Q: Why has this been implemented so suddenly?
A: That's a good question, while we have discussed this topic multiple
times in the past and also issued a concrete proposal to arch-dev-
public (plus its follow-up summary) no further steps were taken to
get something testable. However, during arch-conf in Berlin, the
general enthusiasm and acceleration of the group was so strong, that
we have warp-10 driven this task. We acknowledge that this could have
been handled more carefully, but in the spirit of arch-conf please
excuse us and bear with us. For instance, we could have noticed
during a test phase the consequences of not having systemd-
sysvcompat. For the time being, we have added it to the base package
as the implications were not explicitly desired and discussed
beforehand, however that topic will be covered in a follow-up.
Q: Why has the group been superseded by a meta-package?
A: The difference with the new base package is that this defines the
level at which we tell you, you have modified your system
sufficiently that you break it, you bought it -- it is your
responsibility to debug issues caused by your overriding basic
assumptions of the system.
This has theoretically already been the case before this change as
packages in the old `base` group were expected to be installed by a
lot of packages. On the other hand, it has never explicitly been
stated that this indeed is a requirement. The consequence is that
packages may potentially misbehave or fail to properly install.
Following this reasoning, it's a very natural choice to use a meta-
package. It allows us to reflect changes (additions or removals) in
the required package set straight on user installs with the next
system upgrade which allows preserving the consistency of those
assumptions. A simple example would be the systemd package, no matter
what use-case you have (like a tiny container) we assume you must
have systemd installed as its being relied on (sysusers, tmpdirs,
Q: Why has it been shrunk down and doesn't contain $package anymore?
A: The base meta-package is meant to be the lowest common denominator
across different use-cases like desktop usage, bare-metal server or
container runtime and therefore defines the foundation people can
build their desired environment upon. The old base group did not just
contain too many unnecessary packages for us to reasonably be able to
call it the lowest common denominator, it was also inconsistent (for
example it shipped reiserfsprogs but not btrfs-progs).
After reflecting on all of the feedback so far, it also became quite
clear that the old base group was kind of misused as a poor man's
installer. There is nothing wrong with such a use case per se, but if
there is a desire to solve this on our side, we should find the
optimal solution instead of just sticking to some legacy. Be it an
actual installer, something like a base-extras group, or nothing at
all, it is out of scope here, but we should look into it separately
-- so maybe its a bit of xkcd#1172.
Q: Why couldn't we just add a new package and leave the base group as it
A: The primary reason is that base (as the name indicates) shall be the
minimal foundation and denominator of different use-cases. If we
would introduce anything other than 'base' to fulfill this job, like
base-system, base-minimal, base-foundation, or whatever one could
come up with, it wouldn't get the point across as clearly as the name
'base' itself. The historic knowledge is mostly responsible for our
bias here, but it is clearly easier to grasp the intention compared
to the difference between having `base` and `base-minimal`.
Q: Should we update the dependencies of our packages to depend on base
instead of on its members?
A: Nothing has changed in that regard and there is no strict rule for do
or don't yet. This is in general more related to the topic of
transitive dependencies which needs to be discussed eventually. My
recommendation is to just keep the members that are a first-level
dependency because there is absolutely no point in making every
single package in the universe depend on the base meta-package.
However, it can be correct, clean and useful to list your first-level
dependencies explicitly instead of not at all.
Q: What happens next, are there any plans?
A: The most important part was getting this abstract out to provide
information and aid in resolving some of the confusion. The next
steps will include multiple follow-up topics to iterate over and
decide case by case -- for example how we want to handle systemd-
sysvcompat and other topics that emerged.