transferables

NPM | GitHub | Licence

A utility library that lists out all transferable objects that can be moved between Workers and the main thread*.

* There are many asterisks involved in transferable objects, the transferables library is able sort out a large number of these asterisks, but it can't sort all of them. Those it can't, have been listed in #limitations, you should do your own research before using.

Installation

npm install transferables

yarn add transferables

or

pnpm install transferables

Usage

import { hasTransferables, getTransferables } from "transferables";

You can also use it directly through a script tag:

<script src="https://unpkg.com/transferables" type="module"></script>
<script type="module">
  // You can then use it like this
  const { hasTransferables, getTransferables } = window.Transferables;
</script>

You can also use it via a CDN, e.g.

import { hasTransferables, getTransferables } from "https://cdn.skypack.dev/transferables";
// or
import { hasTransferables, getTransferables } from "https://cdn.jsdelivr.net/npm/transferables";
// or any number of other CDN's

Showcase

A couple sites/projects that use transferables:

• Your site/project here...

API

The API of transferables is pretty straight forward,

• hasTransferables quickly checks if the input contains at least one transferable object.
• getTransferable returns an iterator that contains the transferable objects from the input.
• getTransferables generates an array of transferable objects from the input.
• isSupported tests what transferable objects are actually supported (support isn't always guranteed) and returns a Promise which resolves to an object that represent if messagechannel and streams are supported.
• isObject, isTypedArray, isStream, isMessageChannel, isTransferable, and filterOutDuplicates are utility functions that are used internally by transferables, but can be used externally to customize transferables to match other use cases the transferables library itself doesn't.

You use the exported methods from the API like so,

import { hasTransferables, getTransferables, getTransferable } from "transferables";

// data is an object that contains transferable objects
const data = { /* ... */ }

// Quick check for transferable object
const containsTransferables = hasTransferables(data);

// Send postMessage with transferables, if they exist
const transferables = containsTransferables ? getTransferables(data) : undefined;
postMessage(data, transferables);

// Clone data with transferables, if they exist
const transferablesIterator = containsTransferables ? Array.from(getTransferable(data)) : undefined;
structuredClone(data, transferablesIterator);

import { 
  isSupported, 
  isObject, 
  isTypedArray, 
  isStream, 
  isMessageChannel, 
  isTransferable, 
  filterOutDuplicates 
} from "transferables";

// isSupported
isSupported(); // Promise<{ channel: true, streams: true }>

// isObject
isObject(data); // true

// isTypedArray
isTypedArray(data); // false

// isStream
isStream(data); // false

// isMessageChannel
isMessageChannel(data); // false

// isTransferable
isTransferable(data); // false

// filterOutDuplicates
filterOutDuplicates([1, 2, 3, 3, 4, 5, 5]); // [1, 2, 3, 4, 5]

Advanced Usage

/**
 * Quickly checks to see if input contains at least one transferable object, up to a max number of iterations
 * 
 * @param obj Input object
 * @param streams Include streams as transferable
 * @param maxCount Maximum number of iterations
 * @returns Whether input object contains transferable objects
 */
hasTransferables(data: unknown, streams: boolean, maxCount: number): boolean


/**
 * Creates an array of transferable objects which exist in a given input, up to a max number of iterations
 * ...
 * @returns An array of transferable objects
 */
getTransferables(data: unknown, streams: boolean, maxCount: number): TypeTransferable[]


/**
 * An iterator that contains the transferable objects from the input, up to a max number of iterations
 * ...
 * @returns Iterator that contains the transferable objects from the input
 */
getTransferable(data: unknown, streams: boolean, maxCount: number): Generator<TypeTransferable | TypeTypedArray | MessageChannel | DataView>

Look through the benchmark/ folder for complex examples, and multiple ways to use transferables across different js runtimes.

Note: (Readable/Writeable/Transform)streams and MessagePort aren't transferable in all js runtimes; devs can decide based off the runtime whether to support streams and message channel/port or not

Note: depending on how large your object is you may need go over the maxCount (max iteration count), if you need to change the max number of iterations remember that--that might cause the thread to be blocked while it's computing.

Benchmarks

Machine: GitHub Action ubuntu-latest

• 2-core CPU (x86_64)
• 7 GB of RAM
• 14 GB of SSD space

JS Runtimes:

• Node 19 - Run using vitest
• Deno 1.28.3
• Bun v0.2.2 - Run using vitest (it's basically a clone of the nodejs benchmark)
• Chrome (latest)
• Firefox (latest)
• Safari (latest)

To determine just how useful the transferables library was, I ran a benchmark, here are the results.

• [Node - Result][node-benchmark]
• Deno - Result
• Bun - Result
• Chrome - Result
• Firefox - Result
• Safari - Result

The benchmark ran using the 3 different types of object transfer.

We ran the benchmark with

1. structuredClone (All)
2. MessageChannel (All)
3. Worker (Deno, Chrome, Firefox, and Safari)

Note: WebWorker's aren't supported in all runtimes

Each type ran for 5 cycles, with a transfer list ranging from 108 - 168 objects per run (depending on the js environement). With 21 different data sizes ranging from 1 B to 1,049 MB in the transfer list, each cycle also has 5 variants.

The variants are,

• hasTransferables
• structuredClone | postMessage (no transfers) - postMessage doesn't actually require listing out objects in the transfer list, only structuredClone requires that; TIL
• structuredClone | postMessage (manually)
• structuredClone | postMessage (getTransferables)
• structuredClone | postMessage (getTransferable*)

Note: postMessage is for the MessageChannel and Worker types of object transfer.

