[arch-projects] [mkinitcpio] [PATCH] Add sd-resume hook.

[Dave Reisner]

On Wed, Aug 27, 2014 at 02:05:41AM +0400, Ivan Shapovalov wrote:
> On Tuesday 26 August 2014 at 17:42:54, Dave Reisner wrote:	
> > On Wed, Aug 27, 2014 at 12:47:10AM +0400, Ivan Shapovalov wrote:
> > > This depends on the systemd hook. It adds systemd-hibernate-resume at .service
> > > and systemd-hibernate-resume-generator which perform userspace parsing of
> > > resume= kernel parameter, allowing to specify the resume device by its
> > > persistent path ("resume=/dev/disk/by-foo/bar") or fstab-like specifier
> > > ("resume=FOO=bar"), just like the root= parameter.
> > > ---
> > > 
> > > Given that relevant functionality has just landed in systemd git, it makes
> > > sense to pipeline things a bit (even if actual applying will be delayed until
> > > systemd-217).
> > > 
> > > Also, does this need to be a separate hook or can be merged into 'systemd'
> > > hook itself?
> > 
> > This belongs in mkinitcpio proper, along with the systemd hooks
> > currently residing in the core/systemd package. You can blame me for the
> > latter not yet happening. Given that, we should probably add this to the
> > systemd package for now, as to not give people the idea that sd-resume
> > can be used without the systemd hook.
> > 
> > Thanks so much for pushing this work upstream!
> 
> I'm confused a bit.

Apparently, I am too. My brain is largely fried from being sick for a
week (and then traveling/partying copiously this past weekend).

Let's try this again -- this time with a bit more context/detail.

> Do you say that the 'systemd' hook currently in core/systemd shall be
> eventually moved to core/mkinitcpio?

Actually, no. I'm incorrectly recalling my own plan. What I should have
said is that the add_systemd_unit and add_udev_rule shell functions were
to be matured in core/systemd before adding them to the "functions" file
of mkinitcpio. I've not seen any bug reports since Tom Gundersen and I
wrote those (maybe a year ago?) so perhaps it's time to just go ahead
and do that.

I'm actually against moving the whole systemd hook into core/mkinitcpio
because it then puts mkinitcpio in a situation where it has to be
reactionary to backwards incompatible changes to systemd. This kind of
dependency has bitten me in the past (though offhand, I don't recall
which hook this was relevant to). Maybe Thomas recalls.

> Either way, there are 'sd-vconsole' and 'sd-shutdown' already in core/mkinitcpio.
> I think that if 'sd-resume' will be a separate hook, then either it should be
> placed here, or all three sd-* hooks should be moved to core/systemd.

Yes, you're right -- this contradicts my previous reply. Ultra
long-term, mkinitcpio may lose its shell-based early userspace and only
support systemd. That would mean that I ignore my paranoia about
backwards incompatible changes just deal with mkinitcpio sometimes
needing to move in lockstep with systemd.

> (The global alternative -- viability unknown -- is not to make 'sd-resume'
> a separate hook but to add all this directly from 'systemd' hook.)

Certainly viable. The extra binaries are quite small, and it'd be nice
if this "just worked" out of the box.

Hope this makes things a bit clearer...

d