avocado-framework-plugin-varianter-yaml-to-mux

.. _yaml-to-mux-plugin:

YAML to Mux plugin

avocado_varianter_yaml_to_mux

This plugin utilizes the multiplexation mechanism to produce variants out of a yaml file. This section is example-based, if you are interested in test parameters and/or multiplexation overview, please take a look at test-parameter.

As mentioned earlier, it inherits from the avocado_varianter_yaml_to_mux.mux.MuxPlugin and the only thing it implements is the argument parsing to get some input and a custom yaml parser (which is also capable of parsing json).

The YAML file is perfect for this task as it's easily read by both, humans and machines. Let's start with an example (line numbers at the first columns are for documentation purposes only, they are not part of the multiplex file format):

.. code-block:: yaml

 1  hw:
 2      cpu: !mux
 3          intel:
 4              cpu_CFLAGS: '-march=core2'
 5          amd:
 6              cpu_CFLAGS: '-march=athlon64'
 7          arm:
 8              cpu_CFLAGS: '-mabi=apcs-gnu -march=armv8-a -mtune=arm8'
 9      disk: !mux
10          scsi:
11              disk_type: 'scsi'
12          virtio:
13              disk_type: 'virtio'
14  distro: !mux
15      fedora:
16          init: 'systemd'
17      mint:
18          init: 'systemv'
19  env: !mux
20      debug:
21          opt_CFLAGS: '-O0 -g'
22      prod:
23          opt_CFLAGS: '-O2'

.. warning:: On some architectures misbehaving versions of CYaml Python library were reported and Avocado always fails with unacceptable character #x0000: control characters are not allowed. To workaround this issue you need to either update the PyYaml to the version which works properly, or you need to remove the python2.7/site-packages/yaml/cyaml.py or disable CYaml import in Avocado sources. For details check out the Github issue <https://github.com/avocado-framework/avocado/issues/1190>_

There are couple of key=>value pairs (lines 4,6,8,11,13,...) and there are named nodes which define scope (lines 1,2,3,5,7,9,...). There are also additional flags (lines 2, 9, 14, 19) which modifies the behavior.

Nodes

They define context of the key=>value pairs allowing us to easily identify for what this values might be used for and also it makes possible to define multiple values of the same keys with different scope.

Due to their purpose the YAML automatic type conversion for nodes names is disabled, so the value of node name is always as written in the YAML file (unlike values, where yes converts to True and such).

Nodes are organized in parent-child relationship and together they create a tree. To view this structure use avocado variants --tree -m <file>::

┗━━ run ┣━━ hw ┃ ┣━━ cpu ┃ ┃ ╠══ intel ┃ ┃ ╠══ amd ┃ ┃ ╚══ arm ┃ ┗━━ disk ┃ ╠══ scsi ┃ ╚══ virtio ┣━━ distro ┃ ╠══ fedora ┃ ╚══ mint ┗━━ env ╠══ debug ╚══ prod

You can see that hw has 2 children cpu and disk. All parameters defined in parent node are inherited to children and extended/overwritten by their values up to the leaf nodes. The leaf nodes (intel, amd, arm, scsi, ...) are the most important as after multiplexation they form the parameters available in tests.

Keys and Values

Every value other than dict (4,6,8,11) is used as value of the antecedent node.

Each node can define key/value pairs (lines 4,6,8,11,...). Additionally each children node inherits values of it's parent and the result is called node environment.

Given the node structure below:

.. code-block:: yaml

devtools:
    compiler: 'cc'
    flags:
        - '-O2'
    debug: '-g'
    fedora:
        compiler: 'gcc'
        flags:
            - '-Wall'
    osx:
        compiler: 'clang'
        flags:
            - '-arch i386'
            - '-arch x86_64'

And the rules defined as:

• Scalar values (Booleans, Numbers and Strings) are overwritten by walking from the root until the final node.
• Lists are appended (to the tail) whenever we walk from the root to the final node.

The environment created for the nodes fedora and osx are:

• Node //devtools/fedora environment compiler: 'gcc', flags: ['-O2', '-Wall']
• Node //devtools/osx environment compiler: 'clang', flags: ['-O2', '-arch i386', '-arch x86_64']

Note that due to different usage of key and values in environment we disabled the automatic value conversion for keys while keeping it enabled for values. This means that the key is always a string and the value can be YAML value, eg. bool, list, custom type, or string. Please be aware that due to limitation None type can be provided in yaml specifically as string 'null'.