Asterisks* & Limitations

There are things to be aware of when using transferables.

1. Not all transferable objects are supported in all browsers.
2. Not all transferable objects can be transfered between Workers and the main thread.
3. structuredClone when trying to clone an object that is transferable will crashes if the transferable objects aren't listed in the transfer list.
4. Only use this library when you don't know the shape of the object to be transfered. The reason for this is, traversing the input object adds a noticeable delay, you notice the delay as you go through the #benchmark.

Also, there are compatability issues js runtimes, here are the ones I've found so far,

• Safari does not support transferable objects with TransformStream, ReadableStream, and WritableStream
• AudioData & VideoFrame are not supported on Firefox and Safari
• OffscreenCanvas is not supported on Safari
• In a reverse uno card, only Safari supports RTCDataChannel being transferable
• Deno doesn't support transferable MessagePort

Note: isSupported() should help with some of the compatability issues, but not all transferable objects have been tested for compatability.

Transferable objects

The following are transferable objects:

• ArrayBuffer
• MessagePort
• ImageBitmap
• ReadableStream
• WritableStream
• TransformStream
• DataView
• AudioData
• ImageBitmap
• VideoFrame
• OffscreenCanvas
• RTCDataChannel

From the brief research I've done on the topic, I've found that

• ArrayBuffer: Can be transferred between Workers and the main thread. It's really the only type of transferable object that can be transferred reliably on all major js runtimes.
• TypedArray: A data view of an ArrayBuffer (e.g. Uint8Array, Int32Array, Float64Array, etc.). They can't directly be transferred between Workers and the main thread, but the ArrayBuffer they contain can. Due to this fact, it's possible if you have multiple TypedArray's that all share the same ArrayBuffer, that only that ArrayBuffer is transfered.
• MessagePort (~): A port to communicate with other workers. Can be transferred between Workers and the main thread. Support for this isn't guranteed in all js runtimes, and can be finicky in Deno
• ImageBitmap (^): An image that can be transferred between Workers and the main thread. It represents a bitmap image which can be drawn to a <canvas> without undue latency. It can also be used as textures in WebGL.
• OffscreenCanvas (^): A canvas that can be transferred between Workers and the main thread. It can also be used as a texture in WebGL.
• (Readable/Writable/Transform)Stream (~): A stream that can be transferred between Workers and the main thread. They can also be used to create Response objects. Support across js runtimes is very spotty

^ unverified/untested - Make sure to do your own research for this specific use case.

~ spotty support - Check below for js runtimes where it's ok to use

Here is a support matrix that might help your decision making process,

| | Chrome | Firefox | Safari | Node | Deno | Bun | | ---------------------------- | ------ | ------- | ------ | ------ | ------ | ------ | | structuredClone (channel) | false | false | false | true | true | true |
| structuredClone (streams) | true | true | false | true | false | true | | Worker.postMessage (channel) | false | false | false | - | true | - |
| Worker.postMessage (streams) | false | false | false | - | false | - |

FAQ & Glossary

What are transferable objects?

Transferable objects are objects that can be transferred between Workers and the main thread. It works sort of like ploping out the piece of memory attached to the Worker for the transferable object (e.g. an ArrayBuffer) and then moving that piece of memory to the main-thread for use by a newly created transferable objects and vice-versa. You can read more about them on the MDN docs.

Note: Notable exceptions to the transferable objects list are Blob and File objects, which are not transferable, but can be cloned.

Why should I use this?

The main use case of the transferables library is for determining when there is a transferable object and/or then listing said transferable objects out. A good example of when to use this is when working with structuredClone. structuredClone errors out when using transferables objects as they are not cloneable, e.g.

Warning: Remember the previous thread transferable objects are transfered from lose all access to the transfered data.

Warning: There is a performance threshold for transferable objects, before which using transferable objects becomes genuinly slower, it's probably not worth it to use this library if you reach that threshold #benchmark.

What is the difference between transferable objects and cloneable objects?

Transferable objects are objects that can be transferred between Workers and the main thread. They can be transferred from the main thread to a Worker, and vice versa. Cloneable objects are objects that can be cloned using the structured clone algorithim, due to not all objects being cloneable we use transferable objects to move transfer uncloneable object to the new cloned object, MDN - structured clone algorithim.

Browser Support

| Chrome | Edge | Firefox | Safari | | ------ | ---- | ------- | ------ | | 7+ | 12+ | 41+ | 5+ |

Native support for transferables is rather good, but due to not all browsers supporting all transferable objects actually determing browser support is more complex, #astericks covers these limitations.

Contributing

Thanks @aaorris for the helping optimizing the performance of the transferables library.

I encourage you to use pnpm to contribute to this repo, but you can also use yarn or npm if you prefer.

Install all necessary packages

npm install

Then run tests

npm test

Build project

npm run build

You can also run the benchmarks

npm run benchmark:node:all

To run the browser benchmarks,

npm run playwright:init &&
npm run benchmark:browser:all

To run the deno & bun benchmarks (install deno & bun)

npm run benchmark:deno:all &&
npm run benhmark:bun:all

Note: This project uses Conventional Commits standard for commits, so, please format your commits using the rules it sets out.

Licence

See the LICENSE file for license rights and limitations (MIT).