[Date Prev][Date Next] [Thread Prev][Thread Next] [Date Index] [Thread Index]

Re: Bug#260763: [powerpc] Problems with logical volume setup during installation



On Mon, Jan 10, 2005 at 11:13:07AM +0100, Michael Schmitz wrote:
> > > > Parted converts flags to funky partition type names, i think. Not idea, but it
> > > > works, like  said, i think parted is inherently broken on this, but there is
> > > > not much we can do at this time.
> > >
> > > They might have done this following a comment of mine (my memory id fuzzy
> > > on this, not sure they ever asked). Using the partition type (which is
> > > just a string) is indeed broken IMO.
> >
> > Why ? take for example the bootstrap partition, which is Apple_bootstrap on
> > mac, and 0x41 on chrp/prep MBRs ? The same goes for LVM or RAID partitions.
> 
> Why? Because we don't know how MockOS will react to a partition _type_ set

I don't know any MockOS, please provide real examples.

> to anything but Apple_Bootstrap, Apple_HFS, Apple_UNIX_SVR2 or Apple_Free.
> If you're sure (or just reasonably confident) Linux_LVM will be ignored by

Just give it a try and enlighten us.

> Apple boot and disk mount code, go ahead. I can only test this for a
> fairly narrow range of OS X versions, and maybe a resurrected 8.6. Since
> there's always the partition name (for Apple_UNIX_SVR2 partitions which
> MacOS ignores since prehistoric times), I didn't really want to bother
> testing this.

Yeah, but imagine that the user wants to use the partition name for something
sensible, and all this goes down.

Notice that i was approached about RAID on sparc, which is the exact same
issue too.

> > Furthermore, i believe that the partition type is the right place to store
> > this info, confirmed by the fact that on MBR a type is used for both RAID and
> > LVM. After all, the fact that a partition does RAID or LVM cannot be
> > compatible by it being a linux swap partition for example, in the way the
> > bootable flag is set for example.
> >
> > I would suggest we use 'Linux_RAID' and 'Linux_LVM' for those two.
> 
> And libfdisk (for example) used to look at the partition _name_ field in
> Mac partition tables for clues where to find swap. What's the problem with
> that?

well, suppose that a french user wants to name it "echange" instead of swap ?

> > > So we should press this point a bit more forcefully :-)
> >
> > Like said, i will gladly write the (trivial) code, if some consensus is
> > reached.
> 
> We seem to have trouble reaching a consensus even here :-) Seriously
> though, let's just try with 'Linux_RAID' and 'Linux_LVM' partition _type_
> fields in Mac partition tables, maybe add Linux_Swap later and keep plain
> data partitons as before. That should keep the comaptibility code to a
> minimum (look for 'swap' in type and name fields) and get the scheme some
> limited testing.

Ok, but as said, Colin Watson vetoed me on that, and as we already have a
parted in experimental, ...

> > > If parted currently encodes the LVM flag in the partition type we can
> > > still put it into the partition name on top of that, and get the Debian
> > > LVM maintainer to accept either flag?
> >
> > No, i would vote against this. After all, a parted user can change the name of
> > a partition through the parted and d-i/partman interface.
> 
> A mac-fdisk user can change either at will. Linux always reserves your

The future belongs to libparted frontends though. And this is not linux, but
the new user-friendly debian-installer partman, so ...

> right to shoot yourself into the foot. But for the sake of saving users
> from their own stupidity, go ahead with the scheme you suggested.

I would have for ages, but Colin Watson vetoed it without larger consensus,
and since he is one of the RMs, ... And beside i sort of agree with his
arguments about compatibility problems.

Friendly,

Sven Luther



Reply to: