Kaleidoscope Parse Tree Visitors

The LLVM.NET Kaleidoscope tutorials all use ANTLR4 to parse the Kaleidoscope language When ANTLR processes the language grammar description to generate the lexer and parser it also generates a base "visitor" class for applying the Visitor pattern to the parse tree. This is the primary mechanism for generating the AST for these examples. The AST transformation classes derive from the ANTLR generated base visitor class.

There are only a few top level rules in the grammar to consider (although each has many sub rules).

repl
    : DEF prototype expression[0] # FunctionDefinition
    | EXTERN prototype            # ExternalDeclaration
    | expression[0]               # TopLevelExpression
    | SEMICOLON                   # TopLevelSemicolon
    ;

The parse tree consist of a tree of nodes generated for each rule in the grammar. The rule classes are generated at build time when the antlr grammar file is parsed. The C# target for ANTLR4 will generate the rule classes with the partial keyword, allowing the application code to add application specific support to the rule classes, and thus to the parse tree. For Kaleidoscope, this is used to provide simpler access to the information parsed from the Kaleidoscope language input simplifying generation of the AST. For simplicity and clarity of understanding each of the extended partial classes are placed into their own source file.

Generally speaking, the use of ANTLR and the ParseTree is a hidden internal implementation detail of the Llvm.NET Kaleidoscope tutorials. The actual code generation deals only with the AST so the parsing could be done with some other technology. (Though with all the functionality that ANTLR4 provides it would take a strong argument to justify something else)