Oedipus Lex - This is not your father’s lexer

home

github.com/seattlerb/oedipus_lex

rdoc

docs.seattlerb.org/oedipus_lex

DESCRIPTION

Oedipus Lex is a lexer generator in the same family as Rexical and Rex. Oedipus Lex is my independent lexer fork of Rexical. Rexical was in turn a fork of Rex. We’ve been unable to contact the author of rex in order to take it over, fix it up, extend it, and relicense it to MIT. So, Oedipus was written clean-room in order to bypass licensing constraints (and because bootstrapping is fun).

Oedipus brings a lot of extras to the table and at this point is only historically related to rexical. The syntax has changed enough that any rexical lexer will have to be tweaked to work inside of oedipus. At the very least, you need to add slashes to all your regexps.

Oedipus, like rexical, is based primarily on generating code much like you would a hand-written lexer. It is not a table or hash driven lexer. It uses StrScanner within a multi-level case statement. As such, Oedipus matches on the first match, not the longest (like lex and its ilk).

This documentation is not meant to bypass any prerequisite knowledge on lexing or parsing. If you’d like to study the subject in further detail, please try [TIN321] or the [LLVM Tutorial] or some other good resource for CS learning. Books… books are good. I like books.

Syntax:

lexer           = (misc_line)*
                  /class/ class_id
                  (option_section)?
                  (inner_section)?
                  (start_section)?
                  (macro_section)?
                  (rule_section)?
                  /end/
                  (misc_line)*

misc_line       = /.*/

class_id        = /\w+.*/

option_section  = /options?/ NL (option)*
option          = /stub/i
                | /debug/i
                | /do_parse/i
                | /lineno/i
                | /column/i

inner_section   = /inner/ NL (misc_line)*

start_section   = /start/ NL (misc_line)*

macro_section   = /macros?/ NL (macro)*
macro           = name regexp
name            = /\w+/
regexp          = /(\/(?:\\.|[^\/])+\/[io]?)/

rule_section    = /rules?/ NL (rule|group)*
rule            = (state)? regexp (action)?
group           = /:/ regexp NL (rule)+
state           = label
                | predicate
label           = /:\w+/
predicate       = /\w+\?/
action          = name
                | /\{.*\}.*/

Basic Example

class Calculator
macros
  NUMBER /\d+/
rules
       /rpn/       :RPN # sets @state to :RPN
       /#{NUMBER}/ { [:number, text.to_i] }
       /\s+/
       /[+-]/      { [:op, text] }

  :RPN /\s+/
  :RPN /[+-]/      { [:op2, text] }
  :RPN /#{NUMBER}/ { [:number2, text.to_i] }
  :RPN /alg/       nil # clears state
end

Header

Anything before the class line is considered the “header” and will be added to the top of your file. This includes extra lines like module namespacing.

Class Line

The class line, like a regular ruby class declaration, specifies what class all of the lexer code belongs to. You may simply specify a class name like:

class MyLexer

or it may specify a superclass as well:

class MyLexer < MyParser

You might do this latter case to mix your lexer and your racc parser together.

Personally, I recommend keeping them apart for cleanliness and testability.

Options

All options are opt-in and can be specified either in the grammar or via an options hash in ‘OedipusLex#initialize`.

Specify ‘debug` to turn on basic tracing output.

Specify ‘stub` to create a generic handler that processes all files specified on the commandline with a rather generic handler. This makes it easy to get up and running before you have the rest of your system in place.

Specify ‘do_parse` to generate a generic do_parse method that automatically dispatches off to `lex_<token_type>` methods.

Specify ‘lineno` to generate automatic line number handling at the beginning of `next_token`. This was the default in 1.0.0 and you must now activate it.

Specify ‘column` to generate automatic column number handling.

Inner

The inner section is just code, like header or footer, but inner gets put inside the class body. You can put extra methods here.

Personally, I recommend you don’t use inner and you put all of your extra methods and class code in a separate file. This makes lexer generation faster and keeps things separate and small.

