Can we do this? [Was: Re: How to tell users that ia32-libs will go away]

[Goswin von Brederlow <goswin-v-b@web.de>]

Goswin von Brederlow <goswin-v-b@web.de> writes:

> Dear ftp-master,
>
> I wonder if the solution below for transitioning ia32-libs to multiarch
> would be OK in regards to DAK and testing transition etc. Any technical
> problems why we couldn't make an exception for the 3 ia32-libs* packages
> for this?

Since ftp-master has not raised any objection ("who stays quiet agrees") 
the same question goes to the release team.

Specifically can we set up an exception for ia32-libs to go into testing
(and then stable) with the dependency being unfullfillable on amd64
(without i386 enabled)?

> Steve Langasek <vorlon@debian.org> writes:
>
>> On Thu, Feb 09, 2012 at 04:43:04PM +0100, Thibaut Paumard wrote:
>>> Le 09/02/12 15:53, Goswin von Brederlow a écrit :
>>
>>> > now that a multiarch dpkg has been uploaded to experimental it looks
>>> > like we can finaly get rid of ia32-libs* for wheezy.
>>
>>> > !!!HURAY!!!
>>
>>> > The problem now is the transition:
>>
>>> > 1) multiarch and ia32-libs are incompatible
>>> >[...]
>>> > What this means is that users that want to use multiarch should remove
>>> > ia32-libs (and lib32* really) soonest.
>>
>>> Couldn't you make ia32-libs a meta-package pulling the multiarch version
>>> of the libs it used to include ?
>>
>> This would require something like
>>
>>   Depends: libpam0g:i386, libssl098:i386, [...]
>>
>> and this syntax is not yet supported (intentionally, because there's a lot
>> of policy that needs to be put in place before we allow such things).
>>
>> Ubuntu, faced with the same issue, kludged a bit to make upgrades possible.
>>
>>   Package: ia32-libs
>>   Architecture: amd64
>>   Depends: ia32-libs-multiarch
>>
>>   Package: ia32-libs-multiarch
>>   Architecture: i386
>>   Multi-Arch: foreign
>>   Depends: libpam0g, [...]
>>
>> This doesn't require us to support :arch syntax for dependencies anywhere
>> yet; it just requires that the i386 arch is enabled via multiarch, and that
>> the package manager is able to resolve the fact that ia32-libs' dependency
>> is satisfied by the only copy of ia32-libs-multiarch available, the i386
>> one.
>>
>> However, this still introduces at least some of the same policy problems -
>> for instance, britney has to be taught that this is ok if you want to be
>> able to migrate this package to testing automatically.  And you need a
>> multiarch-capable package manager installed and configured *before* you can
>> upgrade this package, so that requires a two-step upgrade of some variety:
>> either holding ia32-libs back until after the dist-upgrade, or upgrading the
>> package manager before the dist-upgrade.
>
> MfG
>         Goswin