PetitParser for Dart

Grammars for programming languages are traditionally specified statically. They are hard to compose and reuse due to ambiguities that inevitably arise. PetitParser combines ideas from scannerless parsing, parser combinators, parsing expression grammars (PEG) and packrat parsers to model grammars and parsers as objects that can be reconfigured dynamically.

This library is open source, stable and well tested. Development happens on GitHub. Feel free to report issues or create a pull-request there. General questions are best asked on StackOverflow.

The package is hosted on dart packages. Up-to-date API documentation is created with every release.

Tutorial

Below are step-by-step instructions of how to write your first parser. More elaborate examples (JSON parser, LISP parser and evaluator, Prolog parser and evaluator, etc.) are included in the example repository. Try out the running demos at petitparser.github.io.

Installation

Follow the installation instructions on dart packages.

Import the package into your Dart code using:

import 'package:petitparser/petitparser.dart';

It is also possible to more selectively import only certain parts of this library, i.e. package:petitparser/core.dart and package:petitparser/parser.dart for core infrastructure and the basic parsers.

[!IMPORTANT] This library makes extensive use of static extension methods. If you import the library using a library prefix or only selectively show classes you might miss some of the functionality.

Writing a Simple Grammar

Writing grammars with PetitParser is as simple as writing Dart code. For example, the following code creates a parser that can read identifiers (a letter followed by zero or more letter or digits):

final id = letter() & (letter() | digit()).star();  // (0): Parser<List<dynamic>>

If you inspect the object id in the debugger, you'll notice that the code above builds a tree of parser objects:

• SequenceParser: This parser accepts the sequence of its child parsers.
  • SingleCharacterParser: This parser accepts a single letter.
  • PossessiveRepeatingParser: This parser accepts zero or more times its child parsers.
    • ChoiceParser: This parser accepts the first of its succeeding child parsers, or otherwise fails.
      • SingleCharacterParser: This parser accepts a single letter.
      • SingleCharacterParser: This parser accepts a single digit.

The operators & and | are overloaded and create a sequence and a choice parser respectively. In some contexts it might be more convenient to use chained function calls, or the extension methods on lists. All of the following parsers accept the same inputs as the parser above:

final id1 = letter().seq(letter().or(digit()).star());  // (1): Parser<List<dynamic>>
final id2 = [letter(), [letter(), digit()].toChoiceParser().star()].toSequenceParser();  // (2): Parser<List<Object>>
final id3 = seq2(letter(), [letter(), digit()].toChoiceParser().star());  // (3): Parser<(String, List<String>)>

[!NOTE] The inferred type of the 3 parsers is not equivalent: Due to github.com/dart-lang/language/issues/1557 the inferred type of sequence and choice parsers created with operators (0) or chained function calls (1) is Parser<dynamic>. The parser built from lists (2) provides the most generic type, List<Object> in this example. The last variation (3) is the only one that doesn't lose type information and produces a record (tuple) with two typed elements String and List<String>.

[!IMPORTANT] Parsers read input greedily: if a parser fails, it does not retry. Only ChoiceParser provides backtracking by trying its children in order until one succeeds; if none do, the whole choice fails.

Parsing Some Input

To actually consume an input string we use the method Parser.parse:

final result1 = id.parse('yeah');
final result2 = id.parse('f12');

The method Parser.parse returns a Result, which is either an instance of Success or Failure. In both examples we are successful and can retrieve the resulting value using Success.value:

print(result1.value);                   // ['y', ['e', 'a', 'h']]
print(result2.value);                   // ['f', ['1', '2']]

While it seems odd to get these nested arrays with characters as a return value, this is the default decomposition of the input into a parse-tree. We'll see in a while how that can be customized.

If we try to parse something invalid we get an instance of Failure and we can retrieve a descriptive error message using Failure.message:

final result3 = id.parse('123');
print(result3.message);                 // 'letter expected'
print(result3.position);                // 0

Trying to retrieve result by calling Failure.value would throw the exception ParserException. Pattern matching can be used to decide if the parse result was a success or a failure:

switch (id.parse(input)) {
  case Success(value: final value):
    print('Success: $value');
  case Failure(message: final message, position: final position):
    print('Failure at $position: $message');
}

If you are only interested if a given string is valid you can use the helper method Parser.accept:

print(id.accept('foo'));                // true
print(id.accept('123'));                // false

Different Kinds of Parsers

PetitParser provides a large set of ready-made parser that you can compose to consume and transform arbitrarily complex languages.