Variants

In the end all leaves are gathered and turned into parameters, more specifically into AvocadoParams:

.. code-block:: yaml

setup:
    graphic:
        user: "guest"
        password: "pass"
    text:
        user: "root"
        password: "123456"

produces [graphic, text]. In the test code you'll be able to query only those leaves. Intermediary or root nodes are available.

The example above generates a single test execution with parameters separated by path. But the most powerful multiplexer feature is that it can generate multiple variants. To do that you need to tag a node whose children are meant to be multiplexed. Effectively it returns only leaves of one child at the time.In order to generate all possible variants multiplexer creates cartesian product of all of these variants:

.. code-block:: yaml

cpu: !mux
    intel:
    amd:
    arm:
fmt: !mux
    qcow2:
    raw:

Produces 6 variants::

/cpu/intel, /fmt/qcow2
/cpu/intel, /fmt/raw
...
/cpu/arm, /fmt/raw

The !mux evaluation is recursive so one variant can expand to multiple ones:

.. code-block:: yaml

fmt: !mux
    qcow: !mux
        2:
        2v3:
    raw:

Results in::

/fmt/qcow2/2
/fmt/qcow2/2v3
/raw

.. _yaml-to-mux-resolution-order:

Resolution order

You can see that only leaves are part of the test parameters. It might happen that some of these leaves contain different values of the same key. Then you need to make sure your queries separate them by different paths. When the path matches multiple results with different origin, an exception is raised as it's impossible to guess which key was originally intended.

To avoid these problems it's recommended to use unique names in test parameters if possible, to avoid the mentioned clashes. It also makes it easier to extend or mix multiple YAML files for a test.

For multiplex YAML files that are part of a framework, contain default configurations, or serve as plugin configurations and other advanced setups it is possible and commonly desirable to use non-unique names. But always keep those points in mind and provide sensible paths.

Multiplexer also supports default paths. By default it's /run/* but it can be overridden by --mux-path, which accepts multiple arguments. What it does it splits leaves by the provided paths. Each query goes one by one through those sub-trees and first one to hit the match returns the result. It might not solve all problems, but it can help to combine existing YAML files with your ones:

.. code-block:: yaml

qa:         # large and complex read-only file, content injected into /qa
    tests:
        timeout: 10
    ...
my_variants: !mux        # your YAML file injected into /my_variants
    short:
        timeout: 1
    long:
        timeout: 1000

You want to use an existing test which uses params.get('timeout', '*'). Then you can use --mux-path '/my_variants/*' '/qa/*' and it'll first look in your variants. If no matches are found, then it would proceed to /qa/*

Keep in mind that only slices defined in mux-path are taken into account for relative paths (the ones starting with *)

Injecting files

You can run any test with any YAML file by::

avocado run sleeptest.py --mux-yaml file.yaml

This puts the content of file.yaml into /run location, which as mentioned in previous section, is the default mux-path path. For most simple cases this is the expected behavior as your files are available in the default path and you can safely use params.get(key).

When you need to put a file into a different location, for example when you have two files and you don't want the content to be merged into a single place becoming effectively a single blob, you can do that by giving a name to your YAML file::

avocado run sleeptest.py --mux-yaml duration:duration.yaml

The content of duration.yaml is injected into /run/duration. Still when keys from other files don't clash, you can use params.get(key) and retrieve from this location as it's in the default path, only extended by the duration intermediary node. Another benefit is you can merge or separate multiple files by using the same or different name, or even a complex (relative) path.

Last but not least, advanced users can inject the file into whatever location they prefer by::

avocado run sleeptest.py --mux-yaml /my/variants/duration:duration.yaml

Simple params.get(key) won't look in this location, which might be the intention of the test writer. There are several ways to access the values:

• absolute location params.get(key, '/my/variants/duration')
• absolute location with wildcards params.get(key, '/my/*) (or /*/duration/*...)
• set the mux-path avocado run ... --mux-path /my/* and use relative path

It's recommended to use the simple injection for single YAML files, relative injection for multiple simple YAML files and the last option is for very advanced setups when you either can't modify the YAML files and you need to specify custom resolution order or you are specifying non-test parameters, for example parameters for your plugin, which you need to separate from the test parameters.

