Streams Standard

1. Introduction

This section is non-normative.

Large swathes of the web platform are built on streaming data: that is, data that is created, processed, and consumed in an incremental fashion, without ever reading all of it into memory. The Streams Standard provides a common set of APIs for creating and interfacing with such streaming data, embodied in readable streams, writable streams, and transform streams.

These APIs have been designed to efficiently map to low-level I/O primitives, including specializations for byte streams where appropriate. They allow easy composition of multiple streams into pipe chains, or can be used directly via readers and writers. Finally, they are designed to automatically provide backpressure and queuing.

This standard provides the base stream primitives which other parts of the web platform can use to expose their streaming data. For example, [FETCH] exposes Response bodies as ReadableStream instances. More generally, the platform is full of streaming abstractions waiting to be expressed as streams: multimedia streams, file streams, inter-global communication, and more benefit from being able to process data incrementally instead of buffering it all into memory and processing it in one go. By providing the foundation for these streams to be exposed to developers, the Streams Standard enables use cases like:

Video effects: piping a readable video stream through a transform stream that applies effects in real time.
Decompression: piping a file stream through a transform stream that selectively decompresses files from a .tgz archive, turning them into img elements as the user scrolls through an image gallery.
Image decoding: piping an HTTP response stream through a transform stream that decodes bytes into bitmap data, and then through another transform that translates bitmaps into PNGs. If installed inside the fetch hook of a service worker, this would allow developers to transparently polyfill new image formats. [SERVICE-WORKERS]

Web developers can also use the APIs described here to create their own streams, with the same APIs as those provided by the platform. Other developers can then transparently compose platform-provided streams with those supplied by libraries. In this way, the APIs described here provide unifying abstraction for all streams, encouraging an ecosystem to grow around these shared and composable interfaces.

2. Model

A chunk is a single piece of data that is written to or read from a stream. It can be of any type; streams can even contain chunks of different types. A chunk will often not be the most atomic unit of data for a given stream; for example a byte stream might contain chunks consisting of 16 KiB Uint8Arrays, instead of single bytes.

2.1. Readable streams

A readable stream represents a source of data, from which you can read. In other words, data comes out of a readable stream. Concretely, a readable stream is an instance of the ReadableStream class.

Although a readable stream can be created with arbitrary behavior, most readable streams wrap a lower-level I/O source, called the underlying source. There are two types of underlying source: push sources and pull sources.

Push sources push data at you, whether or not you are listening for it. They may also provide a mechanism for pausing and resuming the flow of data. An example push source is a TCP socket, where data is constantly being pushed from the OS level, at a rate that can be controlled by changing the TCP window size.

Pull sources require you to request data from them. The data may be available synchronously, e.g. if it is held by the operating system’s in-memory buffers, or asynchronously, e.g. if it has to be read from disk. An example pull source is a file handle, where you seek to specific locations and read specific amounts.

Readable streams are designed to wrap both types of sources behind a single, unified interface. For web developer–created streams, the implementation details of a source are provided by an object with certain methods and properties that is passed to the ReadableStream() constructor.

Chunks are enqueued into the stream by the stream’s underlying source. They can then be read one at a time via the stream’s public interface, in particular by using a readable stream reader acquired using the stream’s getReader() method.

Code that reads from a readable stream using its public interface is known as a consumer.

Consumers also have the ability to cancel a readable stream, using its cancel() method. This indicates that the consumer has lost interest in the stream, and will immediately close the stream, throw away any queued chunks, and execute any cancellation mechanism of the underlying source.

Consumers can also tee a readable stream using its tee() method. This will lock the stream, making it no longer directly usable; however, it will create two new streams, called branches, which can be consumed independently.

For streams representing bytes, an extended version of the readable stream is provided to handle bytes efficiently, in particular by minimizing copies. The underlying source for such a readable stream is called an underlying byte source. A readable stream whose underlying source is an underlying byte source is sometimes called a readable byte stream. Consumers of a readable byte stream can acquire a BYOB reader using the stream’s getReader() method.

