On Wed, 2024-11-06 at 09:37 -0800, Chuck Wolber wrote:
> On Wed, Nov 6, 2024 at 9:18 AM Richard Purdie
> <[email protected]> wrote:
> > On Wed, 2024-11-06 at 08:31 -0800, Chuck Wolber wrote:
> > > On Wed, Nov 6, 2024 at 6:18 AM Richard Purdie via
> > > lists.openembedded.org
> > > <[email protected]>
> > > wrote:
>  
> > I'd have thought quite differently about that since "include"
> > already means "include all of the stuff inside of foo.inc".
> 
> I do not disagree, and it is how I would think of it too. But from my
> vantage point, I have seen a lot of people very new to Yocto/OE make
> broad assumptions about the face value of stuff that surprised me.
> This feels like it would be one of them, and I worry about surprising
> side effects (see below).
>  
> 
> > > Also, is it safe to assume that when you say "each
> > > maintainers.inc
> > > file it finds", it does not actually mean full layer search, but
> > > something that is path relevant:
> > > 
> > > import os
> > > for DIR in BBPATH.split(":"):
> > >     F = DIR + "/" + INC_FILE
> > >     if os.path.exists(F):
> > >         include F
> > 
> > Correct, the idea is it would mimic the behaviour of the existing
> > require/include statements but instead of matching one file, it
> > would
> > match all (hence the naming).
> > 
> 
> 
> Sorry if this question is redundant (I feel like it might be given
> what others have asked) - what about side effects like pulling in a
> layer that has include_all in a fragment somewhere. Would this pull
> things together in a non-obvious way almost creating a "backdoor
> bbappend" effect?

There is certainly some risk here but I'm not sure we enable anything
terrible that you can't already do in some way. We're not adding
something which acts as a bbclass extension (something we definitely
want to avoid). I did wonder about putting in some kind of variable
filtering so for example maintainers.inc was only expected to set the
MAINTAINERS* variables but it seemed complex and fragile. We could also
tweak the yocto-check-layer script to "fail" on poor behaviours. That
test will already identify cases where adding a layer changes behaviour
and this addition wouldn't change that.

Cheers,

Richard



-=-=-=-=-=-=-=-=-=-=-=-
Links: You receive all messages sent to this group.
View/Reply Online (#2068): 
https://lists.openembedded.org/g/openembedded-architecture/message/2068
Mute This Topic: https://lists.openembedded.org/mt/109425270/21656
Group Owner: [email protected]
Unsubscribe: https://lists.openembedded.org/g/openembedded-architecture/unsub 
[[email protected]]
-=-=-=-=-=-=-=-=-=-=-=-

Reply via email to