[Pkg-sysvinit-devel] Bug#648433: Handling of non-conffiles

Roger Leigh rleigh at codelibre.net
Sun Jan 29 20:29:55 UTC 2012


On Sun, Jan 29, 2012 at 11:27:07AM +0100, Julien Cristau wrote:
> On Wed, Jan 11, 2012 at 21:19:15 +0000, Roger Leigh wrote:
> 
> > On Wed, Jan 11, 2012 at 06:07:21PM -0200, Henrique de Moraes Holschuh wrote:
> > > On Wed, 11 Jan 2012, Roger Leigh wrote:
> > > > One problem mentioned in a few bugreports is the handling of
> > > > updates to /etc/default/rcS.  Because it's not a conffile, we
> > > > don't provide updates to it.  While there are good reasons for
> > > > this approach, I was thinking of ways of improving the situation.
> > > > 
> > > > While we don't want to lose any user customisation of this file,
> > > > would it make sense to update the file to the latest version if
> > > > it hasn't been modified?  We can do this quite simply: take the
> > > 
> > > Better check ucf, no reason to reinvent the wheel...
> > 
> > I've attached a patch to do this.  I haven't used ucf before, and
> > didn't realise it was equivalent.  Is this OK with you?
> > 
> initscripts is priority required, ucf is standard.  I don't think this
> wishlist bug is a good reason to make the Essential set larger.

While the bug was originally filed as only wishlist, this does solve
a rather more important issue: any change to the defaults would never
take effect on upgrade; it would require a fresh install, because it's
only set once on initial install.

With the changes to e.g. RAM* for the /run transition, this introduces
a rather large discrepancy between wheezy upgrades and new installs,
which using ucf solves.  I originally wrote this using sha256 hashes
of all previous conffiles and implementing ucf-like behaviour directly
in the maintainer script, but it was (rightly) suggested that
reimplementing ucf was suboptimal, and that I should be using ucf
directly.

Another alternative would be to make /etc/default/rcS a conffile.  The
only niggle is that this file is modified by the installer, which is
not ideal.  However, this would make the use of ucf redundant, if we
can find a way to make this possible.


Regards,
Roger

-- 
  .''`.  Roger Leigh
 : :' :  Debian GNU/Linux             http://people.debian.org/~rleigh/
 `. `'   Printing on GNU/Linux?       http://gutenprint.sourceforge.net/
   `-    GPG Public Key: 0x25BFB848   Please GPG sign your mail.



More information about the Pkg-sysvinit-devel mailing list