Macros

Macros define named regexps that you can use via interpolation inside other subsequent macros or within rule matchers.

Start

The lexer runs in a loop until it finds a match or has to bail. Use the ‘start` section to place extra code at the top of your `next_token` method, before the loop. Eg:

start
  space_seen = false

This code will get expanded into the very top of the lexer method. Do note that this code gets run before _every token_, not just on lexer initialization.

Rules

The rule section is the meat of the lexer. It contains one or more rule lines where each line consists of:

  • a required state (as a ‘:symbol`), a predicate method, or nothing.

  • a regular expression.

  • an action method, an action block, or nothing.

More often than not, a rule should not specify a required state. Only use them when you’re convinced you need them.

So a rule can very simple, including just a regexp:

rules
  /#.*/  # ignore comments

or can contain any combination of state checks or action types:

rules
  :state     /token/    action_method
  predicate? /another/  { do_something }
States and Predicates

In order for the tokenizer to determine if the rule’s regexp should even be considered, a rule may specify a required state, a predicate method to call, or leave it blank.

If the rule does not specify a state, it can be used whenever ‘@state` is nil or a symbol that starts lowercase (an inclusive rule). If the rule specifies a symbol that starts uppercase (an exclusive rule), it will only use those rules when `@state` matches.

Alternatively, a rule may specify a predicate method to check. If that method returns a truthy value, the rule is currently valid. This is equivalent to setting the required state to nil, as it will be used with inclusive and nil states, and ignored for exclusive states.

End & Footer

Like the header, anything after the end line is considered the “footer” and will be added to the bottom of your file.

Suggested Structure

Here’s how I suggest you structure things:

Rakefile

You only need a minimum of dependencies to wire stuff up if you use the supplied rake rule.

Rake.application.rake_require "oedipus_lex"

task :lexer  => "lib/mylexer.rex.rb"
task :parser => :lexer # plus appropriate parser rules/deps
task :test   => :parser

lib/mylexer.rex

Put your lexer definition here. It will generate into ‘“lib/mylexer.rex.rb”`.

class MyLexer
macros
  # ...
rules
  # ...
end

lib/mylexer.rb

require "new_ruby_lexer.rex"

class MyLexer
  # ... predicate methods and stuff
end

lib/myparser.rb

Assuming you’re using a racc based parser, you’ll need to define a ‘next_token` method that bridges over to your lexer:

class MyParser
  def next_token
    lexer.next_token # plus any sanity checking / error handling...
  end
end

Differences with Rexical

If you’re already familiar with rexical, this might help you get up and running faster. If not, it could provide an overview of the value-added.

Additions or Changes

A generic rake rule is defined for rex files.

Oedipus defines a rake rule that allows you simply define a file-based dependency and rake will take care of the rest. Eg:

file "lib/mylexer.rex.rb" => "lib/mylexer.rex"

task :generated => %w[lib/mylexer.rex.rb]

task :test => :generated

All regular expressions must be slash delimited.

Basically, regexps are now plain slashed ruby regexps. This allows for regexp flags to be provided individually, rather than specifying an entire grammar is case-insensitive, you can have a single rule be case insensitive.

Right now only ‘/i` and `/o` are properly handled.

Regular expressions now use ruby interpolation.

Instead of ‘aaa{macro}ccc` it is `/aaa#macroccc/`.

Macros define class constants.

Macros simply become class constants inside the lexer class. This makes them immediately available to other macros and to the regexps in the rules section.

This also implies that they must start uppercase, since that is required by ruby.

Rules can be activated by predicate methods.

Instead of just switching on state, rules can now check predicate methods to see if they should trigger. Eg:

rules
  sad?   /\w+/    { [:sad,   text] }
  happy? /\w+/    { [:happy, text] }
end

# elsewhere:
def sad?
  # ...
end

def happy?
  not sad?
end

Rule actions are only a single-line.

