[OE-core] [PATCH] rootfs_ipk: respect ONLINE_PACKAGE_MANAGEMENT

[Richard Purdie]

On Tue, 2011-05-24 at 14:59 +0100, Phil Blundell wrote:
> On Thu, 2011-05-19 at 16:08 +0100, Richard Purdie wrote:
> > On Thu, 2011-05-19 at 16:04 +0100, Phil Blundell wrote:
> > > On Thu, 2011-05-19 at 13:16 +0100, Richard Purdie wrote:
> > > > So if we:
> > > > 
> > > > a) Only add ROOTFS_PKGMANAGE_BOOTSTRAP if postinstalls were present
> > > > b) Add the read-only-rootfs option we discussed which errors if 
> > > >    postinstalls are present
> > > > 
> > > > we end up a lot closer to where you want to be.
> > > 
> > > Yes, sounds reasonable.  And I think we could then eliminate
> > > remove_packaging_data_files() altogether, in favour of having the right
> > > thing happen automatically during rootfs construction, which would
> > > probably be a good thing too.
> > 
> > Agreed, I think we have a plan :)
> 
> One other thing that occurred to me is that ONLINE_PACKAGE_MANAGEMENT
> (in classic oe) is a DISTRO feature rather than an image one.  This is
> significant because, for example, update-rc.d.bbclass doesn't include
> update-rc.d in RDEPENDS if it knows that the package will never be
> installed on a running target.  If we're going to make package
> management into an IMAGE_FEATURE then obviously this isn't going to work
> as it stands.
> 
> I guess we could work around it by letting update-rc.d add its
> dependency as normal, and then adding code to the rootfs constructor to
> stop it taking effect (and/or substitute a dummy update-rc.d package
> with no files in) if an image with no package management is being
> generated.  That doesn't seem terribly elegant but, short of going back
> to a DISTRO-based selection, I can't think of any better way of fixing
> it.

I think allowing selection of this at image generation time is the more
powerful way to handle this. It could be we go through a step of
forcibly removing packages we don't want from the rootfs such as
update-rc.d, or we can tell the package manager to ignore the dependency
which is probably neater.

I have to admit the update-rc.d change was concerning and this does feel
like a better way to handle it.

Cheers,

Richard