Re: apt2ostree - build ostree images of apt based distributions

[Will Manley <will williammanley net>]

On Mon, 10 Dec 2018, at 20:28, Colin Walters wrote:
> You mention that you *could* be using rofiles-fuse; was there a specific 
> reason you didn't?
> Requirement on FUSE?

The second-stage implementation is really a demo rather than something that I'll actually be using, so I 
wanted to keep it as simple as possible as an example for others.  I'm not sure what the interaction between 
bwrap, fuse and sudo will be.  I'm sure you know better than I do based on your involvement with 
https://github.com/projectatomic/bubblewrap/issues/132 .  Also I don't think rofiles-fuse would save us 
anything here, we still need to checkout as root for the permissions to be correct.

For our real second-stage we avoid and copying root by using:

* a patched bwrap with overlayfs support: https://github.com/projectatomic/bubblewrap/pull/167
* fakeroot patched to understand ostree xattrs: https://github.com/stb-tester/fakeroot/commits/ostree
* Corresponding ostree patch which I need to get back to and tidy up for merging: 
https://github.com/ostreedev/ostree/pull/1678

Maybe this could be pushed into apt2ostree in the future, but it's pretty dirty and I think it makes more 
sense to concentrate on getting the core apt2ostree functionality reliable for now.  I also suspect that each 
user will have different constraints in this regard and may want to use docker for example rather than bwrap.

> It looks like a design goal of this is independence from the host, but 
> it'd be good to note somewhere what environment it's being tested/
> developed on (are you using VMs or containers?).

Right now I'm developing on Ubuntu 18.04 although using the word "tested" would be premature at this point 
;).  I'll add this to the README.

> I'd be happy moving this under the ostreedev/ organization if you like 
> that idea.
>
> Also, no harm in sending a PR to link to it in the README.md now.

I do like that idea, but it's not ready yet.  I want to gain a bit of confidence in it before publicising it 
more widely.  The minimum bar for that is us using it instead of multistrap in the build process of our 
product. Hopefully this shouldn't take too long.  Once I know it's producing usable images I'll let you know 
and we can move it over, mention it in READMEs and publicise it on debian-devel, etc.

Thanks

Will