Re: Incremental target images

[Claudius Heine <claudius.heine.ext@siemens.com>]

On 2018-08-27 10:29, Jan Kiszka wrote:
> On 2018-08-27 09:44, Claudius Heine wrote:
>> On 2018-08-27 09:25, [ext] Claudius Heine wrote:
>>> Hi Jan,
>>>
>>> On 2018-08-27 08:41, [ext] Jan Kiszka wrote:
>>>> Hi all,
>>>>
>>>> I was wondering if / how we could model the increasingly common case 
>>>> of building very similar target images more efficiently.
>>>>
>>>> A devel image, e.g., likely consists of almost the same base package 
>>>> set as the release image. It may only add further packages and maybe 
>>>> replace very few (like customization packages). When building both, 
>>>> we could save time - specifically when doing cross - by 
>>>> bootstrapping the baseline only once. We already do that for the 
>>>> debootstrap step, but not yet for further packages.
>>>>
>>>> What do you think? And how could that be modeled from user perspective?
>>>
>>> That is not very easy.
>>>
>>> A complex way this could be done by having a `common image` recipe 
>>> that depends on other image recipes to deploy their required package 
>>> list,
>>
>> What I meant by that is each image recipe writes its package list into 
>> its own text file in the 'deploy' directory.
>>
>>> then figure out the common dependencies of all of them and build the 
>>> custom image.
>>
>> s/custom/common/
>>
>> Problem scenario here is:
>>
>>   Image A: depends on A which depends on D
>>   Image B: depends on B which depends on D
>>
>> Since D is not named in any recipes, but should be installed in the 
>> `common image` as automatically selected package (not as manual). This 
>> scripting is a bit tricky.
>>
>>> Then those other image recipes depend of the `common image` recipe in 
>>> turn to create it, copy it and base their own customizations on top.
>>>
>>>  From a user perspective they would have to add their image recipes 
>>> into a global variable or bbappend, so that is available in the 
>>> `common image` recipe. The rest could be done in the image classes.
>>>
>>> Cheers,
>>> Claudius
>>
> 
> Hmm, I'm not seeing yet where the complication could come from. I would 
> have expected the following:
> 
> - write a new image recipe include that takes a preexisting image in
>    form of a rootfs as input (rather than using setup_root_file_system)
> 
> - create specialized images on top of that which depend on some base
>    image recipe as well as additional IMAGE_INSTALL recipes and may come
>    with own IMAGE_PREINSTALLs (always on top of base images)
> 
> - push all common IMAGE_PREINSTALLs into base image recipes, same for
>    all IMAGE_INSTALLs
> 
> And done. What am I missing?

Ok, then I misunderstood you. I thought you wanted a mostly transparent 
mechanism in the background and not just an inheritance scheme for images.

Cheers,
Claudius

-- 
DENX Software Engineering GmbH,      Managing Director: Wolfgang Denk
HRB 165235 Munich, Office: Kirchenstr.5, D-82194 Groebenzell, Germany
Phone: (+49)-8142-66989-54 Fax: (+49)-8142-66989-80 Email: ch@denx.de