>> One of my objection is the absolutely pc-centric attitude of both
>> documents, which can be smoothed at no cost. [...]
> I disagree. The mechanism to boot up a computer varies from
> architecture to architecture very much, so basically any boot loader
> depends on the system architecture the boot loader supports too
> much.
Yes and no. I think 50% of the code is unique to the platform, but the
other 50% is not. Being forced to deal with 10 different boot loaders
on 10 different platforms is a real pain. SILO and MILO (sparc and
alpha) offer filesystem access like grub, but the syntax and interface
is different; MILO has menus like GRUB, but SILO lacks it. On the ppc
and m68k (mac) you need to go through macOS as you lack a second stage,
the netwinder has its own user interface (with filesystem access), ...
While I'm not able to port GRUB to all of these platforms, and some of
them may be plain impossible as you destroy your loading routines as
soon as you leave the previous OS, the door should be left open. GRUB
is so much superior to all other loaders, that real hackers (as
opposed to me) may be willing to port the tool, and the port is more
likely to happen and smoother to accomplish if the infrastructure is
in place. Well, I *might* be able to port it to Sparc, given enough time.
> For example, most workstations and some personal computers (such
> as SparcStation and iMac) has a built-in firmware program, while IBM
> PC has BIOS instead.
There's no conceptual difference, they offer the same services.
> Some of the architectures have the concept "MBR", while others
> don't.
It's a detail. When the stage2 is loaded, platform dependencies may
be confined to the library of utility functions used by the stage2 itself.
> The format of partition tables is not the same.
Again, a detail. If the Linux kernel deals with all of them, can't
GRUB do that too?
> And so on. Therefore, the design of GRUB is specialized to PC
> architecture (e.g. there is no need to split it into two stages, if
> you write a boot loader for HP workstations).
Well, I don't have an HP at home (yet), but I remember that the first
message is "loading IPL", and then the IPL loads the kernel accessing
the filesystem. So can't the IPL be replaced with GRUB's stage 2?
The design of GRUB is, in my opinion, "offering a decent user
interface and menu system to make booting both easy for novices and
configurable for technicians". The split in two stages is an accident.
If there is some interest (and time to discuss about it), I can study
a migration to a multi-platform source tree and patch the docs to
clearly identify platform dependent information.
Actually, page 15 of the manual states that "support for non-PC
hardware architectures is also planned" :)
/alessandro, guessing how many developers are currently hacking grub..
