[OE-core] [oe-core][PATCH 1/2] defaultsetup.conf: enable select init manager

[richard.purdie at linuxfoundation.org]

On Sat, 2019-07-06 at 12:53 +0300, Adrian Bunk wrote:
> On Thu, Jul 04, 2019 at 09:45:19PM +0800, kai.kang at windriver.com
> wrote:
> > From: Kai Kang <kai.kang at windriver.com>
> > 
> > Introduce a new variable INIT_MANAGER and create 3 init-manager-
> > *.inc
> > files to configure init manager settings. Available values of
> > INIT_MANAGER are sysvinit, systemd and mdev-busybox. 'sysvinit' is
> > set
> > by default for compatibility.
> > ...
> > --- a/meta/conf/distro/defaultsetup.conf
> > +++ b/meta/conf/distro/defaultsetup.conf
> > @@ -23,3 +23,6 @@ PACKAGE_CLASSES ?= "package_ipk"
> >  INHERIT_BLACKLIST = "blacklist"
> >  INHERIT_DISTRO ?= "debian devshell sstate license remove-libtool"
> >  INHERIT += "${PACKAGE_CLASSES} ${USER_CLASSES} ${INHERIT_DISTRO}
> > ${INHERIT_BLACKLIST}"
> > +
> > +INIT_MANAGER ??= "sysvinit"
> > +require conf/distro/include/init-manager-${INIT_MANAGER}.inc
> > ...
> > --- /dev/null
> > +++ b/meta/conf/distro/include/init-manager-sysvinit.inc
> > @@ -0,0 +1,6 @@
> > +# Use sysvinit for system initialization
> > +DISTRO_FEATURES_append = " sysvinit"
> > +DISTRO_FEATURES_BACKFILL_CONSIDERED_append = " systemd"
> > +VIRTUAL-RUNTIME_init_manager = "sysvinit"
> > +VIRTUAL-RUNTIME_initscripts = "initscripts"
> > +VIRTUAL-RUNTIME_login_manager = "busybox"
> 
> I am not sure whether this can be fixed better, but this does break 
> existing configurations that use a non-default init system.
> 
> I just ran into a build issue with
>   VIRTUAL-RUNTIME_init_manager = "systemd"
> since this now resulted in both sysvinit and systemd being attempted
> to 
> be installed to the image.
> 
> This was fixable in my configuration with
>   -VIRTUAL-RUNTIME_init_manager = "systemd"
>   +INIT_MANAGER = "systemd"
> 
> This at least needs to be properly documented as a breaking change.

This change also didn't quite work out for some of our selftests. I've
a patch in master-next which tried to improve things and that fixed the
selftests.

The only other way we could do this is to default INIT_MANAGER to say
"unset" and have an empty file for that. That will of course break my
patch and need the defaults adding back into the recipe. Unless we put
those defaults into the unset version.

Not sure quite what we should do here...

Cheers,

Richard