Parser

Gem Version Build Status Code Climate Coverage Status

Parser is a production-ready Ruby parser written in pure Ruby. It performs on par or better than Ripper, Melbourne, JRubyParser or ruby_parser.

You can also use unparser to produce equivalent source code from Parser’s ASTs.

Sponsored by Evil Martians.

Installation

$ gem install parser

Usage

Parse a chunk of code:

require 'parser/current'

p Parser::CurrentRuby.parse("2 + 2")
# (send
#   (int 2) :+
#   (int 2))

Access the AST’s source map:

p Parser::CurrentRuby.parse("2 + 2").loc
# #<Parser::Source::Map::Send:0x007fe5a1ac2388
#   @dot=nil,
#   @begin=nil,
#   @end=nil,
#   @selector=#<Source::Range (string) 2...3>,
#   @expression=#<Source::Range (string) 0...5>>

p Parser::CurrentRuby.parse("2 + 2").loc.selector.source
# "+"

Traverse the AST: see the documentation for gem ast.

Parse a chunk of code and display all diagnostics:

parser = Parser::CurrentRuby.new
parser.diagnostics.consumer = lambda do |diag|
  puts diag.render
end

buffer = Parser::Source::Buffer.new('(string)')
buffer.source = "foo *bar"

p parser.parse(buffer)
# (string):1:5: warning: `*' interpreted as argument prefix
# foo *bar
#     ^
# (send nil :foo
#   (splat
#     (send nil :bar)))

If you reuse the same parser object for multiple #parse runs, you need to #reset it.

You can also use the ruby-parse utility (it’s bundled with the gem) to play with Parser:

$ ruby-parse -L -e "2+2"
(send
  (int 2) :+
  (int 2))
2+2
 ~ selector
~~~ expression
(int 2)
2+2
~ expression
(int 2)
2+2

$ ruby-parse -E -e "2+2"
2+2
^ tINTEGER 2                                    expr_end     [0 <= cond] [0 <= cmdarg]
2+2
 ^ tPLUS "+"                                    expr_beg     [0 <= cond] [0 <= cmdarg]
2+2
  ^ tINTEGER 2                                  expr_end     [0 <= cond] [0 <= cmdarg]
2+2
  ^ false "$eof"                                expr_end     [0 <= cond] [0 <= cmdarg]
(send
  (int 2) :+
  (int 2))

Features

Documentation

Documentation for parser is available online.

Node names

Several Parser nodes seem to be confusing enough to warrant a dedicated README section.

(block)

The (block) node passes a Ruby block, that is, a closure, to a method call represented by its first child, a (send), (super) or (zsuper) node. To demonstrate:

$ ruby-parse -e 'foo { |x| x + 2 }' (block (send nil :foo) (args (arg :x)) (send (lvar :x) :+ (int 2)))

(begin) and (kwbegin)

TL;DR: Unless you perform rewriting, treat (begin) and (kwbegin) as the same node type.

Both (begin) and (kwbegin) nodes represent compound statements, that is, several expressions which are executed sequentally and the value of the last one is the value of entire compound statement. They may take several forms in the source code:

and so on.

$ ruby-parse -e '(foo; bar)' (begin (send nil :foo) (send nil :bar)) $ ruby-parse -e 'def x; foo; bar end' (def :x (args) (begin (send nil :foo) (send nil :bar)))

Note that, despite its name, kwbegin node only has tangential relation to the begin keyword. Normally, Parser AST is semantic, that is, if two constructs look differently but behave identically, they get parsed to the same node. However, there exists a peculiar construct called post-loop in Ruby:

begin body end while condition

This specific syntactic construct, that is, keyword begin..end block followed by a postfix while, behaves very unlike other similar constructs, e.g. (body) while condition. While the body itself is wrapped into a while-post node, Parser also supports rewriting, and in that context it is important to not accidentally convert one kind of loop into another.

$ ruby-parse -e 'begin foo end while cond' (while-post (send nil :cond) (kwbegin (send nil :foo))) $ ruby-parse -e 'foo while cond' (while (send nil :cond) (send nil :foo)) $ ruby-parse -e '(foo) while cond' (while (send nil :cond) (begin (send nil :foo)))

(Parser also needs the (kwbegin) node type internally, and it is highly problematic to map it back to (begin).)

Compatibility with Ruby MRI

Unfortunately, Ruby MRI often changes syntax in patchlevel versions. This has happened, at least, for every release since 1.9; for example, commits c5013452 and 04bb9d6b were backported all the way from HEAD to 1.9. Moreover, there is no simple way to track these changes.

This policy makes it all but impossible to make Parser precisely compatible with the Ruby MRI parser. Indeed, at September 2014, it would be necessary to maintain and update ten different parsers together with their lexer quirks in order to be able to emulate any given released Ruby MRI version.

As a result, Parser chooses a different path: the parser/rubyXY parsers recognize the syntax of the latest minor version of Ruby MRI X.Y at the time of the gem release.

Known issues

Adding support for the following Ruby MRI features in Parser would needlessly complicate it, and as they all are very specific and rarely occuring corner cases, this is not done.

Parser has been extensively tested; in particular, it parses almost entire Rubygems corpus. For every issue, a breakdown of affected gems is offered.

Void value expressions

Ruby MRI prohibits so-called “void value expressions”. For a description of what a void value expression is, see this gist and this Parser issue.

It is unknown whether any gems are affected by this issue.

Invalid characters inside comments

Ruby MRI permits arbitrary non-7-bit characters to appear in comments regardless of source encoding.

As of 2013-07-25, there are about 180 affected gems.

\u escape in 1.8 mode

Ruby MRI 1.8 permits to specify a bare \u escape sequence in a string; it treats it like u. Ruby MRI 1.9 and later treat \u as a prefix for Unicode escape sequence and do not allow it to appear bare. Parser follows 1.9+ behavior.

As of 2013-07-25, affected gems are: activerdf, activerdf_net7, fastreader, gkellog-reddy.

Invalid Unicode escape sequences

Ruby MRI 1.9+ permits to specify invalid Unicode codepoints in Unicode escape sequences, such as \u{d800}.

As of 2013-07-25, affected gems are: aws_cloud_search.

Dollar-dash

(This one is so obscure I couldn’t even think of a saner name for this issue.) Pre-2.1 Ruby allows to specify a global variable named $-. Ruby 2.1 and later treat it as a syntax error. Parser follows 2.1 behavior.

No known code is affected by this issue.

Contributors

Acknowledgements

The lexer testsuite is derived from ruby_parser.

The Bison parser rules are derived from Ruby MRI parse.y.

Contributing

  1. Make sure you have Ragel ~> 6.7 installed
  2. Fork it
  3. Create your feature branch (git checkout -b my-new-feature)
  4. Commit your changes (git commit -am 'Add some feature')
  5. Push to the branch (git push origin my-new-feature)
  6. Create new Pull Request