node package manager

nearley

nearley JS.ORG

Simple parsing for node.js.

What is nearley?

nearley uses the Earley parsing algorithm augmented with Joop Leo's optimizations to parse complex data structures easily. nearley is über-fast and really powerful. It can parse literally anything you throw at it.

nearley is used by artificial intelligence and computational linguistics classes at universities, as well as file format parsers, markup languages and complete programming languages. It's an npm staff pick.

Why do I care?

nearley can parse what other JS parse engines cannot, because it uses a different algorithm. The Earley algorithm is general, which means it can handle any grammar you can define in BNF. In fact, the nearley syntax is written in itself (this is called bootstrapping).

PEGjs and Jison are recursive-descent based, and so they will choke on a lot of grammars, in particular left recursive ones.

nearley also has capabilities to catch errors gracefully, and detect ambiguous grammars (grammars that can be parsed in multiple ways).

Installation and Usage

Note: For beginners, Guillermo Webster's nearley-playground is a wonderful way to explore nearley interactively in your browser:

A screenshot of the playground

Enjoy!

To use nearley, you need both a global and a local installation. The two types of installations are described separately below.


To compile a nearley parser (a .ne file), you need to install the nearleyc command from npm:

$ npm install -g nearley
$ nearleyc parser.ne

nearley ships with three additional tools:

  • nearley-test lets you quickly test a grammar against some input and see the results. It also lets you explore the internal state of nearley's Earley table, in case you find that interesting.
  • nearley-unparse inverts a parser into a generator, allowing you to create random strings that match your grammar.
  • nearley-railroad generates pretty railroad diagrams from your parser. This is mainly helpful for creating documentation, as (for example) on json.org.

You can uninstall the nearley compiler using npm uninstall -g nearley.


To use a generated grammar, you need to install nearley as a per-project dependency via npm (note that there is no -g in the first command):

$ npm install nearley
$ node
> var nearley = require("nearley");
> var grammar = require("./my-generated-grammar.js");

Alternatively, to use a generated grammar in a browser runtime, include the nearley.js file in a <script> tag.

<script src="nearley.js"></script>
<script src="my-generated-grammar.js"></script>

Parser specification

This is a basic overview of nearley syntax and usage. For an advanced styleguide, see this file.

A parser consists of several nonterminals, which are constructions in a language. A nonterminal is made up of a series of either other nonterminals or strings. In nearley, you define a nonterminal by giving its name and its expansions.

Strings are the terminals, which match those string literals (specified as JSON-compatible strings).

The following grammar matches a number, a plus sign, and another number:

expression -> number "+" number

Anything from a # to the end of a line is ignored as a comment:

expression -> number "+" number # sum of two numbers

A nonterminal can have multiple expansions, separated by vertical bars (|):

expression ->
      number "+" number
    | number "-" number
    | number "*" number
    | number "/" number

The parser tries to parse the first nonterminal that you define in a file. However, you can (and should!) introduce more nonterminals as "helpers". In this example, we would have to define the expansion of number.

Postprocessors

Each meaning (called a production rule) can have a postprocessing function, that can format the data in a way that you would like:

expression -> number "+" number {%
    function (data, location, reject) {
        return ["sum", data[0], data[2]];
    }
%}

data is an array whose elements match the nonterminals in order. The postprocessor id returns the first token in the match (literally function(data) {return data[0];}).

location is the index at which that rule was found. Retaining this information in a syntax tree is useful if you're writing an interpreter and want to give fancy error messages for runtime errors.

If, after examining the data, you want to force the rule to fail anyway, return reject. An example of this is allowing a variable name to be a word that is not a string:

variable -> word {%
    function(data, location, reject) {
        if (KEYWORDS.indexOf(data[0]) === -1) {
            return data[0]; // It's a valid name 
        } else {
            return reject;  // It's a keyword, so reject it 
        }
    }
%}

You can write your postprocessors in CoffeeScript by adding @preprocessor coffee to the top of your file. Similarly, you can write them in TypeScript by adding @preprocessor typescript to the top of your file. If you would like to support a different postprocessor language, feel free to file a PR!

Epsilon rules

The epsilon rule is the empty rule that matches nothing. The constant null is the epsilon rule, so:

a -> null
    | a "cow"

will match 0 or more cows in a row.

Charsets

You can use valid RegExp charsets in a rule:

not_a_letter -> [^a-zA-Z]

The . character can be used to represent "any character".

