Re: Adding support for HP Service Partition GUID for GPT

[Richard Hirst]

On Tue, Mar 19, 2002 at 04:54:06PM +1100, Andrew Clausen wrote:
> On Mon, Mar 18, 2002 at 07:03:05PM +0000, Richard Hirst wrote:
> > Hi,
> >   HP have defined a Service Partition GUID for their ia64 systems.  It
> > is a partition with FAT filesystem, holding various diagnostic and
> > configuration tools (like compaq-diag, dell-diag, etc).
> 
> Any objections to having pairs of flags instead?  i.e. have a
> compaq and a diag flag.  If you set compaq on, it will also set
> diag on (unless, say, "sleep" was already on).
> 
> That way we get a few less flags...
> 
> So, we would have, presumebly:
> 
>       company: hp, compaq, dell, (award?)
>       type: sleep, diag
> 
> Any others?

HP have a GUID for HPUX filesystems too, that it would be nice to
recognise one day.  Also HP refer to this as a service partition, not a
diagnostics partition.  Maybe one day they will want a diagnostics
partition that is different from their service partition (unlikely, I
guess).

I have a problem with this idea, in that these things are not really
flags at all.  I found yesterday that "set N lvm off" on MSDOS was
broken, and pretending that we have two flags that between them control
what this one GUID is set to has similar problems.  For example, there
is nowhere on disk to record that one of these flags has been set but
not the other, so if I set compaq and diag, then restart parted and
unset diag, then restart parted and set diag, what would I end up with?
The disk is currently written each time I change the state of a flag, as
there is no concept of "I've now finished playing, it is ok to commit my
changes".  That would have to change somehow, unless we dictated that
these two flags have to always be set/unset together - then we may as
well have one "flag".  I wonder if we shouldn't have a concept of flags
for things that are really flags (like boot under msdos, and 'this is a
h/w partition do not delete it' under GPT), and a concept of type for
those fields which are multi-valued (like GUID under GPT).  Of course,
then 'boot' would be a flag under msdos and a type under GPT, which
wrecks the consistent user interface.  Maybe they can be managed in a
similar way, but attempts to set 'off' for a muti-valued field returns
an error saying "Sorry, can't do that. You have to turn on some other
flag to clear this, e.g. 'data'."

Richard