json-ext

A set of utilities designed to extend JSON’s capabilities, especially for handling large JSON datasets (over 100MB) efficiently and streaming JSONL/NDJSON processing:

• parseChunked() – Parses JSON and JSONL/NDJSON incrementally; similar to JSON.parse(), but processing data in chunks.
• stringifyChunked() – Converts JavaScript objects to JSON or JSONL incrementally; similar to JSON.stringify(), but returns a generator that yields strings in parts.
• stringifyInfo() – Estimates the size of the JSON or JSONL stringify result and identifies circular references without generating the output.
• parseFromWebStream() – A helper function to parse JSON chunks directly from a Web Stream.
• createStringifyWebStream() – A helper function to generate JSON data as a Web Stream.

Key Features

• Optimized to handle large JSON data with minimal resource usage (see benchmarks)
• Built-in JSONL/NDJSON support for parsing and serializing newline-delimited JSON
• Works seamlessly with browsers, Node.js, Deno, and Bun
• Supports both Node.js and Web streams
• Available in both ESM and CommonJS
• TypeScript typings included
• No external dependencies
• Compact size: 9.0Kb (minified), 4.0Kb (min+gzip)

Why json-ext?

• Handles large JSON files: Overcomes the limitations of V8 for strings larger than ~500MB, enabling the processing of huge JSON data.
• Prevents main thread blocking: Distributes parsing and stringifying over time, ensuring the main thread remains responsive during heavy JSON operations.
• Reduces memory usage: Traditional JSON.parse() and JSON.stringify() require loading entire data into memory, leading to high memory consumption and increased garbage collection pressure. parseChunked() and stringifyChunked() process data incrementally, optimizing memory usage.
• Size estimation: stringifyInfo() allows estimating the size of resulting JSON before generating it, enabling better decision-making for JSON generation strategies.
• JSONL/NDJSON streaming: Native support for parsing and serializing newline-delimited JSON, enabling efficient processing of log streams, data pipelines, and large datasets without loading everything into memory.

Install

npm install @discoveryjs/json-ext

API

parseChunked()

Functions like JSON.parse(), iterating over chunks to reconstruct the result object, and returns a Promise.

function parseChunked(input: Iterable<Chunk> | AsyncIterable<Chunk>, reviver?: Reviver): Promise<any>;
function parseChunked(input: Iterable<Chunk> | AsyncIterable<Chunk>, options?: ParseOptions): Promise<any>;
function parseChunked(input: () => (Iterable<Chunk> | AsyncIterable<Chunk>), reviver?: Reviver): Promise<any>;
function parseChunked(input: () => (Iterable<Chunk> | AsyncIterable<Chunk>), options?: ParseOptions): Promise<any>;

type Chunk = string | Buffer | Uint8Array;
type Reviver = (this: any, key: string, value: any) => any;
type ParseOptions = {
    reviver?: Reviver;
    mode?: 'json' | 'jsonl' | 'auto';
    onRootValue?: (value: any, state: ParseChunkedState) => void;
    onChunk?: (chunkParsed: number, chunk: string | null, pending: string | null, state: ParseChunkedState) => void;
};
type ParseChunkedState = {
    mode: 'json' | 'jsonl';
    returnValue: any;
    currentRootValue: any;
    rootValuesCount: number;
    consumed: number;
    parsed: number;
};

Benchmark

Usage:

import { parseChunked } from '@discoveryjs/json-ext';

const data = await parseChunked(chunkEmitter);

Parameter chunkEmitter can be an iterable or async iterable that iterates over chunks, or a function returning such a value. A chunk can be a string, Uint8Array, or Node.js Buffer.

You can pass reviver either as the second argument (parseChunked(input, reviver)) or inside options (parseChunked(input, { mode, reviver })). reviver works the same way as in JSON.parse().

options.mode controls JSON Lines support:

• 'json' (default): parse as regular JSON;
• 'jsonl': parse as JSONL (Newline Delimited JSON) and always return an array of parsed lines;
• 'auto': parse as regular JSON, but switch to JSONL mode when an additional value appears after a newline.

options.onRootValue is called when a root value is parsed and finalized. When onRootValue is specified, parseChunked() resolves to the number of processed root values (instead of returning parsed value(s)), which allows processing huge or infinite streams without accumulating all values in memory.

options.onChunk is called after each input chunk is processed and once at the end with chunk = null. It provides parsing progress and parser state as chunks are processed.

The state object passed to onRootValue and onChunk callbacks has the following properties: - consumed – number of characters consumed so far - parsed – number of characters parsed so far (not necessarily the same when a chunk ends in the middle of a token) - mode – current parsing mode (json or jsonl) - rootValuesCount – number of root values parsed so far - currentRootValue – current root value being parsed - returnValue – current return value state, i.e. what parseChunked() will return when finished (either the parsed value or the number of root values, depending on whether onRootValue is specified)

Examples:

• Generator:

    parseChunked(function*() {
        yield '{ "hello":';
        yield Buffer.from(' "wor'); // Node.js only
        yield new TextEncoder().encode('ld" }'); // returns Uint8Array
    });
  

• Async generator:

    parseChunked(async function*() {
        for await (const chunk of someAsyncSource) {
            yield chunk;
        }
    });
  

• Array:

    parseChunked(['{ "hello":', ' "world"}'])
  

• Function returning iterable:

    parseChunked(() => ['{ "hello":', ' "world"}'])
  

• Node.js Readable stream:

    import fs from 'node:fs';
  
    parseChunked(fs.createReadStream('path/to/file.json'))
  

• Web stream (e.g., using fetch()):

  Note: Iterability for Web streams was added later in the Web platform, not all environments support it. Consider using parseFromWebStream() for broader compatibility.

    const response = await fetch('https://example.com/data.json');
    const data = await parseChunked(response.body); // body is ReadableStream
  

stringifyChunked()

Functions like JSON.stringify(), but returns a generator yielding strings instead of a single string.

