cli: Add build-commit

[jmarrero]

Closes #216

[jmarrero]

mm I think I need to use the ostree-ext crate directly instead of just calling ostree.

[vrothberg]

LGTM

[sallyom]

@jmarrero do you want to convert to PR?

[cgwalters]

So one big thing we could do here in general is a lot of what happens today in the rpm-ostree postprocessing phase; for example, a big one is handling /etc/passwd and generating sysusers fragments.

Another interesting one is kernel handling; erroring out if you have multiple kernels, cleaning up the duplicate content the Fedora kernel packages drop in /boot etc.

However...lately I'm also thinking just for maximum compatibility we should start to do that stuff on the client too automatically.

[jmarrero]

Did not get it out of Draft because of background chats with Colin, sorry for not updating sooner. I think we wanted to make this bootc native code instead of calling back to ostree to reduce the dep on ostree and on top of that, the logic with /var changes needs to be revised.

What do you all think about:

1. check for stuff in /tmp and error out
2. relax the /var errors make it warning
3. any other check?

[cgwalters]

Or maybe what would be best is something like a generalized "hooks" framework, i.e. we have /usr/lib/bootc/container-commit.d that contains drop-ins that can execute at build time. That would make everything easily extensible...and today we have sysusers code in rpm-ostree that we could keep where it is, and have e.g.

cat /usr/lib/bootc/container-commit.d/10-rpmostree-sysusers
#!/bin/bash
exec rpm-ostree postprocess sysusers

or so?

[jmarrero]

I guess that would work, but we would need also to scan the dropins and skip anything calling binaries not present on the system? Or just fail with a warning? Because, with your example we would fail the build on an image that does not have rpm-ostree.