JSOX – JavaScript Object eXchange format.

100% Compatible reader for JSON. JSOX.stringify cannot generate JSON compatible output; it would lose all the features anyway; use existing JSON.stringify() if required, all JSON(JSON3/JSON5/JSON6) is valid JSOX.

JSOX adds BigInt, Date, and TypeArray transport support, and includes keywords (5)'Infinity', (5)'NaN', (6)'undefined'.

JSOX adds optional processing of typed data. Type names can be applied to Objects, Arrays and Strings. Type names are defined and provided with to/from JSOX handlers by users of this library. The data, decoded as the object '{}', array '[]', or string '""' is passed to the fromJSOX handler, and the resulting value returned as the decoded object.

Typed-objects may also be emitted as a class-defintition and then class-references. A class-defintition defines the fields in the object, and a class-reference would provide the values for each field respectively.

A typed-object example: v{ x, y } { a : v{1,2} }, which decodes as { a : {x:1,y:2} }. It defines a template/class of object that has fields 'x', and 'y'. Then defines an object with a field A what is a object of type 'v', with values (1), and (2). This example does not gain any visible savings; savings comes when you have a lot of the same sort of record with the same field names repeated often.

• adds macro/class support for object field names.
• adds support for bigint numbers; indicated with an 'n' suffix.
• adds support for Date parsing and stringification.
• o === JSOX.parse(JSON.stringify(o)) should always be exactly true.
• adds support for circular references.
• typed-strings, typed-arrays, and typed-objects, for user defined types and to and from JSOX methods. more

Example Encoding

r = JSOX.stringify( o = { 
    a: "simple object"
    , b:3
    , c:new Date()
    , d:123n
    , e:null
    , f:undefined
    , g:NaN
    , h:Infinity
    , i:-Infinity
    , j:-0.302 
    , k:new Uint8Array(9)}, null, 3 );
cnsole.log( "pretty:", o, "=\n", r );

// -- output --
pretty:
{
      a: "simple object",
      b: 3,
      c: 2018-09-14T02:55:27-07:00,
      d: 123n,
      e: null,
      g: NaN,
      h: Infinity,
      i: -Infinity,
      j: -0.302,
      k: u8[AAAAAAAAAAA=]
}

JSOX is a proposed replacement to JSON that aims to make it easier for humans to write and maintain by hand, while also transporting the correct type of the data. Humans, for example, wouldn't hand-code a base64 encoding for a TypedArray; however sending a mesh from a server to a client already processed as a typed buffer ready for WebGL consumption may be of use.

The class/tag support is entirely optional, and while it's goal is to reduce redundancy, which for large datasets of similar records can benefit, it has been argued that gzip could just be used to reduce the size; However, this also reduces the size of the data to be parsed on input, which gzip does not do.

JSOX is a (super-sub)set of JavaScript, although adds no new data types, and works with all existing JSON content. Some features allowed in JSOX are not directly supported by Javascript; although all javascript parsable features can be used in JSOX, except functions or any other code construct, transporting only data save as JSON. Most ES6 structure can be parsed, with the extension of classes/tags the reverse is not true. It was true for JSON6.

JSOX is a proprosal for an official successor to JSON, and JSOX stringified content will not work with existing JSON parsers. For this reason, JSOX files use a new .jsox extension. (TODO: new MIME type needed too.)

The code is a reference JavaScript implementation for both Node.js and all browsers. The code is derrived from JSON-6 sources.

Why

Beyond the existing reasons for JSON5/JSON6 for their modifications; this addresses the biggest shortcoming of JSON, which is the repetitive and redundant specification of field names; especially when lots of the same sort of object is represented.

This also aims to provide support for BigInt and Date format for less work at the application layer. A method for handling typed array object members should also be impelemented

(Historic Why below)

JSON isn’t the friendliest to write. Keys need to be quoted, objects and arrays can’t have trailing commas, and comments aren’t allowed — even though none of these are the case with regular JavaScript today.

That was fine when JSON’s goal was to be a great data format, but JSON’s usage has expanded beyond machines. JSON is now used for writing configs, manifests, even tests — all by humans.

There are other formats that are human-friendlier, like YAML, but changing from JSON to a completely different format is undesirable in many cases. JSON6’s aim is to remain close to JSON and JavaScript.

Features

The following is the exact list of additions to JSON’s syntax introduced by JSOX. All of these are optional.

• Concise representation of dates and times including as much information as is available for the timestamp(timezone).

