Re: manual selection of devices
Adam Di Carlo wrote:
> Glenn McGrath <Glenn.McGrath@jcu.edu.au> writes:
> > 1) to manually select a device i think the best approach is to add a
> > manual option to the list of valid partitions (in select_not_mounted.c)
> > when this is selected a box allows you to browse the deivce subdirectory
> > and select the device you want (like tzconfig) then generate a partition
> > struct for it and parse it back to select not mounted, from then on it
> > can automatically be used by the rest of dbootstrap.
> Yup -- as a quick hack, you can just allow the user to type in the device.
Yea, i thought of that, but whasnt sure how to implement that with
existing dbootstrap routines (i didnt look too hard though)
> > 2) timezone configuration (tzconfig.c) is basically a filesystem browser
> > that passes the selected file to another function. If we pull theis
> > function appart (and move it to utils.c ?) we can make it a general
> > purpose filesystem browser/selector that could be used by other sections
> > of dbootstrap.
> Yeah, that would be nice. I was thinking the same thing for the
> "install kernel and modules" and "install base" manual list selection
> options. It would be really nice if the user could browse the dirs
> there rather than typing them in manually.
Cool, sounds like having a seperate routine would be good then.
The general filesystem browser is basically all there in tzconfig.c,
tzconfig would be simplified heaps.
functions that would be "poached" into the filebrowser would be
configure_timezone_step1 from tzconfig.c, this has to be modified as its
hardcoded to call write_timezone which is where it uses the selected
read_directory from tzconfig.c this wouldnt have to be changed at all.
> > 3) If manual selection option is always there then it will break NFS as
> > NFS option only occurs when there are no options. If manual selection
> > isnt always present as an installation target, then how should it
> > behave?
> Actually, NFS appears if and only if the network has been configured.
> The reminds me that I feel that we will need to move network selection
> higher up in the list -- here is where we might have some options,
> like "configure network manually", "configure with dhcp", "configure
> with bootp", or "don't configure the network yet".
> > 4) Being able to manually select the partition should enable installing
> > to loop and raid(with some manual work from console) and other
> > non-standard block devices.
> That would be great...
> Are you implementing the stuff where you are sniffing around in
> /proc/ide and /proc/scsi for devices? I'm a little skeptical we could
> get this in time but it would be really nice.
No, i wasnt doing anything fancy, i was thinking fairly superficially,
if you needed to install to a block device that dbootstrap doesnt
otherwise recognise (e.g md? ) then you could go to the console prepare
the device (for raid, make an //etc/raidtab and run mkraid) then select
it manually as an install target.
Im not totally sure about how loop devices, so maybe work would be
required for that... dunno.
I didnt think of looking through /proc.
I didnt even think of looking around /proc for the block devices.
> > This should be pretty easy, but i have discovered how bad my c skills
> > are, i got it to the stage where you browse the filesysytem, you can
> > select a file, but it segfaults when i try and change dir.
> > Anyway its getting pretty messy, when i get back ill try pulling
> > tzconfig appart first and making that re-usable.
> Sounds like a plan!
> .....Adam Di Carlo....adam@onShore.com.....<URL:http://www.onShore.com/>