Re: Adapting OSTree to Yocto pseudo

[Colin Walters <walters verbum org>]

On Tue, Oct 4, 2016, at 02:24 PM, Anton Gerasimov wrote:
> Hi all,
>
> I'm now working on integration of OSTree into Yocto build process.
> Basically, it lets user automatically commit generated rootfs image to a
> chosen OSTree repository and later deploy them on target devices. If you
> are interested in details, you can look at Automotive Grade Linux
> repository
> (https://git.automotivelinux.org/gerrit/#/admin/projects/AGL/meta-agl-extra),
> namely at meta-agl-extra/meta-sota/rescipes-sota/ostree_git.bb recipe
> and meta-agl-extra/meta-sota/classes/image_types_ostree.bbclass.

Cool, that's great to hear.  Supporting embedded systems in general,
and Yocto in particular is very important to me.  Although my employer's
use of OSTree is for more general purpose systems, I think OSTree has
strengths for classical embedded versus the alternatives (package updates, dual partitions),
and just like the Linux kernel, there's a lot of benefit in sharing code like this across
both "big" and small systems.

So I'm happy to make changes to improve Yocto and embedded in general.

> So I wonder if you consider it a good idea and are interested in
> compatibility with fake root enviroments or it is a non-goal.

Absolutely, and for anyone interested in the details, see:
https://github.com/GNOME/libglnx/pull/27

I will say in the bigger picture though, Yocto (well really wrpseudo) at some point would
gain a lot of benefit from taking a hard dependency on using
modern Linux containers (e.g. CLONE_NEWUSER or https://github.com/projectatomic/bubblewrap)
and likely FUSE as well.

It's a lot more performant than LD_PRELOAD and gets one out
of the never-ending game of tracking kernel system calls.

Also, I want to note that OSTree was explicitly designed to support
unprivileged use as Yocto is doing - things like the --owner-uid argument
to commit allow writing files into the repo that are root-owned in
the data, just like fakeroot + tar.