Special values

As you might have noticed, we are using mapping/dicts to define the structure of the params. To avoid surprises we disallowed the smart typing of mapping keys so:

.. code-block:: yaml

on: on

Won't become True: True, but the key will be preserved as string on: True.

You might also want to use dict as values in your params. This is also supported but as we can't easily distinguish whether that value is a value or a node (structure), you have to either embed it into another object (list, ..) or you have to clearly state the type (yaml tag !!python/dict). Even then the value won't be a standard dictionary, but it'll be collections.OrderedDict and similarly to nodes structure all keys are preserved as strings and no smart type detection is used. Apart from that it should behave similarly as dict, only you get the values ordered by the order they appear in the file.

Multiple files

You can provide multiple files. In such scenario final tree is a combination of the provided files where later nodes with the same name override values of the preceding corresponding node. New nodes are appended as new children:

.. code-block:: yaml

file-1.yaml:
    debug:
        CFLAGS: '-O0 -g'
    prod:
        CFLAGS: '-O2'

file-2.yaml:
    prod:
        CFLAGS: '-Os'
    fast:
        CFLAGS: '-Ofast'

results in:

.. code-block:: yaml

debug:
    CFLAGS: '-O0 -g'
prod:
    CFLAGS: '-Os'       # overridden
fast:
    CFLAGS: '-Ofast'    # appended

It's also possible to include existing file into another a given node in another file. This is done by the !include : $path directive:

.. code-block:: yaml

os:
    fedora:
        !include : fedora.yaml
    gentoo:
        !include : gentoo.yaml

.. warning:: Due to YAML nature, it's mandatory to put space between !include and the colon (:) that must follow it.

The file location can be either absolute path or relative path to the YAML file where the !include is called (even when it's nested).

Whole file is merged into the node where it's defined.

Advanced YAML tags

There are additional features related to YAML files. Most of them require values separated by ":". Again, in all such cases it's mandatory to add a white space (" ") between the tag and the ":", otherwise ":" is part of the tag name and the parsing fails.

!include ^^^^^^^^

Includes other file and injects it into the node it's specified in:

.. code-block:: yaml

my_other_file:
    !include : other.yaml

The content of /my_other_file would be parsed from the other.yaml. It's the hardcoded equivalent of the -m $using:$path.

Relative paths start from the original file's directory.

!using ^^^^^^

Prepends path to the node it's defined in:

.. code-block:: yaml

!using : /foo
bar:
    !using : baz

bar is put into baz becoming /baz/bar and everything is put into /foo. So the final path of bar is /foo/baz/bar.

!remove_node ^^^^^^^^^^^^

Removes node if it existed during the merge. It can be used to extend incompatible YAML files:

.. code-block:: yaml

os:
    fedora:
    windows:
        3.11:
        95:
os:
    !remove_node : windows
    windows:
        win3.11:
        win95:

Removes the windows node from structure. It's different from filter-out as it really removes the node (and all children) from the tree and it can be replaced by you new structure as shown in the example. It removes windows with all children and then replaces this structure with slightly modified version.

As !remove_node is processed during merge, when you reverse the order, windows is not removed and you end-up with /windows/{win3.11,win95,3.11,95} nodes.

!remove_value ^^^^^^^^^^^^^

It's similar to !remove_node_ only with values.

!mux ^^^^

Children of this node will be multiplexed. This means that in first variant it'll return leaves of the first child, in second the leaves of the second child, etc. Example is in section Variants_

!filter-only

Defines internal filters. They are inherited by children and evaluated during multiplexation. It allows one to specify the only compatible branch of the tree with the current variant, for example::

cpu:
    arm:
        !filter-only : /disk/virtio
disk:
    virtio:
    scsi:

will skip the [arm, scsi] variant and result only in [arm, virtio]

Note: It's possible to use !filter-only multiple times with the same parent and all allowed variants will be included (unless they are filtered-out by !filter-out)

Note2: The evaluation order is 1. filter-out, 2. filter-only. This means when you booth filter-out and filter-only a branch it won't take part in the multiplexed variants.

!filter-out

Similarly to !filter-only_ only it skips the specified branches and leaves the remaining ones. (in the same example the use of !filter-out : /disk/scsi results in the same behavior). The difference is when a new disk type is introduced, !filter-only still allows just the specified variants, while !filter-out only removes the specified ones.

