YANG Tools Developer Guide


YANG Tools is set of libraries and tooling providing support for use YANG for Java (or other JVM-based language) projects and applications. The YANG Tools provides following features in OpenDaylight:

  • Parsing of YANG sources and semantic inference of relationship across YANG models as defined in RFC6020

  • Representation of YANG-modeled data in Java

    • Normalized Node representation - DOM-like tree model, which uses conceptual meta-model more tailored to YANG and OpenDaylight use-cases than a standard XML DOM model allows for.

  • Serialization / deserialization of YANG-modeled data driven by YANG models


YANG tools consist of the following logical subsystems:

YANG Tools




Set of general purpose code, which is not specific to YANG, but is also useful outside YANG Tools implementation.

YANG Model and Parser

YANG semantic model and lexical and semantic parser of YANG models, which creates in-memory cross-referenced representation of YANG models, which is used by other components to determine their behavior based on the model.

YANG data

Definition of Normalized Node APIs and Data Tree APIs, reference implementation of these APIs and implementation of XML and JSON codecs for Normalized Nodes.

YANG Maven Plugin

Maven plugin which integrates YANG parser into Maven build life-cycle and provides code-generation framework for components, which wants to generate code or other artefacts based on YANG model.


Project defines base concepts and helper classes which are project-agnostic and could be used outside of YANG Tools project scope.


  • yang-common

  • yang-data-api

  • yang-data-codec-gson

  • yang-data-codec-xml

  • yang-data-impl

  • yang-data-jaxen

  • yang-data-transform

  • yang-data-util

  • yang-maven-plugin

  • yang-maven-plugin-it

  • yang-maven-plugin-spi

  • yang-model-api

  • yang-model-export

  • yang-model-util

  • yang-parser-api

  • yang-parser-impl


Class diagram of yang model API



YANG Parser

YANG Statement Parser works on the idea of statement concepts as defined in RFC6020, section 6.3. We come up here with basic ModelStatement and StatementDefinition, following RFC6020 idea of having sequence of statements, where every statement contains keyword and zero or one argument. ModelStatement is extended by DeclaredStatement (as it comes from source, e.g. YANG source) and EffectiveStatement, which contains other sub-statements and tends to represent result of semantic processing of other statements (uses, augment for YANG). IdentifierNamespace represents common superclass for YANG model namespaces.

Input of the YANG Statement Parser is a collection of StatementStreamSource objects. StatementStreamSource interface is used for inference of effective model and is required to emit its statements using supplied StatementWriter. Each source (e.g. YANG source) has to be processed in three steps in order to emit different statements for each step. This package provides support for various namespaces used across statement parser in order to map relations during declaration phase process.

Currently, there are two implementations of StatementStreamSource in YANGtools:

  • YangStatementSourceImpl - intended for yang sources

  • YinStatementSourceImpl - intended for yin sources


Class diagram of yang data API



YANG data Codecs

Codecs which enable serialization of NormalizedNodes into YANG-modeled data in XML or JSON format and deserialization of YANG-modeled data in XML or JSON format into NormalizedNodes.

YANG Maven Plugin

Maven plugin which integrates YANG parser into Maven build life-cycle and provides code-generation framework for components, which wants to generate code or other artefacts based on YANG model.

How to / Tutorials

Working with YANG Model

First thing you need to do if you want to work with YANG models is to instantiate a SchemaContext object. This object type describes one or more parsed YANG modules.

In order to create it you need to utilize YANG statement parser which takes one or more StatementStreamSource objects as input and then produces the SchemaContext object.

StatementStreamSource object contains the source file information. It has two implementations, one for YANG sources - YangStatementSourceImpl, and one for YIN sources - YinStatementSourceImpl.

Here is an example of creating StatementStreamSource objects for YANG files, providing them to the YANG statement parser and building the SchemaContext:

StatementStreamSource yangModuleSource == new YangStatementSourceImpl("/example.yang", false);
StatementStreamSource yangModuleSource2 == new YangStatementSourceImpl("/example2.yang", false);

CrossSourceStatementReactor.BuildAction reactor == YangInferencePipeline.RFC6020_REACTOR.newBuild();
reactor.addSources(yangModuleSource, yangModuleSource2);

SchemaContext schemaContext == reactor.buildEffective();

First, StatementStreamSource objects with two constructor arguments should be instantiated: path to the yang source file (which is a regular String object) and a Boolean which determines if the path is absolute or relative.

