...

<extension point="de.cau.cs.kieler.kiml.layoutProviders">
    <layoutOption
        id="de.cau.cs.kieler.nodeLabelPlacement"
        name="Node Label Placement"
        description="Hints for where node labels are to be placed; if empty, the node label's position is not modified."
        advanced="true"
        appliesTo="nodes"
        type="enumset"
        class="de.cau.cs.kieler.kiml.options.NodeLabelPlacement"
        default="">
    </layoutOption>
</extension>

Such declarations are provided by layout algorithm developers, but not by tool developers who merely want to connect the layout infrastructure to their diagram viewers. Let's walk through the attributes available for layout options (not every available attribute appears in the example above):

...

This tells KIML that the defined layout algorithm supports the border spacing option. And even more, it overrides the default value declared by the layout option and sets it to 20.

The

...

By now, we have an idea of what layout options meta data gathered from the extension point are made available through LayoutMetaDataService. For direct programmatic access, some of that information is duplicated with constants in the class LayoutOptions. The layout option declared above, for example, is available as LayoutOptions.NODE_LABEL_PLACEMENT.

The Layout Option Manager

By now, we have an idea of what layout options do and why they are important in the first place. However, we haven't looked at how layout options end up on KGraph elements yet. This is where the LayoutOptionsManager comes in.After a diagram layout manager has LayoutOptionManager comes in. If you are not interested in the internal details, but want to configure automatic layout for your diagram viewer or editor, you may skip this section and proceed to programmatically setting layout options.

After a diagram layout manager has finished turning a given diagram into its KGraph representation, the layout options option manager is asked to enrich the KGraph elements with layout options. The option values can come from different sources: the user might have set some using the layout view; there might be some defaults for certain kinds of diagrams; or the programmer might have decided to attach some layout options to certain elements for just this one layout run. Whatever the source, the options manager is in charge of collecting all these layout option values and making sure they find their way to the correct KGraph element. To start off with a clean plate, it first makes sure there are no layout options attached to the KGraph elements. It then does two things: collect every eligible source of layout options, and transfer values of layout options to the associated KGraph elements. Sounds easy enough.

The question remains how the layout options sources work. Each source is represented by a class that implements the ILayoutConfig interface, called a layout configurator. KIML currently provides the following layout configurators, each representing a particular source of layout options, listed here in order of increasing priority:

...

This may seem complicated, and it is, but the good news is that the vast majority of developers will not need to dig that deep into the layout configuration infrastructure. There are easier ways to specify configurations, as described in the following section.

Programmatically Setting Layout Options

Anchor
programmatic-config programmatic-config

So with all these layout configurators available, how do you actually go about setting values for layout options programmatically? Well, as always: it depends.

...

...

Using the Extension Point

The recommended way to configure your layout is to use the layoutConfigs extension point. It offers three different kinds of configurations, explained in the following.

staticConfig

A staticConfig element can set one value for one layout option in the context of a particular diagram element type. Let's see an example:

      <staticConfig
            class="org.eclipse.emf.ecore.EReference"
            option="de.cau.cs.kieler.edgeType"
            value="ASSOCIATION">
      </staticConfig>

Here class refers to a domain model class, in this case the EReference class from the Ecore meta model defined by EMF, and option refers to a layout option through its identifier. The meaning of this declaration is that whenever automatic layout is requested for an Ecore class diagram, the edgeType option is set to ASSOCIATION for all edges linked to instances of EReference. Since the domain model (abstract syntax) is independent of the specific diagram viewer (concrete syntax), this configuration is applied to all diagram viewers that use the Ecore meta model.

Alternatively to domain model elements, staticConfig may also reference concrete syntax elements:

      <staticConfig
            class="org.eclipse.emf.ecoretools.diagram.edit.parts.EClassESuperTypesEditPart"
            option="de.cau.cs.kieler.edgeType"
            value="GENERALIZATION">
      </staticConfig>

This layout option value is applied only to edges linked to instances of EClassESuperTypesEditPart, which is a concrete syntax element of the Ecore Tools class diagram editor. Other Ecore meta model editors are not affected by this declaration. This distinction is particularly useful for meta models that are accessed with multiple different editors, as is often the case for UML tools.

