Subject: RE: GPT support still needed? (was: RE: Recursive partitioning)
To: Jason Thorpe <thorpej@shagadelic.org>
From: De Zeurkous <zeurkous@nichten.info>
List: tech-kern
Date: 06/06/2007 17:37:57
Haai,

On Wed, June 6, 2007 17:11, Jason Thorpe wrote:
>
>>[snip]
>
> Yes, a new, simplified one LIKE GPT.
>
> Maybe we can advance this discussion a little further by you actually
> outlining what you don't like about GPT.
>

1) It includes features which are not necessary (GUIDs for example);
2) It's practically a superset of MBR (calling it 'protective' doesn't
change that fact);
3) It deals with a load of IBM PC-specific (mostly Windoze, even) stuff;
4) Worse, it deals with some stuff that is likely to change (fixed
checksum format, little-endianness, other constants which should be
copy-on-write)
5) Doesn't seem to have non-specific metadata fields, even optional.
6) M$ and Intel seem to have pushed it, which most likely means the
current format is gone faster than it takes to be actually standardized.

In all, I think we can do a lot better :)

>>[snip]
>
> When I mean "layout", I mean location on the disk of the partition
> map, number of partitions that can be described by the partition map,
> byte-order of the fields within the partition map, etc.

Those should be standardized to network byte order in the new disklabel
format.

>
>>[snip]
>
> ...but, as a matter of practicality, are possible to generate with
> sufficient "uniqueness" as to make the theoretical chance of a
> collision very unlikely.

Still, why not remove the impact of the collision entirely?

>
>>[snip]
>
> How nice for you to live in a bubble.

You don't want to know what that means to someone who has played Metroid
(FDS/NES)...

...anyway, I don't. However, thanks to me /not/ living in a bubble, I've
build up enough stubbornness over the years, resulting in not giving up
the fight for sensible decisions easily.

>
>>[snip]
>
> Ah, you mean one like, let's see ... GPT.

Hmmyes. They work just fine, thank you.

>
>>[snip]
>
> You are confused.

That's a pretty bold remark to make to the Whining Bastard.

> I guess you missed the part in the EFI spec that
> says the "protective MBR" is optional (it is, however, recommended)?
> Most GPT partition map parsers don't even look for one.  The MBR is a
> footnote, and nothing more.  It is not relevant to the actual usage of
> GPT.  Get over it.

Footnote or not, when M$ and Intel are involved you can be /sure/ that it
will be come practically required soon enough.

>
>>[snip]
>
> What on earth are you talking about?

The Windoze-specificness and legacy support built into GPT, and, as a side
note, the fact that I prefer Sun workstations running NetBSD over IBM PCs.

>
>> Hm -- I haven't seen _any_ clever way as of yet.
>
> Then I guess you haven't been looking very hard.

Its pretty difficult to look harder when the issue is both as clear and
predictable as /dev/zero.

Baai,

De Zeurkous
-----------

Friggin' Machines!

>
> -- thorpej
>
>