Apologies if I may be repeating information as we were concurrently working on
those messages.
On Mon, Oct 27, 2014 at 11:34:06 +0100, Santiago Vila wrote:
[...]
> I just expect debootstrap maintainers to cooperate with the release
> managers to ensure that the version in stable is able to debootstrap
> the testing distribution, whenever it is possible to do so.
>
This should, however, ideally work both ways: any package modification expected
to break debootstrap would have to be communicated. But arguably this will not
necessarily be obvious, as can be seen here.
> I've heard that the version in wheezy-backports does not have this problem.
> Maybe it could be just a matter of making an upload for the next point release.
> I don't know.
>
That's because of the following workaround, which is being disputed:
debootstrap (1.0.56) unstable; urgency=low
.
[ Tollef Fog Heen ]
* Install base-passwd and base-files in two calls rather than one to
avoid problems with home-built media with different ordering in
Packages. Thanks to Jo Shields for pointing this out and providing
the workaround. Closes: #601670. LP: #1001131.
Though see my earlier mail: if debootstrap ends up being the package to finally
close this bug, then, yes, this might be the route to go.
[...]
> > > If a tool like apt-get or dpkg really behaves in a different way when
> > > I add a Depends field which was already implicit, I think that there
> > > is something fundamentally wrong here.
> >
> > Does dpkg really add the "essential" information into its dependency
> > information? Wouldn't this rather be seen as "ah, essential, so it must be
> > there?" At least briefly looking at dpkg's source code I cannot seem to see dpkg
> > considering this implicit dependency at all (unless attempting to remove an
> > essential package).
>
> In the general case we don't have to worry about that because once
> that essential packages are properly installed and configured, they
> will continue to work all the time unless apt-get does some really
> weird things.
>
Yes and no: mind that that "essential" only provides guarantees for
functionality being available as soon as the package is unpacked. The
"configured" bit is irrelevant for features deemed "essential."
> But how are we expected to have all the essential packages properly
> installed and configured?
>
> Should base-files worry about base-passwd being properly installed and
> configured before making a chown? Certainly not, this is the work of
> debootstrap!
>
No. As you are using the "base-passwd is essential" argument, base-files can
only rely on functionality provided by base-passwd when that is unpacked. The
order of packages being configured is not covered by "essential" and neither
should it be debootstrap's job to sort this out (but debootstrap 1.0.56 does do
this).
[...]
> In either case, I'm going to re-examine carefully what I did in
> base-files 7.7 and see if there is a simple workaround that may be
> done to avoid this problem.
>
Much appreciated, but obviously we should, as you suggested, see what the right
way of fixing this is.
Best,
Michael
Attachment:
pgpEFBzZXCfE7.pgp
Description: PGP signature