• Supports encode and decode of BigInt numbers with no application overhead.

• reduces overhead from none-requires quotes for identifiers.

• can further reduce overall output size by using class tags ( Needs internal implementation; although users can generate output simply ).

Caveats

JSOX.stringify will produce output that JSON.parse cannot handle; JSOX.parse can always handle JSON.stringify.

Summary of Changes from JSON6

• BigInt encoding
• ISO date/time Encoding/decoding (as part of Number format)
• Adds classes/tags to reduce redundant information.

Summary of Changes from JSON5/JSON

• Keyword undefined
• Objects/Strings back-tick quoted strings (no template support, just quotes); Object key names can be unquoted.
• Strings - generous multiline string definition; all javascript character escapes work. (\x##, \0###, \u####, \u{} )
• Numbers - underscore digit separation in numbers, octal and binary formats; all javascript number notations. Addtionally support leading 0 to interpret as octal as C, C++ and other languages support.
• Arrays - empty members
• Streaming reader interface
• (Twice the speed of JSON5; subjective)

Additional support above JSON base

All items listed below are JSON5 additions if not specifed as JSON6.

ArrayBuffer/TypedArray

• (JSOX) Support transporting ArrayBuffer and TypedArray fields. This is implemented with constants as tags applied prefixing and opening brace '[' and encoding the binary data as a base64 string(without quotes) before the closing ']'.
  • these are prefix tags that can be applied. u8, u16, cu8, u32, s8,s16, s32, f32, f64, ab; the array is a base64 string without quotes.
  • Base64 is as dense as is feasible; it's a 33% loss; where utf8 encoding of random bytes is 50% loss. Something like base127 would be 7 bytes to 8 encoded bytes; and potential length penalty of 5 bytes.

Objects

• Object keys can be unquoted if they do not have ':', ']', '[', '{', '}', ',', any quote or whitespace, and do not begin like a number.

• Object keys can be single-quoted, (JSON6) or back-tick quoted; any valid string

• Object keys can be double-quoted (original JSON).

• Objects can have a single trailing comma. Excessive commas in objects will cause an exception. '{ a:123,,b:456 }' is invalid.

Arrays

• Arrays can have trailing commas. If more than 1 is found, additional empty elements will be added.

• (JSON6) Arrays can have comma ( ['test',,,'one'] ), which will result with empty values in the empty places.

Strings

• Strings can be double-quoted (as per original JSON).

• Strings can be single-quoted.

• Strings can be back-tick (`) (grave accent) -quoted.

• Strings can be split across multiple lines; just prefix each newline with a backslash. [ES5 §7.8.4]

• (JSON6) all strings will continue keeping every character between the start and end, this allows multi-line strings and keep the newlines in the string; if you do not want the newlines they can be escaped as previously mentioned.

Numbers

• (JSOX) BitInt numbers are stringified with suffix of 'n' as in ES(?), and implemented in V8(google/chrome/node) 2018/09/12. BigInt number parsed with 'n' suffix.

• (JSON6) Numbers can have underscores separating digits '_' these are treated as zero-width-non-breaking-space. (Proposal with the exception that _ can preceed or follow . and may be trailing.)

• Numbers can be hexadecimal (base 16). ( 0x prefix )

• (JSON6) Numbers can be binary (base 2). (0b prefix)

• (JSON6) Numbers can be octal (base 8). (0o prefix)

• (JSON6) Numbers can be octal (base 8). (0 prefix followed by more numbers, without a decimal)

• Numbers can begin or end with a (leading or trailing) decimal point.

• Numbers can include Infinity, -Infinity, NaN.

• Numbers can begin with an explicit plus sign.

• Numbers can begin with multiple minus signs. For example '----123' === 123.

Dates

• (JSOX) Encodes date time with local timestamp information to recover as much information as the original date contained. Is treated as a subtype of Number parsing; and are stored without quotes.

Keyword Values

• (JSON6) supports 'undefined' in addition to 'true', 'false', 'null'.

Comments

• Both inline (single-line using '//' (todo:or '#'?) ) and block (multi-line using /* */ ) comments are allowed.