Next comes the initiation of new yang parsing cycle - which is represented by CrossSourceStatementReactor.BuildAction object. You can get it by calling method newBuild() on CrossSourceStatementReactor object (RFC6020_REACTOR) in YangInferencePipeline class.

Then you should feed yang sources to it by calling method addSources() that takes one or more StatementStreamSource objects as arguments.

Finally, you call the method buildEffective() on the reactor object which returns EffectiveSchemaContext (that is a concrete implementation of SchemaContext). Now you are ready to work with contents of the added YANG sources.

Let us explain how to work with models contained in the newly created SchemaContext. If you want to get all the modules in the schemaContext, you have to call method getModules() which returns a Set of modules. If you want to get all the data definitions in schemaContext, you need to call method getDataDefinitions, etc.

Set<Module> modules == schemaContext.getModules();
Set<DataSchemaNodes> dataSchemaNodes == schemaContext.getDataDefinitions();

Usually you want to access specific modules. Getting a concrete module from SchemaContext is a matter of calling one of these methods:

  • findModuleByName(),

  • findModuleByNamespace(),

  • findModuleByNamespaceAndRevision().

In the first case, you need to provide module name as it is defined in the yang source file and module revision date if it specified in the yang source file (if it is not defined, you can just pass a null value). In order to provide the revision date in proper format, you can use a utility class named SimpleDateFormatUtil.

Module exampleModule == schemaContext.findModuleByName("example-module", null);
// or
Date revisionDate == SimpleDateFormatUtil.getRevisionFormat().parse("2015-09-02");
Module exampleModule == schemaContext.findModuleByName("example-module", revisionDate);

In the second case, you have to provide module namespace in form of an URI object.

Module exampleModule == schema.findModuleByNamespace(new URI("opendaylight.org/example-module"));

In the third case, you provide both module namespace and revision date as arguments.

Once you have a Module object, you can access its contents as they are defined in YANG Model API. One way to do this is to use method like getIdentities() or getRpcs() which will give you a Set of objects. Otherwise you can access a DataSchemaNode directly via the method getDataChildByName() which takes a QName object as its only argument. Here are a few examples.

Set<AugmentationSchema> augmentationSchemas == exampleModule.getAugmentations();
Set<ModuleImport> moduleImports == exampleModule.getImports();

ChoiceSchemaNode choiceSchemaNode == (ChoiceSchemaNode) exampleModule.getDataChildByName(QName.create(exampleModule.getQNameModule(), "example-choice"));

ContainerSchemaNode containerSchemaNode == (ContainerSchemaNode) exampleModule.getDataChildByName(QName.create(exampleModule.getQNameModule(), "example-container"));

The YANG statement parser can work in three modes:

  • default mode

  • mode with active resolution of if-feature statements

  • mode with active semantic version processing

The default mode is active when you initialize the parsing cycle as usual by calling the method newBuild() without passing any arguments to it. The second and third mode can be activated by invoking the newBuild() with a special argument. You can either activate just one of them or both by passing proper arguments. Let us explain how these modes work.

Mode with active resolution of if-features makes yang statements containing an if-feature statement conditional based on the supported features. These features are provided in the form of a QName-based java.util.Set object. In the example below, only two features are supported: example-feature-1 and example-feature-2. The Set which contains this information is passed to the method newBuild() and the mode is activated.

Set<QName> supportedFeatures = ImmutableSet.of(
    QName.create("example-namespace", "2016-08-31", "example-feature-1"),
    QName.create("example-namespace", "2016-08-31", "example-feature-2"));

CrossSourceStatementReactor.BuildAction reactor = YangInferencePipeline.RFC6020_REACTOR.newBuild(supportedFeatures);

In case when no features should be supported, you should provide an empty Set<QName> object.

Set<QName> supportedFeatures = ImmutableSet.of();

CrossSourceStatementReactor.BuildAction reactor = YangInferencePipeline.RFC6020_REACTOR.newBuild(supportedFeatures);

When this mode is not activated, all features in the processed YANG sources are supported.

Mode with active semantic version processing changes the way how YANG import statements work - each module import is processed based on the specified semantic version statement and the revision-date statement is ignored. In order to activate this mode, you have to provide StatementParserMode.SEMVER_MODE enum constant as argument to the method newBuild().

