[parted-devel] A short rant about parted

[Jim Meyering]

Richard W.M. Jones wrote:
> As you may not know, we have written a library for virtual machine
> disk manipulation called libguestfs (http://libguestfs.org/).  Part of
> this library allows you to manipulate the partition table.  For this
> we use parted.  However parted is very annoying to use in practice.
>
> (1) You can't get and set individual aspects of a partition.  For
> example: You can set the bootable flag, but getting it involves using
> 'parted print' to get the entire table and parsing the whole output.
>
> (2) You can't get and set the MSDOS type byte or the GPT label at all.
> Although real operating systems don't care about the type byte, it is
> *essential* for making Windows bootable.
>
> (3) The mkpart command is ill-defined.
>
>   mkpart [PART-TYPE FS-TYPE NAME] START END
>
> but this depends on the partition type.  eg. If you do:
>
>   mkpart primary 10 20
>
> then this makes a primary partition if the partition table is MBR, but
> it makes a partition called "primary" if the partition table is GPT.
>
> (4) You can add a partition, which (AFAIK) adds it in the first
> available slot.  You don't know which slot was used except by a very
> complex process of parsing 'parted print'.  And as mentioned in point
> (1) almost any update operation (except for the boot flag) involves a
> complex process of parsing 'parted print'.
>
> (5) Machine-readable format is not very machine-readable.
>
> (6) The library part of parted is complex and ill-documented, and
> doesn't solve any of the above.  You still can't set the type byte
> even at the library level.
>
> Honestly I'm sick of parted and I'd rather have a small library that
> can just do MBR and GPT partition tables (those are the only ones that
> anyone cares about).

Hi Rich,

It appears (I wasn't involved) that the program we know as "parted"
was originally intended solely as a convenient way to exercise the
functionality provided by libparted.

I would welcome a new command-line tool with a better interface and
additional functionality.  Alternatively, we could add to the existing
command-line interface.  In particular, "mkpart" (your #3) would need
a replacement, say "make-partition". I haven't looked at the code, but
suspect it could be arranged to print some indication of the resulting
partition number for each one it creates.

I think someone was working on an interface to get/set aspects
of a partition and/or partition table.

For #5, provide a new sub-command, or a new option to "print".

I agree wholeheartedly that using parted's command-line interface
is no fun.  However, most of the issues you raise can be addressed
without sacrificing compatibility.
It's just a SMOP ;-)