On Mon, 2004-08-23 at 07:46 -0400, Alan Cox wrote:
On Sun, Aug 22, 2004 at 09:37:56PM -0400, Havoc Pennington wrote:
> BTW, when using NetworkManager for desktops, do we really need network-
> scripts and sysconfig/networking at all?
Define "desktop". If you mean "dhcp managed disposable splat" then
probably
not, but desktop also covers a large multitude of other things including
secure IPsec networks, static addressing, cable modems, ppp, ...
I'm not trying to say "there are no config files," rather "maybe
there's
no point writing one out if it just says BOOTPROTO=dhcp and nothing
else, as the ones kudzu made for me all currently do - just leave them
blank by default and let NetworkManager decide what to do"
The kernel names interfaces it finds eth0, eth1, ... etc (or ppp0 and
so on).
It provides the ability to rename them as you like (so you can rename them
to "ADSL" "DMZ" "DavesRoom" and so on...). The current
process is meant to be
Driver gets loaded
We peer at the MAC
We decide what eth* device it was before from the MAC (optional)
We rename it
We run the scripts
What's the story on this "wifi0" "sit0" stuff, btw?
> One policy we might try to get closer to: clearly split autogenerated
> information from manually-edited information, and machine-readable
> information from human-only information such as scripts.
The how does the user edit it - and don't say "with the config tools" they
are not accessible or complete enough.
You edit it by changing the manually-edited information. It's the same
idea as the way many config files have a "<include>site-
local.conf</include>" type of statement, where you are supposed to
modify site-local.conf instead of the stock file.
In an ideal system, you might have a file (probably in /var) that
defines the autoprobe results; you have a config file (in /etc) that is
fully machine-readable and defines how to use and/or massage/override
the autoprobe results; from the config file, you point to script hooks
called at well-defined times. The config file should not itself be a
script, because that is not machine-editable.
I don't think we're that far from this now. hwconf is the autoprobe
results for example. kudzu seems to also look at the manual-edited files
though. I guess what I'm saying is have some default behavior as a
function of autoprobe results, and view the manual-edited files as
override of the defaults - then machine-writing defaults into the
manual-edited files isn't necessary.
It's not really a big deal, I'm just wondering how to make things more
robust.
Havoc