Re: [PATCH v2 1/1] copy dtbs into a subdirectory based on image name

[Gylstorff Quirin <quirin.gylstorff@siemens.com>]

On 4/17/24 10:25 AM, Baurzhan Ismagulov wrote:
> On 2024-04-16 17:36, 'Gylstorff Quirin' via isar-users wrote:
>> On 4/16/24 12:07 PM, Nicusor Liviu Huhulea wrote:
>>> There are cases when multiple images are build and because the same dtbs
>>> are copied in deploydir we have a build failure. The build error is related
>>> to files that are installed into a shared area where those files already exists.
>>>
>>> To solve this situation we will copy each dtb in a subdirectory based on
>>> image name thus avoiding the conflicts. Therefore the other areas needs to be
>>> updated on the dtbs paths.
>>
>> This is a special use case, which breaks all downstream layers which
>> currently use devicetrees. Coping to the directory dtbs should be disabled
>> by default as for most uses it is unnecessary.
> 
> The origin of the problem is bitbake QA with multiple images. The kernel recipe
> builds the dtb which is deployed by the image. In case of multiple images, the
> dtbs are considered as different from bitbake PoV and it complains about it --
> this is the problem we are trying to address. In our experience, multiple
> images are common -- and we would still need a solution for a supported feature
> even if they weren't.
> 
>  From the user PoV, multiple directories with the same dtbs are not necessary at
> all. So, I'm not a fan of the per-image dtb approach although we also have a
> similar patch to address the immediate failures that we are experiencing. If we
> go with this approach, I wouldn't welcome switching per-image dtb on and off --
> we already have many variables and are missing coverage for untested
> combinations; maybe some configurable deployment path at most.
> 
> That said, I think we should really try to bringe the bitbake model closer to
> the reality and look at moving this to the kernel recipe as Uladzimir has
> suggested. Maybe there are cases that need to be verified (like different
> kernels building the same dtbs) and might still need further adjustments.
> 
For me the kernel approach would be fine - from the looks the kernel 
provides the DTB in yocto:
https://git.yoctoproject.org/poky/tree/meta/classes-recipe/devicetree.bbclass

Best regards
Quirin

> With kind regards,
> Baurzhan
>