systemd/sysvinit switches (was [meta-systemd][PATCH 1/2] Delete systemd class)

systemd/sysvinit switches (was [meta-systemd][PATCH 1/2] Delete systemd class)

[Andreas Müller]

On Thu, Apr 11, 2013 at 2:12 AM, Andreas Müller
<schnitzeltony@googlemail.com> wrote:
> On Tue, Apr 9, 2013 at 10:01 PM, Martin Jansa <martin.jansa@gmail.com> wrote:
>> On Tue, Mar 05, 2013 at 11:59:07PM +0100, Martin Jansa wrote:
>>> On Tue, Mar 05, 2013 at 11:50:41PM +0100, Andreas Müller wrote:
>>> > On Tue, Mar 5, 2013 at 11:18 PM, Martin Jansa <martin.jansa@gmail.com> wrote:
>>> > > On Wed, Feb 13, 2013 at 09:25:34AM -0800, Khem Raj wrote:
>>> > >> Signed-off-by: Khem Raj <raj.khem@gmail.com>
>>> > >
>>> > > good old meta-systemd should now parse again,
>>> > Thanks - will come back to test HEADs soon.
>>>
>>> But be aware that I've fixed only parsing of avahi and wpa-supplicant.
>>>
>>> I haven't tested it properly in runtime because my upgrade path is a bit
>>> broken (because I was building images with khem's patches for a while to
>>> test other stuff) and even clean build with failed to boot with
>>> qemux86-64 and I don't have time to debug it now.
>>>
>>> > > I wont apply this 2 commits
>>> > > until systemd discussion in oe-core is finished and future of
>>> > > meta-systemd decided.
>>> > >
>>> > After systemd regression is worked around yocto 1.4 release regression
>>> > is knocking at the door...
>>>
>>> yes :/
>>>
>>> If we need to maintain meta-systemd bbappends for upgrade path we can
>>> also maintain them with PN-systemd packages and good old systemd bbclass.
>>
>> Khem pinged me about this patch, so here are my thoughts:
>>
>> PN-systemd discussion in oe-core haven't moved a bit since this patch
>> was proposed, I think we have 2-3 options, lets vote about meta-systemd future.
>>
>> 1) keep PN-systemd separation in meta-systemd and fixup .bbappends to
>> resolve issues created by oe-core (I've fixed some issues, but still a
>> lot of issues in postinst/postrm/prerm scripts when both inits are
>> sharing the same PN)
>>
>> 2) give up on clean separation of PN-<init> and provide upgrade path
>> from PN-systemd to PN (mostly done by these 2 commits from khem).
>>
>> 3) give up on clean separation of PN-<init> and upgrade path (there are
>> many issues in other parts of oe-core, so upgrade path from danny to
>> dylan doesn't look usable.
>>
>> A) migrate existing .bbappends to their recipes to .bb files and
>> remove this layer. Any volunteer for that?
>>
>> My vote: 3) + A) later in 1.5
> A bit off-topic and mentioned here & there: I am working on
> network-manager(-applet) 0.9.8.0 update and xinput-calibrator. I could
> add 3) + A) for them if agreed. Hope to send them beginning of next
> week.
>
OK - just in case I have spare cycles to start this: I am afraid I
lost overview regarding distro features. I found systemd and sysvinit
which can be set both - still correct?

Ross could you please link me to the information how the use case 'I
want to build for both init-systems but let image decide which one to
use' is set up.
Sorry I didn't follow the last changes in detail because of other
issues (and personally I just use sysvinit).

Andreas

Re: systemd/sysvinit switches (was [meta-systemd][PATCH 1/2] Delete systemd class)

[Burton, Ross]

Hi Andreas,

On 12 April 2013 09:56, Andreas Müller <schnitzeltony@googlemail.com> wrote:
> OK - just in case I have spare cycles to start this: I am afraid I
> lost overview regarding distro features. I found systemd and sysvinit
> which can be set both - still correct?
>
> Ross could you please link me to the information how the use case 'I
> want to build for both init-systems but let image decide which one to
> use' is set up.
> Sorry I didn't follow the last changes in detail because of other
> issues (and personally I just use sysvinit).

It's in the manual!  Okay, only just, and it's still being worked on:

http://www.yoctoproject.org/docs/1.4/dev-manual/dev-manual.html#selecting-an-initialization-manager

(obviously the DISTRO_FEATURES = "systemd" bit is wrong, but you can
see what the intention is)

Ross

Re: systemd/sysvinit switches (was [meta-systemd][PATCH 1/2] Delete systemd class)

[Martin Jansa]

[-- Attachment #1: Type: text/plain, Size: 1343 bytes --]

On Fri, Apr 12, 2013 at 11:46:16AM +0100, Burton, Ross wrote:
> Hi Andreas,
> 
> On 12 April 2013 09:56, Andreas Müller <schnitzeltony@googlemail.com> wrote:
> > OK - just in case I have spare cycles to start this: I am afraid I
> > lost overview regarding distro features. I found systemd and sysvinit
> > which can be set both - still correct?
> >
> > Ross could you please link me to the information how the use case 'I
> > want to build for both init-systems but let image decide which one to
> > use' is set up.
> > Sorry I didn't follow the last changes in detail because of other
> > issues (and personally I just use sysvinit).
> 
> It's in the manual!  Okay, only just, and it's still being worked on:
> 
> http://www.yoctoproject.org/docs/1.4/dev-manual/dev-manual.html#selecting-an-initialization-manager
> 
> (obviously the DISTRO_FEATURES = "systemd" bit is wrong, but you can
> see what the intention is)

for systemd-only case it should also recommend:

# sysvinit scripts are often missing in recipe
BAD_RECOMMENDATIONS += "update-rc.d"

Because at least dbus/udev are missing sysvinit scripts without sysvinit in
DISTRO_FEATURES and postinst scripts are failing when there is
udpate-rc.d but not script which it is trying to enable..

-- 
Martin 'JaMa' Jansa     jabber: Martin.Jansa@gmail.com

[-- Attachment #2: Digital signature --]
[-- Type: application/pgp-signature, Size: 205 bytes --]