Hello Racc

30 May 2007

When I said HelloCup I was looking at a yacc based parser in a language that didn't require me to handle my dirty pointers. Another alternative to play with is Ruby which now has a yaccish parser built in to the standard library - inevitably called racc.

Racc has an interesting interplay between ruby and grammar syntax. You define the grammar with a racc file which will generate a parser class.

Again I'll do my simple hello world case. The input text is

item camera
item laser

I'll populate item objects inside a catalog, using the following model classes.

class Item
  attr_reader :name
  def initialize name
    @name = name
  end
end

class Catalog 
  extend Forwardable
  def initialize
    @items = []
  end
  def_delegators :@items, :size, :<<, :[] 
end

Forwardable is a handy library that allows me to delegate methods to an instance variable. In this case I delegate a bunch of methods to the @items list.

I test what I read with this.

class Tester < Test::Unit::TestCase
  def testReadTwo
    parser = ItemParser.new
    parser.parse "item camera\nitem laser\n"
    assert_equal 2, parser.result.size
    assert_equal 'camera', parser.result[0].name
    assert_equal 'laser', parser.result[1].name
  end
  def testReadBad
    parser = ItemParser.new
    parser.parse "xitem camera"
    fail
    rescue #expected
  end   
end

To build the file and run the tests I use a simple rake file.

# rakefile...
task :default => :test

file 'item.tab.rb' => 'item.y.rb' do
  sh 'racc item.y.rb'
end

task :test => 'item.tab.rb' do 
  require 'rake/runtest'
  Rake.run_tests 'test.rb'
end

The racc command needs to be installed on your system. I did it the easy way on Ubuntu with apt-get. It takes the input file and creates one named inputFileName.tab.rb.

The parser grammar class is a special format, but one that's pretty familiar to yaccish people. For this simple example it looks like this:

#file item.y.rb...
class ItemParser
  token 'item'  WORD
  rule
    catalog: item | item catalog;
    item: 'item' WORD {@result << Item.new(val[1])};
end

The tokens clause declares the token's we get from the lexer. I use the string 'item' and WORD as a symbol. The rule clause starts the production rules which are in the usual BNF form for yacc. As you might expect I can write actions inside curlies. To refer to the elements of the rule I use the val array, so val[1] is the equivalent to $2 in yacc (ruby uses 0 based array indexes, but I've forgiven it). Should I wish to return a value from the rule (equivalent to yacc's $$) I assign it to the variable result.

The most complicated part of using racc is to sort out the lexer. Racc expects to call a method that yields tokens, where each token is a two-element array with the first element being the type of token (matching the token declaration) and the second element the value (what shows up in val - usually the text). You mark the end of the token stream with [false, false]. The sample code with racc uses regular expression matching on a string. A better choice for most cases is to use StringScanner, which is in the standard ruby library.

I can use this scanner to convert a string into an array of tokens.

#file item.y.rb....
---- inner
def make_tokens str
  require 'strscan'
  result = []
  scanner = StringScanner.new str
  until scanner.empty?
    case
      when scanner.scan(/\s+/)
        #ignore whitespace
      when match = scanner.scan(/item/)
        result << ['item', nil]
      when match = scanner.scan(/\w+/)
        result << [:WORD, match]
      else
        raise "can't recognize  <#{scanner.peek(5)}>"
    end
  end
  result << [false, false]
  return result
end

To integrate the scanner into the parser, racc allows you to place code into the generated parser class. You do this by adding code to the grammar file. The declaration ---- inner marks the code to go inside the generated class (you can also put code at the head and foot of the generated file). I'm calling a parse method in my test, so I need to implement that.

#file item.y.rb....
---- inner
attr_accessor :result

def parse(str)
  @result = Catalog.new
  @tokens = make_tokens str
  do_parse
end

The do_parse method initiates the generated parser. This will call next_token to get at the next token, so we need to implement that method and include it in the inner section.

#file item.y.rb....
---- inner
def next_token
  @tokens.shift
end

This is enough to make racc work with the file. However as I play with it I find the scanner more messy than I would like. I really just want it to tell the lexer what patterns to match and what to return with them. Something like this.

#file item.y.rb....
---- inner
def make_lexer aString
  result = Lexer.new
  result.ignore /\s+/
  result.keyword 'item'
  result.token /\w+/, :WORD
  result.start aString
  return result
end

To make this work I write my own lexer wrapper over the base functionality provided by StringScanner. Here's the code to set up the lexer and and handle the above configuration.

class Lexer...
  require 'strscan'
  def initialize 
    @rules = []
  end
  def ignore pattern
    @rules << [pattern, :SKIP]
  end
  def token pattern, token
    @rules << [pattern, token]
  end
  def keyword aString
    @rules << [Regexp.new(aString), aString]
  end
  def start aString
    @base = StringScanner.new aString
  end

To perform the scan I need to use StringScanner to compare the rules against the input stream.

class Lexer...
  def next_token
    return [false, false] if @base.empty?
    t = get_token
    return (:SKIP == t[0]) ? next_token : t
  end
  def get_token
    @rules.each do |key, value|
      m = @base.scan(key)
      return [value, m] if m
    end 
    raise  "unexpected characters  <#{@base.peek(5)}>"
  end  

I can then alter the code in the parser to call this lexer instead.

#file item.y.rb....
---- inner
def parse(arg)
  @result = Catalog.new
  @lexer = make_lexer arg
  do_parse
end

def next_token
  @lexer.next_token
end

As well as giving me a better way to define the rules, this also allows the grammar to control the lexer because it's only grabbing one token at a time - this would give me a mechanism to implement lexical states later on.

On the whole racc is pretty easy to set up and use - providing you know yacc. The documentation is on the minimal side of sketchy. There's a simple manual on the website and some sample code. There's also a very helpful presentation on racc. I also got a few tips from our Mingle team who've used it for a nifty customization language inside Mingle.