>> The partition type as stored in the partition
>> table is only a hint, not authoritative information.
> I disagree to that partition types are not authoritative
They definitely are not. If they were there would be an ID for
Reiserfs, for ext3 etc, and there would be no "this or that" entries.
And there would be an organization assigning those numbers.
> Additionally, magic numbers alone are not always
> reliable (e.g. a swap partition may have the same magic number
> accidentally).
Yes, you have a probability of 1/64k to detect a minixfs in a
partition whose initial sector is random, and 1/4G to get a
false-positive on ext2. But why, then, doesn't reiserfs and nftl
depend on a partition ID?
I am not strongly against your choice, but it adds unneeded
complication, in my opinion.
> The patch by Stefan Ondreji enables GRUB to modify partition tables,
> so you won't have to invoke FDISK just for changing a partition
> type.
Please leave this code conditional with a configure option, I don't
think grub should do that because it's not its task, just like "ls"
doesn't print free disk space. To get a full recovery tool you'll load
a kernel and filesystem from floppy. Grub is aboot loader, not an OS.
, and I don't want to enlarge my grub image to support this.
Besides, this would add yet another PC-specific feature, and it would be
almost impossible to port it forward to multiplatform. I agree that
multiplatform support should be based on a complete reorganization of
code and this should be delayed after this code base is stable and
released. But the new version should offer backwards compatibility: I
think adding PC-specific stuff at this point is a bad move.
/alessandro
