Re: Not really a bug...

[Andrew Clausen]

address@hidden wrote:
> 
> ... rather user-stupidity.
> 
> I don't know exactly which version of parted I was working with,
> since currently I'm not at the same system, it's the one shipped
> with SuSE 7.1 anyway, so probably it is rather recent.

One with a major bug.  Err.  1.4.5, IIRC
 
> I still can't quite believe I have done so, but yesterday I called
> parted without parameters and didn't realize I wasn't working on
> hdd but on hda, the drive with both a windoze and a linux
> installation (four partitions total). The first thing I did was to create
> an entirely new label, which parted did not complain about,
> although three of the partitions were mounted read/write (including
> the root) - this part of the problem really does look like a bug to
> me.

I suspect the problem is that the format of /proc/mounts changed
with your kernel (what kernel do you have?), so Parted couldn't
see that you had used partitions.

Could you give me the output of cat /proc/mounts?  thanks.

> However, when I finally had parted print the partition table and
> realized, what I had done, what I seriously missed was an option to
> exit parted without writing the changes (or a restore or undo
> function or whatever, I imagine it shouldn't be too complicated to
> back up the old partition table prior to any changes). Now thanks to
> the friendly gpart tool, I could restore my harddrive, but it was some
> 45 minutes of terror, anyway.

Yeah, I know how you feel!  One of these days, I will add gpart
functionality to parted... the hard bit is the user-interface, hehe.

> Therefore I'd suggest implementing the following new features:
> - Don't default to anything when called without device-parameter!
> Print a usage message instead (like fdisk does).

Parted does display "Using [device-name]".  You really think
I should remove this guessing stuff?  What does everyone else
think?

> - Create a backup of the partition table (for all I care it's enough to
> keep the backup in RAM, while parted is running, so you can
> easily discard changes before exiting, but perhaps the option to
> back up on disk could come in handy in some situation (e.g. the
> vamos boot-manager (shareware) has the option to hide partitions
> from M$ by changing their system id, which might get you into a
> bit of trouble if you have lots of hidden partitions and M$ decides to
> overwrite the mbr where the boot-manager resides)).

I'm working on operation queue stuff ATM, so it will work by
[do all operations], then you type commit, where it will display
a scary message.  I think this is better in the long run.

> - Have parted check whether any of the partitions are mounted
> before creating a new label.

This already exists.

> I hope this comment is useful.

Of course!  Thanks!
Andrew Clausen