Re: amiga partitioning support for parted

[Sven Luther]

On Wed, Jul 30, 2003 at 05:35:16PM +1000, Andrew Clausen wrote:
> On Thu, Jul 24, 2003 at 09:42:18AM +0200, Sven Luther wrote:
> > Hello, ...
> > 
> > I have a question concerning <partition_type>_alloc.
> > 
> > It seems to me that this function has two purposes, that it allocate the
> > memory needed to create a partition table, but also that it fills in the
> > fields of said partition table for an empty partition.
> 
> Correct.
> 
> > In case we read the partition table from disk, these values are
> > immediately overwritten by the read data, so in a sense, would it not
> > be better to separate this functionality into an _alloc function which
> > allocate the partition table, and a _init, which fills in a blank
> > partition table ?
> 
> I don't think so... I like having everything in a consistent state
> all the time.  Also, your proposed way would need more lines of code.

Maybe not calling _alloc before the read, but calling amiga_alloc
directly from amiga_read would be a solution, no ?

> > Also, would it make sense to add a new filesystem type (affs) with only
> > the probe function for now, so that parted can list filesystems of this
> > type ?
> 
> Yes.

I have done that, i will try to add support for all the amiga related
filesystems i can find info about (ffs, sfs, pfs, maybe others).

> > Also, most partition table have a feature to set the filesystem type in
> > the partition table, but these seems not to be used at all in parted. I
> > have a disk where parted was not able to probe the filesystems, altough
> > i know that there are ext2 filesystems on it. Is it ok to fall back to
> > the partition table filesystem type if the filesystem was badly detected ?
> 
> I think this is rather difficult, given that there has historically
> been lots of confusion over the allocation of partition numbers.

??? I don't understand, what has the partition number to do with this ?
The amiga partitioning scheme is a linked list, the partition number is
not really all that important, in fact, you can modify it by changing
the way the blocks are chained in the list.

> > Also, what happens in the case a partition can be detected has having
> > more than one filesystem on it ?
> 
> Parted uses extra heuristics, such as what size the file system
> believes it is (compared to the partition size), and whether there
> are file system consistency errors.

The problem is about filesystem it doesn't know about, like affs.

> > Maybe because an older filesystem of a
> > different type still is present, because it was not overwritten enough
> > when the new filesystem was created, and the _probe function for the
> > old filesystem can detect it, and thus potentially destroy the new
> > filesystem ? 
> 
> If you try to resize, yes.

:(((

> > What if the filesystem probed and the filesystem type written on the
> > partition table doesn't correspond ? Should i fix it, or just ignore the
> > partition table value for the probed filesystem ?
> 
> It probably doesn't matter... I'd ignore it.

The problem is that other OSes which cohabit on the same disk may use
them. They are also linked to a sector list rooted in the Rigid Disk
Block which contains filesystem drivers and such for AmigaOS/MorphOS.
Not to speak about the builtin ROM of the original amiga hardware.

BTW, what do you think i should do about partition sectors with bad
checksums ? Correct them or reject the block ? Or is there a way to give
a warning about that to whatever higher level is using libparted. My
tentatives to use the ped exceptions failed miserably.

Friendly,

Sven Luther
> 
> Cheers,
> Andrew