[yocto] [opkg-devel] [opkg-utils] Question: why update-alternatives from opkg-utils chooses /usr/lib to hold database?

[Alex Kiernan]

On Wed, Mar 6, 2019 at 11:37 PM Mark Hatle <mark.hatle at windriver.com> wrote:
>
> On 3/5/19 10:44 PM, Alex Kiernan wrote:
> > On Tue, Mar 5, 2019 at 10:50 PM Richard Purdie
> > <richard.purdie at linuxfoundation.org> wrote:
> >>
> >> On Tue, 2019-03-05 at 16:05 +0000, Alejandro Del Castillo wrote:
> >>>
> >>> On 3/5/19 12:11 AM, ChenQi wrote:
> >>>> Hi All,
> >>>>
> >>>> Recently I'm dealing with issue from which some discussion raises.
> >>>> I'd like to ask why update-alternatives from opkg-utils chooses
> >>>> /usr/lib
> >>>> to hold its alternatives database?
> >>>> I looked into debian, its update-alternatives chooses /var/lib by
> >>>> default.
> >>>> Is there some design consideration? Or some historical reason?
> >>>
> >>> Update-alternatives used to be on the opkg repo. I did a search
> >>> there
> >>> all the way to the first commit on 2008-12-15 [1], but even then
> >>> /usr/lib was used. I can't think of a design consideration that
> >>> would
> >>> make /usr/lib more palatable than the Debian default.
> >>>
> >>> Maybe someone with more knowledge of the previous history can chime
> >>> in?
> >>>
> >>> [1]
> >>> http://git.yoctoproject.org/clean/cgit.cgi/opkg/commit/?id=8bf49d16a637cca0cd116450dfcabc4c941baf6c
> >>
> >> I think the history is that the whole of /var was considered volatile
> >> and we wanted the alternatives data to stick around so it was put under
> >> /usr.
> >>
> >> That decision doesn't really make sense now since only parts of /var
> >> are volatile..
> >>
> >
> > I don't use opkg (or in fact any package manager on a target), but I
> > do use OSTree, where my /var isn't part of what gets deployed to a
> > device (https://ostree.readthedocs.io/en/latest/manual/adapting-existing/#adapting-existing-package-managers)
> > so having the option to keep it in /usr would be important to anyone
> > who has mechanisms like that.
> >
>
> Do you allow a device that has been sent an update via OSTree to then run
> update-alternatives to change what has been set by the update mechanism?
>

No.

> In my own uses of both OSTree and update-alternatives, I set this on a global
> basis and use it that way.  So no individual user (device) would be different
> then what was globally sent out.
>

Sounds like our use case is similar... variation is exactly what we're
trying to avoid, so we don't allow changes like that.

> If this is desired, then continuing to have a mechanism to allow it to be
> overridden for legacy or your purposes seems reasonable... but I think moving
> the default still makes a lot of sense.
>

I'd agree, it's clearly valuable for some, it's just how you accommodate both.

I guess so long as it's still changeable, it's not a problem.

-- 
Alex Kiernan