[bug#27327] [PATCH] bootloader: Add u-boot.

[Danny Milosavljevic]

Hi Ludo,

On Fri, 16 Jun 2017 10:40:39 +0200
address@hidden (Ludovic Courtès) wrote:

> > +   (package #f)
> > +   (installer #f)))  
> 
> I still find it weird to use #f for these two fields.
> 
> I would find it more reasonable to have, say, a ‘make-u-boot-bootloader’
> that returns a <bootloader> with all the fields appropriately set (not
> #f).  Otherwise it sounds like we’re going to have to deal with support
> requests #about “wrong type to apply: #f”, and I’d like to avoid that.
> :-)

Yeah, but I think Mathieu added explicit support for leaving those #f.

> Can’t we just say that ‘package’ is always a <package> 

Difficult.  U-Boot is more like a BIOS is on PCs.  That means in normal 
operation you'd not flash it.  If there was a mistake in the flashed U-Boot 
often you can't fix it again without electronics knowledge and gear (if at all 
- serial pads must be available on the board).  There's no extra BIOS.  U-Boot 
does it all:  RAM initialization, keyboard initialization, display 
initialization etcetc.

Sometimes flashing is only possible via serial cable (or worse) from an 
external machine.

Also, there are lots of forks of U-Boot.  Since U-Boot is licensed under GPL 
the package should be available somewhere - but not necessarily in the U-Boot 
master branch (and often it's in fact not available there).

In short, we should add U-Boot packages gradually, and only after testing each 
U-Boot package on the hardware.

That means in the beginning we'd have a LOT of (package #f) - but that's still 
preferrable to bricking.

>and that
> ‘installer’ is always a procedure?

Yeah, the installation procedure could probably check (if package (invoke 
"...")).  

I've submitted this minimal patch mostly to make clear how U-Boot would extend 
extlinux.  I think it's good to have it in place as a guard against 
too-much-syslinux-bias :)

That said, it should actually make GuixSD boot on a machine with U-Boot already 
installed.