EBNF

nearley compiles some higher-level constructs into BNF for you. In particular, the *, ?, and + operators from Regexes (or EBNF) are available as shown:

batman -> "na":* "batman" # nananana...nanabatman

You can also use capture groups with parentheses. Its contents can be anything that a rule can have:

banana -> "ba" ("na" {% id %} | "NA" {% id %}):+

Macros

You can create "polymorphic" rules through macros:

match3[X] -> $X $X $X
quote[X]  -> "'" $X "'"

main -> match3[quote["Hello?"]]
# matches "'Hello?''Hello?''Hello?'"

Macros are dynamically scoped:

foo[X, Y] -> bar[("moo" | "oink" | "baa")] $Y
bar[Z]    -> $X " " $Z # 'remembers' $X from its caller
main -> foo["Cows", "."]
# matches "Cows oink." and "Cows moo."

Macros cannot be recursive (nearleyc will go into an infinite loop trying to expand the macro-loop).

Additional JS

For more intricate postprocessors, or any other functionality you may need, you can include parts of literal JavaScript between production rules by surrounding it with @{% ... %}:

@{% var makeCowWithString = require('./cow.js') %}
cow -> "moo" {% function(d) {makeCowWithString(d[0]); } %}

Note that it doesn't matter where you define these; they all get hoisted to the top of the generated code.

Importing

You can include the content of other parser files:

@include "../misc/primitives.ne" # path relative to file being compiled
sum -> number "+" number

There are also some built-in parsers whose contents you can include:

@builtin "cow.ne"
main -> cow:+

See the builtin/ directory for an index of this library. Contributions are welcome here!

Including a parser imports all of the nonterminals defined in the parser, as well as any JS, macros, and config options defined there.

Custom lexers

You can pass a lexer instance to Parser, which must have the following interface:

  • reset(chunk, Info): set the internal buffer to chunk, and restore line/col/state info taken from save().
  • next() -> Token return e.g. {type, value, line, col, …}. Only the value attribute is required.
  • save() -> Info -> return an object describing the current line/col etc. This allows us to preserve this information between feed() calls, and also to support Parser#rewind(). The exact structure is lexer-specific; nearley doesn't care what's in it.
  • formatError(token) -> return a string with an error message describing the line/col of the offending token. You might like to include a preview of the line in question.
  • has(tokenType) -> return true if the lexer can emit tokens with that name. Used to resolve %-specifiers in compiled nearley grammars.

If Parser isn't given a lexer option, it will look for a .lexer attribute on its Grammar. The @lexer directive allows exporting a lexer object from your .ne grammar file. (See json.ne for an example.)

Custom tokens

Nearley assumes by default that your fundamental unit of parsing, called a token, is a character. That is, you're parsing a list of characters. However, sometimes you want to preprocess your string to turn it into a list of lexical tokens. This means, instead of seeing "1", "2", "3", the nearley might just see a single list item "123". This is called tokenizing, and it can bring you decent performance gains. It also allows you to write cleaner, more maintainable grammars and to prevent ambiguous grammars.

Tokens can be defined in two ways: literal tokens and testable tokens. A literal token matches exactly, while a testable token runs a function to test whether it is a match or not.

@{%
var print_tok  = {literal: "print"};
var number_tok = {test: function(x) {return x.constructor === Number; }}
%}
 
main -> %print_tok %number_tok

Now, instead of parsing the string "print 12", you would parse the array ["print", 12].

You can write your own tokenizer using regular expressions, or choose from several existing tokenizing libraries for node.

Using a parser

nearley exposes the following API:

var grammar = require("generated-code.js");
var nearley = require("nearley");
 
// Create a Parser object from our grammar. 
var p = new nearley.Parser(grammar.ParserRules, grammar.ParserStart);
 
// Parse something 
p.feed("1+1");
// p.results --> [ ["sum", "1", "1"] ] 

The Parser object can be fed data in parts with .feed(data). You can then find an array of parsings with the .results property. If results is empty, then there are no parsings. If results contains multiple values, then that combination is ambiguous.

