native python recipes using setuptools

native python recipes using setuptools

[Konrad Weihmann]

Hi,

I'm a bit puzzled at the moment - after 
fd6a007efa7cb45101a66f294af81d9d33bb3fab in meta-openembedded I faced a 
lot of changes in the behavior of native-only recipes (@RP yes the ones 
that don't use BBCLASSEXTEND :-))

after this change I had a couple of dependency loops like

python3-my-recipe-native > python3 > ... > python3-my-recipe-native

and I could track it down to distutils3-base.bbclass [1].
There for all recipes no matter what the set (I guess BBCLASSEXTEND 
magic will solves that in the end) "python3-native python3" is set in 
DEPENDS.

My question is, is that intended behavior or would it make sense to be 
more explicit here, like

DEPENDS += "python3-native"
DEPENDS_class-target += "python3"

?

Mainly this happens when a globally inherited class requests a native 
python3 package via DEPENDS (a not too uncommon use case for me).

Thoughts?

Cheers
Konrad

[1] 
https://git.yoctoproject.org/cgit/cgit.cgi/poky/tree/meta/classes/distutils3-base.bbclass#n1

Re: [OE-core] native python recipes using setuptools

[Richard Purdie]

On Tue, 2021-02-16 at 10:59 +0100, Konrad Weihmann wrote:
> I'm a bit puzzled at the moment - after 
> fd6a007efa7cb45101a66f294af81d9d33bb3fab in meta-openembedded I faced a 
> lot of changes in the behavior of native-only recipes (@RP yes the ones 
> that don't use BBCLASSEXTEND :-))
> 
> after this change I had a couple of dependency loops like
> 
> python3-my-recipe-native > python3 > ... > python3-my-recipe-native
> 
> and I could track it down to distutils3-base.bbclass [1].
> There for all recipes no matter what the set (I guess BBCLASSEXTEND 
> magic will solves that in the end) "python3-native python3" is set in 
> DEPENDS.
> 
> My question is, is that intended behavior or would it make sense to be 
> more explicit here, like
> 
> DEPENDS += "python3-native"
> DEPENDS_class-target += "python3"
> 
> ?
> 
> Mainly this happens when a globally inherited class requests a native 
> python3 package via DEPENDS (a not too uncommon use case for me).
> 
> Thoughts?

Definitely a bug, I've sent a patch. The DEPENDS would be needed for
nativesdk too and since it doesn't appear to require python3-native in
the native case, I didn't change that.

Cheers,

Richard