[arch-dev-public] definition of the core repo, putting nilfs-utils, btrfs-progs and dosfstools in core

Thomas Bächler thomas at archlinux.org
Sat Jan 8 15:45:13 EST 2011


Am 22.12.2010 20:26, schrieb Pierre Schmitz:
> I think the core repo and its sign-off policy has proven itself to be a
> good idea. But we could make a little adjustment to our signoff policy
> to solve your filesystem package problem.

Agreed.

> [core]: This contains everything you need to boot up, connect to the
> internet and install additional packages from e.g. [extra]

Sorry for being late on this, but this is what I remember from when we
redefined the repos - [current] and [extra] was not completely
well-defined, so we came up with this (which is mostly what Pierre said):

[core] was meant to be the core of Arch, which was supposed to be on the
installer. This means:

* Packages that are needed to boot (C library, kernel, init scripts,
filesystem, system logger, at least one bootloader).
* Packages that may be needed to connect to the internet (dhcp client,
wireless management tools, pp(t)p, various common VPN clients) - for
installing more packages.
* Essential package building: default compiler, fakeroot, other support
tools for makepkg
* Packages for file system management: mkfs and fsck tools for common
file systems (this would include btrfs tools, if btrfs is popular enough).
* Packages that do not match any of the above criteria, but that
virtually anyone will want or need early in the system setup process.
The only one I can think of is openssh.
* Dependencies (but not necessarily makedepends) of the above.

This was never written down this explicitly, but this is how I remember
the consensus we made back then. tpowa and andy were iirc two of the
driving forces behind the repository transition, I compiled the initial
list of packages.

> base group: A smaller subset of [core] that include packages that
> should be installed on every Arch system.

That was the idea. Apart from wpa_supplicant, this is okay the way it is
(wpa_supplicant is afaik still in base, but shouldn't be).

> base-devel group: Additional packages needed to build our base
> packages. (This group is indeed questionable and one might consider
> moving them to [extra]

The idea was to run 'pacman -S base-devel' and have all essential
support tools for makepkg. I'd keep them in core - at least gcc will
stay in core, unless we want to make two separate PKGBUILDs for gcc and
gcc-libs.

Having them in core is a good idea, as someone might want to do a core
install and makepkg something before being able to continue (to compile
an important network access tool).

> Everything else in core are packages that are not needed by everybody
> but required by some to "boot up, connect to the internet and install
> additional packages"; e.g. file system packages, firmware for your
> wireless card, wireless_tools etc..

Agreed in principle, but I went into the details more above.

> I would suggest to change our policy to this:
> * packages in the base group and its dependencies still need the usual
> two sign-off per architecture

One sign-off is implicit by the package builder, so we always had one
extra signoff from someone else.

I would also only require one sign-off for -any packages, instead of one
per architecture.

> * sign-offs for all other packages in core are optional; they still
> need to enter testing first, but can be moved to core without any
> sign-off after 3 days (or one week or whatever)

Nice idea.

> The install CD would than contain the full core repo.

As was originally intended.

> What do you think of this proposal?

+1

-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 262 bytes
Desc: OpenPGP digital signature
URL: <http://mailman.archlinux.org/pipermail/arch-dev-public/attachments/20110108/f1362143/attachment.asc>


More information about the arch-dev-public mailing list