CrossSourceStatementReactor.BuildAction reactor == YangInferencePipeline.RFC6020_REACTOR.newBuild(StatementParserMode.SEMVER_MODE);

Before you use a semantic version statement in a YANG module, you need to define an extension for it so that the YANG statement parser can recognize it.

module semantic-version {
    namespace "urn:opendaylight:yang:extension:semantic-version";
    prefix sv;
    yang-version 1;

    revision 2016-02-02 {
        description "Initial version";
    sv:semantic-version "0.0.1";

    extension semantic-version {
        argument "semantic-version" {
            yin-element false;

In the example above, you see a YANG module which defines semantic version as an extension. This extension can be imported to other modules in which we want to utilize the semantic versioning concept.

Below is a simple example of the semantic versioning usage. With semantic version processing mode being active, the foo module imports the bar module based on its semantic version. Notice how both modules import the module with the semantic-version extension.

module foo {
    namespace foo;
    prefix foo;
    yang-version 1;

    import semantic-version { prefix sv; revision-date 2016-02-02; sv:semantic-version "0.0.1"; }
    import bar { prefix bar; sv:semantic-version "0.1.2";}

    revision "2016-02-01" {
        description "Initial version";
    sv:semantic-version "0.1.1";

module bar {
    namespace bar;
    prefix bar;
    yang-version 1;

    import semantic-version { prefix sv; revision-date 2016-02-02; sv:semantic-version "0.0.1"; }

    revision "2016-01-01" {
        description "Initial version";
    sv:semantic-version "0.1.2";


Every semantic version must have the following form: x.y.z. The x corresponds to a major version, the y corresponds to a minor version and the z corresponds to a patch version. If no semantic version is specified in a module or an import statement, then the default one is used - 0.0.0.

A major version number of 0 indicates that the model is still in development and is subject to change.

Following a release of major version 1, all modules will increment major version number when backwards incompatible changes to the model are made.

The minor version is changed when features are added to the model that do not impact current clients use of the model.

The patch version is incremented when non-feature changes (such as bugfixes or clarifications of human-readable descriptions that do not impact model functionality) are made that maintain backwards compatibility.

When importing a module with activated semantic version processing mode, only the module with the newest (highest) compatible semantic version is imported. Two semantic versions are compatible when all of the following conditions are met:

  • the major version in the import statement and major version in the imported module are equal. For instance, 1.5.3 is compatible with 1.5.3, 1.5.4, 1.7.2, etc., but it is not compatible with 0.5.2 or 2.4.8, etc.

  • the combination of minor version and patch version in the import statement is not higher than the one in the imported module. For instance, 1.5.2 is compatible with 1.5.2, 1.5.4, 1.6.8 etc. In fact, 1.5.2 is also compatible with versions like 1.5.1, 1.4.9 or 1.3.7 as they have equal major version. However, they will not be imported because their minor and patch version are lower (older).

If the import statement does not specify a semantic version, then the default one is chosen - 0.0.0. Thus, the module is imported only if it has a semantic version compatible with the default one, for example 0.0.0, 0.1.3, 0.3.5 and so on.

Working with YANG data

If you want to work with YANG data, you are going to need NormalizedNode objects that are specified in the YANG data API. NormalizedNode is an interface at the top of the YANG data hierarchy. It is extended through sub-interfaces which define the behavior of specific NormalizedNode types like AnyXmlNode, ChoiceNode, LeafNode, ContainerNode, etc. Concrete implementations of these interfaces are defined in yang-data-impl module. Once you have one or more NormalizedNode instances, you can perform CRUD operations on YANG data tree which is an in-memory database designed to store normalized nodes in a tree-like structure.

In some cases it, is clear which NormalizedNode type belongs to which yang statement (e.g. AnyXmlNode, ChoiceNode, LeafNode). However, there are some normalized nodes which are named differently from their yang counterparts. They are listed below:

Normalized Nodes






Leaf-list that is ordered-by user


Concrete entry in a leaf-list


Keyed list


Keyed list that is ordered-by user


Concrete entry in a keyed list


Unkeyed list


Concrete entry in an unkeyed list

To create a concrete NormalizedNode object, use the utility class Builders or ImmutableNodes. These classes can be found in yang-data-impl module and they provide methods for building each type of normalized node. Here is a simple example of building a normalized node:

