[Piuparts-devel] Bug#1036705: override: adduser:admin/required

[Helmut Grohne]

On Wed, May 24, 2023 at 06:54:01PM +0200, Cyril Brulebois wrote:
> Watching from the sideline, this seems to come in horribly late.

How am I not to agree with this?

> > apt used to depend on adduser and apt is required, so adduser is
> > transitively required in bullseye. Johannes and myself worked towards
> > making apt not depend on adduser and that work succeeded.
> 
> FSVO “success” then, given the rest of the mail…

I'm really sorry about this. None of us saw the deluser breakage coming.
After all, we were "just" killing a dependency. We should have noticed
that it was the last and thus possibly having bad effects, yes. We did
not. When I caught one of Andreas' bug reports about this, I immediately
informed the release team to not loose any further time. It was already
horribly late back then. :-(

> Via olasd/#debian-release: adduser got that field, not apt.

Thanks.

> Same question as before, why not just add the dependency back?

That dependency is conceptually wrong now. apt does not need adduser
anymore. I think the initial idea was to add it back, but Julian rightly
pushed back on this.

A major technical goal was to push adduser out of the essential+apt
package set (which hints that we should have paid more attention,
sorry). Adding this dependency breaks that goal while adding protected
or required does not, so we'd actually get what we wanted.

> Aren't we risking a redux of “we turned another knob, and now we're
> discovering yet another issue”?

It is very difficult to disagree with this one given that I thought
"Protected: yes" to be harmless earlier.

> But I'm very much worried about possible side effects at this critical
> stage of the freeze.

I will not stand in the way of turning this back and adding the
dependency back to apt. It seemed to me though that this was not the
preferred solution and that a (FSVO) better solution was available.

In theory, "Protected: yes" should solve the issue for purging. It just
happens that piuparts does deal well with this, so the remaining issue
is one of having broken a QA tool rather than having broken something
for real. I can try talking to Nicolas about possibilites of adapting
piuparts instead.

Helmut