[Pkg-uml-devel] Re: [Pkg-uml-pkgs] Bug#388076: [Pkg-uml-commit] r227 - in trunk/src/rootstrap: debian modules

[Stefano Melchior]

On Wed, Oct 04, 2006 at 12:30:35PM +0200, Mattia Dongili wrote:
Dear all,
> On Wed, October 4, 2006 10:38 am, Stefano Melchior said:
> [...]
> > -  * add amd64 architecture, build and unstable rootfs image creation
> > tested.
> 
> please remove also the build-architecture from the debian/control file.
ops, I forgot, but I shall do it.
> 
> > +  * add uml IP, hostname and domain in /etc/hosts (Closes: #388076)
> >    * correctly write the timezone file in the guest FS (Closes: #388055)
> >    * try to print descriptive error messages when a module fails
> >      (Closes: #388061)
> >
> > Modified: trunk/src/rootstrap/modules/network
> > ===================================================================
> > --- trunk/src/rootstrap/modules/network	2006-10-03 18:47:11 UTC (rev 226)
> > +++ trunk/src/rootstrap/modules/network	2006-10-04 08:38:50 UTC (rev 227)
> > @@ -22,16 +22,20 @@
> >  echo "netmask=$netmask" >> $CONF
> >  echo "gateway=$gateway" >> $CONF
> >
> > +echo "127.0.0.1  localhost " > /etc/hosts
> > +
> > +if [ -z "$hostname" ]; then
> > +  hostname=rootstrap
> > +fi
> > +
> > +hostname "$hostname"
> > +
> >  if test -n "$domain"; then
> >      echo "domain $domain" > /etc/resolv.conf
> > +    echo "$uml $hostname.$domain $hostname" >> /etc/hosts
> >  fi
> 
> I'm not very happy with this implementation (as I wasn't with the
> suggestion in the bug report):
> 1. $uml can also have the value "dhcp" which would break /etc/hosts.
> 2. how does this work with a DHCP guest anyway?
> 3. it sucks that we are doing something that may be unwanted (eg: I
> wouldn't want it), IOW make it _at_least_ configurable.
Possible idea (for Etch+1): a second stage for configuration where you can
decide how to config something that at the moment is not
mandatory/essential for the guest.
For instance, you need to configure the $uml $hostname.$domain $hostname
for Apache (that is only a possible example), then you may need additional
(and useful) parameter.
Shall we need an extra module?
What do you think?
> 
> Anyway this bug would be better dealt with within a larger picture, for
> example support for config file templates allowing to build multiple root
> images, something I'd like to add in the near feature. Just to give an
> idea I'd like to be able to tell rootstrap to create images for the whole
> 10.1.1.0/24 subnet (and even with the uml COW support).

I definitively agree, at the moment my test is not strong enough .
It is a nonsense to add in Etch's rootstrap.

Among the TODOs, do you think it would be good to add code to debug python
and to make the process a bit more verbose?
Let's discuss later the latter.

SteX

-- 
Stefano Melchior, GPG key = D52DF829 - <stefano.melchior at openlabs.it>
http://www.openlabs.it/~stex    --     <stefano.melchior at openlabs.it>
http://etinarcadiaego.dyndns.org  --  <ste at etinarcadiaego.dyndns.org>
Skype ID "stefanomelchior"
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 307 bytes
Desc: Digital signature
Url : http://lists.alioth.debian.org/pipermail/pkg-uml-devel/attachments/20061004/135df959/attachment.pgp