Message-ID: <20040529145843.GA3550@treasure>
Date: Sat, 29 May 2004 16:58:43 +0200
From: Bastian Kleineidam <email address hidden>
To: <email address hidden>
Subject: Re: Bug#250528: Partition table interpretation error on kernel 2.6

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

Hi Sven,

On Tue, May 25, 2004 at 07:40:41AM +0200, Sven Luther wrote:
> > Apart from that, this error message is pretty shitty, it does not say
> > anything:
> > - does not tell what units the numbers 7476/255/63 are
> 
> Come on, anyone handling partition table would know that, it is in
> 512byte sectors naturally.
> 
> > - "there are many reasons", but does not tell what they are
> 
> Because it is an open ended list.
>
> > - "most likely" is a wild guess and not helpful
> 
> Sure.

It would be helpful if the error message told me how to verify if the
Linux system indeed detected the wrong numbers. Instead this is left
as an exercise to the reader, who would have to know about the /proc
file system, read a little in linux/Documentation/filesystems.txt,
and eventually find /proc/ide/hda/geometry. After that, the user would
have to know that those magic a/b/c numbers in the geometry file are
also written on top of most modern harddisks, and compare them with the
file contents.
This is exactly the stuff I found so frustrating in my early days:
lots of programs expected me to magically know something, because
everyone should know about it, right? Well, I know now but can easily
imagine people who don't. And expecting from them to magically gain
knowledge is not user friendly.

Anyway, it is not that hard to write units behind numbers, and to give
a little help on how to get disk geometry information both from Linux
and from the harddisk.

> 
> > - I can not see that "Linux detected the BIOS incorrectly", instead of
> >   that I see two more weird number pairs:
> >   $ cat /proc/ide/hda/geometry
> >   physical     16383/16/63
> >   logical      65535/16/63
> 
> So, what is the real geometry of your disk ? And what did it show with
> 2.4 kernels ? 

The physical data (as read from the harddisk label) for my IBM 60GB
harddisk is:
CHS: 16383/16/63
which is correctly detected by the 2.6.6 kernel (see above).
So it seems what parted detected (7476/255/63) is wrong, assuming that
parted indeed meant physical cylinder/header/sector format and not
something else.

I have no 2.4 kernel anymore, so I cannot test it.

> 
> > - what should the user input after seeing a prompt "Ignore/Cancel" ?
> >   better would be "[i]gnore/[C]ancel:"
> 
> The capital letter is the one to use, is that not enough ? 

IMO the most secure thing is trying to make everything explicit.
So my suggestion above is also flawed. A better approach could be:
"Press c or C to cancel, and i or I to ignore. Any other key aborts the
command."
This way the user does not have to guess the answer.


Greetings, Bastian

PS: why am I so hyper-correct about this error message? Easy, it gives
me the creeps when harddisk-changing programs are not clear and concise.
What I really did when seeing the parted error message was to kill -9
the parted process to make sure it does not fuck up my harddisk, and
purge the parted package. You can be sure I don't run this program
again until the author gives more about helpful error messages :)

- -- 
  ,''`.    Bastian Kleineidam  .  calvin (at) debian.org
 : :' :
 `. `'    GnuPG Schl�http://kampfwurst.net/gpgkey.txt
   `-

Men are from Mars. Women are from Venus. Computers are from hell.

-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.2.4 (GNU/Linux)

iD8DBQFAuKUjeBwlBDLsbz4RAtA+AKCsbID9dHEQzcynzkBeSqbmgRwIYwCfa+WC
WEsRvzlMTxQ91SI5HmivrEo=
=DLr4
-----END PGP SIGNATURE-----