// simple example, array buffer with 8 bytes
var ab = new ArrayBuffer([0,1,2,3,4,5,6,7]);
console.log( JSOX.stringify( {ab:new Float32Array(ab)} ) );

// example output
{ab:f32[AAECAwQFBgc=]}

Base64 vs UTF-8 Encoding

UTF-8, for character 0-127 requires 1 byte; 128-255 requires 2 bytes. For random data 0-255, 1.5 bytes will, on average, bt used to represent the string. So this is 150% larger than the original string. Even if like a base 2^40 bits, which would encode 5 bytes into a single (very extended) utf8 encoding, each byte has the prefix of 2 bits 10xx xxxx, which gives 6 bits per byte used.

Base64 is 6 bits per byte used, so instead of having a complex encoder, base64 is the optimal of 3:4 byte expasion (133%) which is the ideal that extra-long UTF8 encoding coule reach.

An alternative might be a base128 encoding, which would be close to utf-8, but would actually require 129 characters, one to indicate the bytes that are unused. 7 bytes expand to 8, gathering the top bit of each of the 7 bytes of each one into one more byte; it could be an optimal encoding using code points 0-128 (128 being the end terminator like '=' in base64). But, this also mean that up to 5 bytes of waste may be included. That is incomplete values at the end of the string have to be marked as unused.

Example conversion

The following is a contrived example, but it illustrates most of the features:

{
    foo: 'bar',
    while: true,
    nothing : undefined, // why not?

    this: 'is a \
multi-line string',

    thisAlso: 'is a
multi-line string; but keeps newline',

    // this is an inline comment
    here: 'is another', // inline comment

    /* this is a block comment
       that continues on another line */

    hex: 0xDEAD_beef,
    binary: 0b0110_1001,
    decimal: 123_456_789,
    octal: 0123,
    half: .5,
    delta: +10,
    negative : ---123,
    to: Infinity,   // and beyond!

    finally: 'a trailing comma',
    oh: [
        "we shouldn't forget",
        'arrays can have',
        'trailing commas too',
    ],
}

This implementation’s own package.jsox is more realistic:

// This file is written in JSOX syntax, naturally, but npm needs a regular
// JSON file, so compile via `npm run build`. Be sure to keep both in sync!

{
    name: 'JSOX',
    version: '0.1.105',
    description: 'JSON for the ES6 era.',
    keywords: ['json', 'es6'],
    author: 'd3x0r <d3x0r@github.com>',
    contributors: [
        // TODO: Should we remove this section in favor of GitHub's list?
        // https://github.com/d3x0r/JSOX/contributors
    ],
    main: 'lib/JSOX.js',
    bin: 'lib/cli.js',
    files: ["lib/"],
    dependencies: {},
    devDependencies: {
        gulp: "^3.9.1",
        'gulp-jshint': "^2.0.0",
        jshint: "^2.9.1",
        'jshint-stylish': "^2.1.0",
        mocha: "^2.4.5"
    },
    scripts: {
        build: 'node ./lib/cli.js -c package.JSOX',
        test: 'mocha --ui exports --reporter spec',
            // TODO: Would it be better to define these in a mocha.opts file?
    },
    homepage: 'http://github.com/d3x0r/JSOX/',
    license: 'MIT',
    repository: {
        type: 'git',
        url: 'https://github.com/d3x0r/JSOX',
    },
}

Community

Join the Google Group if you’re interested in JSOX news, updates, and general discussion. Don’t worry, it’s very low-traffic.

The GitHub wiki (will be) a good place to track JSOX support and usage. Contribute freely there!

GitHub Issues is the place to formally propose feature requests and report bugs. Questions and general feedback are better directed at the Google Group.

Usage

This JavaScript implementation of JSOX simply provides a JSOX object just like the native ES5 JSON object.

To use from Node:

npm install jsox

var JSOX = require('jsox');

To use in the browser (adds the JSOX object to the global namespace):

<script src="node_modules/jsonx/lib/jsox.js"></script>

Then in both cases, you can simply replace native JSON calls with JSOX:

var obj = JSOX.parse('{unquoted:"key",trailing:"comma",}');
var str = JSOX.stringify(obj); /* uses JSON stringify, so don't have to replace */

| registerToJSOX | (name,prototype,toCb) | For each object that matches the prototype, the name is used to prefix the type; and the cb is called to get toJSOX | | registerFromJSOX| (name,fromCb) | fromCb is called whenever the type 'name' is revived. The type of object following the name is passd as 'this'. | | registerToFrom | (name,prototype,toCb, fromCb) | register both to and from for the same name |

registerToJSOX

Registers a handler to convert a type to JSOX. This method is used to avoid modification of prototypes; would require instead that ojects that have a toJSOX know of the JSOX module instead. The result of the callback should be a string, and is up to the toJSOX method to include quotes if it is a string value. Any string may result that is valid JSOX.