2.2. Writable streams

A writable stream represents a destination for data, into which you can write. In other words, data goes in to a writable stream. Concretely, a writable stream is an instance of the WritableStream class.

Analogously to readable streams, most writable streams wrap a lower-level I/O sink, called the underlying sink. Writable streams work to abstract away some of the complexity of the underlying sink, by queuing subsequent writes and only delivering them to the underlying sink one by one.

Chunks are written to the stream via its public interface, and are passed one at a time to the stream’s underlying sink. For web developer-created streams, the implementation details of the sink are provided by an object with certain methods that is passed to the WritableStream() constructor.

Code that writes into a writable stream using its public interface is known as a producer.

Producers also have the ability to abort a writable stream, using its abort() method. This indicates that the producer believes something has gone wrong, and that future writes should be discontinued. It puts the stream in an errored state, even without a signal from the underlying sink, and it discards all writes in the stream’s internal queue.

2.3. Transform streams

A transform stream consists of a pair of streams: a writable stream, known as its writable side, and a readable stream, known as its readable side. In a manner specific to the transform stream in question, writes to the writable side result in new data being made available for reading from the readable side.

Concretely, any object with a writable property and a readable property can serve as a transform stream. However, the standard TransformStream class makes it much easier to create such a pair that is properly entangled. It wraps a transformer, which defines algorithms for the specific transformation to be performed. For web developer–created streams, the implementation details of a transformer are provided by an object with certain methods and properties that is passed to the TransformStream() constructor. Other specifications might use the GenericTransformStream mixin to create classes with the same writable/readable property pair but other custom APIs layered on top.

An identity transform stream is a type of transform stream which forwards all chunks written to its writable side to its readable side, without any changes. This can be useful in a variety of scenarios. By default, the TransformStream constructor will create an identity transform stream, when no transform() method is present on the transformer object.

Some examples of potential transform streams include:

A GZIP compressor, to which uncompressed bytes are written and from which compressed bytes are read;
A video decoder, to which encoded bytes are written and from which uncompressed video frames are read;
A text decoder, to which bytes are written and from which strings are read;
A CSV-to-JSON converter, to which strings representing lines of a CSV file are written and from which corresponding JavaScript objects are read.

2.4. Pipe chains and backpressure

Streams are primarily used by piping them to each other. A readable stream can be piped directly to a writable stream, using its pipeTo() method, or it can be piped through one or more transform streams first, using its pipeThrough() method.

A set of streams piped together in this way is referred to as a pipe chain. In a pipe chain, the original source is the underlying source of the first readable stream in the chain; the ultimate sink is the underlying sink of the final writable stream in the chain.

Once a pipe chain is constructed, it will propagate signals regarding how fast chunks should flow through it. If any step in the chain cannot yet accept chunks, it propagates a signal backwards through the pipe chain, until eventually the original source is told to stop producing chunks so fast. This process of normalizing flow from the original source according to how fast the chain can process chunks is called backpressure.

Concretely, the original source is given the controller.desiredSize (or byteController.desiredSize) value, and can then adjust its rate of data flow accordingly. This value is derived from the writer.desiredSize corresponding to the ultimate sink, which gets updated as the ultimate sink finishes writing chunks. The pipeTo() method used to construct the chain automatically ensures this information propagates back through the pipe chain.

When teeing a readable stream, the backpressure signals from its two branches will aggregate, such that if neither branch is read from, a backpressure signal will be sent to the underlying source of the original stream.

Piping locks the readable and writable streams, preventing them from being manipulated for the duration of the pipe operation. This allows the implementation to perform important optimizations, such as directly shuttling data from the underlying source to the underlying sink while bypassing many of the intermediate queues.

2.5. Internal queues and queuing strategies

Both readable and writable streams maintain internal queues, which they use for similar purposes. In the case of a readable stream, the internal queue contains chunks that have been enqueued by the underlying source, but not yet read by the consumer. In the case of a writable stream, the internal queue contains chunks which have been written to the stream by the producer, but not yet processed and acknowledged by the underlying sink.

