[LEDE-DEV] Multi-Profile selection support

[John Crispin]

On 23/05/2016 14:02, Rafał Miłecki wrote:
> On 23 May 2016 at 13:59, Felix Fietkau <nbd at nbd.name> wrote:
>> On 2016-05-23 13:56, Rafał Miłecki wrote:
>>> On 23 May 2016 at 13:38, Felix Fietkau <nbd at nbd.name> wrote:
>>>> after a long delay, I finally got around to looking into multi profile
>>>> selection again. After reviewing your code and thinking about it some
>>>> more, I decided to go with a slightly different approach from the one
>>>> that you took.
>>>>
>>>> I'm working towards not only being able to build an image with support
>>>> for a specific set of devices, but also being able to optionally build
>>>> multiple images at once with the per-image rootfs being affected by the
>>>> device profile as well.
>>>>
>>>> This can currently only work properly with the new image building code
>>>> along with having the device profile data specified in the Device
>>>> section, so I limited multi-profile selection to that.
>>>>
>>>> As a next step, I'm going to add some wrapper code to allow the legacy
>>>> image building code to be incrementally converted as well without having
>>>> to adjust everything to the new Build/* chunks right away.
>>>
>>> Is it worth it? If there are some targets really wanting this feature
>>> I guess we could convert them to the new Device/* building system. My
>>> knowledge is surely limited but I can try to help with that. I somehow
>>> managed to convert brcm47xx and improve bcm53xx.
>> The multi-profile stuff is something we want to have for all targets, so
>> that we can avoid hackish scripts involving the image builder and
>> multiple build passes.
>> Converting ar71xx is going to be a lot of work, especially testing all
>> those different devices - unless you can find a way to automate that by
>> proving that the output stays the same with the old and the new code,
>> but I don't know if our reproducible build support is quite there yet.
> 
> I don't have experience with ar71xx, but when I was converting
> brcm47xx I simply used sha1sum on resulting images. We don't expect
> any change in final images from the switch to the new building system,
> do we?
> 

with the amount of profiles that need to be converted, i would estimate
this to be anywhere between 50 and 100 hours of work for the full tree.

	John