Terminal Parsers

Terminal parsers are the simplest. We've already seen a few of those:

• any() parses any character.
• anyOf('abc') parses any of the characters, a, b or c.
• char('a') (or 'a'.toParser()) parses the character a.
• digit() parses a single digit from 0 to 9.
• letter() parses a single letter from a to z and A to Z.
• noneOf('abc') parses none of the characters, i.e. d, 1, or A.
• newline() parses a newline character sequence, i.e. LF (Unix) or CR+LF (Windows).
• pattern('a-f') (or 'a-f'.toParser(isPattern: true)) parses a single character between a and f.
• string('abc') (or 'abc'.toParser()) parses the string abc.
• whitespace() parses a whitespace character, i.e. ␣ or ↦.
• word() parses a single letter, digit, or the underscore character.

By default all parsers use an automatically generated descriptive error message, match case-sensitive, and work on 16-bit UTF-16 code units. To change this default behavior use the named arguments (where appropriate):

• message: 'expected a special character' to define a custom error message,
• ignoreCase: true to accept both lower- and uppercase variations, and
• unicode: true to decode surrogate pairs and read Unicode code-points.

Combinator Parsers

The next set of parsers are used to combine other parsers together:

• p1 & p2, p1.seq(p2), [p1, p2].toSequenceParser(), seq2(p1, p2) or (p1, p2).toSequenceParser() parse p1 followed by p2 (sequence). The first two produce a result of type List<dynamic>, the third one a List<P1 & P2>, and the last two a strictly typed record type (P1, P2).
• p1 | p2, p1.or(p2), or [p1, p2].toChoiceParser() parse p1, if that doesn't work parse p2 (ordered choice). The first two produce a result of type dynamic, the last one a result of type P1 & P2.

The following parsers repeat another parser a configured amount of times, and produce a list of parsed results. Check the documentation for other repeaters that are lazy or greedy, or that can handle separators.

• p.star() parses p zero or more times.
• p.plus() parses p one or more times.
• p.times(n) parses p exactly n times.
• p.repeat(n, m) parses p between n and m times.

A variation of the parsers above is the optional operator, it produces the value of p or null.

• p.optional() parses p and returns its result, otherwise returns null.
• p.optionalWith(v) parses p and returns its result, otherwise returns the argument v.

More complicated combinators that can come in handy at times are:

• p.and() parses p, but does not consume its input.
• p.not() parses p and succeeds when p fails, but does not consume its input.
• p.end() parses p and succeeds at the end of the input.

Transforming Parsers

The last type of parsers are actions or transformations we can use as follows:

• p.map((value) => ...) performs a transformation using the provided callback on the result of p.
• p.where((value) => ...) fails the parser p if its result does not satisfy the predicate.
• p.pick(n) returns the n-th element of the list p returns.
• p.cast<T>() casts the result of p to the type T.
• p.flatten() creates a string from the consumed input of p.
• p.token() creates a token that encapsulates the begin and end position, and result of p.
• p.trim() trims whitespaces before and after p.
• p.skip(before: p1, after: p2) consumes p1, p, and p2 in sequence, but only returns the result of p.

[!TIP] Various other parsers for more specific use-cases are available, browse the subclasses and extensions of the Parser class.

To return a string of the parsed identifier, we can modify our parser like this:

final id = (letter() & pattern('a-zA-Z0-9').star()).flatten();

To conveniently find all matches in a given input string you can use Parser.allMatches:

final matches = id.allMatches('foo 123 bar4');
print(matches);                         // ['foo', 'bar4']

Writing a More Complicated Grammar

Now we are able to write a more complicated grammar for evaluating simple arithmetic expressions. Within a file we start with the grammar for an integer:

final number = digit().plus().flatten().trim().map(int.parse);

Then we define the productions for addition and multiplication in order of precedence. Note that we instantiate the productions with undefined parsers upfront, because they recursively refer to each other. Later on we can resolve this recursion by setting their reference:

final term = undefined();
final prod = undefined();
final prim = undefined();

final add = (prod & char('+').trim() & term)
    .map((values) => values[0] + values[2]);
term.set(add | prod);

final mul = (prim & char('*').trim() & prod)
    .map((values) => values[0] * values[2]);
prod.set(mul | prim);

final parens = (char('(').trim() & term & char(')').trim())
    .map((values) => values[1]);
final number = digit().plus().flatten().trim().map(int.parse);
prim.set(parens | number);

To make sure our parser consumes all input we wrap it with the end() parser in the start production:

final parser = term.end();

That's it, now we can test our parser and evaluator:

parser.parse('1 + 2 * 3');              // 7
parser.parse('(1 + 2) * 3');            // 9

Using Parser References

Defining and reusing complex grammars can be cumbersome, particularly if the grammar is large and recursive (such as the example above). PetitParser provides building blocks to conveniently define and build complex grammars with possibly hundreds of productions.

To create a new grammar definition subclass GrammarDefinition. In our case we call the class ExpressionDefinition. For every production create a new method returning the primitive parser defining it. The method called start is supposed to return the start production of the grammar. To refer to a production defined in the same definition use ref(Function) with the function reference as the argument.

class ExpressionDefinition extends GrammarDefinition {
  Parser start() => ref(term).end();

  Parser term() => ref(add) | ref(prod);
  Parser add() => ref(prod) & char('+').trim() & ref(term);

  Parser prod() => ref(mul) | ref(prim);
  Parser mul() => ref(prim) & char('*').trim() & ref(prod);

  Parser prim() => ref(parens) | ref(number);
  Parser parens() => char('(').trim() & ref(term) & char(')').trim();

  Parser number() => digit().plus().flatten().trim();
}

To create a parser with all the references correctly resolved call build().

final definition = ExpressionDefinition();
final parser = definition.build();
parser.parse('1 + 2 * 3');              // ['1', '+', ['2', '+', '3']]

Again, since this is plain Dart, common code refactorings such as renaming a production updates all references correctly. Also code navigation and code completion works as expected.

[!TIP] The use of ref(Function) is not limited to subclasses of GrammarDefinition, it can be called from anywhere in Dart. To build the resulting parser use resolve(Parser) on the root node of the grammar.

[!TIP] The function ref takes positional arguments to parametrize the created parser, if the referenced function takes arguments. While ref supports an arbitrary amount of arguments, it can neither infer nor check return or argument types at compile time. The variations ref0, ref1, ref2, ... solve this problem, but require you to specify the number of arguments.

To attach custom production actions you might want to further subclass your grammar definition and override the necessary productions defined in the superclass:

class EvaluatorDefinition extends ExpressionDefinition {
  Parser add() => super.add().map((values) => values[0] + values[2]);
  Parser mul() => super.mul().map((values) => values[0] * values[2]);
  Parser parens() => super.parens().castList<num>().pick(1);
  Parser number() => super.number().map((value) => int.parse(value));
}

Similarly, build the evaluator parser like so:

final definition = EvaluatorDefinition();
final parser = definition.build();
parser.parse('1 + 2 * 3');              // 7

[!TIP] Subclassing of definitions only works well, if you keep your parsers dynamic like in the example above (Parser or Parser<dynamic>). While this might increase reusability of your parser definitions, it might also increase your code size and come with extra run-time cost.

To use just a part of the parser you can specify the start production when building. For example, to reuse the number parser one would write:

final definition = EvaluatorDefinition();
final parser = definition.build(start: definition.number);
parser.parse('42');                     // 42

Check out the documentation for more examples.

Using the Expression Builder

Writing such expression parsers is pretty common and can be tricky to get right. To simplify things, PetitParser comes with a builder that can help you to define such grammars easily. It supports the definition of operator precedence; and prefix, postfix, left- and right-associative operators.

The following code creates the empty ExpressionBuilder producing values of type num:

final builder = ExpressionBuilder<num>();

Every ExpressionBuilder needs to define at least one primitive type to parse. In this example these are the literal numbers. This time we accept floating-point numbers, not just integers. The mapping function converts the string input into an actual number.

builder.primitive(digit()
    .plus()
    .seq(char('.').seq(digit().plus()).optional())
    .flatten()
    .trim()
    .map(num.parse));

Then we define the operator-groups in descending precedence. The highest precedence have parentheses. The mapping function receives both the opening parenthesis, the value, and the closing parenthesis as arguments:

builder.group().wrapper(
    char('(').trim(), char(')').trim(), (left, value, right) => value);

Then come the normal arithmetic operators. We are using cascade notation to define multiple operators on the same precedence-group. The mapping functions receive both, the terms and the parsed operator in the order they appear in the parsed input:

// Negation is a prefix operator.
builder.group().prefix(char('-').trim(), (operator, value) => -value);

// Power is right-associative.
builder.group().right(
    char('^').trim(), (left, operator, right) => math.pow(left, right));

// Multiplication and addition are left-associative, multiplication has
// higher priority than addition.
builder.group()
  ..left(char('*').trim(), (left, operator, right) => left * right)
  ..left(char('/').trim(), (left, operator, right) => left / right);
builder.group()
  ..left(char('+').trim(), (left, operator, right) => left + right)
  ..left(char('-').trim(), (left, operator, right) => left - right);

Finally, we can build the parser:

final parser = builder.build().end();

After executing the above code we get an efficient parser that correctly evaluates expressions like:

parser.parse('-8');                     // -8
parser.parse('1+2*3');                  // 7
parser.parse('1*2+3');                  // 5
parser.parse('8/4/2');                  // 1
parser.parse('2^2^3');                  // 256

Check out the documentation for more examples.