A queuing strategy is an object that determines how a stream should signal backpressure based on the state of its internal queue. The queuing strategy assigns a size to each chunk, and compares the total size of all chunks in the queue to a specified number, known as the high water mark. The resulting difference, high water mark minus total size, is used to determine the desired size to fill the stream’s queue.

For readable streams, an underlying source can use this desired size as a backpressure signal, slowing down chunk generation so as to try to keep the desired size above or at zero. For writable streams, a producer can behave similarly, avoiding writes that would cause the desired size to go negative.

Concretely, a queuing strategy for web developer–created streams is given by any JavaScript object with a highWaterMark property. For byte streams the highWaterMark always has units of bytes. For other streams the default unit is chunks, but a size() function can be included in the strategy object which returns the size for a given chunk. This permits the highWaterMark to be specified in arbitrary floating-point units.

A simple example of a queuing strategy would be one that assigns a size of one to each chunk, and has a high water mark of three. This would mean that up to three chunks could be enqueued in a readable stream, or three chunks written to a writable stream, before the streams are considered to be applying backpressure.

In JavaScript, such a strategy could be written manually as { highWaterMark: 3, size() { return 1; }}, or using the built-in CountQueuingStrategy class, as new CountQueuingStrategy({ highWaterMark: 3 }).

2.6. Locking

A readable stream reader, or simply reader, is an object that allows direct reading of chunks from a readable stream. Without a reader, a consumer can only perform high-level operations on the readable stream: canceling the stream, or piping the readable stream to a writable stream. A reader is acquired via the stream’s getReader() method.

A readable byte stream has the ability to vend two types of readers: default readers and BYOB readers. BYOB ("bring your own buffer") readers allow reading into a developer-supplied buffer, thus minimizing copies. A non-byte readable stream can only vend default readers. Default readers are instances of the ReadableStreamDefaultReader class, while BYOB readers are instances of ReadableStreamBYOBReader.

Similarly, a writable stream writer, or simply writer, is an object that allows direct writing of chunks to a writable stream. Without a writer, a producer can only perform the high-level operations of aborting the stream or piping a readable stream to the writable stream. Writers are represented by the WritableStreamDefaultWriter class.

Under the covers, these high-level operations actually use a reader or writer themselves.

A given readable or writable stream only has at most one reader or writer at a time. We say in this case the stream is locked, and that the reader or writer is active. This state can be determined using the readableStream.locked or writableStream.locked properties.

A reader or writer also has the capability to release its lock, which makes it no longer active, and allows further readers or writers to be acquired. This is done via the defaultReader.releaseLock(), byobReader.releaseLock(), or writer.releaseLock() method, as appropriate.

3. Conventions

This specification depends on the Infra Standard. [INFRA]

This specification uses the abstract operation concept from the JavaScript specification for its internal algorithms. This includes treating their return values as completion records, and the use of ! and ? prefixes for unwrapping those completion records. [ECMASCRIPT]

This specification also uses the internal slot concept and notation from the JavaScript specification. (Although, the internal slots are on Web IDL platform objects instead of on JavaScript objects.)

The reasons for the usage of these foreign JavaScript specification conventions are largely historical. We urge you to avoid following our example when writing your own web specifications.

In this specification, all numbers are represented as double-precision 64-bit IEEE 754 floating point values (like the JavaScript Number type or Web IDL unrestricted double type), and all arithmetic operations performed on them must be done in the standard way for such values. This is particularly important for the data structure described in § 8.1 Queue-with-sizes. [IEEE-754]

4. Readable streams

4.1. Using readable streams

The simplest way to consume a readable stream is to simply pipe it to a writable stream. This ensures that backpressure is respected, and any errors (either writing or reading) are propagated through the chain:

readableStream.pipeTo(writableStream)
  .then(() => console.log("All data successfully written!"))
  .catch(e => console.error("Something went wrong!", e));

If you simply want to be alerted of each new chunk from a readable stream, you can pipe it to a new writable stream that you custom-create for that purpose:

readableStream.pipeTo(new WritableStream({
  write(chunk) {
    console.log("Chunk received", chunk);
  },
  close() {
    console.log("All data successfully read!");
  },
  abort(e) {
    console.error("Something went wrong!", e);
  }
}));

By returning promises from your write() implementation, you can signal backpressure to the readable stream.

