[Parted-maintainers] Bug#757417: libparted2-udeb: breaks auto-lvm in d-i

[Colin Watson]

clone 757417 -1 -2 -3 -4 -5 -6
reassign -1 partman-auto-lvm
retitle -1 partman-auto-lvm: cope with automatic creation of partitions in loop labels
reassign -2 partman-base
retitle -2 partman-base: make tests cope with automatic creation of partitions in loop labels
severity -2 normal
reassign -3 partman-crypto
retitle -3 partman-crypto: cope with automatic creation of partitions in loop labels
reassign -4 partman-lvm
retitle -4 partman-lvm: cope with automatic creation of partitions in loop labels
reassign -5 partman-md
retitle -5 partman-md: cope with automatic creation of partitions in loop labels
reassign -6 partman-zfs
retitle -6 partman-zfs: cope with automatic creation of partitions in loop labels
thanks

On Fri, Aug 08, 2014 at 12:53:07AM +0200, Cyril Brulebois wrote:
> 3.2-2 fixes the non-LVM case, while it doesn't fix the second one, since
> we're now getting a different issue, apparently about primary partition
> count. I'll try to post more details soon.

There are two problems here, both caused by apparently-intentional
upstream changes:

  1) parted 3.2 now automatically creates a partition when you create a
     loop label.  Various bits of partman expect there to be free space
     after creating a loop label, and fail or misbehave when this isn't
     the case.

  2) parted 3.2 no longer probes LVM logical volumes, or indeed any
     device-mapper devices other than dmraid whole disks.  As a result,
     those devices never show up in partman at all.

2) is a distinctly unhelpful change at least in the context of d-i,
although I guess it might be helpful elsewhere (parted -l tends to be
pretty noisy if you have device-mapper devices present).  For now I'm
reverting it since it's a one-line change to undo just this bit.

However, 1) is trickier.  Reverting that change to parted results in
different breakage, and it looks like I would have to do some quite
complicated disentangling to revert it successfully.  I'm not convinced
this would result in something more stable.  On the other hand, the end
result is actually more convenient for d-i because it needs to do less
work, so I'm minded to leave it in place and adjust the calling code
instead.  I've successfully tested changes to partman-auto-lvm and
partman-lvm; as indicated by the control commands above there are a few
other things to change too.

> Given this particular bug was also present in 3.2-1 (even if masked by
> the alignment issue), I'm versioning it as found in 3.2-1, so that 3.2-2
> has a chance to migrate when it's old enough. If that doesn't sound like
> a good idea, please adjust version [and explain why ;)].

That sounds quite sensible, yes.

Apologies for the inconvenience casued by all this; I clearly didn't
test the new upstream version of parted well enough.  I will sort all
this out as quickly as I can.

-- 
Colin Watson                                       [cjwatson at debian.org]