On 08/31/2015 11:41 AM, Simo Sorce wrote:
On Mon, 2015-08-31 at 10:18 -0700, Brendan Conoboy wrote:
On 08/31/2015 08:17 AM, Harald Hoyer wrote:
[snip]
Minutes:
<http://meetbot.fedoraproject.org/fedora-meeting-2/2015-08-31/fedora_base_design_working_group.2015-08-31-14.15.html>
Minutes (text):
<http://meetbot.fedoraproject.org/fedora-meeting-2/2015-08-31/fedora_base_design_working_group.2015-08-31-14.15.txt>
Log:
<http://meetbot.fedoraproject.org/fedora-meeting-2/2015-08-31/fedora_base_design_working_group.2015-08-31-14.15.log.html>
For today's meeting we didn't really use zodbot minute keeping
features, so in the interest of sparking some discussion I'd like to
recap.
At Flock 2015 there was a 2 hour session on the subject of rings which
are basically policy zones inhabited by packages. Right now fedora
packaging has 1 policy, so the entire OS is in a single ring.
Creating more rings means creating more policies. By doing this
Fedora can become adopt the flexible to appeal to diverse development
communities and thus grow. The general consensus at Flock was that
Environments and Stacks should take the lead in helping to define new
rings, and especially how the rings interact. As a side note,
everyone agreed the word "rings" breaks down the further you get away
from the center, but nobody has come up with something better yet
(Venns? Blobs? Zones?). A week or so ago, a small Environments and
Stacks meeting took place where it was generally agreed that the Base
working group was the right place to define Ring 0. That brings us to
this morning's BaseWG meeting. I talked a lot so here is a rough
recap integrating a few of the questions and comments people had (Do
read the log if you have time!)
Right now the Fedora distribution is 1 ring, let's call it ring 1.
The distribution contains an operating system and numerous
applications that run on that operating system. When we talk about
defining ring 0 we're really talking about distinguishing between the
operating system and the applications that run on top of it.
We want to go from this:
Ring 1: The Fedora Distribution
To this:
The Fedora Distribution:
Ring 0: The Linux Operating System
Ring 1: The Applications and Stacks
It seems quite modest, but working through the details on what this
means is hard. What is an operating system in the Linux context?
Ring 0 will likely have the strictest set of policies of all the
rings, so we want to keep it as small as possible, but it is more than
a minimal install. These are the traits of rings in general and ring
0 in particular as I see it:
1. Ring 0 is a repository of rpm packages built in koji.
2. Ring 0 contains, but is not limited to, the minimal install of
packages to go from Power On to a login prompt.
3. Ring 0 passes repoclosure on its own (Packages listed as hard
"Requires" in a ring 0 spec file are themselves are implicitly ring 0).
4. Ring 0 is not self hosting. Packages listed in "BuildRequires" do
not need to be members of Ring 1. This isn't ideal, but it's a
practical consideration.
5. Ring membership is at the source package level, not the binary
package. If one source package's binary/noarch sub-package is in ring
0, all sub-packages are in ring 0.
Can you elaborate more on this point (5) ?
I can totally see how a package may be critical and therefore deserve to
be in ring 0 and yet have optional features in terms of subpackages that
are not necessarily ring 0.
For example some library that offers optionally bindings for somewhat
rarely used languages (say ocaml). The subpackages for those bindings
shouldn't necessarily be ring 0.
Or did I misunderstand the point ?
Let's take gcc for example. The gcc package produces numerous
subpackages including various compilers and libraries. One of those
libraries, libgcc, is linked into nearly every dynamically linked ELF
executable on your system (Run ldd to confirm), including /sbin/init.
Since we really want /sbin/init to function we need to include
libgcc in ring 0. Since ring membership is at a source level rather
than a sub-package level, gcc is ring 0. There are other good
arguments for the executable-generating tools living in ring 0, but
this one illustrates the point.
Ring policies might include things like how long the package is
supported, what build system can generate it, what release cycle it is
on, what the packaging guidelines are, what the updates rebase policy
is, and so forth. These types of policies apply to source rpms as a
whole. You can't apply one to gcc-c++ and another to libgcc.
I think the question you are posing might be, in the context of gcc,
might be: Does gcc-g++ need to appear in the same repository as
libgcc? IE, can we decouple what we build from what is presented to
users? If the threshold for must-include is repoclosure then no, we
don't need to include gcc-c++, we only need to include libgcc and make
gcc-c++ available elsewhere. In talking to Langdon about this, he had
the idea of "ring 0" and "ring 0 prime" where prime is ring 0 plus
something. We weren't sure what the something might be, but perhaps
it could be the optional sub-packages.
6. Ring 0 contains the minimal libraries that define the OS API/ABI,
such as glibc. This probably happens implicitly by #3, repoclosure.
7. Ring 0 contains the tools needed to update existing packages and
install new packages.
That's the starting point, but it is by no means comprehensive. The
OS probably provides specific services beyond the ability to login,
for instance. Which styles of boot are supported? Where does
installation infrastructure like anaconda land? This is equal parts
philosophy and practicality. Also, policies for ring 0 may differ
from what Base has previously adopted: Do we create a ring 0 minimal
compose since we already need to check repoclosure? This might be a
great way to refactor primary/secondary such that we can gracefully
transition i686 down and secondary arches up. Lots of opportunities,
much to consider.
--
Brendan Conoboy / Red Hat, Inc. / blc@xxxxxxxxxx
--
Brendan Conoboy / Red Hat, Inc. / blc@xxxxxxxxxx
--
devel mailing list
devel@xxxxxxxxxxxxxxxxxxxxxxx
https://admin.fedoraproject.org/mailman/listinfo/devel
Fedora Code of Conduct: http://fedoraproject.org/code-of-conduct