In order to push complexity down, ‘{ rule actions }` may only be a single line.

Rules can invoke methods.

For more complex actions, use a method by specifying its name:

rules
  /\w+/    process_word
end

And then define the handler method to return a result pair:

def process_word text
  # do lots of normalization...
  [:word, token]
end

This strikes a good balance between readability and maintainability. It also makes it much easier to write unit tests for the complex actions.

Rules can define state.

There are shortcuts built in to define or clear state:

rules
       /rpn/  :RPN  # sets @state to :RPN
  # ...
  :RPN /alg/   nil  # clears @state

Use a ‘start` section to define pre-lex code.

The lexer runs in a loop until it finds a match or has to bail. Sometimes more complex lexers need to set some local state. You can now do this in a ‘start` section. Eg:

start
  space_seen = false

This code will get expanded into the very top of the lexer method. Do note that this code gets run before _every token_, not just on initialization.

Rule state can be inclusive or exclusive.

This actually isn’t new from rexical… It just wasn’t really well documented.

Exclusive states start with an uppercase letter (and are generally all uppercase). Inclusive states start with a lowercase letter. Exclusive states will only try their own matchers. Inclusive states will also try any matcher w/o a state.

In both cases, the order of generated matchers is strictly defined by the source file. Nothing is re-ordered, ever. Eg:

rules
        /\d+/
        /\s+/  # used in both nil-state and :rpn state
        /[+-]/

  :rpn  /\d+/  # won't hit, because of nil-state matcher above

  :OP   /\s+/  # must define its own because no-nil-state matchers are used
  :OP   /\d+/
end

Default ‘do_parse` will dispatch to lex_xxx automatically.

The method ‘do_parse` is generated for you and automatically dispatches off to user-defined methods named `lex_<token-type>` where token-type is the first value returned from any matching action. Eg:

rules
  /\s*(\#.*)/  { [:comment, text] }

# elsewhere:

def lex_comment line
  # do nothing
end

‘text` is passed in, or use `match` or `matches`

You can use the ‘text` variable for the entire match inside an action, or you can use `match` to access a specific match group, or `matches` to get an array of all match groups. Eg:

/class ([\w:]+)(.*)/ { [:class, *matches] }

In this case, the action will return something like: ‘[:class, “ClassName” “< Superclass”]`.

You can override the scanner class by defining ‘scanner_class`.

Oedipus will define the method ‘scanner_class` to return `StringScanner` unless you define one yourself. Because it uses reflection to figure out whether you’ve defined it or not, you may need to require the generated lexer AFTER you’ve defined ‘scanner_class`. Eg:

class MyLexer
  # ...

  def scanner_class
    CustomStringScanner
  end

  # ...
end

require "my_lexer.rex"

NOTE: I’m totally open to better ways of doing this. I simply needed to get stuff done and this presented itself as viable-enough.

Removals

There is no command-line tool.

There is no command-line tool. Instead, use the rake rule described above.

There are only two options: debug and stub.

All other options from rexical have been removed because they don’t make sense in Oedipus.

Probably others…

It’s hard to think about what I took out. What I added is plain as day. :P

[TIN321]: www.cse.chalmers.se/edu/year/2011/course/TIN321/lectures/proglang-04.html [LLVM Tutorial]: llvm.org/docs/tutorial/LangImpl1.html#language

Requirements:

  • ruby version 1.8.x or later.

Install

  • sudo gem install oedipus_lex

License

(The MIT License)

Copyright © Ryan Davis, seattle.rb

Permission is hereby granted, free of charge, to any person obtaining a copy of this software and associated documentation files (the ‘Software’), to deal in the Software without restriction, including without limitation the rights to use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies of the Software, and to permit persons to whom the Software is furnished to do so, subject to the following conditions:

The above copyright notice and this permission notice shall be included in all copies or substantial portions of the Software.

THE SOFTWARE IS PROVIDED ‘AS IS’, WITHOUT WARRANTY OF ANY KIND, EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.