Re: question about fstab in squeeze and uuid

[Paul E Condon <pecondon@mesanetworks.net>]

On 20100313_095320, Tom H wrote:
> >> I believe a UUID is generated when the partition is "formatted", either
> >> with
> >> mkfs or mkswap.
> 
> > I confirm - just tried shrinking and growing back an extfs. UUID is left
> > untouched (as expected); that Mint article is BS or just obsolete.
> 
> I have never come across the problem described by the Mint link.

I have seen a UUID change in a way that disrupts my work:

When I install a 2nd/3rd distrib on a HD, I have made it a practice
to set up fstab so the existing distrib are mounted automatically.
Repeated use leads to all functioning distrib to be crosslinked.
But when a distrib must be reinstalled because something drasticly
wrong happened, or whatever, access to that replacement distrib in
all the older distrib is broken because the UUID of the partition
is changed.

This is not a killer objection for me. But it is a use case expample
that belies some broad claims about UUIDs.

If there were a place on every partition where a UUID of the partition
could be recorded, and if the partitioning software could be trained 
to preserve that record whenever the partition is simply wiped in
preparation for a reinstall, then ... maybe this use case would behave
more to my liking. But I am aware that there are a lot of individual
requirements for partition identifiers, so I hesitate to advocate
this proposal in this most simple form. As I say, I understand enough
about the scheme that I think I can live with it in its current form. 

Also. There seems not to be a standard algorithm for computing UUID
bit strings. Different developer/users of UUID ideas seem to be free
to choose what algorithm and what input data they use to get their
UUIDs. So, the actual behavior of UUID technology in Debian may change
over time. Future versions of Debian may silently change to address
old problems (and to introduce new problems).

> 
> The closest is having swap's UUID change when installing a 2nd/3rd
> distrib on the same HD.
> 
> Someone said earlier
> 
> "I can't imagine any case where an filesystem UUID change "by the
> face", I think this is not possible if you didn't execute any
> filesystem command to do it."
> 
> and cannot but +1 the comment because the UUID is held in a
> partition's superblock.
> 

My original post has generated a lot of very interesting discussion.
UUIDs are no longer so mysterious. I think that there is more work to
be done, but I have no good ideas to pursue and I have reached my
attention span limit. I don't want to post with a subject line
"Solved", but it is true that I want to stop discussing this.

Thanks to all.

-- 
Paul E Condon           
pecondon@mesanetworks.net