Yoga

gem "yoga"

module MyLanguage
  class Scanner
    # All of the behavior from Yoga for scanners.  This provides the
    # `match/2` method, the `call/0` method, the `match_line/1` method,
    # the `location/1` method, and the `emit/2` method.  The major ones that
    # are used are the `match/2`, the `call/0`, and the `match_line/1`
    # methods.
    include Yoga::Scanner

    # This must be implemented.  This is called for the next token.  This
    # should only return a Token, or true.
    def scan
      # Match with a string value escapes the string, then turns it into a
      # regular expression.
      match("[") || match("]") ||
      # Match with a symbol escapes the symbol, and turns it into a regular
      # expression, suffixing it with `symbol_negative_assertion`.  This is
      # to prevent issues with identifiers and keywords.
      match(:class) || match(:func) ||
      # With a regular expression, it's matched exactly.  However, a token
      # name is highly recommended.
      match(/[a-z][a-zA-Z0-9_]*[!?=]?/, :IDENT)
    end
  end
end

source = "class alpha [func a []]"
MyLanguage::Scanner.new(source).call # => #<Enumerable ...>

MyLanguage::Scanner.new(source).call.to_a # => [...]

MyLanguage::Scanner.new(source, "foo").call.first.location.to_s # => "foo:1.1-6"

module MyLanguage
  class Parser
    class Root < Yoga::Node
      # An attribute on the node.  This is required for Yoga nodes since the
      # update syntax requires them.  The type for the attribute is optional.
      attribute :statements, type: [Yoga::Node]
    end

    class Expression < Yoga::Node
    end

    class Operation < Expression
      attribute :operator, type: ::Symbol
      attribute :left, type: Expression
      attribute :right, type: Expression
    end

    class Literal < Expression
      attribute :value, type: ::Integer
    end
  end
end

module MyLanguage
  class Parser
    # This provides all of the parser helpers.  This is the same as adding
    # `Yoga::Parser::Helpers` as an include statement as well.
    include Yoga::Parser

    # Like the `scan/0` method on the scanner, this must be implemented.  This
    # is the entry point for the parser.  However, public usage should use the
    # `call/0` method.  This should return a node of some sort.
    def parse_root
      # This "collects" a series of nodes in sequence.  It iterates until it
      # reaches the `:EOF` token (in this case).  The first parameter to
      # collect is the "terminating token," and can be any value that
      # `expect/1` or `peek?/1` accepts.  The second, optional parameter to
      # collect is the "joining token," and is required between each node.
      # We're not using the semicolon as a joining token because that is
      # required for _all_ statements.  The joining token can be used for
      # things like argument lists.  The parameter can be any value that
      # `expect/1` or `peek?/1` accepts.
      children = collect(:EOF) { parse_statement }

      # "Unions" the location of all of the statements in the list.
      location = children.map(&:location).inject(:union)
      Parser::Root.new(statements: children, location: location)
    end

    # Parses a statement.  This is the same as the <statement> rule as above.
    def parse_statement
      expression = parse_expression
      # This says that the next token should be a semicolon.  If the next token
      # isn't, it throws an error with a detailed error message, denoting
      # what was expected (in this case, a semicolon), what was given, and
      # where the error was located in the source file.
      expect(:";")

      expression
    end


    # A switch statement, essentially.  This is defined beforehand to make it
    # _faster_ (not really; it's just useful).  The first parameter to the
    # switch function is the name of the switch.  This is used later to
    # actually perform the switch; it is also used to define a first set with
    # the allowed tokens for the switch.  The second parameter defines a key
    # value pair.  The keys are the tokens that are allowed; a symbol or an
    # array of symbols can be used.  The value is the block or the method that
    # is executed upon encountering that token.
    switch(:Operation,
      "=": proc { |left| parse_operation(:"=", left) },
      "+": proc { |left| parse_operation(:"+", left) },
      "-": proc { |left| parse_operation(:"-", left) },
      "*": proc { |left| parse_operation(:"*", left) },
      "/": proc { |left| parse_operation(:"/", left) },
      "^": proc { |left| parse_operation(:"^", left) },
      "%": proc { |left| parse_operation(:"%", left) })

    def parse_expression
      # Parse a literal.  All expressions must contain a literal of some sort;
      # we're just going to use a numeric literal here.
      left = parse_expression_literal

      # Whenever the `.switch` function is called, it creates a
      # "first set" that can be used like this.  The first set consists of
      # a set of tokens that are allowed for the switch statement.  In this
      # case, it just makes sure that the next token is an operator.  If it
      # is, it parses it as an operation.
      if peek?(first(:Operation))
        # Uses the switch defined below.  If a token is found as a key, its
        # block is executed; otherwise, it errors, giving a detailed error of
        # what was expected.
        switch(:Operation, left)
      else
        left
      end
    end

    def parse_operation(op, left)
      token = expect(op)
      right = parse_expression

      Parser::Operation.new(left: left, op: op, right: right, location:
        left.location | op.location | right.location)
    end

    def parse_expression_literal
      token = expect(:NUMERIC)
      Parser::Literal.new(value: token.value, location: token.location)
    end
  end
end

source = "a = 2;\nb = a + 2;\n"
scanner = MyLanguage::Scanner.new(source).call
MyLanguage::Parser.new(scanner).call # => #<MyLanguage::Parser::Root ...>