Re: diskles-image-*
>>>>> "Adam" == Adam Di Carlo <adam@onshore.com> writes:
>> debconf-tiny sounds like it might be the best approach. More
>> standard. I have no idea though what debconf-tiny is, or how to
>> program debconf even - these are issues I am going to find out
>> soon.
Adam> The nice thing is that we could run 'dpkg-reconfigure
Adam> <whatever>' to rerun the debconf configuration whenever.
Not that I fully understand debconf yet, but agree this would be
an advantage.
(Does dpkg-reconfigure re-invoke the preinst and/or postinst scripts?)
I made a fully debconf version of my diskless-image-* packages, when I
suddenly realized, this will only work if the base*.tgz file contains
some sort of debconf support. What should I do? Try and manually
install debconf (this is inconvenient to the user), or does the
base*.tgz file already have some support for debconf?
What is the difference between debconf and debconf-tiny? Are there any
things I should watch out for so my packages are debconf-tiny
compatable?
Adam> Ideally, base packages themselves should all support
Adam> debconf. However, that is not yet so. I am conflicted
Adam> whether we should try to take at least the postinstallation
Adam> tasks and debconf-tiny'ify them, and try to either:
Adam> (a) get the upstream base pkg maintainers to add the
Adam> scripts (probably not possible)
I think this sounds the ideal solution. When you say "probably not
possible", do you mean for technical reasons, or practical reasons (ie
updating every base package).
Adam> (b) make a new little package, base-config which has all
Adam> this stuff, then you cuold do 'dpkg-reconfigure base-config'
Adam> or some such
I think this is the next best thing to (a).
Adam> (c) just have the scripts be part of the boot-floppies
Adam> themselves
I don't think I see any value in this approach, but it would work...
>> It has been a while since I looked at this configuration stuff.
>> Wondering aloud: Are there any parts that are not appropriate
>> for NFS-Root systems? Probably the network configuration is
>> redundant (the NFS-Root kernel automatically detects this at
>> startup).
Adam> Shaleh and Marcel are working on bootp/dhcp for this, which
Adam> is more appropriate.
The difference with diskless boot, is that the network is already fully
configured by the kernel before any user space process (including
/sbin/init) starts to execute.
However, the current kernel versions fail to configure 127.0.0.1
>> The loopback (ie localhost) device must be configured though.
Adam> Ah yes, I think that is done, but only if there is no other
Adam> network....
Sure? I think 127.0.0.1 must always be manually configured, even when
there is another network connection. Otherwise programs that try to
contact localhost will fail.
--
Brian May <bam@debian.org>
Reply to: