Re: [PATCH] Fix when installing on pationless but partionable medium

[Pavel Roskin <proski@gnu.org>]

On Wed, 2009-07-22 at 19:22 +0200, Robert Millan wrote:
> On Sat, Jul 18, 2009 at 11:28:58PM +0200, Vladimir 'phcoder' Serbinenko wrote:
> > > I don't understand what you mean here.
> > Let's take a common example of cdrom. Most of the users and developers
> > are accustomed to a cdrom holding one filesystem. On macs however cds
> > are partitioned and not being able to access all the partitions is a
> > problem for end user. Such situations are probably common. If we ditch
> > has_partitions altogether the only negative side effect will be that
> > in some weird configurations unpartitioned media may appear to have
> > partitions but whole media is still accessible. Additionally it
> > simplifies and makes kernel smaller
> 
> I'm not sure they're so weird.  But we could still do it.  Pavel, what do
> you think?

We don't use grub-setup to install GRUB on CDROM.  I don't know if we
can eliminate has_partitions.  If we can, I'm fine.

I suggest that we start with known safe cases, such are PC partition
table and FAT filesystem.  If we can positively identify the place where
we install the bootsector as either of that, we can install without
--force.  We can extend the rules as we check if other filesystems allow
overwriting the bootsector.

I don't think dumping floppy support would be right at this point.
Maybe five years from now.

> > > I haven't looked at the source code, but what he said is we can determine if
> > > an MBR is valid by checking the bootable flag, and this is not always so.
> > I don't see any problem. He said: checking that bootable flags of all
> > partitions are either set (0x80) or unset (0x0) and not another value
> 
> Oh, that's different.  I think it's fine provided that:
> 
>   - None of the commonly used free partitioning tools uses an illegal value.
> 
>   - We fail gracefully and let the user know why.

I'm fine with extra checks.

-- 
Regards,
Pavel Roskin