Re: gnu parted - re-sorting of logical partition minor numbers

[Andrew Clausen]

Hi Richard,

On Sun, Sep 22, 2002 at 09:45:55PM -0400, Richard Ballantyne wrote:
> As you can see,  the minor numbers for these 4 partitions no longer remained 
> in sequential order.  This is a bad thing because if it causes the following 
> unrecoverable error in Partition Magic:

Partition Magic is broken in this respect, IMHO.

> "#120     The logical drive chain is incompatible 
> This error occurs under some operating systems when logical partitions are 
> not 
> chained together in the expected order. DOS, OS/2, Windows 95, Windows 98, 
> and Windows NT require that logical partitions be chained together in 
> ascending order.

I haven't found any evidence of this.

> using the DOS FDISK in this situation can cause loss of 
> one or more partitions. "

I guess we should check this out, but I haven't heard of any problems
wrt this.  And I've heard plenty of stories about partition tables
in my life time :)

This isn't a simple matter of just "going with the flow", because it's
impossible to implement what Partition Magic suggests it's doing.
(So I'm not sure if I believe it's doing it!)

> New Method:  Resort minor numbers;  the minor number for the new partition = 
> <previous partition minor number + 1>.  The minor numbers for all partitions 
> that exist _after_ the new partition are incremented by 1.

This won't work.  The first logical partition (5) often has it's
entry stored in the extended partition table (not a chained one), and
no chained table has been allocated to it.  So if you want to renumber
it to 6, say, after inserting a new logical partition, there may be no
room to add a new chained partition table.

In this case, you'll get an "Unable to align partition" exception.

> You could allow users to choose the enumeration method by adding an optional 
> parameter to mkpart and mkpartfs;
> 
> mkpart PART-TYPE [FS-TYPE] [ENUM-METHOD] START END
> mkpartfs PART-TYPE FS-TYPE [ENUM-METHOD] START END
> 
> Where ENUM-METHOD can be set to either:
> 'resort'    -   the new method
> 'default'  -   the existing method -- Make this the default method if 
> [ENUM_METHOD] is unspecified.

This later interface change seems complicated, and probably confusing
to users.  Besides, it will break on some partition tables.

Can you find some evidence that Parted is doing something bad?
Got any other suggestions?  I thought about this problem a lot about
two years ago, and I couldn't think of a better solution.

Thanks,
Andrew