As for the speed optimization, currently Avocado is strongly optimized towards fast !filter-out so it's highly recommended using them rather than !filter-only, which takes significantly longer to process.

Complete example

Let's take a second look at the first example::

 1    hw:
 2        cpu: !mux
 3            intel:
 4                cpu_CFLAGS: '-march=core2'
 5            amd:
 6                cpu_CFLAGS: '-march=athlon64'
 7            arm:
 8                cpu_CFLAGS: '-mabi=apcs-gnu -march=armv8-a -mtune=arm8'
 9        disk: !mux
10            scsi:
11                disk_type: 'scsi'
12            virtio:
13                disk_type: 'virtio'
14    distro: !mux
15        fedora:
16            init: 'systemd'
17        mint:
18            init: 'systemv'
19    env: !mux
20        debug:
21            opt_CFLAGS: '-O0 -g'
22        prod:
23            opt_CFLAGS: '-O2'

After filters are applied (simply removes non-matching variants), leaves are gathered and all variants are generated::

$ avocado variants -m selftests/.data/mux-environment.yaml
Variants generated:
Variant 1:    /hw/cpu/intel, /hw/disk/scsi, /distro/fedora, /env/debug
Variant 2:    /hw/cpu/intel, /hw/disk/scsi, /distro/fedora, /env/prod
Variant 3:    /hw/cpu/intel, /hw/disk/scsi, /distro/mint, /env/debug
Variant 4:    /hw/cpu/intel, /hw/disk/scsi, /distro/mint, /env/prod
Variant 5:    /hw/cpu/intel, /hw/disk/virtio, /distro/fedora, /env/debug
Variant 6:    /hw/cpu/intel, /hw/disk/virtio, /distro/fedora, /env/prod
Variant 7:    /hw/cpu/intel, /hw/disk/virtio, /distro/mint, /env/debug
Variant 8:    /hw/cpu/intel, /hw/disk/virtio, /distro/mint, /env/prod
Variant 9:    /hw/cpu/amd, /hw/disk/scsi, /distro/fedora, /env/debug
Variant 10:    /hw/cpu/amd, /hw/disk/scsi, /distro/fedora, /env/prod
Variant 11:    /hw/cpu/amd, /hw/disk/scsi, /distro/mint, /env/debug
Variant 12:    /hw/cpu/amd, /hw/disk/scsi, /distro/mint, /env/prod
Variant 13:    /hw/cpu/amd, /hw/disk/virtio, /distro/fedora, /env/debug
Variant 14:    /hw/cpu/amd, /hw/disk/virtio, /distro/fedora, /env/prod
Variant 15:    /hw/cpu/amd, /hw/disk/virtio, /distro/mint, /env/debug
Variant 16:    /hw/cpu/amd, /hw/disk/virtio, /distro/mint, /env/prod
Variant 17:    /hw/cpu/arm, /hw/disk/scsi, /distro/fedora, /env/debug
Variant 18:    /hw/cpu/arm, /hw/disk/scsi, /distro/fedora, /env/prod
Variant 19:    /hw/cpu/arm, /hw/disk/scsi, /distro/mint, /env/debug
Variant 20:    /hw/cpu/arm, /hw/disk/scsi, /distro/mint, /env/prod
Variant 21:    /hw/cpu/arm, /hw/disk/virtio, /distro/fedora, /env/debug
Variant 22:    /hw/cpu/arm, /hw/disk/virtio, /distro/fedora, /env/prod
Variant 23:    /hw/cpu/arm, /hw/disk/virtio, /distro/mint, /env/debug
Variant 24:    /hw/cpu/arm, /hw/disk/virtio, /distro/mint, /env/prod

Where the first variant contains::

/hw/cpu/intel/  => cpu_CFLAGS: -march=core2
/hw/disk/       => disk_type: scsi
/distro/fedora/ => init: systemd
/env/debug/     => opt_CFLAGS: -O0 -g

The second one::

/hw/cpu/intel/  => cpu_CFLAGS: -march=core2
/hw/disk/       => disk_type: scsi
/distro/fedora/ => init: systemd
/env/prod/      => opt_CFLAGS: -O2

From this example you can see that querying for /env/debug works only in the first variant, but returns nothing in the second variant. Keep this in mind and when you use the !mux flag always query for the pre-mux path, /env/* in this example.

