Parsing Expressions for Ruby
Ruby Vim script
Pull request Compare This branch is 153 commits behind mjackson:master.
Fetching latest commit…
Cannot retrieve the latest commit at this time.
Failed to load latest commit information.


                                  ~* Citrus *~

                          Parsing Expressions for Ruby

Citrus is a compact and powerful parsing library for
[Ruby]( that combines the elegance and expressiveness of
the language with the simplicity and power of
[parsing expressions](

# Installation

Via [RubyGems](

    $ gem install citrus

From a local copy:

    $ git clone git://
    $ cd citrus
    $ rake package install

# Background

In order to be able to use Citrus effectively, you must first understand the
difference between syntax and semantics. Syntax is a set of rules that govern
the way letters and punctuation may be used in a language. For example, English
syntax dictates that proper nouns should start with a capital letter and that
sentences should end with a period.

Semantics are the rules by which meaning may be derived in a language. For
example, as you read a book you are able to make some sense of the particular
way in which words on a page are combined to form thoughts and express ideas
because you understand what the words themselves mean and you understand what
they mean collectively.

Computers use a similar process when interpreting code. First, the code must be
parsed into recognizable symbols or tokens. These tokens may then be passed to
an interpreter which is responsible for forming actual instructions from them.

Citrus is a pure Ruby library that allows you to perform both lexical analysis
and semantic interpretation quickly and easily. Using Citrus you can write
powerful parsers that are simple to understand and easy to create and maintain.

In Citrus, there are three main types of objects: rules, grammars, and matches.

## Rules

A [Rule](api/classes/Citrus/Rule.html) is an object that specifies some matching
behavior on a string. There are two types of rules: terminals and non-terminals.
Terminals can be either Ruby strings or regular expressions that specify some
input to match. For example, a terminal created from the string "end" would
match any sequence of the characters "e", "n", and "d", in that order. Terminals
created from regular expressions may match any sequence of characters that can
be generated from that expression.

Non-terminals are rules that may contain other rules but do not themselves match
directly on the input. For example, a Repeat is a non-terminal that may contain
one other rule that will try and match a certain number of times. Several other
types of non-terminals are available that will be discussed later.

Rule objects may also have semantic information associated with them in the form
of Ruby modules. Rules use these modules to extend the matches they create.

## Grammars

A grammar is a container for rules. Usually the rules in a grammar collectively
form a complete specification for some language, or a well-defined subset

A Citrus grammar is really just a souped-up Ruby
[module]( These modules may be
included in other grammar modules in the same way that Ruby modules are normally
used. This property allows you to divide a complex grammar into more manageable,
reusable pieces that may be combined at runtime. Any grammar rule with the same
name as a rule in an included grammar may access that rule with a mechanism
similar to Ruby's super keyword.

## Matches

Matches are created by rule objects when they match on the input. A
[Match](api/classes/Citrus/Match.html) is actually a
[String]( object with some extra
information attached such as the name(s) of the rule(s) from which it was
generated and any submatches it may contain.

During a parse, matches are arranged in a tree structure where any match may
contain any number of other matches. This structure is determined by the way in
which the rule that generated each match is used in the grammar. For example, a
match that is created from a non-terminal rule that contains several other
terminals will likewise contain several matches, one for each terminal.

Match objects may be extended with semantic information in the form of methods.
These methods should provide various interpretations for the semantic value of a

# Syntax

The most straightforward way to compose a Citrus grammar is to use Citrus' own
custom grammar syntax. This syntax borrows heavily from Ruby, so it should
already be familiar to Ruby programmers.

## Terminals

Terminals may be represented by a string or a regular expression. Both follow
the same rules as Ruby string and regular expression literals.

    'abc'         # match "abc"
    "abc\n"       # match "abc\n"
    /abc/i        # match "abc" in any case
    /\xFF/        # match "\xFF"

Character classes and the dot (match anything) symbol are supported as well for
compatibility with other parsing expression implementations.

    [a-z0-9]      # match any lowercase letter or digit
    [\x00-\xFF]   # match any octet
    .             # match any single character, including new lines

Also, strings may use backticks instead of quotes to indicate that they should
match in a case-insensitive manner.

    `abc`         # match "abc" in any case

See [Terminal](api/classes/Citrus/Terminal.html) and
[StringTerminal](api/classes/Citrus/StringTerminal.html) for more information.

## Repetition

Quantifiers may be used after any expression to specify a number of times it
must match. The universal form of a quantifier is `N*M` where `N` is the minimum
and `M` is the maximum number of times the expression may match.

    'abc'1*2      # match "abc" a minimum of one, maximum of two times
    'abc'1*       # match "abc" at least once
    'abc'*2       # match "abc" a maximum of twice

Additionally, the minimum and maximum may be omitted entirely to specify that an
expression may match zero or more times.

    'abc'*        # match "abc" zero or more times

The `+` and `?` operators are supported as well for the common cases of `1*` and
`*1` respectively.

    'abc'+        # match "abc" one or more times
    'abc'?        # match "abc" zero or one time

See [Repeat](api/classes/Citrus/Repeat.html) for more information.

