RE: [PATCH v2 03/10] rootfs: recursively depend on packages

["Schmidt, Adriaan" <adriaan.schmidt@siemens.com>]

> -----Original Message-----
> From: Jan Kiszka <jan.kiszka@siemens.com>
> Sent: Mittwoch, 13. Oktober 2021 18:02
> To: Schmidt, Adriaan (T RDA IOT SES-DE) <adriaan.schmidt@siemens.com>; isar-
> users@googlegroups.com
> Subject: Re: [PATCH v2 03/10] rootfs: recursively depend on packages
> 
> On 13.10.21 17:18, Schmidt, Adriaan (T RDA IOT SES-DE) wrote:
> >>>>> This fixes a bug.
> >>>>> So far, this only worked when the dependency-of-dependency packages
> >>>>> were in isar-apt because they were just built. This explicitly
> >>>>> enforces do_deploy_deb on *all* dependencies.
> >>>>
> >>>> More curiosity: I'm not yet understanding the error scenario. A
> >>>> rootfs depends on a package recipe A, and that means on the
> >>>> completion of that recipe's do_deploy_deb task. If that recipe A
> >>>> depends on another package recipe B that one needs to be in isar-apt,
> >>>> thus has to finish its do_deploy_deb as well. But why could A have
> >>>> finished its build when B didn't even run do_deploy_deb yet? If A
> >>>> depends on B, it also depends on B's do_deploy_deb task, no?
> >>>
> >>> The rootfs depends on A's do_deploy_deb, but only A's do_prepare_build
> >> depends on B's do_deploy_deb.
> >>> If we can produce A (do_deploy_deb) from cache without running its
> >> do_prepare_build, then B might not be there (which is exactly what
> happened
> >> to me).
> >>>
> >>
> >> OK, this issue only surfaces once you introduce caching. Without caching,
> >> it's impossible to complete do_deploy_deb (or do_prepare_build) without
> all
> >> dependencies' do_deploy_deb.
> >
> > Exactly. In that way also patch 05/10 in this series is similar. Here the
> added dependencies on isar-apt ensure that things work also when stuff is
> taken from cache instead of built.
> >
> >>> Thinking about it I'm wondering if the same thing can happen when
> building
> >> a package, and if actually also do_prepare_build in dpkg-base needs to
> define
> >> recrdeptask instead of its current `do_prepare_build[deptask] =
> >> "do_deploy_deb"`?
> >>
> >> Conceptually, yes, same scenario, just different root.
> >>
> >> How does OE handle such cases? I bet there can be more.
> >
> > An `ack recrdeptask poky/meta` shows
> >
> > classes/package_ipk.bbclass
> > 282:do_build[recrdeptask] += "do_package_write_ipk"
> >
> > classes/base.bbclass
> > 349:do_build[recrdeptask] += "do_deploy"
> >
> > classes/meta.bbclass
> > 4:do_build[recrdeptask] = "do_build"
> >
> > classes/image.bbclass
> > 108:do_rootfs[recrdeptask] += "do_packagedata"
> >
> > Plus some more in specialized images or other package types. My feeling is
> that recrdeptask was created for this exact case we're facing here.
> >
> 
> Sounds like more fun ahead ;)

Yes, indeed.
Adding `do_prepare_build[recrdeptask] = "do_deploy_deb"` creates a dependency loop (somehow it makes recipes' do_prepare_build depend on their own do_deploy_deb!?).

So im V3, I'm adding a `do_deploy_deb[deptask] = "do_deploy_deb"`, which should ensure dependencies are there, even if packages are taken from cache.

Adriaan