The incremental feeding design is inspired so that you can parse data from stream-like inputs, or even dynamic readline inputs. For example, to create a Python-style REPL where it continues to prompt you until you have entered a complete block.

p.feed(prompt_user(">>> "));
while (p.results.length < 1) {
    p.feed(prompt_user("... "));
}
console.log(p.results);

The nearley.Parser constructor takes an optional third parameter, options, which is an object with the following possible keys:

  • keepHistory (boolean, default false): if set to true, nearley will preserve the internal state of the parser in the parser's .table property. Preserving the state has some performance cost (because it can potentially be very large), so we recommend leaving this as false unless you are familiar with the Earley parsing algorithm and are planning to do something exciting with the parse table.

Catching errors

If there are no possible parsings, nearley will throw an error whose offset property is the index of the offending token.

try {
    p.feed("1+gorgonzola");
} catch(parseError) {
    console.log(
        "Error at character " + parseError.offset
    ); // "Error at character 2" 
}

Exploring a parser interactively

The global install will provide nearley-test, a simple command-line tool you can use to inspect what a parser is doing. You input a generated grammar.js file, and then give it some input to test the parser against. nearley-test prints out the output if successful, and also gives you the complete parse table used by the algorithm. This is very helpful when you're testing a new parser.

This was previously called bin/nearleythere.js and written by Robin.

The Unparser

The Unparser takes a (compiled) parser and outputs a random string that would be accepted by the parser.

$ nearley-unparse -s number <(nearleyc builtin/prims.ne)
-6.22E94

You can use the Unparser to...

  • ...test your parser specification by generating lots of random expressions and making sure all of them are "correct".
  • ...generate random strings from a schema (for example, random email addresses or telephone numbers).
  • ...create fuzzers and combinatorial stress-testers.
  • ...play "Mad-Libs" automatically! (Practical application: automatic grammatically valid loremtext.)

The Unparser outputs as a stream by continuously writing characters to its output pipe. So, if it "goes off the deep end" and generates a huge string, you will still see output scrolling by in real-time.

To limit the size of the output, you can specify a bound on the depth with the -d flag. This switches the Unparser to a different algorithm. A larger depth bound corresponds to larger generated strings.

As far as I know, nearley is the only parser generator with this feature. It is inspired by Roly Fentanes' randexp, which does the same thing with regular expressions.

Automagical Railroad Diagrams

nearley lets you convert your grammars to pretty SVG railroad diagrams that you can include in webpages, documentation, and even papers.

$ nearley-railroad regex.ne -o grammar.html

Railroad demo

See a bigger example here.

(This feature is powered by railroad-diagrams by tabatkins.)

Other Tools

This section lists nearley tooling created by other developers. These tools are not distributed with nearley, so if you have problems, please contact the respective author for support instead of opening an issue with nearley.

Atom users can write nearley grammars with this plugin by Bojidar Marinov.

Sublime Text users can write nearley grammars with this syntax by liam4.

Vim users can use this plugin by Andrés Arana.

Visual Studio Code users can use this extension by Pouya Kary.

Python users can convert nearley grammars to Python using lark by Erez.

Browser users can use nearley-playground by Guillermo Webster to explore nearley interactively in the browser. There is also a Mac app by Pouya Kary.

Webpack users can use nearley-loader by Andrés Arana to load grammars directly.

Still confused?

You can read the calculator example to get a feel for the syntax (see it live here). You can read a grammar for tosh over here. There are more sample grammars in the /examples directory. For larger examples, we also have experimental parsers for CSV and Lua.

Contributing

Clone, hack, PR. Tests live in test/ and can be called with npm test. Make sure you read test/profile.log after tests run, and that nothing has died (parsing is tricky, and small changes can kill efficiency).

If you're looking for something to do, here's a short list of things that would make me happy:

  • Optimize. There are still plenty of optimizations that an enterprising JS-savant could implement.
  • Help build the builtins library by PRing in your favorite primitives.
  • Solutions to issues labeled "up for grabs" on the issue tracker.

Nearley is MIT licensed.

A big thanks to Nathan Dinsmore for teaching me how to Earley, Aria Stewart for helping structure nearley into a mature module, and Robin Windels for bootstrapping the grammar. Additionally, Jacob Edelman wrote an experimental JavaScript parser with nearley and contributed ideas for EBNF support. Joshua T. Corbin refactored the compiler to be much, much prettier. Bojidar Marinov implemented postprocessors-in-other-languages. Shachar Itzhaky fixed a subtle bug with nullables.

Further reading

  • Read my blog post to learn more about the algorithm.
  • Read about Marpa to learn more than you ever thought you wanted to know about parsing.
  • A nearley tutorial written by @gajus.