gpt code is broken as designed

[Andrew Clausen]

Hi Matt,

I don't know how you convinced me that computing the on-disk metadata
of a partition at gpt_partition_new() time was sane.  It definitely
isn't :(

You obviously didn't try to resize a partition.  (Weren't you excited
by the idea?! *grin*)

Also, it doesn't look remotely endianness-safe.  Run:

        $ cat disk_gpt.c | grep 'gpt->' | grep -v CPU

for some examples.

I'm backing this out until it's fixed.  Sorry, for not noticing/complaining
earlier.

I would much prefer you structured it properly.  (i.e.  generate the
on-disk layout in gpt_write(), rather than patch up the on-disk layout
on changes, or a similar work-around)

Thanks,
Andrew

P.S. is there any reason why gpt_probe() is so strict?
Shouldn't sanity checks be in gpt_open(), possibly offering to
correct problems...?