
situations where it would have the biggest impact. As it seems you want
to do it for 7.0, you missed this opportunity already, a lot of people
already updated GNOME/KDE (I haven't, but it's not about me). And as we
see on the lists, people start to test the betas. Some of them may
track the way to the release via a fresh install, some via updates. For
the ones which update, you force a complete rebuild.

Again, I don't object to the change, I just throw in some thoughts for
food. If no critical amount of ports break in a experimental ports
build (unfortunately this only covers build time problems, but not run
time changes) and you absolutely want to have this, feel free to go
ahead. As already said, I don't hold you back, I just give you some
input to think about.

Bye,
Alexander.

-- 
http://www.Leidinger.net  Alexander @ Leidinger.net: PGP ID = B0063FE7
http://www.FreeBSD.org     netchild @ FreeBSD.org  : PGP ID = 72077137


More information about the freebsd-current mailing list