Regsitering the same name more than once throws an error.

JSOX.registerToJSOX( "stringTest", stringTest.prototype, function() { return '"' + this.toString() + '"' } );

registerFromJSOX

Registers a handler to convert recovered string, array or object from JSOX. The converted data from the JSOX stream is passed as 'this'. The result of the callback may be any type of value; the resulting value is used instead of the data converted from JSOX.

Regsitering the same name more than once throws an error.

// this epects a string, as indicated by the above toJSOX output.
JSOX.registerFromJSOX( "stringTest", function() {
    console.log( "Resuurect from String:[%s]", this /*string*/ );
    return new stringTest( this );
} );

registerToFrom

Registers both to and from methods or a spsecified name, using the specified prototype to match during stringify. Internally, calls the above functions with the parameters split as appropriate.

Regsitering the same name more than once for From or To throws an error.

JSOX.registerToFrom( "stringTest", stringTest.prototype
    , function() { return '"' + this.toString() + '"' }
    , function() {
        console.log( "Resuurect from String:[%s]", this /*string*/ );
        return new stringTest( this );  // some stringTest class with string initializer
    }
);

JSOX typed-objects, typed-arrays, and typed-strings

typed-data is represented in the JSOX stream as <identifier>[data]. For objects, this is a document compression technique, which reduces the size of data to process. For arrays, internally, fixed types represent ES6 TypedArray types (u8,s8, u16,f32,etc); Another internal type is 'ref' which uses the array to have a list of element identifiers that define the path to the original object reference. Another, simple variation is to implement typed-strings, which allows color"0x1234568" to have a fromJSOX method that is passed the string, and can result with a color object.

In each case, in the following example JSOX, the same 'color' fromJSOX method will be called. It will be invoked with a string, with an array, with an object, with an object, and with an object respectively.

// this are all variations which may be used to revive a color object
color"0x12345678"                   // typed-string
color[0x12,0x34,0x56,0x78]          // typed-array

// at a root level, the first 'color' definition encountered
// is used to create a field-name map.  Then later usages
// sould only specify the values.
color{r,g,b,a}            // typed-object definition
color{0x12,34,0x56,0x78}  // typed-object subsequent usage

// if no typed-object definition is rquired, then 
// the typed-object must never be used at a root level.

{ a : color{r:0x12,g:0x34,b:0x56,a:0x78} } // object containing typed-object fromJSOX only, no pre-field-definition
[ color{r:0x12,g:0x34,b:0x56,a:0x78} ]     // array containing typed-object fromJSOX only, no pre-field-definition

Typed strings have a caveat; at a root level, strings which are typed, MUST have unquoted-identifier strings indicating their type. Because the closing quote is a definitive end-of-data marker, quoted strings at a root level always emit as a completed string; This also requires no space between the unquoted-identifier string and the quoted data string.

Typed-object and typed-arrays also require the identifer or string used for their type information not be followed by a space before the opening '{' or '['.

More on Classes/Tags - typed-object and typed-array

(constant initializers on typed-objects are not yet supported?)

The definition of a class is an identifer at the top level (before the JSON data) followed immediately by an open brace ('{'), whitespace is not allowed. Within the open brace '{' until the close '}' is a list of names seprated by commas, and of constants indicated by an identifier followed by a colon and a value.

All objects created with a class/tag definition shares the same prototype.

tagdef : identifier '{' identifier [ ',' identifier ] ... '}'
tagdef : identifier '{' constant_initializer [ ',' identifier ] ... '}' (TBI)

constant_initializer : identifier ':' value 

Usage of tags is done by specifing their identifer followed by an open brace '{' in the value field of an object definition; or at a top level referencing the same tag name already defined. For each field defined in the class, a value is expected. If a value is not found, the field will not be added, as if receiving field:undefined.

tag usage : ':' identifier '{' value [ ',' value ]... '}' 

//-- the following...
a { firstField, secondField }
a { 1, 2 }
//-- results as
{ firstField : 1, secondField : 2 }


//-- the following...
a { firstField, secondField }
[ a { 1, 2 }, a(5,6), a("val1","val2") ]
//-- results as
[ { firstField : 1, secondField : 2 }, { firstField : 5, secondField : 6 }, { firstField : "val1", secondField : "val2" } ]

Implementation of tags allows apply a class to arrays. Arrays have a class of ArrayBuffer, or other TypedArray type. The representation path in an array and a reference type for the array. This allows circular encoding.

