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.
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.
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.
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.
- ChoiceParser: This parser accepts the first of its succeeding child parsers, or otherwise fails.
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 that 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 loose type information and produces a record (tuple) with two typed elements String
and List<String>
.
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 ParserError
. 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
PetitParser provides a large set of ready-made parser that you can compose to consume and transform arbitrarily complex languages.
Terminal parsers are the simplest. We've already seen a few of those:
any()
parses any character.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.newline()
parses a newline character sequence, i.e. LF (Unix) and CR+LF (Windows).pattern('a-f')
(or'a-f'.toParser(isPattern: true)
) parses a single character between a and f.patternIgnoreCase('a-f')
(or'a-f'.toParser(isPattern: true, caseInsensitive: true)
) parses a single character between a and f, or A and F.string('abc')
(or'abc'.toParser()
) parses the string abc.stringIgnoreCase('abc')
(or'abc'.toParser(caseInsensitive: true)
) parses the strings Abc, aBC, ...word()
parses a single letter, digit, or the underscore character.
So instead of using the letter and digit predicate above, we could have written our identifier parser using one of the equivalent variations below:
final id1 = letter() & word().star();
final id2 = letter() & pattern('a-zA-Z0-9').star();
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 typeList<dynamic>
, the third one aList<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 typedynamic
, the last one a result of typeP1 & 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)
parsers 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, if possible.
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 succeed when p fails, but does not consume its input.p.end()
parses p and succeed at the end of the input.
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 typeT
.p.flatten()
creates a string from the consumed input of p.p.token()
creates a token from the 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.
Various other parsers for more specific use-cases are available, to discover browse the subclasses 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']
These are the basic elements to build parsers. There are a few more well documented and tested factory methods in the Parser
class. If you want browse their documentation and tests.
Now we are able to write a more complicated grammar for evaluating simple arithmetic expressions. Within a file we start with the grammar for a number (actually 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 into 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
Defining and reusing complex grammars can be cumbersome, particularly if the grammar is large and recursive (such as the example above). The class GrammarDefinition
provides the building block 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 ref0
with the function reference as the argument. The 0 at the end of ref0
means that the production reference isn't parametrized (zero argument production method).
class ExpressionDefinition extends GrammarDefinition {
Parser start() => ref0(term).end();
Parser term() => ref0(add) | ref0(prod);
Parser add() => ref0(prod) & char('+').trim() & ref0(term);
Parser prod() => ref0(mul) | ref0(prim);
Parser mul() => ref0(prim) & char('*').trim() & ref0(prod);
Parser prim() => ref0(parens) | ref0(number);
Parser parens() => char('(').trim() & ref0(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.
To attach custom production actions you might want to further subclass your grammar definition and override overriding 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
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 avoid, specify the desired static types or let Dart infer them.
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
This is just the surface of what GrammarDefinition
can do, check out the documentation and the examples using it.
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 expression builder 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.
Real world grammar 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 optional start parameter of the build
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.build(start: 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.
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 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[1:2]: f
PossessiveRepeatingParser<String>[0..*]
SingleCharacterParser[letter or digit expected]
Success[1:3]: 1
SingleCharacterParser[letter or digit expected]
Failure[1:3]: letter or digit expected
Success[1:3]: [1]
Success[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.
The functions profile and progress work similarly: profile
produces a table of activation counts and times of each parser; and progress
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.
petitparser.github.io contains up-to-date information about PetitParser and ports to other languages.
The package comes with a large collection of example grammars and language experiments ready to explore:
- Dart contains an experimental Dart grammar.
- JSON contains a complete JSON grammar and parser.
- Lisp contains a complete LISP grammar, parser and evaluator.
- 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.
PetitParser was originally implemented in Smalltalk. Later on, as a mean 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 practises of the target language.
The MIT License, see LICENSE.