## Lookahead

Both positive and negative lookahead are supported in Citrus. Use the `&` and
`!` operators to indicate that an expression either should or should not match.
In neither case is any input consumed.

    &'a' 'b'      # match a "b" preceded by an "a"
    'a' !'b'      # match an "a" that is not followed by a "b"
    !'a' .        # match any character except for "a"

A special form of lookahead is also supported which will match any character
that does not match a given expression.

    ~'a'          # match all characters until an "a"
    ~/xyz/        # match all characters until /xyz/ matches

See [AndPredicate](api/classes/Citrus/AndPredicate.html),
[NotPredicate](api/classes/Citrus/NotPredicate.html), and
[ButPredicate](api/classes/Citrus/ButPredicate.html) for more information.

## Sequences

Sequences of expressions may be separated by a space to indicate that the rules
should match in that order.

    'a' 'b' 'c'   # match "a", then "b", then "c"
    'a' [0-9]     # match "a", then a numeric digit

See [Sequence](api/classes/Citrus/Sequence.html) for more information.

## Choices

Ordered choice is indicated by a vertical bar that separates two expressions.
When using choice, each expression is tried in order. When one matches, the
rule returns the match immediately without trying the remaining rules.

    'a' | 'b'       # match "a" or "b"
    'a' 'b' | 'c'   # match "a" then "b" (in sequence), or "c"

It is important to note when using ordered choice that any operator binds more
tightly than the vertical bar. A full chart of operators and their respective
levels of precedence is below.

See [Choice](api/classes/Citrus/Choice.html) for more information.

## Labels

Match objects may be referred to by a different name than the rule that
originally generated them. Labels are created by placing the label and a colon
immediately preceding any expression.

    chars:/[a-z]+/  # the characters matched by the regular expression
                    # may be referred to as "chars" in an extension
                    # method

See [Label](api/classes/Citrus/Label.html) for more information.

## Grouping

As is common in many programming languages, parentheses may be used to override
the normal binding order of operators.

    'a' ('b' | 'c')   # match "a", then "b" or "c"

## Extensions

Extensions may be specified using either "module" or "block" syntax. When using
module syntax, specify the name of a module that is used to extend match objects
in between less than and greater than symbols.

    [a-z0-9]5*9 <CouponCode>  # match a string that consists of any lower
                              # cased letter or digit between 5 and 9
                              # times and extend the match with the
                              # CouponCode module

Additionally, extensions may be specified inline using curly braces. Inside the
curly braces you may embed method definitions that will be used to extend match

    # match any digit and return its integer value when calling the
    # #value method on the match object
    [0-9] {
      def value

## Super

When including a grammar inside another, all rules in the child that have the
same name as a rule in the parent also have access to the `super` keyword to
invoke the parent rule.

See [Super](api/classes/Citrus/Super.html) for more information.

## Precedence

The following table contains a list of all Citrus symbols and operators and
their precedence. A higher precedence indicates tighter binding.

Operator  | Name                      | Precedence
--------- | ------------------------- | ----------
''        | String (single quoted)    | 6
""        | String (double quoted)    | 6
``        | String (case insensitive) | 6
[]        | Character class           | 6
.         | Dot (any character)       | 6
//        | Regular expression        | 6
()        | Grouping                  | 6
*         | Repetition (arbitrary)    | 5
+         | Repetition (one or more)  | 5
?         | Repetition (zero or one)  | 5
&         | And predicate             | 4
!         | Not predicate             | 4
~         | But predicate             | 4
:         | Label                     | 4
<>        | Extension (module name)   | 3
{}        | Extension (literal)       | 3
e1 e2     | Sequence                  | 2
e1 | e2   | Ordered choice            | 1

# Example

Below is an example of a simple grammar that is able to parse strings of
integers separated by any amount of white space and a `+` symbol.

    grammar Addition
      rule additive
        number plus (additive | number)

      rule number
        [0-9]+ space

      rule plus
        '+' space

      rule space
        [ \t]*

Several things to note about the above example:

* Grammar and rule declarations end with the `end` keyword
* A Sequence of rules is created by separating expressions with a space
* Likewise, ordered choice is represented with a vertical bar
* Parentheses may be used to override the natural binding order
* Rules may refer to other rules in their own definitions simply by using the
  other rule's name
* Any expression may be followed by a quantifier

## Interpretation

The grammar above is able to parse simple mathematical expressions such as "1+2"
and "1 + 2+3", but it does not have enough semantic information to be able to
actually interpret these expressions.

At this point, when the grammar parses a string it generates a tree of
[Match](api/classes/Citrus/Match.html) objects. Each match is created by a rule
and may itself be comprised of any number of submatches.

Submatches are created whenever a rule contains another rule. For example, in
the grammar above `number` matches a string of digits followed by white space.
Thus, a match generated by this rule will contain two submatches.

We can define methods inside a set of curly braces that will be used to extend
matches when they are created. This works in similar fashion to using Ruby's
blocks. Let's extend the `Addition` grammar using this technique.

    grammar Addition
      rule additive
        (number plus term:(additive | number)) {
          def value
            number.value + term.value

      rule number
        ([0-9]+ space) {
          def value

      rule plus
        '+' space

      rule space
        [ \t]*

In this version of the grammar we have added two semantic blocks, one each for
the additive and number rules. These blocks contain methods that will be present
on all match objects that result from matches of those particular rules. It's
easiest to explain what is going on here by starting with the lowest level
block, which is defined within the number rule.

The semantic block associated with the number rule defines one method, `value`.
Inside this method, we can see that the value of a number match is determined to
be its text value, stripped of white space and converted to an integer.
[Remember](background.html) that matches are simply strings, so the `strip`
method in this case is actually

The `additive` rule also extends its matches with a `value` method. Notice the
use of the `term` label within the rule definition. This label allows the match
that is created by either the additive or the number rule to be retrieved using
the `term` label. The value of an additive is determined to be the values of its
`number` and `term` matches added together using Ruby's addition operator.

Since additive is the first rule defined in the grammar, any match that results
from parsing a string with this grammar will have a `value` method that can be
used to recursively calculate the collective value of the entire match tree.

To give it a try, save the code for the `Addition` grammar in a file called
addition.citrus. Next, assuming you have the Citrus
[gem]( installed, try the following sequence of
commands in a terminal.

    $ irb
    > require 'citrus'
     => true
    > Citrus.load 'addition'
     => [Addition]
    > m = Addition.parse '1 + 2 + 3'
     => #<Citrus::Match ...
    > m.value
     => 6

Congratulations! You just ran your first piece of Citrus code.

One interesting thing to notice about the above sequence of commands is the
return value of [Citrus#load](api/classes/Citrus.html#M000003). When you use
`Citrus.load` to load a grammar file (and likewise
[Citrus#eval](api/classes/Citrus.html#M000004) to evaluate a raw string of
grammar code), the return value is an array of all the grammars present in that

Take a look at
for an example of a calculator that is able to parse and evaluate more complex
mathematical expressions.

## Implicit Value

It is very common for a grammar to only have one interpretation for a given
symbol. For this reason, you may find yourself writing a `value` method for
every rule in your grammar. Because this can be tedious, Citrus allows you to
omit defining such a method if you choose. For example, the `additive` and
`number` rules from the simple calculator example above could also be written

    rule additive
      (number plus term:(additive | number)) {
        number.value + term.value

    rule number
      ([0-9]+ space) {

Since no method name is explicitly specified in the semantic blocks, they may be
called using the `value` method.

# Testing

Citrus was designed to facilitate simple and powerful testing of grammars. To
demonstrate how this is to be done, we'll use the `Addition` grammar from our
previous [example](example.html). The following code demonstrates a simple test
case that could be used to test that our grammar works properly.

    class AdditionTest < Test::Unit::TestCase
      def test_additive
        match = Addition.parse('23 + 12', :root => :additive)
        assert_equal('23 + 12', match)
        assert_equal(35, match.value)

      def test_number
        match = Addition.parse('23', :root => :number)
        assert_equal('23', match)
        assert_equal(23, match.value)

The key here is using the `:root` option when performing the parse to specify
the name of the rule at which the parse should start. In `test_number`, since
`:number` was given the parse will start at that rule as if it were the root
rule of the entire grammar. The ability to change the root rule on the fly like
this enables easy unit testing of the entire grammar.

Also note that because match objects are themselves strings, assertions may be
made to test equality of match objects with string values.

## Debugging

When a parse fails, a [ParseError](api/classes/Citrus/ParseError.html) object is
generated which provides a wealth of information about exactly where the parse
failed including the offset, line number, line text, and line offset. Using this
object, you could possibly provide some useful feedback to the user about why
the input was bad. The following code demonstrates one way to do this.

    def parse_some_stuff(stuff)
      match = StuffGrammar.parse(stuff)
    rescue Citrus::ParseError => e
      raise ArgumentError, "Invalid stuff on line %d, offset %d!" %
        [e.line_number, e.line_offset]

In addition to useful error objects, Citrus also includes a means of visualizing
match trees in the console via `Match#dump`. This can help when determining
which rules are generating which matches and how they are organized in the
match tree.

# Extras

Several files are included in the Citrus repository that make it easier to work
with grammar files in various editors.

## TextMate

To install the Citrus [TextMate]( bundle, simply
double-click on the `Citrus.tmbundle` file in the `extras` directory.

## Vim

To install the [Vim]( scripts, copy the files in
`extras/vim` to a directory in Vim's

# Links

The primary resource for all things to do with parsing expressions can be found
on the original [Packrat and Parsing Expression Grammars page](
at MIT.

Also, a useful summary of parsing expression grammars can be found on

Citrus draws inspiration from another Ruby library for writing parsing
expression grammars, Treetop. While Citrus' syntax is similar to that of
[Treetop](, it's not identical. The link is
included here for those who may wish to explore an alternative implementation.

# License

Copyright 2010 Michael Jackson

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.