Quantal meta package review [RFC]

[Tim Gardner]

On 05/18/2012 11:11 AM, Leann Ogasawara wrote:
> On 05/18/2012 08:49 AM, Tim Gardner wrote:
>> Quantal kernel flavour and meta package carnage
>>
>> Pursuant to UDS blueprints, the kernel team has collapsed multiple 12.04
>> x86 flavours into one 12.10 flavour, e.g. generic. We've modified kernel
>> packaging for x86 such that it is delivered in 2 packages. This is to
>> accommodate the needs of virtual instances where the smallest possible
>> kernel package is desired.
>>
>> The general relationship rules for kernel meta packages are thus:
>>
>> linux-headers-<flavour>  - depends on the kernel headers package.
>> linux-image-<flavour>  - depends on the kernel and extras binary
>> packages.
>> linux-<flavour>  - depends on linux-image-<flavour>  and
>> linux-headers-<flavour>.
>>
>> The exception is linux-image-virtual which only depends on the kernel
>> binary package and not the extras package. Linux-image-extra-virtual
>> depends on both kernel and extras packages.
>>
>> Note the dependency of linux-<flavour>  on the appropriate headers
>> package. This is now common across all architectures.
> 
> All of the above looks good to me.  Just a few comments inlined below...
>> The interesting Quantal kernel meta packages are as follows:
>>
>> [i386 amd64] linux → linux-generic
>> [i386 amd64] linux-image → linux-image-generic
>>
>> [i386 amd64] linux-headers-generic → linux-headers-ABI-generic
>> [i386 amd64] linux-image-generic → linux-image-ABI-generic,
>> linux-image-extra-ABI-generic
>> [i386 amd64] linux-generic → linux-image-generic, linux-headers-generic
>>
>> [i386 amd64] linux-headers-server → linux-headers-generic
>> [i386 amd64] linux-image-server → linux-image-generic
>> [i386 amd64] linux-server → linux-image-server, linux-headers-server
>>
>> [i386 amd64] linux-headers-virtual → linux-headers-generic
>> [i386 amd64] linux-image-virtual → linux-image-ABI-generic
>> [i386 amd64] linux-image-extra-virtual → linux-image-virtual,
>> linux-image-extra-ABI-generic
> 
> This could be simplified to linux-image-extra-virtual →
> linux-image-generic in which case do we really need to have a
> linux-image-extra-virtual meta package?  Seems unnecessary to me or does
> this resolve an upgrade path from Precise?

linux-image-extra-virtual → linux-image-generic has one less level of
abstraction, but doing that kind of violates the general relationship
rules. However, virtual is kind of a violation anyways. I'm ambivalent.

There _is_ a linux-image-extra-virtual meta package in Precise. So, yes,
I think it makes sense to keep it for upgrades. The fewer hacks that we
have to add to update-manager-core, the better.

>> [i386 amd64] linux-virtual → linux-image-virtual, linux-headers-virtual
>>
>> [armel armhf] linux-headers-omap → linux-headers-ABI-omap
>> [armel armhf] linux-image-omap → linux-image-ABI-omap
>> [armel armhf] linux-omap → linux-image-omap, linux-headers-omap
>>
>> [powerpc] linux-headers-powerpc64-smp → linux-headers-ABI-powerpc64-smp
>> [powerpc] linux-image-powerpc64-smp → linux-image-ABI-powerpc64-smp
>> [powerpc] linux-powerpc64-smp →   linux-image-powerpc64-smp,
>> linux-headers-powerpc64-smp
>>
>> [powerpc] linux-headers-powerpc → linux-headers-powerpc-smp
>> [powerpc] linux-image-powerpc → linux-image-powerpc-smp
>> [powerpc] linux-powerpc → linux-powerpc-smp, linux-headers-powerpc
> 
> Since we dropped the non-smp powerpc flavor in Precise, I think we
> should just drop the linux-powerpc, linux-image-powerpc, and
> linux-headers-powerpc meta packages from Quantal and have them reaped
> from the archive just like we'll do for generic-pae.  It doesn't seem
> necessary to carry them if they just re-direct to the powerpc-smp
> flavor.  I believe I already transitioned the non-smp powerpc flavor in
> Precise to the powerpc-smp flavor so the upgrade path should already be
> in place:
> 
> commit 19a10a6c255f5adcc313be3268d75f426187fcfd
> Author: Leann Ogasawara <leann.ogasawara at canonical.com>
> Date:   Mon Mar 26 07:20:47 2012 -0700
> 
>     UBUNTU: Point powerpc meta to powerpc-smp meta
> 
>     Doing so will allow us to completely remove the non-smp powerpc flavor
>     in P+1
> 
>     Signed-off-by: Leann Ogasawara <leann.ogasawara at canonical.com>

Doh! I forgot about the non-smp powerpc flavours. I'll drop them.

>> [powerpc] linux-headers-powerpc-smp → linux-headers-ABI-powerpc-smp
>> [powerpc] linux-image-powerpc-smp → linux-image-ABI-powerpc-smp
>> [powerpc] linux-powerpc-smp →   linux-image-powerpc-smp,
>> linux-headers-powerpc-smp
> 
> Everything else looks fine.  I assume we'll implement this only after we
> re-instate the i386 generic flavor.
> 

I've already pushed master-next with those patches. generic-pae is no
more then a bad memory.

rtg
-- 
Tim Gardner tim.gardner at canonical.com