Although readable streams will usually be used by piping them to a writable stream, you can also read them directly by acquiring a reader and using its read() method to get successive chunks. For example, this code logs the next chunk in the stream, if available:

const reader = readableStream.getReader();

reader.read().then(
  ({ value, done }) => {
    if (done) {
      console.log("The stream was already closed!");
    } else {
      console.log(value);
    }
  },
  e => console.error("The stream became errored and cannot be read from!", e)
);

This more manual method of reading a stream is mainly useful for library authors building new high-level operations on streams, beyond the provided ones of piping and teeing.

The above example showed using the readable stream’s default reader. If the stream is a readable byte stream, you can also acquire a BYOB reader for it, which allows more precise control over buffer allocation in order to avoid copies. For example, this code reads the first 1024 bytes from the stream into a single memory buffer:

const reader = readableStream.getReader({ mode: "byob" });

let startingAB = new ArrayBuffer(1024);
const buffer = await readInto(startingAB);
console.log("The first 1024 bytes: ", buffer);

async function readInto(buffer) {
  let offset = 0;

  while (offset < buffer.byteLength) {
    const { value: view, done } =
     await reader.read(new Uint8Array(buffer, offset, buffer.byteLength - offset));
    buffer = view.buffer;
    if (done) {
      break;
    }
    offset += view.byteLength;
  }

  return buffer;
}

An important thing to note here is that the final buffer value is different from the startingAB, but it (and all intermediate buffers) shares the same backing memory allocation. At each step, the buffer is transferred to a new ArrayBuffer object. The view is destructured from the return value of reading a new Uint8Array, with that ArrayBuffer object as its buffer property, the offset that bytes were written to as its byteOffset property, and the number of bytes that were written as its byteLength property.

Note that this example is mostly educational. For practical purposes, the min option of read() provides an easier and more direct way to read an exact number of bytes:

const reader = readableStream.getReader({ mode: "byob" });
const { value: view, done } = await reader.read(new Uint8Array(1024), { min: 1024 });
console.log("The first 1024 bytes: ", view);

4.2. The `ReadableStream` class

The ReadableStream class is a concrete instance of the general readable stream concept. It is adaptable to any chunk type, and maintains an internal queue to keep track of data supplied by the underlying source but not yet read by any consumer.

4.2.1. Interface definition

The Web IDL definition for the ReadableStream class is given as follows:

[Exposed=*, Transferable]
interface ReadableStream {
  constructor(optional object underlyingSource, optional QueuingStrategy strategy = {});

  static ReadableStream from(any asyncIterable);

  readonly attribute boolean locked;

  Promise<undefined> cancel(optional any reason);
  ReadableStreamReader getReader(optional ReadableStreamGetReaderOptions options = {});
  ReadableStream pipeThrough(ReadableWritablePair transform, optional StreamPipeOptions options = {});
  Promise<undefined> pipeTo(WritableStream destination, optional StreamPipeOptions options = {});
  sequence<ReadableStream> tee();

  async iterable<any>(optional ReadableStreamIteratorOptions options = {});
};

typedef (ReadableStreamDefaultReader or ReadableStreamBYOBReader) ReadableStreamReader;

enum ReadableStreamReaderMode { "byob" };

dictionary ReadableStreamGetReaderOptions {
  ReadableStreamReaderMode mode;
};

dictionary ReadableStreamIteratorOptions {
  boolean preventCancel = false;
};

dictionary ReadableWritablePair {
  required ReadableStream readable;
  required WritableStream writable;
};

dictionary StreamPipeOptions {
  boolean preventClose = false;
  boolean preventAbort = false;
  boolean preventCancel = false;
  AbortSignal signal;
};

4.2.2. Internal slots

Instances of ReadableStream are created with the internal slots described in the following table:

Internal Slot	Description (non-normative)
[[controller]]	A `ReadableStreamDefaultController` or `ReadableByteStreamController` created with the ability to control the state and queue of this stream
[[Detached]]	A boolean flag set to true when the stream is transferred
[[disturbed]]	A boolean flag set to true when the stream has been read from or canceled
[[reader]]	A `ReadableStreamDefaultReader` or `ReadableStreamBYOBReader` instance, if the stream is locked to a reader, or undefined if it is not
[[state]]	A string containing the stream’s current state, used internally; one of "`readable`", "`closed`", or "`errored`"
[[storedError]]	A value indicating how the stream failed, to be given as a failure reason or exception when trying to operate on an errored stream

4.2.3. The underlying source API

The ReadableStream() constructor accepts as its first argument a JavaScript object representing the underlying source. Such objects can contain any of the following properties:

dictionary UnderlyingSource {
  UnderlyingSourceStartCallback start;
  UnderlyingSourcePullCallback pull;
  UnderlyingSourceCancelCallback cancel;
  ReadableStreamType type;
  [EnforceRange] unsigned long long autoAllocateChunkSize;
};

typedef (ReadableStreamDefaultController or ReadableByteStreamController) ReadableStreamController;

callback UnderlyingSourceStartCallback = any (ReadableStreamController controller);
callback UnderlyingSourcePullCallback = Promise<undefined> (ReadableStreamController controller);
callback UnderlyingSourceCancelCallback = Promise<undefined> (optional any reason);

enum ReadableStreamType { "bytes" };

start(controller), of type UnderlyingSourceStartCallback

A function that is called immediately during creation of the ReadableStream.

Typically this is used to adapt a push source by setting up relevant event listeners, as in the example of § 10.1 A readable stream with an underlying push source (no backpressure support), or to acquire access to a pull source, as in § 10.4 A readable stream with an underlying pull source.

If this setup process is asynchronous, it can return a promise to signal success or failure; a rejected promise will error the stream. Any thrown exceptions will be re-thrown by the ReadableStream() constructor.

pull(controller), of type UnderlyingSourcePullCallback

A function that is called whenever the stream’s internal queue of chunks becomes not full, i.e. whenever the queue’s desired size becomes positive. Generally, it will be called repeatedly until the queue reaches its high water mark (i.e. until the desired size becomes non-positive).

For push sources, this can be used to resume a paused flow, as in § 10.2 A readable stream with an underlying push source and backpressure support. For pull sources, it is used to acquire new chunks to enqueue into the stream, as in § 10.4 A readable stream with an underlying pull source.

This function will not be called until start() successfully completes. Additionally, it will only be called repeatedly if it enqueues at least one chunk or fulfills a BYOB request; a no-op pull() implementation will not be continually called.

If the function returns a promise, then it will not be called again until that promise fulfills. (If the promise rejects, the stream will become errored.) This is mainly used in the case of pull sources, where the promise returned represents the process of acquiring a new chunk. Throwing an exception is treated the same as returning a rejected promise.

cancel(reason), of type UnderlyingSourceCancelCallback

A function that is called whenever the consumer cancels the stream, via stream.cancel() or reader.cancel(). It takes as its argument the same value as was passed to those methods by the consumer.

Readable streams can additionally be canceled under certain conditions during piping; see the definition of the pipeTo() method for more details.

For all streams, this is generally used to release access to the underlying resource; see for example § 10.1 A readable stream with an underlying push source (no backpressure support).

If the shutdown process is asynchronous, it can return a promise to signal success or failure; the result will be communicated via the return value of the cancel() method that was called. Throwing an exception is treated the same as returning a rejected promise.

Even if the cancelation process fails, the stream will still close; it will not be put into an errored state. This is because a failure in the cancelation process doesn’t matter to the consumer’s view of the stream, once they’ve expressed disinterest in it by canceling. The failure is only communicated to the immediate caller of the corresponding method.