A third variant is the use of diagram types, as in this example:

      <diagramType
            id="de.cau.cs.kieler.layout.diagrams.classDiagram"
            name="Class Diagram">
      </diagramType>
      <staticConfig
            class="org.eclipse.emf.ecore.EPackage"
            option="de.cau.cs.kieler.diagramType"
            value="de.cau.cs.kieler.layout.diagrams.classDiagram">
      </staticConfig>
      <staticConfig
            class="de.cau.cs.kieler.layout.diagrams.classDiagram"
            option="de.cau.cs.kieler.edgeRouting"
            value="SPLINES">
      </staticConfig>

A diagram type can be declared with a diagramType element and can be associated with an abstract syntax or concrete syntax class using the diagramType option, as shown in the first staticConfig declaration in the example above. The second staticConfig sets an option for the declared diagram type by using its identifier in the class attribute. This kind of indirection is very useful when you have n model classes and you want to set m layout options for each of those classes. Instead of writing n × m static declarations, you assign a diagram type t to each of the n classes and then declare the m layout options for t, resulting in n + m option declarations (in many cases n + m < n × m).

A further use of diagram types is for the selection of layout algorithms: a layout algorithm may declare that is is especially suited to process diagrams of certain type t. If the diagram type t is assigned to a diagram viewer, the most suitable layout algorithm is chosen automatically for that viewer.

The following diagram types are predefined in KIML:

• de.cau.cs.kieler.layout.diagrams.stateMachine – All kinds of state machines, statecharts, etc.
• de.cau.cs.kieler.layout.diagrams.dataFlow – All kinds of data flow diagrams, e.g. actor diagrams, block diagrams, certain component diagrams, etc.
• de.cau.cs.kieler.layout.diagrams.classDiagram – Class diagrams as defined by the UML, but also meta model diagrams such as the Ecore format.
• de.cau.cs.kieler.layout.diagrams.usecaseDiagram – UML use case diagrams.
• de.cau.cs.kieler.layout.diagrams.boxes – Unconnected boxes (graphs with no edges), e.g. parallel regions in statecharts.

semanticConfig

A semanticConfig element registers a subclass of SemanticLayoutConfig:

      <semanticOption
            class="de.cau.cs.kieler.synccharts.Scope"
            config="de.cau.cs.kieler.synccharts.diagram.custom.AnnotationsLayoutConfig">
      </semanticOption>

Similarly to staticConfig entries, the class attribute refers to which model elements the configuration is applied. However, only domain model (a.k.a. semantic model) classes may be referenced. The config attribute names a concrete implementation of the semantic layout configurator.

The advantage of this kind of configuration compared to staticConfig declarations is that it may perform arbitrary analyses of the domain model. For instance, different option values may be computed depending on certain properties of the domain model elements. This approach can be used to enable annotations of domain model elements. When the domain model is stored with a textual format, e.g. defined with Xtext, such annotations can be written in the source file that specifies the model:

@portConstraints FIXED_SIDE
@minWidth 20.0
@minHeight 15.0
entity IdentityActor
{
    @portSide WEST
    port Input;
    
    @portSide EAST
    port Output;
}

The source file annotations can be translated to KIML layout options with a semantic layout configurator, which is registered to each domain model class where annotations can occur.

customConfig

This extension element can be used to register arbitrary implementations of ILayoutConfig. However, this is required only for some experimental configurators used in research. Tool developers normally do not need to use this kind of extension element.

Using Volatile Configurators

The class VolatileLayoutConfig is meant for programmatic layout configuration. It stores layout option values in a hash map. Values are either set globally, that means they are applied to all graph elements, or with a specific context. Global values are easy to configure:

DiagramLayoutEngine.INSTANCE.layout(workbenchPart, diagramPart,
        new VolatileLayoutConfig()
            .setValue(LayoutOptions.ALGORITHM, "de.cau.cs.kieler.klay.layered")
            .setValue(LayoutOptions.SPACING, 30.0f)
            .setValue(LayoutOptions.ANIMATE, true));

If multiple configurators are passed to the DiagramLayoutEngine, the layout is computed multiple times: once for each configurator. This behavior can be used to apply different layout algorithms one after another, e.g. first a node placer algorithm and then an edge router algorithm, as in this example:

DiagramLayoutEngine.INSTANCE.layout(workbenchPart, diagramPart,
        new VolatileLayoutConfig()
            .setValue(LayoutOptions.ALGORITHM, "de.cau.cs.kieler.klay.force"),
        new VolatileLayoutConfig()
            .setValue(LayoutOptions.ALGORITHM, "de.cau.cs.kieler.kiml.libavoid"));

If you want to use multiple configurators in the same layout computation, use a CompoundLayoutConfig:

DiagramLayoutEngine.INSTANCE.layout(workbenchPart, diagramPart,
        CompoundLayoutConfig.of(config1, config2, ...));

Setting layout option values with a specific context is done with this method of VolatileLayoutConfig:

public <T, C> VolatileLayoutConfig setValue(final IProperty<? super T> option, final C contextObj,
            final IProperty<? super C> contextKey, final T value)

Don't be scared by the rather cryptic declaration. The arguments contextKey and contextObj determine in which context the option value is to be applied. For instance, using LayoutContext.DOMAIN_MODEL as context key and a specific domain model element as context object, the value is applied exactly to the graph element that is linked to that model element. If you want to refer to an element of the diagram viewer, i.e. the concrete syntax, use LayoutContext.DIAGRAM_PART as context key. The return value is the volatile layout configurator itself, allowing for a builder pattern.

