Xtext hint: Content assist for multiple consecutive keywords

Keywords play a central role in Xtext DSLs to define the languages general structure. In many cases a single keyword is sufficient to specify what is expected next. Although Xtext generally supports multiple consecutive keywords the default content assist implementation will propose one keyword after another instead of concatenating consecutive keywords. This can lead to confusing proposals for the user. This article will help you to prevent this confusion.

After demonstrating the problem with a simple DSL it will be shown that slighty changing the grammar generates the infrastructure to offer more valuable proposals. The source code for this example is available in our Github repository. Further examples for cases in which a single keyword is sufficient to specify what is expected next can be found in the blog articals regarding controlled natural languages described by Christoph Knauf.

Sample Grammar

As an example we will use the well-known domainmodel example DSL that is shipped with the Xtext SDK, extended by the new concept Relation. A Relation comes with its own semantics that is represented by a sequence of keywords. While depends on carries the semantics that one entity is relying on the existence of a certain other entity, is composed of defines a composition dependency between two entities. Since putting each keyword in its own quote is considered as good grammar style the keyword sequences are separated.

grammar org.eclipse.xtext.example.domainmodel.Domainmodel with org.eclipse.xtext.xbase.Xbase

generate domainmodel "http://www.xtext.org/example/Domainmodel"


    'entity' name=ValidID ('extends' superType=JvmParameterizedTypeReference)? '{'

    ('depends' 'on' | 'is' 'composed' 'of') referencedEntity=[Entity]

name=ValidID ':' type=JvmTypeReference;

As an example we want to define an entity Person and an entity Club where the entity Club is composed of Persons. Within the entity Club the proposal provider is invoked leading to the following suggestions:

Default Proposal Provider

The figure above shows the proposals based on the default implementation of the content assist that offers each first keyword of the Relation rule. As a consequence each single keyword of the sequence will be proposed by an additional hit of CTRL + Space. Especially the beginning of the second keyword sequence is gives no information about the whole keyword sequence making the proposal nearly unusable. To improve the suggestion of multiple consecutive keywords the grammar and the proposal provider will be slightly adjusted.

Grammar Adjustments

To enable the proposal provider to make more valuable suggestions the keyword sequences are moved to their own parser rule that is than referenced from the Relation rule.

    (DependsOn | IsComposedOf) referencedEntity=[Entity]

    'depends' 'on'

    'is' 'composed' 'of'

Although, the change makes no difference to the grammar itself it causes valuable changes in the generated language framework and especially in the proposal provider. Each of the keyword sequences now has its own complete_ method in the AbstractDomainmodelProposalProvider.

    public void complete_DependsOn(EObject model, RuleCall ruleCall, ContentAssistContext context, ICompletionProposalAcceptor acceptor) {
        // subclasses may override
    public void complete_IsComposedOf(EObject model, RuleCall ruleCall, ContentAssistContext context, ICompletionProposalAcceptor acceptor) {
        // subclasses may override

As the by default generated methods suggest we now implement each of these in the DomainmodelProposalProvider to return the whole sequence of keywords as a proposal. First, as preparation we inject the DomainmodelGrammarAccess as an extension to the DomainmodelProposalProvider. The GrammarAccess is used in the overwritten complete_ methods to get easy access to the keyword sequence defined by the grammar.

class DomainmodelProposalProvider extends AbstractDomainmodelProposalProvider {

    @Inject extension DomainmodelGrammarAccess

Second, the complete_ methods are overwritten to create a proposal string that contains all elements of the keyword sequence. The injected DomainmodelGrammarAccess is used to get the group of keywords that is than passed to the createKeywordProposal method that does the real magic.

override complete_DependsOn(EObject model, RuleCall ruleCall, ContentAssistContext context, ICompletionProposalAcceptor acceptor) {
override complete_IsComposedOf(EObject model, RuleCall ruleCall, ContentAssistContext context, ICompletionProposalAcceptor acceptor) {

The following code snippet shows how a coherent keyword sequence is computed and proposed. After checking that the passed in group is not null the proposal string is concatenated. First, the list of elements from the group are retrieved and filtered for instances of org.eclipse.xtext.Keyword. Second, the values of each Keyword are fetched and concatenated using the join method with a single space as delimiter. Finally, the proposalString is turned into a completion proposal that is than handed over to the ICompletionProposalAcceptor calling the accept method.

def createKeywordProposal(Group group, ContentAssistContext context, ICompletionProposalAcceptor acceptor) {
    if (group == null) {
        return null
    val proposalString = group.elements.filter(Keyword).map[value].join(" ") + " "
    acceptor.accept(createCompletionProposal(proposalString, proposalString, null, context))

The following figure shows the newly created proposal strings that now contain the whole sequence of keywords.

Enhanced Proposal Provider


As demonstrated above the default keyword support of the Xtext language workbench is sufficient for single keywords. However, there are situations in which multiple consecutive keywords are required to define thorough and comprehensive language statements. To cope with sequences of keywords as a whole small changes to the grammar are required. In addition to solely proposing consecutive keywords the overwritten methods can also be used to filter, e.g. allow only one depends on relation per entity. All in all, the rather small changes to the grammar make the suggested consecutive keywords more valuable and in addition create new opportunities to improve the overall proposals.


Share this blog post


About The Author

Hendrik is an IT consultant at itemis AG in Bonn. He works on projects utilizing Xtext to build integrated toolchains. He is interested in model driven software development, DSLs and software architecture.