On Wednesday, 20 January 2021 12:23:34 NZDT Richard Purdie wrote:

On Wed, 2021-01-20 at 12:13 +1300, Paul Eggleton wrote:

quoted

On Wednesday, 20 January 2021 09:52:41 NZDT Richard Purdie wrote:

quoted

I think the one remaining issue here is the need to change the DEPENDS
of so many other recipes, likely not just here in this patch but in
other layers. I think if util-linux DEPENDS on util-linux-uuid that
might remove the need for those changes? That should still allow you to
break the circular dependency problem?

I have to admit to a gap in my own knowledge of how our build system
handles transitive dependencies. Of course the recipe sysroot should
still get everything it needs in it even if the dependency is only
indirectly included, in the back of my mind I have the impression that
there are expectations that all dependencies are explicitly called out
and there are subtle issues if they aren't, but that could be a mistaken
impression on my part.

I do wonder a little about that as well. As you say, sysroot
dependencies should handle this. Anything linking against libuuid
should also establish an package level runtime dependency through the
linkage so I think this should work.

We definitely don't explicitly list every dependency in every recipe.

If this can work, it makes the migration path for people easier so I
think its at least worth investigating/testing.

OK, sure thing.

Just while I'm thinking, the PACKAGES_remove also bothers me a little.
Can we rearrange the variables so libuuid is only added in the libuuid
recipe variant?

[the idea being that since we control the metadata in oe-core, we
shouldn't need to use _remove and can restructure so we don't need to,
they're hard to undo. I know we do use it in places sadly even in core]

Yes, that is a good point. We should be able to avoid using _remove and I 
don't like using it either given how heavy a hammer it is.

quoted

I agree that it would be better being separate, FWIW.

quoted

I am also worried this is going to break AUH and mean we have to
manually handle this recipe but again, I suspect there is little to be
done and we just have to deal with it.

Could we perhaps fix the AUH to handle this properly? Do we need some kind
of mechanism to get it to always upgrade the two recipes together or is
that only part of the issue?

I don't know for sure that AUH won't handle it, I just worry about it.
If it doesn't it definitely could be something we could fix there. I
just don't know of anyone with the time to spend on what is a marginal
corner case if it doesn't work.

Well, on the other hand if we're "causing" a problem with the AUH with this 
change it does behoove us to try to resolve that. It is something I'd be 
prepared to look into at least.

Cheers
Paul