[OE-core] [PATCH v2] kernel-module-split: rrecommend kernel-image instead of rdepend

[Martin Hundebøll]

On 2017-09-08 23:29, Paul Barker wrote:
> On Fri, Sep 8, 2017 at 10:07 PM, Martin Hundebøll
> <martin.hundeboll at prevas.dk> wrote:
>>
>>
>> On September 8, 2017 10:37:26 PM GMT+02:00, Denys Dmytriyenko <denis at denix.org> wrote:
>>> On Fri, Sep 08, 2017 at 09:06:17PM +0100, Paul Barker wrote:
>>>>
>>>> This breaks the method described in the kernel dev FAQ to omit the
>>>> kernel image from the rootfs which is to set the following in a conf
>>>> file:
>>>>
>>>>      RDEPENDS_kernel-base = ""
>>>>
>>>> See
>>> http://www.yoctoproject.org/docs/2.3.1/kernel-dev/kernel-dev.html#idm45690538384880.
>>>>
>>>> By recommending kernel-image directly instead of recommending the
>>>> top-level kernel-<version> package which was actually removed from
>>>> extra_depends, we skip the inclusion of the kernel-base package. I'm
>>>> not familiar enough with the details of kernel module loading to know
>>>> if the 2 files in kernel-base (modules.builtin and modules.order) are
>>>> strictly needed but I think they should at least be included in via
>>>> recommendation.
>>>>
>>>> As we already have a documented way to omit the kernel image, can we
>>>> revert this?
>>>>
>>>> As an alternative I can send in a patch which changes the recommend
>>> to
>>>> the top-level kernel-<version> package instead of kernel-image.
>>>
>>> FWIW, this was discussed at #yocto on IRC today. When I first saw the
>>> change
>>> several days ago, I was puzzled as well...
>>> The existing method has been in use by many of us for a very long time
>>> and
>>> seems to be well documented - I also vote for reverting the change.
>>
>> Does setting `RDEPENDS_kernel-base = ""` in an image recipe work too?
>>
> 
> No, you'd need it to be set in a conf file so it takes effect when the
> kernel-base package is created.
> 
> I can see the issue here though. Can you set `RDEPENDS_kernel-base =
> ""` globally and explicitly install kernel-image into the appropriate
> image? If not then yes we'll need to find an alternative solution to
> simply reverting the change.
> 

Seems like the inverted approach of disabling install of kernel-image 
globally, and enabling it per-image works just fine.

Reverting the patch is fine by me - sorry for the noise.

// Martin