[Mageia-dev] [Mageia 2 specifications] Systemd or not systemd

[Olivier Blin]

Colin Guthrie <mageia at colin.guthr.ie> writes:

>> Secondly, what should be the correct way of supporting systemd in a
>> package? In Mandriva, I thought on adding a --with flag to
>> enable/disable systemd, but in most cases it does (almost) nothing. All
>> services which want to support systemd only need to place their files
>> into /lib/systemd - and that's it. Should we support opting-out of
>> systemd in specs? I believe fcrozat is having the same dilemma in SuSE
>> now as well, and he settled on some common packaging macros.
>
> IMO, unless something goes horribly wrong with systemd (which I very
> seriously doubt will happen - it's actually been amazing to watch this
> project from birth to wide adoption) I'd say we just jump right in and
> package the units unconditionally. They don't really take up much space
> and do nothing if you choose to not use it for the time being. I think
> by mga3 we'll likely *only* offer systemd (if not in mga2), so I'd say
> putting in effort to opt out now is pretty much not going to be worth it.

I agree, it does not seem useful to avoid shipping systemd services.

Though, when systemd is enabled, what happens if one has both systemd +
sysvinit services files installed for a given package?
Will the initscript be run in addition to the systemd service?

>> Almost finally, should the systemd files belong to the main package, the
>> same way as they do with initscripts-based one (e.g., the package would
>> provide /lib/systemd/system/%{name}.service together with
>> %_sysconfig/rc.d/init.d/%{name} for example), with no extra subpackages
>> or flags - or should all systemd-specific files go into %{name}-systemd
>> package for example? What do you think?
>
> I vote to put them in the same place as the initscripts-based ones. As I
> predicted above, I reckon it won't take too long for systemd to be the
> only system offered and obsoleting all those subpackages would be a bit
> of a PITA down the line. And if systemd isn't used those files do
> nothing, so no harm done.

Agreed as well.

>> And finally, what does seems to be the best way of starting to use
>> systemd in cauldron? I have thought on 3 alternatives:
>>  - easy way, only having it packaged, but not
>> providing/obsoleting/conflicting with sysvinit. This way, it will work
>> when kernel is booted with init=/bin/systemd (the least invasive way)
>>  - compatible way (like in Mandriva) - it is available, systemd-sysvinit
>> conflicts with sysvinit, so if someone installs systemd-sysvinit,
>> sysvinit goes away and systemd is run by default. This seems to be the
>> most sane way to me (but I could be biased), and it is easiest one for
>> testing
>>  - ultimate way - systemd provides and obsoletes sysvinit and its
>> goodies. This way, systemd will be the only one (e.g., highlander
>> style). This is how fedora did it if I am not mistaken, but I am not
>> sure if it the best way.
>> 
>> So, that's it for now from my part..
>
> I'd like option 2 please! :) It doesn't exclude using option 1 along
> with it (I was doing this for a while), so it's most flexible.

Option 2 seems good as well, but does it really have to uninstall
sysvinit? Isn't it enough to put some alternatives symlinks for
/sbin/init?

-- 
Olivier Blin - blino