The installation of the above mentioned dracut module package
is required for oem images which uses dracut as initrd system
and have the creation of an installation image requested.
This is related to Issue #576
If an image description only contains package requests
from a bootstrap section but no image packages, it's not
required to install a package manager package into the
system
The new rule allows to check for required attributes for a specific
image type and is used for the filesystem attribute which is required
for the image type oem, vmx and pxe. This Fixes#476
Make sure the default initrd_system is set to dracut
for the vmx image type and to kiwi if no initrd_system
is specified. This allows to simplify other code parts
as we can trust there is a value set
Custom disk format options passed in the formatoptions
attribute were not handled. In addition options with a
value passed to qemu were handled in the wrong way.
This commit addresses both problems and Fixes#463
This commit rearranges some method and variable names and includes
a new runtimecheck to ensure the mediacheck attribute is not set
when building non x86 images.
Patch related to bsc#1049304
If an element like displayname or oem-boot-title contains
characters outside of the ascii table this causes trouble
when kiwi writes out the profile file and the code is called
through python2. The reason here is that the default encoding
on write() (and other methods) is set to ascii in python2
and when it receives unicode characters outside of the
ascii spec a UnicodeEncodeError is thrown. Now all of kiwi
is using Unicode which means this does not produce a problem
when calling the code through python3 because the default
encoding is utf-8 there. This patch introduces a method
which allows to change python's default encoding and calls
it at the code point where we write the profile because we
got unicode and we want to write unicode in any case.
This fixes at least one situation for python2-kiwi to
allow the use of non ascii characters in the XML setup.
If other places will be found the same approach should allow
to fix it for python2
Updates all XML descriptions to latest schema version and
also includes adaptions to the test XML descriptions for
testing the new Xen guest and server setup
* Delete kiwi internal ibs: and suse: types
* Delete handling for --obs-repo-internal and provide a
compatibility message to the user
* Buildservice download server url and scope can be configured
via ~/.config/kiwi/config.yml
* Translate obs urls to http in import_repositories_marked_as_imageinclude
* Use new Uri.is_public method in renamed runtime check
check_image_include_repos_publicly_resolvable
This commit fixes#395, with it, using --ignore-repos, does not
delete imageonly repositories from the description file. This way
imageonly is prepared to be used in the buildservice even when using
the 'obsrepositories:/' reference style.
This commit adds support for the rpm-check-signatures flag, which
sets the package manager to verify or not each package signature.
By default KIWI assumes no gpg checks are done.
This commit adds the possibility of clearing asny subcommand or
entrypoint. This is relevant for docker derived images, as they
inherit the configuration and it might lead to some bad behavior.
This commit includes the root import feature. A `derived_from`
attribute has been included with the <type> section to make
reference to the image file to import. The image format to import
is assumed to be the same as the build type to import.
The current format supported is docker
If a kiwi initrd is used, the kernel used to build the kiwi
initrd and the kernel used in the system image must be the
same in order to avoid an inconsistent boot setup. This is
related to (bsc#1027610)
Runtime check to check for required docker tools umoci and
skopeo. As long as tools tools are not available in the
core distro of all distributions we avoid spec file requirements
and check at runtime if they can be found
Instead of creating a simple tarball the tools umoci and skopeo
from the docker tool chain are used to create official docker
images. Along with those tools more information to describe a
container has been added to the schema. A complete container
setup now consists of the following elements
<type image="docker">
<containerconfig name="..." maintainer="..." user="..." workingdir="...">
<entrypoint execute="command">
<argument name="option"/>
...
</entrypoint>
<subcommand execute="command">
<argument name="option"/>
...
</subcommand>
<expose>
<port number="..."/>
...
</expose>
<volumes>
<volume name="..."/>
...
</volumes>
<environment>
<env name="variable" value="value"/>
...
</environment>
<labels>
<label name="..." value="..."/>
...
</labels>
</containerconfig>
</type>
spare_part receives a size value with an optional unit (M|G)
and is used for disk images to allow the creaton of a spare
partition of the configured size. Issue #234
The vboot firmware setup was a workaround on arm for the
chromebook platform several years ago. From a todays
perspective the setup and the way to boot would be different
and obsoletes the special case implementation. Thus this
is gone now. Related to Issue #234
The packages subsection <ignore> only accepted a package
name to become ignored. However a <package> subsection
to install a package allows for an arch selection. The
same should apply to ignore a package. In addition the
allowed content for the arch attribute has been changed
from simple text to a regular expression. Allowed is a
comma seperated list of reserved architecture words like
they are displayed by uname -m
Zypper supports the al (add lock) command which allows to ignore
a package in the dependecny resolution process. This is useful
to prevent installation of a package which was pulled in by e.g
a recommendation flag from the spec file. Packages marked to be
ignored are not handled for apt and yum right now. Using this
feature together with an unsupported package manager backend
results in a warning to the user
The configuration files /etc/sysconfig/bootloader and
/etc/default/grub needs to be created/updated with the
relevant values regarding the bootloader setup done by
kiwi. This Fixes#226
The root volume in a systemdisk setup is handled in a special
way. It is not allowed to setup a custom name or mountpoint for
the root volume. Therefore the size of the root volume can be
setup via the @root volume name. This check looks up the volume
setup and searches if there is a configuration for the '/'
mountpoint which would cause the image build to fail
If an extension section exists the xml data in this section
is validated against the RelaxNG schema which must be part
of the xmlcatalog for the used extension namespace. If the
data validates the method get_extension_xml_data can be used
to access the etree parse result for each of the extension
configurations
This commit ensure that the arch attribute of each package section
is parsed as a comma separated list. This way, as in previous kiwi
versions, each package might be suitable for one or more specified
architectures.
This commit adds a tester method to check if there is any
repository marked as imageinclude. The method is used to
execute import_repositories_marked_as_imageinclude
method only when it is actually needed.
This addresses the following issues
* An image based on btrfs using the entire boot as a subvolume
causes grub to be unable to find its config file. Thus kiwi
now raises an exception in this situation
* Don't prefix the bootpath with the snapshot path if the
btrfs root is placed in a snapshot. Instead the file
etc/default/grub must be written/updated with the
SUSE_BTRFS_SNAPSHOT_BOOTING variable set to true. Once
this is done the bootpath is consistently set to /boot
no matter which snapshot is active
Some attributes needed to be changed in the test xmls. With this
commit all the description files are consistent with the new co-
occurrence constraints.
As part of the machine section it is possible to setup a custom
entry which is stored in the machine configuration file. The
evaluation of such an entry for the vmdk (.vmx) config file
was still missing. This Fixes#122
Get_users method has been updated in order to be able to append the
groups attribute when a user is defined in multiple <users> sections in
the description file. Appart from groups, all other user attributes
are ignored if the user was already processed when parsing the description
file.
Get_users_groups method has been also updated so it returns a list of
groups with unique values, keeping the description file order.
Now each user item may or may not have a groups attribute. The
groups attribute is a comma separated list of group names. If
present the first group name will be the login or primary group,
if not present, the underlying toolchain will assign the default
login group for that user.
