The base class is named Output. Having a method with the same makes
the code confusing.

If qemu-nbd is not present at the host system, only accept raw images
as input media. This is a workaround for ancient systems like CentOS 6.x
where the nbd module is not compiled in the default kernel and the qemu
package does not include the qemu-nbd executable.

Do not use qcow2 to snapshot images that are raw. qemu-nbd is not
present in some old systems like CentOS 6.x. All we can do to support
those systems is to use the device mapper in as many cases as possible.

This complies better with the python conventions

If you print the snapshotting message before disk.file is referenced,
you may end up outputting messages like this:

Snapshotting media source ... Examining source media `<image_file>' ...

Instead of:

Examining source media `<image_file>' ... looks like a block device
Snapshotting media source ... done

Only use the device mapper when snapshotting block devices. If the
input media is a regular file, then create a qcow2 image and define
the original media as backing_file.

When snapshotting make sure that the size of the underlying block
device is equal to the input media's virtual size. If the input media
is not raw, the QEMU image driver may grow the underlying image. The
block device should be big enough to allow this kind of operations.

This can be used to act on the original media instead of a device-mapper
snapshot.

This resolves #10

There is no need to associate a loop device with the input media if we
are not snapshotting the input media. The loop device is only needed in
order to use the media with the device-mapper.

This attribute hosts the block device associated with this Disk
instance. This can be the original media if the source media is a block
device, a loop device of the original media if the original media is a
regular file or a loop device of a temporary object if the original
media is a directory.

According to the decision of the GRNET Board of Directors,
switch license to GPLv3.

This commit will be propagated to the release
and master branches based on git flow, and the next
release will be licensed as GPLv3.

The user may define parameters needed by the os_type classes to perform
the system preparation tasks.

A new needed_sysprep_params method is added to os_type.OSBase that
returns a list with parameters that should be defined by the user.

 * Prefix all private methods with _
 * Write missing docstrings
 * Move all public methods above the private ones
 * Add module docstrings where missing
 * Define source code encodings on all modules

If the image gets registered as public allow anyone to read the
metadata and the md5sum file.

Rename DiskDevice class to Image and move it to a seperate module.
Also, don't let the user access the os_type/* modules directly. In
Image class create the `os' member variable that will host an
appropriate instance of one of the OSBase classes.

Don't let libguestfs guess the format of the provided image. Enforce
it to treat the image as raw.

snf-image-creator removed /var/tmp if the directory was empty

Support both versions (1.2.x and 2.x) of python-sendfile. In
v1.2.x (py-sendfile) the returning value of the sendfile function is a
tuple, where in v2.x (pysendfile) it's a simple integer.

The cow file when creating images or the temporary image file when
bundling the host system may be large. Allow the user to overwrite
the directory under which those files get created.

Handle correct extended partitions with id 0xf

Add a new try_fail_repeat function that wraps cleanup system commands
and retries them if they fail. There are cases where commands like
dmsetup, umount and losetup -d may fail with device is busy. In most
of the cases, sleeping for halve a second and retrying does the job.

In Guestfs you get notifications about progress on some functions like
launch() by registring callback functions. When using guestfs from python,
this is fragile. If the callback functions raises an exception, you end up
with segmentation fault. It is better to completely remove the progress
monitoring from launch().

Credits go to Giannis Spiliopoulos <gspilio@admin.grnet.gr>

Also do a code cleanup

The partition table of the image is created correct. The partitions
are not yet filled.

Nothing is working yet

Check the file system usage and accordingly reduce the size of the
last partition in the image.

Create a sparse file under /mnt in the size of the root hard disk
and copy the mbr into this file. Not yet workable.