Note: Returns "null" when JSON.stringify() returns undefined (since a chunk cannot be undefined).

function stringifyChunked(value: any, replacer?: Replacer, space?: Space): Generator<string, void, unknown>;
function stringifyChunked(value: any, options: StringifyOptions): Generator<string, void, unknown>;

type Replacer =
    | ((this: any, key: string, value: any) => any)
    | (string | number)[]
    | null;
type Space = string | number | null;
type StringifyOptions = {
    replacer?: Replacer;
    space?: Space;
    mode?: 'json' | 'jsonl';
    highWaterMark?: number;
};

Benchmark

Usage:

• Getting an array of chunks:

    const chunks = [...stringifyChunked(data)];
  

• Iterating over chunks:

    for (const chunk of stringifyChunked(data)) {
        console.log(chunk);
    }
  

• Specifying the minimum size of a chunk with highWaterMark option:

    const data = [1, "hello world", 42];
  
    console.log([...stringifyChunked(data)]); // default 16kB
    // ['[1,"hello world",42]']
  
    console.log([...stringifyChunked(data, { highWaterMark: 16 })]);
    // ['[1,"hello world"', ',42]']
  
    console.log([...stringifyChunked(data, { highWaterMark: 1 })]);
    // ['[1', ',"hello world"', ',42', ']']
  

• JSONL output mode:

    const rows = [{ id: 1 }, { id: 2 }, { id: 3 }];
    const jsonl = [...stringifyChunked(rows, { mode: 'jsonl' })].join('');
  
    // {"id":1}\n{"id":2}\n{"id":3}
  

• Streaming into a stream with a Promise (modern Node.js):

    import { pipeline } from 'node:stream/promises';
    import fs from 'node:fs';
  
    await pipeline(
        stringifyChunked(data),
        fs.createWriteStream('path/to/file.json')
    );
  

• Wrapping into a Promise streaming into a stream (legacy Node.js):

    import { Readable } from 'node:stream';
  
    new Promise((resolve, reject) => {
        Readable.from(stringifyChunked(data))
            .on('error', reject)
            .pipe(stream)
            .on('error', reject)
            .on('finish', resolve);
    });
  

• Writing into a file synchronously:

  Note: Slower than JSON.stringify() but uses much less heap space and has no limitation on string length ```js import fs from ‘node:fs’;

  const fd = fs.openSync(‘output.json’, ‘w’);

  for (const chunk of stringifyChunked(data)) { fs.writeFileSync(fd, chunk); }

  fs.closeSync(fd); ```

• Using with fetch (JSON streaming):

  Note: This feature has limited support in browsers, see Streaming requests with the fetch API

  Note: ReadableStream.from() has limited support in browsers, use createStringifyWebStream() instead.

    fetch('http://example.com', {
        method: 'POST',
        duplex: 'half',
        body: ReadableStream.from(stringifyChunked(data))
    });
  

• Wrapping into ReadableStream:

  Note: Use ReadableStream.from() or createStringifyWebStream() when no extra logic is needed ```js new ReadableStream({ start() { this.generator = stringifyChunked(data); }, pull(controller) { const { value, done } = this.generator.next();

        if (done) {
            controller.close();
        } else {
            controller.enqueue(value);
        }
    },
    cancel() {
        this.generator = null;
    }   });   ```
  

stringifyInfo()

export function stringifyInfo(value: any, replacer?: Replacer, space?: Space): StringifyInfoResult;
export function stringifyInfo(value: any, options?: StringifyInfoOptions): StringifyInfoResult;

type StringifyInfoOptions = {
    replacer?: Replacer;
    space?: Space;
    mode?: 'json' | 'jsonl';
    continueOnCircular?: boolean;
}
type StringifyInfoResult = {
    bytes: number;      // size of JSON in bytes
    spaceBytes: number; // size of white spaces in bytes (when space option used)
    circular: object[]; // list of circular references
};

Functions like JSON.stringify(), but returns an object with the expected overall size of the stringify operation and a list of circular references.

Example:

import { stringifyInfo } from '@discoveryjs/json-ext';

console.log(stringifyInfo({ test: true }, null, 4));
// {
//   bytes: 20,     // Buffer.byteLength('{\n    "test": true\n}')
//   spaceBytes: 7,
//   circular: []    
// }

Options

continueOnCircular

Type: Boolean
Default: false

Determines whether to continue collecting info for a value when a circular reference is found. Setting this option to true allows finding all circular references.

parseFromWebStream()

A helper function to consume JSON from a Web Stream. You can use parseChunked(stream) instead, but @@asyncIterator on ReadableStream has limited support in browsers (see ReadableStream compatibility table).

import { parseFromWebStream } from '@discoveryjs/json-ext';

const data = await parseFromWebStream(readableStream);
// equivalent to (when ReadableStream[@@asyncIterator] is supported):
// await parseChunked(readableStream);

createStringifyWebStream()

A helper function to convert stringifyChunked() into a ReadableStream (Web Stream). You can use ReadableStream.from() instead, but this method has limited support in browsers (see ReadableStream.from() compatibility table).

import { createStringifyWebStream } from '@discoveryjs/json-ext';

createStringifyWebStream({ test: true });
// equivalent to (when ReadableStream.from() is supported):
// ReadableStream.from(stringifyChunked({ test: true }))

License

MIT