nearley - 简单，快速和强大的JavaScript数据结构解析

jopen 9年前

Simple parsing for node.js.

What is nearley?

nearley 使用Earley解析算法来轻松解析复杂的数据结构。nearley超级快和相当强大。It can parse anything you throw at it.

nearley 已经用于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

To compile a parser, use the nearleyc command:

npm install -g nearley  nearleyc parser.ne

Run nearleyc --help for more options.

To use a generated grammar in a node runtime, install nearley as a module:

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

To use a generated grammar in a browser runtime, include nearley.js (you can hardlink from Github if you want):

<script src="https://raw.githubusercontent.com/Hardmath123/nearley/master/lib/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.

The following grammar matches a number, a plus sign, and another number (anything from a # to the end of a line is ignored as a comment):

expression -> number "+" number

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

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) {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. 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.

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);

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.

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.

Still confused?

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

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.
Implement Joop Leo's right-recursion algorithm.
Ports to other languages! pearley for Python and cearley for C would be awesome.
Help build the builtins library by PRing in your favorite primitives.

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.

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