Bug#389430: Updated patch; make mirror/udeb/suite support multiple suites

[dann frazier <dannf@dannf.org>]

On Sat, Nov 03, 2007 at 02:04:28AM +0100, Frans Pop wrote:
> On Saturday 03 November 2007, dann frazier wrote:
> > Here's an updated patch that implements Joey's suggestion of extending
> > mirror/udeb/suite to support multiple suites. Again I've omitted the
> > indention of the inner loop to make it clear what has changed.
> 
> Has this been tested?

Damn - you had to ask that! Well, yes, it has been tested, but against
the etch version of net-retriever. Let me test against the trunk....
.... yep, still works.

> >  * It doesn't solve the issue Frans highlighted about components
> >    coming from separate mirrors. A proper solution for that problem
> >    would probably involve letting users preseed with arbitrary sets of
> >    sources.list style entries, meaning components can be fetched from
> >    multiple types of media within one installation instance. That'd
> >    certainly be a better paradigm, but would require much more
> >    invasive changes.
> 
> I don't really like the idea of leaving this open if we're going to make 
> changes here anyway.
> 
> OTOH...
> > + ? ?is available from multiple places, anna's current behavior is to use
> > + ? ?the last one, not necessarily the one with the greatest version.
> 
> This is completely broken of course if we want to use this for experimental 
> as that _can_ have older versions than the real suite being used.

Yeah, anna could use improvement here; is there version comparison
code in d-i somewhere already?

> 
> Regarding the removal of this comment:
> > -???????# Setting codename to a suite is not very nice, but can do no harm
> >         if ! db_get mirror/udeb/suite || [ -z "$RET" ]; then
> >                 if [ -f /etc/udebs-source ]; then
> 
> Note that what you are doing now is just the opposite: in the _default_ 
> situation the contents of /etc/udebs-source is used and that _is_ a 
> codename (set at D-I images build time). All the rest are the exceptions!
> This should at least remain clear from the code or comments.

Ok. I thought the comment was explaining that the variable named 'codename'
was being abused to possibly include a non-codename, and that renaming
the variable to 'udebsuites' solved that problem. But anyway, I agree
it could use some clarifying comments.

What I'm hearing is:

Changes this patch needs:
 * comments around the suite selection bit
 * proper indention of the loop

Additional changes we want:
 * anna version selection smarts
 * refactoring the nested loops into subroutines
 * deprecate m/u/suite for m/u/suites, providing backwards
   compatability

Does that look accurate?

-- 
dann frazier