On Wed, 2021-09-01 at 10:52 -0400, Bruce Ashfield wrote:
> On Wed, Sep 1, 2021 at 10:46 AM Peter Kjellerstedt
> <[email protected]> wrote:
> > 
> > > -----Original Message-----
> > > From: [email protected] <openembedded-
> > > [email protected]> On Behalf Of Bruce Ashfield
> > > Sent: den 1 september 2021 15:24
> > > To: Michael Opdenacker <[email protected]>
> > > Cc: Patches and discussions about the oe-core layer <openembedded-
> > > [email protected]>
> > > Subject: Re: [OE-core] [PATCH] meta: stop using "virtual/" in RPROVIDES
> > > and RDEPENDS
> > > 
> > > On Wed, Sep 1, 2021 at 5:20 AM Michael Opdenacker
> > > <[email protected]> wrote:
> > > > 
> > > > Fixes [YOCTO #14538]
> > > > 
> > > > Recipes shouldn't use the "virtual/" string in RPROVIDES and RDEPENDS.
> > > > 
> > > > That's confusing because "virtual/" has no special meaning in
> > > > RPROVIDES and RDEPENDS (unlike in PROVIDES and DEPENDS).
> > > > 
> > > > Instead, using "virtual-" instead of "virtual/"
> > > > as already done in the glibc recipe.
> > > 
> > > We have quite a few cases in meta-virtualization, where the same
> > > unclear syntax is used.
> > > 
> > > I can take care of updating it, if you aren't looking at it already.
> > > 
> > > Bruce
> > 
> > Can anyone enlighten me to what benefit changing "virtual/" to
> > "virtual-" for runtime dependencies gives? AFAIU, this makes no
> > technical difference and everything will still work exactly as
> > before, but with a different provided name.
> 
> Yup.
> 
> > 
> > If you then have virtual/foo for build dependencies and virtual-foo
> > for the corresponding runtime dependencies, then I only see increased
> > confusion with a change like this.
> 
> I don't disagree :D
> 
> I wanted to take care of meta-virt for exactly that reason. I want to
> have a closer look at things, and decide what is the best path to take
> (from a consistency point of view, or leaving it .. if it isn't broken
> :)
> 
> > 
> > And yes, I know about the problems with virtual runtime dependencies,
> > but in my experience, as long as you do not try to create a global
> > package repository with packages from multiple builds, but only use
> > the packages created by bitbake to generate an image, they work as
> > expected.
> 
> That has also been my experience. We've had our virtual/ providers for
> years in meta-virt, and haven't even run into an issue.

Have a look at the discussion on the docs list where people are getting really
confused about whether virtual/ is allowed in the runtime space or not.

I maintain that it is not and never has been supported. I'd like to be able to
continue to say that the "virtual/" namespace is DEPENDS/PROVIDES only. I
therefore suggested the small number of problem entries that sneaked into OE-
Core get fixed. I'd actually got further and suggest we add a QA check that
errors if it sees these in RDEPENDS/PRORIVDES.

The change is focused on us being able to clearly say "don't do it, it isn't
expected to work like virtual/". The alternative is just more confusion and docs
which can't be as clear.

Cheers,

Richard



-=-=-=-=-=-=-=-=-=-=-=-
Links: You receive all messages sent to this group.
View/Reply Online (#155596): 
https://lists.openembedded.org/g/openembedded-core/message/155596
Mute This Topic: https://lists.openembedded.org/mt/85297733/21656
Group Owner: [email protected]
Unsubscribe: https://lists.openembedded.org/g/openembedded-core/unsub 
[[email protected]]
-=-=-=-=-=-=-=-=-=-=-=-

Reply via email to