Content assistant of Axiom Query language

Content assistant with the help of the ANTLR4 framework for code analysis and processing provides semantic validation and code completion for midpoint query language. Semantic analysis of the language is performed after syntax analysis. For semantic analysis, a visitor (AxiomQueryContentAssistantVisitor) was created, which performs semantic validation of the input code fragment and processes the cursor position by traversing the abstract syntax tree (AST). The semantics of the language is partially implemented in the visitor. While traversing the AST, values for code completion in the place of the cursor position are collected. The described solution concept is shown in the diagram and individual aspects of the content assistant are described below. The diagram effectively demonstrates the architecture of content assistant.

The Visitor Pattern is a behavioral design pattern used to separate algorithms from the objects they operate on. In the context of ANTLR4, it allows you to traverse a parse tree and perform actions depending on the type of node encountered. This is particularly useful for interpreting or evaluating expressions, performing semantic validation, or translating DSLs into executable code.

Visitor AxiomQueryContentAssistantVisitor implementations semantics validation of Axiom Query language.Validation takes place by traversing the analysis tree, and at each node a check of the semantic rules of the input query is performed. This control of individual nodes of tree is implemented in the corresponding visit() methods.

The definition of PRISM is very important for semantics validation of axiom query language. Item definition can be updated several times during tree traversal. The initial definition is the input parameter rootItemDefinition of the constructor AxiomQueryContentAssistantVisitor, which is provided by the schemaContext annotation from schema. Previous definitions will also be needed to validation some concepts of language, therefore it cannot overwrite the definition, because of that it’s necessary to save each changed definition. Definitions do save to table hash itemDefinitions of type HashMap<ParseTree, Definition> in the subchapter below, is shown the mapping of the item definition based on the abstract syntactic tree (AST).

Unsuccessful validation of the semantic rule is recorded in the errorList, which is later provided to the visitor as a result of the semantic validation

Code completion is the second service provided by the content assistant for Axiom Query Language. There are two ways to solve code completion with antlr4.

Correct localization of the cursor is an essential part of code completions. The input parameter postiion curtor specifies the number of characters after which code completions are called. Based on the number of the cursor position, a node is searched for in the analysis tree, which contains a branch that exactly corresponds to the cursor position. This node is represented with a PositonContext data structure. PositionContex contains two parameters first node from parse tree and index which present branch in the node. Invoke state is obtained based on the position context. From Invoke state stated searching expected tokens in ATN network. During the transition, the rules are also recorded, which later correspond to the found token.

Expected tokens are searched for in 2 phases, in the first phase, superior language concepts such as root and filter are processed, and next, low level language concepts such as itemFILter are processed. It is not correct to use AST and nice functions that ANTLR4 provides us for working with the tree because we have to rely on the position of the cursor and not that the code ends at the position of the cursor. Because we need the context of the left and right side of the cursor for correctly searching for the definition in the scheme. The definitions are used from the table hash itemDefifnition after semantic validation, but in some cases it is necessary to find a better definition with better of the context from the right side of the code from the cursor.

In case the processed code is incomplete or incorrect from the point of view of syntax, we need to obtain an analysis tree. In this case, challenge the default antlr error strategy, which continues the analysis despite syntax errors. Then we have access to the analysis tree and can identify the node in which the syntax error is located.

After all epected tokens have been found, tokens are modified with semantic rules in the generateSugestion method. Semantic rules must also be taken into account in searching the ATN network. Then the method returns a generated suggestion of code completions from cursor position.

Content assistant is currently integrated in the plugin of Intellij IDEA Midpoint studio and Midpoint web application (search panel and query playground).

Semantic validation runs automatically and provides quick responses with underlined code and error messages. Validation return list of semantic and sytantic errors together.