[Date Prev][Date Next] [Thread Prev][Thread Next] [Date Index] [Thread Index]

Re: debconf apt configurator

Adam Di Carlo <adam@onshore.com> writes:

> Okay, sounds good for the network part.  Regarding CD-ROM, I think it
> should *first* check in some reasonable locations on whether a CD-ROM
> is present.  I've CC'd <debian-cd>, perhaps they can give us some
> clues on platform-independant ways of doing this.  I would suppose
> checking for ISO filesystems and well-known top-level files which
> appear on the CD in well-known devices locations, such as /dev/hdc,
> /dev/cdrom, /dev/scd[01].  I wouldn't think it'd be *too* hard to do
> this.  A fallback would be to ask for the device where the CD-ROM is.

/cdrom should be the default location for the CD and the installation
process should set that up in the fstab acordingly. During
installation all available devices could be checked easily and the
files used to detect the right CD in dpkg-multicd can be used to
detect the drive as well.

> This whole "autosense" scenario, if doable, is attractive to me
> because then we could conceive of an automated installation media
> detection for a great number of cases (probably 90% of CD users).

I´m working on an autosense and CD find script that scanns /proc/pci
to detect what modules for ide or scsi should be used and scans for
the right CD there. The script is for my "Demo-Linux" filesystem that
boots directly from CD or any recognised filesystem. The according deb 
package should appear within this week (before university starts
again) but I need a sponsor for it.

> So assuming the autosense fails, perhaps then we would ask whether to
> use:
>   * network (see Joey's stuff)

Booting over NFS is an option of my "Demo-Linux" filesystem. The
Network will autoconfig/test itself. Should work find if dhcp or some
Win is running somewhere.

>   * CD-ROM in a location not autosensed, which would prompt for mount
>     point(s).  It should test that this is valid.
>   * NFS (could we inherit the location of the server used for TFTP in
>     case of that install method?), which needs 'server:share-name' and
>     possibly 'subdir/in/share'.  It should test that this is valid.

Hmm, when using tftp the Network should be configured already and a
ping to broadcast should show all hosts on that network segment. From
that list the DNS can be guessed, tried or selected and then the
nfs-server can be selected via a whiptail dialog.

>   * local disk, which needs 'device-name' and possibly
>     'subdir-in-partition'

I would realy like that. That way you can test a Debian CD before
burning it without the need to wipe an ide drive to simulate the CD.

>   * floppy (don't use apt at all, just emit a big ole' warning)
> One nicety would be overlaying network stuff over possibly staler
> CD-ROM archives, etc., but that's just fluff at this point.

Doesn´t apt do that already? or what do you mean?

> > Using debconf for the frontend might be controversial, but I think it can
> > handle all the types of prompting we might want to add to this, except
> > perhaps ordering a list of sources be priority, which I doubt we want to
> > trouble people with anyway.
> Well, its only controversial in that debconf would have to be added to
> base.  What do you think of that?  Are there any non-base depends that
> debconf has?

debconf should be in base.

> > One nice features comes to light if you test it with this command:
> > 
> > make clean test FRONTEND=dialog PACKAGE=aptconf PRIORITY=critical
> > 
> > That makes it skip some of the less important questions, which happen to be
> > those RMS wouldn't like and those I don't think we should trouble a newbie
> > with. :-)
> That is pretty sweet.

And is exaclty whats needed for the base installation.

May the Source be with you.

Reply to: