~* Citrus *~ Parsing Expressions for Ruby Citrus is a compact and powerful parsing library for [Ruby](http://ruby-lang.org/) that combines the elegance and expressiveness of the language with the simplicity and power of [parsing expressions](http://en.wikipedia.org/wiki/Parsing_expression_grammar). # Installation Via [RubyGems](http://rubygems.org/): $ gem install citrus From a local copy: $ git clone git://github.com/mjijackson/citrus.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 thereof. A Citrus grammar is really just a souped-up Ruby [module](http://ruby-doc.org/core/classes/Module.html). 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](http://ruby-doc.org/core/classes/String.html) 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 match. # 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' "abc\n" /\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 anything, even new lines See [Terminal](api/classes/Citrus/Terminal.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 The + and ? operators are supported as well for the common cases of 1* and *1 respectively. 'abc'+ # match "abc" at least once 'abc'? # match "abc" a maximum of once 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 a "b" that is not preceded by an "a" !'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. Note that any operator binds more tightly than the bar. 'a' | 'b' # match "a" or "b" 'a' 'b' | 'c' # match "a" then "b" (in sequence), or "c" See [Choice](api/classes/Citrus/Choice.html) for more information. ## 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. ## 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 a block method See [Label](api/classes/Citrus/Label.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 [] | 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) end rule number [0-9]+ space end rule plus '+' space end rule space [ \t]* end end 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 end } end rule number ([0-9]+ space) { def value strip.to_i end } end rule plus '+' space end rule space [ \t]* end end 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 [String#strip](http://ruby-doc.org/core/classes/String.html#M000820). 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](https://rubygems.org/gems/citrus) installed, try the following sequence of commands in a terminal. $ irb > require 'citrus' => true > Citrus.load 'addition' => [Addition] > m = Addition.parse '1 + 2 + 3' => # 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 file. Take a look at [examples/calc.citrus](http://github.com/mjijackson/citrus/blob/master/examples/calc.citrus) 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 as: rule additive (number plus term:(additive | number)) { number.value + term.value } end rule number ([0-9]+ space) { strip.to_i } end 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(match) assert_equal('23 + 12', match) assert_equal(35, match.value) end def test_number match = Addition.parse('23', :root => :number) assert(match) assert_equal('23', match) assert_equal(23, match.value) end end The key here is using the `root` [option](api/classes/Citrus/GrammarMethods.html#M000031) 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. 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] end In addition to useful error objects, Citrus also includes a special file that should help grammar authors when debugging grammars. To get this extra functionality, simply `require 'citrus/debug'` instead of `require 'citrus'` when running your code. When debugging is enabled, you can visualize parse trees in the console as XML documents. This can help when determining which rules are generating which matches and how they are organized in the output. Also when debugging, each match object automatically records its offset in the original input, which can also be very helpful in keeping track of which offsets in the input generated which matches. # Links The primary resource for all things to do with parsing expressions can be found on the original [Packrat and Parsing Expression Grammars page](http://pdos.csail.mit.edu/~baford/packrat) at MIT. Also, a useful summary of parsing expression grammars can be found on [Wikipedia](http://en.wikipedia.org/wiki/Parsing_expression_grammar). Citrus draws inspiration from another Ruby library for writing parsing expression grammars, Treetop. While Citrus' syntax is similar to that of [Treetop](http://treetop.rubyforge.org), 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. 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.