[Multiarch-devel] Proper upgrade path for lib32v4l-{0,dev}?

[Sven Joachim]

[ CC'ing ia32-libs maintainers for their opinion. ]

On 2012-07-01 16:04 +0200, Gregor Jasny wrote:

> Hello,
>
> currently my source package v4l-utils builds lib32v4l-0 and
> lib32v4l-dev packages on amd64. I'd like to get rid of them sooner
> than later and provide a proper upgrade path to multiarch
> packages. What's the best way to achieve this?

Getting rid of reverse dependencies so these packages can be removed as
unused, I think.

> As far as I can see this always involves manual action due to the
> required addition of the foreign i386 architecture.

Correct, but the same is true for ia32-libs which is usually the reason
why people install lib32v4l-0.

> So should I simply drop the lib32 packages? The only dependent users
> of these packages are Skype and Google talk plugin users. They usually
> preload the libs via LD_PRELOAD. Due to the changed library paths when
> migrating from lib32 packages to multiarch they need to perform manual
> actions anyway.
>
> Or should I change the packages from:
>
> Package: lib32v4l-0
> Section: libs
> Architecture: amd64
> Depends: libv4l-0 (= ${binary:Version}),
>          ${shlibs:Depends},
>          ${misc:Depends}
>
> to virtual transition packages:
>
> Package: lib32v4l-0
> Section: libs
> Architecture: i386
> Pre-Depends: multiarch-support
> Depends: libv4l-0 (= ${binary:Version}),
>          ${misc:Depends}

The Pre-Depends: multiarch-support is not necessary, but you would need
to mark lib32v4l-0 as "Multi-Arch: foreign".

However, since lib32v4l-0 is usually installed because ia32-libs depends
on it, and the skype-debian_4.0.0.7-1_amd64.deb package I downloaded
from Skype's website depends on ia32-libs but not lib32v4l-0: maybe a
better upgrade path would be given by letting ia32-libs-i386 depend on
libv4l-0?

The ia32-libs-i386 package depends on all libraries formerly included in
ia32-libs (well, as far as they are still available), but not on the
i386 counterparts of the various lib32* packages ia32-libs depended
upon.  This seems like a bug to me.

Cheers,
       Sven