// example 1
ContainerNode containerNode == Builders.containerBuilder().withNodeIdentifier(new YangInstanceIdentifier.NodeIdentifier(QName.create(moduleQName, "example-container")).build();

// example 2
ContainerNode containerNode2 == Builders.containerBuilder(containerSchemaNode).build();

Both examples produce the same result. NodeIdentifier is one of the four types of YangInstanceIdentifier (these types are described in the Javadoc of YangInstanceIdentifier). The purpose of YangInstanceIdentifier is to uniquely identify a particular node in the data tree. In the first example, you have to add NodeIdentifier before building the resulting node. In the second example it is also added using the provided ContainerSchemaNode object.

ImmutableNodes class offers similar builder methods and also adds an overloaded method called fromInstanceId() which allows you to create a NormalizedNode object based on YangInstanceIdentifier and SchemaContext. Below is an example which shows the use of this method.

YangInstanceIdentifier.NodeIdentifier contId == new YangInstanceIdentifier.NodeIdentifier(QName.create(moduleQName, "example-container");

NormalizedNode<?, ?> contNode == ImmutableNodes.fromInstanceId(schemaContext, YangInstanceIdentifier.create(contId));

Let us show a more complex example of creating a NormalizedNode. First, consider the following YANG module:

module example-module {
    namespace "opendaylight.org/example-module";
    prefix "example";

    container parent-container {
        container child-container {
            list parent-ordered-list {
                ordered-by user;

                key "parent-key-leaf";

                leaf parent-key-leaf {
                    type string;

                leaf parent-ordinary-leaf {
                    type string;

                list child-ordered-list {
                    ordered-by user;

                    key "child-key-leaf";

                    leaf child-key-leaf {
                        type string;

                    leaf child-ordinary-leaf {
                        type string;

In the following example, two normalized nodes based on the module above are written to and read from the data tree.

TipProducingDataTree inMemoryDataTree ==     InMemoryDataTreeFactory.getInstance().create(TreeType.OPERATIONAL);

// first data tree modification
MapEntryNode parentOrderedListEntryNode == Builders.mapEntryBuilder().withNodeIdentifier(
new YangInstanceIdentifier.NodeIdentifierWithPredicates(
parentOrderedListQName, parentKeyLeafQName, "pkval1"))
new YangInstanceIdentifier.NodeIdentifier(parentOrdinaryLeafQName))

OrderedMapNode parentOrderedListNode == Builders.orderedMapBuilder().withNodeIdentifier(
new YangInstanceIdentifier.NodeIdentifier(parentOrderedListQName))

ContainerNode parentContainerNode == Builders.containerBuilder().withNodeIdentifier(
new YangInstanceIdentifier.NodeIdentifier(parentContainerQName))
new NodeIdentifier(childContainerQName)).withChild(parentOrderedListNode).build()).build();

YangInstanceIdentifier path1 == YangInstanceIdentifier.of(parentContainerQName);

DataTreeModification treeModification == inMemoryDataTree.takeSnapshot().newModification();
treeModification.write(path1, parentContainerNode);

// second data tree modification
MapEntryNode childOrderedListEntryNode == Builders.mapEntryBuilder().withNodeIdentifier(
new YangInstanceIdentifier.NodeIdentifierWithPredicates(
childOrderedListQName, childKeyLeafQName, "chkval1"))
new YangInstanceIdentifier.NodeIdentifier(childOrdinaryLeafQName))

OrderedMapNode childOrderedListNode == Builders.orderedMapBuilder().withNodeIdentifier(
new YangInstanceIdentifier.NodeIdentifier(childOrderedListQName))

ImmutableMap.Builder<QName, Object> builder == ImmutableMap.builder();
ImmutableMap<QName, Object> keys == builder.put(parentKeyLeafQName, "pkval1").build();

YangInstanceIdentifier path2 == YangInstanceIdentifier.of(parentContainerQName).node(childContainerQName)
.node(parentOrderedListQName).node(new NodeIdentifierWithPredicates(parentOrderedListQName, keys)).node(childOrderedListQName);

treeModification.write(path2, childOrderedListNode);

DataTreeSnapshot snapshotAfterCommits == inMemoryDataTree.takeSnapshot();
Optional<NormalizedNode<?, ?>> readNode == snapshotAfterCommits.readNode(path1);
Optional<NormalizedNode<?, ?>> readNode2 == snapshotAfterCommits.readNode(path2);

First comes the creation of in-memory data tree instance. The schema context (containing the model mentioned above) of this tree is set. After that, two normalized nodes are built. The first one consists of a parent container, a child container and a parent ordered list which contains a key leaf and an ordinary leaf. The second normalized node is a child ordered list that also contains a key leaf and an ordinary leaf.

