UUIDs on drives

[Derek Broughton]

ghe wrote:

> Derek Broughton wrote:
> 
> | I don't see that the two statements are contradictory.  The _hardware_
> | identifies itself - but the hardware never identified itself as
> /dev/sdb3.
> | The software does that...
> 
> Yes. That's true. My complaint is that the software sometimes decides
> that Derek should get a new phone number. Then Derek's friends can't
> find him anymore. And sometimes he's got the kernel in his back pocket.

THAT's why we're using UUID - because software _can_ find that.

> | But why would they want to make artificial distinctions between
> hardwares?
> | The simple fact is that there's nothing intrinsically special about
> | being the first SATA disk - especially if you really are wanting to boot
> | off
> USB.
> 
> Beg to differ. If the boot block is on SATA1, grub needs to be able to
> find it -- or failing that, go look for it.

Yes, but that boot block may just as easily be on a removable USB stick. 
Again, there's nothing intrinsically special about being the first SATA
disk.

> | For that matter so would hd, but what would it mean?  The whole point is
> | that those device names are _meaningless_, and so _shouldn't_ be tagged
> | with artificial and misleading distinctions.
> 
> They aren't meaningless when they're written into config files like
> menu.lst or fstab. 

Where they shouldn't be written!  That's exactly why Ubuntu went to pains to
make sure both fstab _and_ menu.lst were using UUID before the change to
device names occurred.  Now you're complaining that UUID is a bad idea,
because UUID doesn't work if you don't use UUID!

> The UUIDs are consistent because they're created when 
> the fs is built (and because udev never gets ahold of them :-).
> 
> But if the installers and updaters are writing sda or (hd0) in the
> configs, 

They're not.  

> There may have been good and sufficient reasons for udev (and the
> changes to the kernel's block drivers), but the way it was implemented
> broke a system that had been working, literally, for decades.

No, it didn't.  Refusing to use the new methods is what breaks the system.
> 
> | And you couldn't fix that from a live CD?  I've managed to make systems
> | unbootable before, but you go to a live CD (the first time, I didn't
> | even use an Ubuntu one - I don't think they had a live CD then), and
> | run "grub-install".
> 
> No, I couldn't. Because I've never learned how. I thought about that
> route, 

Well, then, I just taught you.  It's that simple.
-- 
derek