[Libosinfo] Resources inheritance for OSes, I need some help/feedback!

[Fabiano Fidêncio]

On Tue, 2018-11-13 at 09:58 +0000, Daniel P. Berrangé wrote:
> On Tue, Nov 13, 2018 at 10:10:17AM +0100, Fabiano Fidêncio wrote:
> > People,
> > 
> > I'm working on making OsinfoResources inherited for OSes that
> > derives-
> > from/clones another OSes. The inheritance, IMO, should work like:
> > - os1 has:
> >   <resources arch="all">
> >     <minimum>
> >       <cpu>1000000000</cpu>
> >       <ram>1073741824</ram>
> >     </minimum>
> >   </resources>
> > 
> > - os2, which inherits the values from os1, has:
> >   <resources arch="all">
> >     <minimum>
> >       <n-cpus>2</n-cpus>
> >     </minimum>
> >   </resources>
> 
> I wonder about making inheritance explicit, eg for os2, have
> 
>    <resources arch="all" inherit="yes">
>      <minimum>
>        <n-cpus>2</n-cpus>
>      </minimum>
>    </resources>
> 
> the benefit of this is that....  [1]

Okay, this idea makes things easier, I like it.

> 
> > - When calling osinfo_resources_get_cpu(os2), I'd expect to get:
> > 1000000000.
> > 
> > 
> > In order to achieve so, we'll need a few more changes in the way
> > resources are added to OSes:
> > - Do *not* have duplicated resources for the same OS;
> >   - This can be easily done in osinfo_os_add_*_resources() + tests
> > to
> > catch this situation;
> > - Do *not* have more than one resources with the same arch for the
> > same
> > OS;
> >   - Although this can be easily done  in
> > osinfo_os_add_*_resources() as
> > well, I'd go only for tests to catch this situation;
> 
> These points look tangential to inheritance to me. Or to put it
> another
> way, shouldn't we ensure uniqueness regardless of whether we have
> inheritance or not.

What would be the use case of having different resources set for the
very same architecture in an OS?

> 
> 
> > - Have a way to specify that a resource field shouldn't be
> > inherited;
> >   - Here I can basically see two different approaches, and my
> > personal
> > preference would be the former:
> >     - A specific value to be set to the field that will just be
> > checked
> > when merging the resources' content when doing the inheritance
> > (suggestions are welcome and remember we can't use -1);
> >     - Add an extra attribute to the elements and have a new
> > structure
> > (similar to DeviceLinks) and play with it in the same way I
> > proposed
> > for removed devices*.
> 
> [1]...you solve this problem more attractively IMHO.
> 
> Regards,
> Daniel

Best Regards,
-- 
Fabiano Fidêncio