// this is a string with a reference.
{company:{name:"Example.com",employees:[{name:"bob"},{name:"tom"}],manager:ref["company","employees",0]}}

// The above 'ref[]' gets resolved into the same employee object...
OUT: ./file6.jsox { company:
   { name: 'Example.com',
     employees: [ /*a*/{ name: 'bob' }, { name: 'tom' } ],
     manager: /*a*/{ name: 'bob' } } }

JSOX Streaming

A Parser that returns objects as they are encountered in a stream can be created. JSON.begin( dataCallback, reviver ); The callback is called for each complete object in a stream of data that is passed.

JSOX.begin( cb, reviver ) returns an object with a few methods.

   // This is (basically) the internal loop that write() uses.
   var result
   for( result = this._write(msg,false); result > 0; result = this._write() ) {
      var obj = this.value();
      // call reviver with (obj)
      // call callback with (obj)
   }

// Example code using write
function dataCallback( value ) {
    console.log( "Value from stream:", value );
}
var parser = JSON.begin( dataCallback );

parser.write( '"Hello ' );   // a broken simple value string, results as 'Hello World!' 
parser.write( 'World!"' );
parser.write( '{ first: 1,' );   // a broken structure
parser.write( ' second : 2 }' );
parser.write( '[1234,12');  // a broken array across a value
parser.write( '34,1234]'); 
parser.write( '1234 456 789 123 523');  // multiple single simple values that are numbers
parser.write( '{a:1} {b:2} {c:3}');  // multiple objects

parser.write( '1234' );  // this won't return immediately, there might be more numeric data.
parser.write( '' ); // flush any pending numbers; if an object or array or string was split, throws an error; missing close.

parser.write( '1234' ); 
parser.write( '5678 ' );  // at this point, the space will flush the number value '12345678' 

Extras

If you’re running this on Node, you can also register a JSOX require() hook to let you require() .jsox files just like you can .json files:

require('jsox/lib/require');
require('./path/to/foo');   // tries foo.jsox after foo.js, foo.json, etc.
require('./path/to/bar.jsox');

This module also provides a jsox executable (requires Node) for converting JSOX files to JSON:

jsox -c path/to/foo.jsox    # generates path/to/foo.json

Other Implementations

This is also implemented as part of npm sack.vfs as a native code node.js addon. This native javascript version allows usage in browsers.

Possible release of amagamated source, native Node Addon (standalone).

Working on a WebAssembly version.

Benchmarks

This is as fast as the javascript version of Douglas Crockford's reference implementation JSON implementation for JSON parsing.

This is nearly double the speed of JSON5 implementation that inspired this (which is half the speed of Crockford's reference implementation).

This is half the speed of the sack.vfs native C++ node addon implementation (which itself is half the speed of V8's native code implementation, but they can cheat and build strings directly).

Development

git clone https://github.com/d3x0r/jsox
cd jsox
npm install
npm test

As the package.jsox file states, be sure to run npm run build on changes to package.jsox, since npm requires package.json.

Feel free to file issues and submit pull requests — contributions are welcome. If you do submit a pull request, please be sure to add or update the tests, and ensure that npm test continues to pass.

Changelog

• 1.0.5
  • Fix streaming ability
  • consequtive strings only have whitespace to separate them, so identifiers for defining typed-objects cannot have whitespace between them and '{'.
  • implement test for non-identifier characters to quote field strings (or not). Implement reading non-identifier characters, and fault if identifier is unquoted and has such a character.
  • implement typed-strings, which can be used to trigger constructors which accept single strings.
  • update readme with typed-strings, typed-arrays, and typed-objects.
  • performance fix; was keeping parser objects forever.
• 1.0.4 - Be more forgiving about platforms not having BigInt native support.
• 1.0.3 - Add ability to register prototypes to use for decoding.
• 1.0.2 - Issue with mutiple leading and trailing spaces. Fix collecting streams of numbers. Fix an issue with nested classes. Add circular reference support.
• 1.0.1 - Removed modification of object prototypes; instead track object prototype to formatting function in a WeakMap(). Fixed class expansion. Make objects of a class share the same prototype.
• 1.0.0 - Intial Release.

License

MIT. See LICENSE.md for details.

Credits

(http://github.com/json5/json5) Inspring this project.