In order to add a child node to a node, method withChild() is used. It takes a NormalizedNode as argument. When creating a list entry, YangInstanceIdentifier.NodeIdentifierWithPredicates should be used as its identifier. Its arguments are the QName of the list, QName of the list key and the value of the key. Method withValue() specifies a value for the ordinary leaf in the list.

Before writing a node to the data tree, a path (YangInstanceIdentifier) which determines its place in the data tree needs to be defined. The path of the first normalized node starts at the parent container. The path of the second normalized node points to the child ordered list contained in the parent ordered list entry specified by the key value "pkval1".

Write operation is performed with both normalized nodes mentioned earlier. It consists of several steps. The first step is to instantiate a DataTreeModification object based on a DataTreeSnapshot. DataTreeSnapshot gives you the current state of the data tree. Then comes the write operation which writes a normalized node at the provided path in the data tree. After doing both write operations, method ready() has to be called, marking the modification as ready for application to the data tree. No further operations within the modification are allowed. The modification is then validated - checked whether it can be applied to the data tree. Finally, we commit it to the data tree.

Now you can access the written nodes. In order to do this, you must create a new DataTreeSnapshot instance and call the method readNode() with path argument pointing to a node in the tree.

Serialization / deserialization of YANG data

If you want to deserialize YANG-modeled data that has the form of an XML document, you can use the XML parser found in the module yang-data-codec-xml. The parser walks through the XML document containing YANG-modeled data based on the provided SchemaContext and emits node events into a NormalizedNodeStreamWriter. The parser disallows multiple instances of the same element except for leaf-list and list entries. The parser also expects that the YANG-modeled data in the XML source are wrapped in a root element. Otherwise it will not work correctly.

Here is an example of using the XML parser.

InputStream resourceAsStream == ExampleClass.class.getResourceAsStream("/example-module.yang");

XMLInputFactory factory == XMLInputFactory.newInstance();
XMLStreamReader reader == factory.createXMLStreamReader(resourceAsStream);

NormalizedNodeResult result == new NormalizedNodeResult();
NormalizedNodeStreamWriter streamWriter == ImmutableNormalizedNodeStreamWriter.from(result);

XmlParserStream xmlParser == XmlParserStream.create(streamWriter, schemaContext);

NormalizedNode<?, ?> transformedInput == result.getResult();

The XML parser utilizes the javax.xml.stream.XMLStreamReader for parsing an XML document. First, you should create an instance of this reader using XMLInputFactory and then load an XML document (in the form of InputStream object) into it.

In order to emit node events while parsing the data you need to instantiate a NormalizedNodeStreamWriter. This writer is actually an interface and therefore you need to use a concrete implementation of it. In this example it is the ImmutableNormalizedNodeStreamWriter, which constructs immutable instances of NormalizedNodes.

There are two ways how to create an instance of this writer using the static overloaded method from(). One version of this method takes a NormalizedNodeResult as argument. This object type is a result holder in which the resulting NormalizedNode will be stored. The other version takes a NormalizedNodeContainerBuilder as argument. All created nodes will be written to this builder.

Next step is to create an instance of the XML parser. The parser itself is represented by a class named XmlParserStream. You can use one of two versions of the static overloaded method create() to construct this object. One version accepts a NormalizedNodeStreamWriter and a SchemaContext as arguments, the other version takes the same arguments plus a SchemaNode. Node events are emitted to the writer. The SchemaContext is used to check if the YANG data in the XML source comply with the provided YANG model(s). The last argument, a SchemaNode object, describes the node that is the parent of nodes defined in the XML data. If you do not provide this argument, the parser sets the SchemaContext as the parent node.

The parser is now ready to walk through the XML. Parsing is initiated by calling the method parse() on the XmlParserStream object with XMLStreamReader as its argument.

Finally, you can access the result of parsing - a tree of NormalizedNodes contains the data as they are defined in the parsed XML document - by calling the method getResult() on the NormalizedNodeResult object.