Injecting values

Beyond the values injected by YAML files specified it's also possible inject values directly from command line to the final multiplex tree. It's done by the argument --mux-inject. The format of expected value is [path:]key:node_value.

.. warning:: When no path is specified to --mux-inject the parameter is added under tree root /. For example: running avocado passing --mux-inject my_key:my_value the parameter can be accessed calling self.params.get('my_key'). If the test writer wants to put the injected value in any other path location, like extending the /run path, it needs to be informed on avocado run call. For example: --mux-inject /run/:my_key:my_value makes possible to access the parameters calling self.params.get('my_key', '/run')

A test that gets parameters without a defined path, such as examples/tests/multiplextest.py::

os_type = self.params.get('os_type', default='linux')

Running it::

$ avocado --show=test run -- examples/tests/multiplextest.py | grep os_type PARAMS (key=os_type, path=*, default=linux) => 'linux'

Now, injecting a value, by default will put it in /, which is not in the default list of paths searched for::

$ avocado --show=test run --mux-inject os_type:myos -- examples/tests/multiplextest.py | grep os_type PARAMS (key=os_type, path=*, default=linux) => 'linux'

A path that is searched for by default is /run. To set the value to that path use::

$ avocado --show=test run --mux-inject /run:os_type:myos -- examples/tests/multiplextest.py | grep os_type PARAMS (key=os_type, path=*, default=linux) => 'myos'

Or, add the / to the list of paths searched for by default::

$ avocado --show=test run --mux-inject os_type:myos --mux-path / -- examples/tests/multiplextest.py | grep os_type PARAMS (key=os_type, path=*, default=linux) => 'myos'

.. warning:: By default, the values are parsed for the respective data types. When not possible, it falls back to string. If you want to maintain some value as string, enclose within quotes, properly escaped, and eclose that again in quotes. For example: a value of 1 is treated as integer, a value of 1,2 is treated as list, a value of abc is treated as string, a value of 1,2,5-10 is treated as list of integers as 1,2,-5. If you want to maintain this as string, provide the value as "\"1,2,5-10\""

.. _mutliplexer:

Multiplexer

avocado_varianter_yaml_to_mux.mux

Multiplexer or simply Mux is an abstract concept, which was the basic idea behind the tree-like params structure with the support to produce all possible variants. There is a core implementation of basic building blocks that can be used when creating a custom plugin. There is a demonstration version of plugin using this concept in avocado_varianter_yaml_to_mux which adds a parser and then uses this multiplexer concept to define an Avocado plugin to produce variants from yaml (or json) files.

Multiplexer concept ^^^^^^^^^^^^^^^^^^^

As mentioned earlier, this is an in-core implementation of building blocks intended for writing varianter-plugins based on a tree with Multiplex domains_ defined. The available blocks are:

• MuxTree_ - Object which represents a part of the tree and handles the multiplexation, which means producing all possible variants from a tree-like object.
• MuxPlugin_ - Base class to build varianter-plugins
• MuxTreeNode - Inherits from tree-node and adds the support for control flags (MuxTreeNode.ctrl) and multiplex domains (MuxTreeNode.multiplex).

And some support classes and methods eg. for filtering and so on.

Multiplex domains ^^^^^^^^^^^^^^^^^

A default avocado-params tree with variables could look like this::

Multiplex tree representation: ┣━━ paths ┃ → tmp: /var/tmp ┃ → qemu: /usr/libexec/qemu-kvm ┗━━ environ → debug: False

The multiplexer wants to produce similar structure, but also to be able to define not just one variant, but to define all possible combinations and then report the slices as variants. We use the term Multiplex domains_ to define that children of this node are not just different paths, but they are different values and we only want one at a time. In the representation we use double-line to visibly distinguish between normal relation and multiplexed relation. Let's modify our example a bit::

Multiplex tree representation: ┣━━ paths ┃ → tmp: /var/tmp ┃ → qemu: /usr/libexec/qemu-kvm ┗━━ environ ╠══ production ║ → debug: False ╚══ debug → debug: True

The difference is that environ is now a multiplex node and it's children will be yielded one at a time producing two variants::

Variant 1: ┣━━ paths ┃ → tmp: /var/tmp ┃ → qemu: /usr/libexec/qemu-kvm ┗━━ environ ┗━━ production → debug: False Variant 2: ┣━━ paths ┃ → tmp: /var/tmp ┃ → qemu: /usr/libexec/qemu-kvm ┗━━ environ ┗━━ debug → debug: False

Note that the multiplex is only about direct children, therefore the number of leaves in variants might differ::

Multiplex tree representation: ┣━━ paths ┃ → tmp: /var/tmp ┃ → qemu: /usr/libexec/qemu-kvm ┗━━ environ ╠══ production ║ → debug: False ╚══ debug ┣━━ system ┃ → debug: False ┗━━ program → debug: True

Produces one variant with /paths and /environ/production and other variant with /paths, /environ/debug/system and /environ/debug/program.

