[pkg-fso-maint] GSoC 2010: Debian Installer on Freerunner

["Steffen Möller"]

Hello,

> > The minimal plan (IMHO) is to produce an SD that a regular user can put
> into the device and boot from. If you say that you want to produce the
> device with the OpenMoko and that you have a good idea how this will work,
> fine. My hunch is, though, that it will be much easier to achieve with a
> virtual machine that has access to the SD. You will be sufficiently busy to set
> this up (newer kernels make problems with ARM QEMU on amd64, at least for
> me) and then there is enough to change to have d-i ask the right questions
> and generally for to it what is in install.sh .  Once you have set this all
> up, then I suggest to prepare a Debian package that automates your setup,
> allows the user to push an SD into their desktop's slot and prepare the image
> with the d-i.
> > 
> > If you want to do it differently, then this is just fine with me,
> really.
> 
> Well, yeah, my initial goal was to have d-i running on the Freerunner,
> but the QEMU approach would be a good start.
> 
> > Also, I expect some other readers of this list to be not really happy
> with what I was proposing here, preferring just something that works from SHR
> or whatever boots the device, directly.  So, dear all, speak out, maybe
> the perfect mentor is crystallizing from this now pending discussion .... or
> Thibaut will just be community-mentored :)
> 
> Well, for that, I would have tried kexec, but it would require to stop
> the host system anyway.
> But if I get to run d-i on the FreeRunner via a µSD card, that's no more
> than a small additional step!

There is another side to it: we need to explain what we are doing to the average Joe Debian enthusiast so that individual can contribute without becoming an expert in embedded computing. We won't beat install.sh on this front, but whatever we come up with, it should allow some success within two hours for a Debian programmer wishing to modify the setup.

I admit to like the dfu-util solution if this means taking readily available Kernels and merge it with a minimal busybox initrd for d-i. And it means that you cannot brick the device, really. Our users will understand the approach. We should probably have that approach implemented. But I like it more for its elegance that I like it for its ease to extend it.

There is yet another thought. When we go for a virtual environment going for an SD, then we are also very close to have a partial emulation environment for the complete FreeRunner. And with that, we would further extend our user base.

Many greetings

Steffen