Ruby_parser 3.4.0 Released

Ruby
#1

ruby_parser version 3.4.0 has been released!

ruby_parser (RP) is a ruby parser written in pure ruby (utilizing
racc–which does by default use a C extension). RP’s output is
the same as ParseTree’s output: s-expressions using ruby’s arrays and
base types.

As an example:

def conditional1 arg1
  return 1 if arg1 == 0
  return 0
end

becomes:

s(:defn, :conditional1, s(:args, :arg1),
  s(:if,
    s(:call, s(:lvar, :arg1), :==, s(:lit, 0)),
    s(:return, s(:lit, 1)),
    nil),
  s(:return, s(:lit, 0)))

Tested against 801,039 files from the latest of all rubygems (as of
2013-05):

  • 1.8 parser is at 99.9739% accuracy, 3.651 sigma
  • 1.9 parser is at 99.9940% accuracy, 4.013 sigma
  • 2.0 parser is at 99.9939% accuracy, 4.008 sigma

Changes:

3.4.0 / 2014-02-04

  • 1 major enhancement:

    • Replaced hand-written/optimized f’d-up lexer with an oedipus_lex
      generated lexer. This makes it roughly 40-50% faster.

  • 30 minor enhancements:

    • 2.0: Added support for a.b c() do d end.e do |f| g end
    • 2.0: Added support for a.b c() do d end.e f do |g| h end
    • Added -s flag to ruby_parse_extract_error to output timings.
    • Added RubyLexer #command_state and #last_state to deal with
      oedipus_lex differences.
    • Added String#lineno and #lineno= because I’m a bad bad person.
    • Added a bunch of RubyLexer scanning methods: beginning_of_line?,
      check, scan, etc.
    • Added a bunch of process_* methods extracted from old yylex.
      process_amper, etc.
    • Added lib/.document to save my laptop’s battery from pain and
      suffering
    • Adjust lineno when we lex a bunch of blank lines.
    • Attach lineno to tIDENTIFIER values (strings, ugh)
    • Cleaned up and re-ordered node_assign to be faster (ordered by
      actual occurrance).
    • Extend RubyParserStuff#gettable to set the lineno if it comes in
      with the id.
    • Extended RubyParserStuff#new_case to take line number.
    • Finally dropped RPStringScanner’s BS #current_line.
    • Finally dropped RPStringScanner’s BS line number calculation
      (lineno).
    • Implemented Sexp#add_all since we now have a test case for it.
    • Removed :call case of node_assign. I don’t think it is possible.
    • Removed RubyLexer #extra_lines_added. No longer used. Complex
      heredoc lineno’s possible screwed up.
    • Removed RubyLexer#parse_number. Handled by oedipus_lex.
    • Removed RubyLexer#yacc_value now that next_token returns pairs.
    • Removed RubyLexer’s @src. Now taken care of by oedipus_lex.
    • Removed RubyParser#advance. RubyParser#next_token takes care of
      everything now.
    • Removed RubyParserExtras#arg_add. (presidentbeef! YAY!)
    • Removed lib/gauntlet_rubyparser.rb. I just don’t use it anymore. Too
      slow.
    • RubyLexer#is_label_possible? doesn’t need an arg
    • RubyLexer#process_token is now a normal oedipal lexer method.
    • RubyParser#next_token now expects RubyLexer#next_token to return a
      pair (type, val).
    • TRYING a new scheme to figure out encodings… but I’m about to
      throw in the towel. I hate this stuff so much.
    • Turned off oedipus_lex’s automatic line counting. (pushing to
      oedipus_lex soon).
    • Updated to oedipus_lex 2.1+.

  • 7 bug fixes:

    • 1.8: Properly parse a (:b, :c, :d => :e). (presidentbeef)
    • Fixed lexing symbol!= vs symbol!. Please use your spacebar. Think of
      the children.
    • Fixed line for dstr spanning multiple lines via backslash.
      (presidentbeef)
    • Fixed line numbers for odd cases with trailing whitespace.
      (presidentbeef)
    • Fixed line numbers on ambiguous calls w/ gvar/ivar args.
      (presidentbeef)
    • Max out unicode hex values to 2-4 or 2-6 chars or pack will overflow
      and puke.
    • Removed ESC_RE from RubyLexer. Must have slipped through.