This is different from the behavior of the close and abort options of a WritableStream's underlying sink, which upon failure put the corresponding WritableStream into an errored state. Those correspond to specific actions the producer is requesting and, if those actions fail, they indicate something more persistently wrong.

type (byte streams only), of type ReadableStreamType

Can be set to "bytes" to signal that the constructed ReadableStream is a readable byte stream. This ensures that the resulting ReadableStream will successfully be able to vend BYOB readers via its getReader() method. It also affects the controller argument passed to the start() and pull() methods; see below.

For an example of how to set up a readable byte stream, including using the different controller interface, see § 10.3 A readable byte stream with an underlying push source (no backpressure support).

Setting any value other than "bytes" or undefined will cause the ReadableStream() constructor to throw an exception.

autoAllocateChunkSize (byte streams only), of type unsigned long long

Can be set to a positive integer to cause the implementation to automatically allocate buffers for the underlying source code to write into. In this case, when a consumer is using a default reader, the stream implementation will automatically allocate an ArrayBuffer of the given size, so that controller.byobRequest is always present, as if the consumer was using a BYOB reader.

This is generally used to cut down on the amount of code needed to handle consumers that use default readers, as can be seen by comparing § 10.3 A readable byte stream with an underlying push source (no backpressure support) without auto-allocation to § 10.5 A readable byte stream with an underlying pull source with auto-allocation.

The type of the controller argument passed to the start() and pull() methods depends on the value of the type option. If type is set to undefined (including via omission), then controller will be a ReadableStreamDefaultController. If it’s set to "bytes", then controller will be a ReadableByteStreamController.

4.2.4. Constructor, methods, and properties

stream = new ReadableStream(underlyingSource[, strategy])

Creates a new ReadableStream wrapping the provided underlying source. See § 4.2.3 The underlying source API for more details on the underlyingSource argument.

The strategy argument represents the stream’s queuing strategy, as described in § 7.1 The queuing strategy API. If it is not provided, the default behavior will be the same as a CountQueuingStrategy with a high water mark of 1.

stream = ReadableStream.from(asyncIterable)

Creates a new ReadableStream wrapping the provided iterable or async iterable.

This can be used to adapt various kinds of objects into a readable stream, such as an array, an async generator, or a Node.js readable stream.

isLocked = stream.locked

Returns whether or not the readable stream is locked to a reader.

await stream.cancel([ reason ])

Cancels the stream, signaling a loss of interest in the stream by a consumer. The supplied reason argument will be given to the underlying source’s cancel() method, which might or might not use it.

The returned promise will fulfill if the stream shuts down successfully, or reject if the underlying source signaled that there was an error doing so. Additionally, it will reject with a TypeError (without attempting to cancel the stream) if the stream is currently locked.

reader = stream.getReader()

Creates a ReadableStreamDefaultReader and locks the stream to the new reader. While the stream is locked, no other reader can be acquired until this one is released.

This functionality is especially useful for creating abstractions that desire the ability to consume a stream in its entirety. By getting a reader for the stream, you can ensure nobody else can interleave reads with yours or cancel the stream, which would interfere with your abstraction.

reader = stream.getReader({ mode: "byob" })

Creates a ReadableStreamBYOBReader and locks the stream to the new reader.

This call behaves the same way as the no-argument variant, except that it only works on readable byte streams, i.e. streams which were constructed specifically with the ability to handle "bring your own buffer" reading. The returned BYOB reader provides the ability to directly read individual chunks from the stream via its read() method, into developer-supplied buffers, allowing more precise control over allocation.

readable = stream.pipeThrough({ writable, readable }[, { preventClose, preventAbort, preventCancel, signal }])

Provides a convenient, chainable way of piping this readable stream through a transform stream (or any other { writable, readable } pair). It simply pipes the stream into the writable side of the supplied pair, and returns the readable side for further use.

Piping a stream will lock it for the duration of the pipe, preventing any other consumer from acquiring a reader.

await stream.pipeTo(destination[, { preventClose, preventAbort, preventCancel, signal }])

Pipes this readable stream to a given writable stream destination. The way in which the piping process behaves under various error conditions can be customized with a number of passed options. It returns a promise that fulfills when the piping process completes successfully, or rejects if any errors were encountered.

