>>>>> "Luca" == Luca Boccassi <bluca@debian.org> writes:
Luca> On Tue, 2021-08-17 at 12:07 +0200, David Kalnischkies wrote:
Luca> If src:usrmerge is made transitively-essential, from that
Luca> point onward it wouldn't matter if a package is no longer
Luca> compatible with the legacy split-usr setup, no?
No, there are upgrades to consider.
We know at the end of the upgrade all the essential packages are going
to be installed.
But especially within the pseudo essential set we do not typically have
ordering guarantees.
So, we generally assume that we need to wait until the release after
such a transition is introduced to depend on it.
So, we can depend on usrmerge for bookworm+1 but not for bookworm.
That is at least Simon's position.
Several people have argued that's not actually what the TC said, but
it's certainly how we normally operate, and at least one prominent
member of the TC appears to be saying that is what the TC meant.
So the current assumption in the discussion is that packages inthe
bookworm development cycle must work both with and without usrmerge.
If you propose a transition faster than that, you have a lot of
difficult questions to answer about corner cases involvind partial
upgrades and what happens during upgrades.
In order to build packages that work on a non-usrmerge system, you need
a build chroot that is not usrmerge.
There are a couple of consequences of that:
1) We need to support non-usrmerge build chroots through the bookworm
cycle.
2) If you are going to automatically upgrade systems for example by
having usrmerge become essential, you need a way to exempt at least
build chroots.
--Sam
Attachment:
signature.asc
Description: PGP signature