[arch-general] [arch-dev-public] nilfs-utils moving in core

[Loui Chang]

On Tue 07 Dec 2010 18:30 +0100, Heiko Baums wrote:
> Am Tue, 07 Dec 2010 16:26:00 +0100
> schrieb Pierre Schmitz <pierre at archlinux.de>:
> 
> > I second this. If the reason for moving a package to core is that the
> > installer cannot handle it otherwise the installer needs to be fixed.
> 
> The question is not that the installer can't handle it if it's not in
> [core]. The question is that if the installer creates a filesystem on
> the harddisk the appropriate tools need to be installed on the system,
> too. And this is only possible if those packages are in [core] because
> the core iso only contains the [core] repo and not [extra] and
> [community]. Also the netinstall iso can only install from [core] as
> far as I know. So if AIF supports nilfs or another filesystem, those
> tools have to be moved to [core].
> 
> > That's why I would vote against moving it to core. I'd even say we
> > should have a look at those packages in core with low usage and see
> > if we should move them to extra. There are already packages for
> > which we don't get any sign-offs which shows that those are no
> > longer needed to be in core.
> 
> That's definitely not the definition of [core]. [core] is not for
> supporting the most used and most popular packages. [core] is for
> providing system relevant packages which are necessary to build a
> minimal system.
> 
> And it's not the dev's task to decide how a user wants to partition
> and format his harddisk.
> 
> If it's done like you're suggesting than you would force a user to
> only format his harddisk with e.g. ext4, because you think that ext4
> is the one and only and most popular filesystem and every other
> filesystem has to be moved to [extra], because they are not necessary
> or not as popular as ext4.
> 
> It's all about choice - the user's, not the dev's choice.

Man this stuff is hard to read.
Anyways, I would like to consider the installer as something separate
from the core system. Sure it can help you install the core system, and
any extras you may want. That shouldn't mean that the core needs to
conform to what the installer enables you to do.


> Instead every filesystem - probably except for dosfsutils - has to be
> moved to [core] and be supported by AIF, because a filesystem is the
> most basic and necessary part of a system which has to be chosen, used
> and installed during installation. You simply can't reformat or
> repartition a harddisk after the installation. Well, it's possible but
> not without having a second harddisk onto which the installed system
> had first to be copied. But that's not the way a system should be
> installed.

Bull.

> So every filesystem - regardless of how much it is used and how popular
> it is - has definitely to be moved to [core] and supported by AIF.
> 
> And, no, ext2 is not the only filesystem which can be used for /boot
> and /.
> 
> But on the other hand every filesystem related package has to be
> removed from (base), while AIF should then be able to recognize which
> filesystems are created on the harddisk and install the appropriate
> filesystem tools automagically.
> 
> > The idea of core was to provide a minimal set of packages that are
> > needed by nearly all users to set up a base system. Our sign-off
> > procedure ensures that we don't put broken packages by accident there.
> 
> But filesystems - all of them - belong to the very minimal set of
> packages that are needed to set up a base system. But every user needs
> a different filesystem for a reason, because every user has different
> requirements.
> 
> experimental != broken
> 
> Just because a filesystem is marked as experimental doesn't mean that
> it's broken. Btrfs e.g. is missing a fsck, yet, but it shall be pretty
> stable until the harddisk gets too cluttered. At least it's said by
> many people in the web. So there are several people who like to use and
> to test it.
> 
> As soon as there's a first stable version of a package it should of
> course not be updated anymore to newer, unstable versions.
> 
> Of course, those experimental packages should explicitly be marked as
> experimental in AIF.
> 
> There have been, btw., several kernel modules in the vanilla kernel
> which have been marked as experimental for many years even if they
> haven been really stable in the meantime. So the word experimental is
> quite relative.
> 
> > I don't think that nilfs matches the criteria needed for inclusion in
> > core. (side note: it has 1.38% usage according to pkgstats)
> 
> Why does it only have 1.38% usage? Just because it's not in [core] and
> not supported by AIF, just because a filesystem is usually chose at
> install time.

And poppycock.
Ask Microsoft to support every filesystem in existence on their standard
install CD and core system and maybe us poor Archers with our limited
time and budget can also rise to your stratospheric expectations.