As mentioned earlier the power is not in producing one variant, but in defining huge scenarios with all possible variants. By using tree-structure with multiplex domains you can avoid most of the ugly filters you might know from Jenkins sparse matrix jobs. For comparison let's have a look at the same example in Avocado::

Multiplex tree representation: ┗━━ os ┣━━ distro ┃ ┗━━ redhat ┃ ╠══ fedora ┃ ║ ┣━━ version ┃ ║ ┃ ╠══ 20 ┃ ║ ┃ ╚══ 21 ┃ ║ ┗━━ flavor ┃ ║ ╠══ workstation ┃ ║ ╚══ cloud ┃ ╚══ rhel ┃ ╠══ 5 ┃ ╚══ 6 ┗━━ arch ╠══ i386 ╚══ x86_64

Which produces::

Variant 1: /os/distro/redhat/fedora/version/20, /os/distro/redhat/fedora/flavor/workstation, /os/arch/i386 Variant 2: /os/distro/redhat/fedora/version/20, /os/distro/redhat/fedora/flavor/workstation, /os/arch/x86_64 Variant 3: /os/distro/redhat/fedora/version/20, /os/distro/redhat/fedora/flavor/cloud, /os/arch/i386 Variant 4: /os/distro/redhat/fedora/version/20, /os/distro/redhat/fedora/flavor/cloud, /os/arch/x86_64 Variant 5: /os/distro/redhat/fedora/version/21, /os/distro/redhat/fedora/flavor/workstation, /os/arch/i386 Variant 6: /os/distro/redhat/fedora/version/21, /os/distro/redhat/fedora/flavor/workstation, /os/arch/x86_64 Variant 7: /os/distro/redhat/fedora/version/21, /os/distro/redhat/fedora/flavor/cloud, /os/arch/i386 Variant 8: /os/distro/redhat/fedora/version/21, /os/distro/redhat/fedora/flavor/cloud, /os/arch/x86_64 Variant 9: /os/distro/redhat/rhel/5, /os/arch/i386 Variant 10: /os/distro/redhat/rhel/5, /os/arch/x86_64 Variant 11: /os/distro/redhat/rhel/6, /os/arch/i386 Variant 12: /os/distro/redhat/rhel/6, /os/arch/x86_64

Versus Jenkins sparse matrix::

os_version = fedora20 fedora21 rhel5 rhel6 os_flavor = none workstation cloud arch = i386 x86_64

filter = ((os_version == "rhel5").implies(os_flavor == "none") && (os_version == "rhel6").implies(os_flavor == "none")) && !(os_version == "fedora20" && os_flavor == "none") && !(os_version == "fedora21" && os_flavor == "none")

Which is still relatively simple example, but it grows dramatically with inner-dependencies.

MuxPlugin ^^^^^^^^^

avocado_varianter_yaml_to_mux.mux.MuxPlugin

Defines the full interface required by avocado.core.plugin_interfaces.Varianter. The plugin writer should inherit from this MuxPlugin, then from the Varianter and call the::

self.initialize_mux(root, paths, debug)

Where:

• root - is the root of your params tree (compound of tree-node -like nodes)
• paths - is the parameter-paths to be used in test with all variants
• debug - whether to use debug mode (requires the passed tree to be compound of TreeNodeDebug-like nodes which stores the origin of the variant/value/environment as the value for listing purposes and is NOT intended for test execution.

This method must be called before the varianter's second stage. The MuxPlugin_'s code will take care of the rest.

MuxTree ^^^^^^^

This is the core feature where the hard work happens. It walks the tree and remembers all leaf nodes or uses list of MuxTrees when another multiplex domain is reached while searching for a leaf.

When it's asked to report variants, it combines one variant of each remembered item (leaf node always stays the same, but MuxTree circles through it's values) which recursively produces all possible variants of different multiplex domains_.