Configuration During Layout Graph Construction

Volatile configurators are also useful for the implementation of diagram layout managers (IDiagramLayoutManager). These implementations are responsible for creating layout graphs following the KGraph meta model from a given diagram viewer (method buildLayoutGraph(…)). For some layout options it is reasonable to determine concrete values while the layout graph is built, e.g. for the minimal width and height of nodes:

KNode childLayoutNode = KimlUtil.createInitializedNode();
KShapeLayout nodeLayout = childLayoutNode.getData(KShapeLayout.class);
Dimension minSize = nodeEditPart.getFigure().getMinimumSize();
nodeLayout.setProperty(LayoutOptions.MIN_WIDTH, (float) minSize.width);
nodeLayout.setProperty(LayoutOptions.MIN_HEIGHT, (float) minSize.height);

The problem is that the layout option manager that applies all configurators to the layout graph removes any option values that have been set directly on the graph elements, hence the configuration done in the previous example has no effect on the layout process. But do not fear, for salvation is near:

mapping.getLayoutConfigs().add(VolatileLayoutConfig.fromProperties(mapping.getLayoutGraph(), PRIORITY));

The variable mapping refers to the LayoutMapping instance created in buildLayoutGraph(…). The static method fromProperties(…) offered by VolatileLayoutConfig creates a configuration that contains all the layout option values that have previously been seen directly on the graph elements. By adding this configuration to the layout mapping, we make sure it is considered by the layout option manager and the options are applied to the graph elements exactly as we have specified. Happy end.

If you are uncertain about which value to use for PRIORITY, take something like 25.

Adding Support for the Layout View

The Layout View empowers users to directly modify the layout configuration for the currently viewed diagram. This power, however, comes with a price. Tool developers implementing a IDiagramLayoutManager additionally have to provide an implementation of IMutableLayoutConfig that loads and saves layout option values in a way that they can be stored persistently with the respective diagram. Good examples of such configurators are GmfLayoutConfig and GraphitiLayoutConfig for GMF and Graphiti diagrams, respectively. The GMF implementation stores option values as styles in the GMF Notation model, while the Graphiti implementation stores the values as properties in the Graphiti Pictogram model. If you are developing an editor based on GMF or Graphiti, simply reuse these implementations and you're fine. Otherwise, read this section to learn how to implement a suitable configurator.

A mutable layout configurator is one that can not only read option values, but also write them. Most interface methods are rather self-explanatory, therefore we will consider only the getContextValue(IProperty, LayoutContext) method here. This method receives a LayoutContext and should return a value that corresponds to the given property, if possible, and null otherwise. The starting point is usually the current value of LayoutContext.DIAGRAM_PART in the given context, called the diagram part, which refers to the currently selected diagram element in the viewer (the abstract syntax element). From this the method should extract more information considering the following other context properties:

• LayoutContext.DOMAIN_MODEL – The domain model element linked to the current diagram part.
• LayoutContext.CONTAINER_DIAGRAM_PART – The diagram part that corresponds to the graph or subgraph that contains the current diagram part. This is called the container. If The current diagram part is already the top-level element of the diagram, then there is no container and null should be returned.
• LayoutContext.CONTAINER_DOMAIN_MODEL – The domain model element linked to the container.
• LayoutContext.OPT_TARGETS – A set containing the kind of graph element that corresponds to the current diagram part, referenced with the enumeration LayoutOptionData.Target. If the diagram part is a node, for example, the set should contain the value NODES. If the node is also a container for a subgraph, the set should additionally contain the value PARENTS.
• DefaultLayoutConfig.HAS_PORTS – If the current diagram part is a node, the returned value for this property should be true or false depending on whether the node has any ports or not. Ports are explicit connection points for edges; they occur frequently in data flow diagrams.
• DefaultLayoutConfig.CONTENT_HINT – If the diagram contains an annotation about which layout algorithm to use for the content of the current diagram part, the returned value for this property should be the identifier of that algorithm. This is the same kind of annotation that is accessed through getOptionValue(…), i.e. a value set by the user with the Layout View.
• DefaultLayoutConfig.CONTAINER_HINT – The same as for CONTENT_HINT, but referring to the container.
• EclipseLayoutConfig.EDITING_DOMAIN – If your diagram editor needs an EMF editing domain in order to modify annotations of layout options, then such an editing domain should be returned for this property.

An instance of your self-made configurator should be returned by the getDiagramConfig() method of your diagram layout manager.