Testing your Grammars

Real world grammars are typically large and complicated. PetitParser's architecture allows one to break down a grammar into manageable pieces, and develop and test each part individually before assembling the complete system.

Start the development and testing of a new grammar at the leaves (or tokens): write the parsers that read numbers, strings, and variables first; then continue with the expressions that can be built from these literals; and finally conclude with control structures, classes and other overarching constructs. At each step add tests and assert that the individual parsers behave as desired, so that you can be sure they also work when composing them to a larger grammar later.

Accessing and testing individual productions is simple: If you organize your grammar in your own code, make sure to expose parts of the grammar individually. If you use a GrammarDefinition, you can build individual productions using the buildFrom method. For example, to test the number production of the EvaluatorDefinition from above you would write:

test('number parsing', () {
  final definition = EvaluatorDefinition();
  final parser = definition.buildFrom(definition.number);
  expect(parser.parse('42').value, 42);
});

Additionally, PetitParser provides a Linter that comes with a collection of predefined rules that can help you find common bugs or inefficient constructs in your code. Among other things, the analyzer detects infinite loops, unreachable parsers, repeated parsers, and unresolved parsers. For an up-to-date list of all available rules check the implementation at linter_rules.dart.

To run the linter as part of your tests include the package petitparser/reflection.dart, call the linter function with the starting parser of your grammar, and assert that there are no findings. With the EvaluatorDefinition from above one would write:

test('detect common problems', () {
  final definition = EvaluatorDefinition();
  final parser = definition.build();
  expect(linter(parser), isEmpty);
});

To exclude certain rules from being reported you can exclude certain rules, i.e. linter(parser, excludedRules: {'Nested choice'}).

Check out the extensive test suites of PetitParser and PetitParser Examples for examples on testing.

Debugging your Grammars

Sometimes parsers might not behave the way you expect them to. The first step should always be to come up with a small reproducible example. If this doesn't already solve the problem, PetitParser comes with a set of built-in tools that can help you understand what is going on.

The function trace(Parser) transforms your grammar so that each parser prints its activation and results:

final parser = letter() & word().star();
trace(parser).parse('f1');

The above snippet produces the following output:

SequenceParser<dynamic>
  SingleCharacterParser[letter expected]
  Success<String>[1:2]: f
  PossessiveRepeatingParser<String>[0..*]
    SingleCharacterParser[letter or digit expected]
    Success<String>[1:3]: 1
    SingleCharacterParser[letter or digit expected]
    Failure[1:3]: letter or digit expected
  Success<List<String>>[1:3]: [1]
Success<List<dynamic>>[1:3]: [f, [1]]

Indentation signifies the activation of a parser object. Reverse indentation signifies the returning of a parse result either with a success or failure context.

Similarly, the function profile(Parser) produces a table of activation counts and run-time tallies of each parser. And, progress(Parser) visualizes how the parsers process (and possibly backtrack) through your input. Both tools can help to understand and optimize the performance characteristics of your parsers.

Misc

petitparser.github.io contains up-to-date information about PetitParser and ports to other languages.

Examples

The package comes with a large collection of example grammars and language experiments ready to explore:

• CSV contains a simple Comma-separated values (CSV) parser.
• Dart contains an experimental Dart grammar.
• JSON contains a complete JSON grammar and parser.
• Lisp contains a complete LISP grammar, parser and evaluator.
• Math contains an mathematical expression parser and evaluator.
• Pascal contains an experimental Pascal grammar.
• Prolog contains a basic Prolog grammar, parser and evaluator.
• Smalltalk contains a complete Smalltalk grammar.
• Uri contains a simple URI parser.

Furthermore, there are numerous open source projects using PetitParser:

• apollovm, a simple VM that can parse, run and generate basic Dart and Java8 code.
• equations is an equation solving library.
• expression_language is a library for parsing and evaluating expressions.
• expressions is a library to parse and evaluate simple expressions.
• json_path is an implementation of JSONPath expressions.
• pem encodes and decodes textual cryptographic keys.
• puppeteer is a library to automate the Chrome browser.
• query implements search queries with support for boolean groups, field scopes, ranges, etc.
• xml is a lightweight library for parsing, traversing, and querying XML documents.

History

PetitParser was originally implemented in Smalltalk. Later on, as a means to learn these languages, I reimplemented PetitParser in Java and Dart. The implementations are very similar in their API and the supported features. If possible, the implementations adopt best practices of the target language.

License

The MIT License, see LICENSE.