Piping a stream will lock it for the duration of the pipe, preventing any other consumer from acquiring a reader.

Errors and closures of the source and destination streams propagate as follows:

An error in this source readable stream will abort destination, unless preventAbort is truthy. The returned promise will be rejected with the source’s error, or with any error that occurs during aborting the destination.
An error in destination will cancel this source readable stream, unless preventCancel is truthy. The returned promise will be rejected with the destination’s error, or with any error that occurs during canceling the source.
When this source readable stream closes, destination will be closed, unless preventClose is truthy. The returned promise will be fulfilled once this process completes, unless an error is encountered while closing the destination, in which case it will be rejected with that error.
If destination starts out closed or closing, this source readable stream will be canceled, unless preventCancel is true. The returned promise will be rejected with an error indicating piping to a closed stream failed, or with any error that occurs during canceling the source.

The signal option can be set to an AbortSignal to allow aborting an ongoing pipe operation via the corresponding AbortController. In this case, this source readable stream will be canceled, and destination aborted, unless the respective options preventCancel or preventAbort are set.

[branch1, branch2] = stream.tee()

Tees this readable stream, returning a two-element array containing the two resulting branches as new ReadableStream instances.

Teeing a stream will lock it, preventing any other consumer from acquiring a reader. To cancel the stream, cancel both of the resulting branches; a composite cancellation reason will then be propagated to the stream’s underlying source.

If this stream is a readable byte stream, then each branch will receive its own copy of each chunk. If not, then the chunks seen in each branch will be the same object. If the chunks are not immutable, this could allow interference between the two branches.

The new ReadableStream(underlyingSource, strategy) constructor steps are:

If underlyingSource is missing, set it to null.
Let underlyingSourceDict be underlyingSource, converted to an IDL value of type UnderlyingSource.

We cannot declare the underlyingSource argument as having the UnderlyingSource type directly, because doing so would lose the reference to the original object. We need to retain the object so we can invoke the various methods on it.
Perform ! InitializeReadableStream(this).
If underlyingSourceDict["type"] is "bytes":
1. If strategy["size"] exists, throw a RangeError exception.
2. Let highWaterMark be ? ExtractHighWaterMark(strategy, 0).
3. Perform ? SetUpReadableByteStreamControllerFromUnderlyingSource(this, underlyingSource, underlyingSourceDict, highWaterMark).
Otherwise,
1. Assert: underlyingSourceDict["type"] does not exist.
2. Let sizeAlgorithm be ! ExtractSizeAlgorithm(strategy).
3. Let highWaterMark be ? ExtractHighWaterMark(strategy, 1).
4. Perform ? SetUpReadableStreamDefaultControllerFromUnderlyingSource(this, underlyingSource, underlyingSourceDict, highWaterMark, sizeAlgorithm).

The static from(asyncIterable) method steps are:

Return ? ReadableStreamFromIterable(asyncIterable).

The locked getter steps are:

Return ! IsReadableStreamLocked(this).

The cancel(reason) method steps are:

If ! IsReadableStreamLocked(this) is true, return a promise rejected with a TypeError exception.
Return ! ReadableStreamCancel(this, reason).

The getReader(options) method steps are:

If options["mode"] does not exist, return ? AcquireReadableStreamDefaultReader(this).
Assert: options["mode"] is "byob".
Return ? AcquireReadableStreamBYOBReader(this).

An example of an abstraction that might benefit from using a reader is a function like the following, which is designed to read an entire readable stream into memory as an array of chunks.

function readAllChunks(readableStream) {
  const reader = readableStream.getReader();
  const chunks = [];

  return pump();

  function pump() {
    return reader.read().then(({ value, done }) => {
      if (done) {
        return chunks;
      }

      chunks.push(value);
      return pump();
    });
  }
}

Note how the first thing it does is obtain a reader, and from then on it uses the reader exclusively. This ensures that no other consumer can interfere with the stream, either by reading chunks or by canceling the stream.

The pipeThrough(transform, options) method steps are:

If ! IsReadableStreamLocked(this) is true, throw a TypeError exception.
If ! IsWritableStreamLocked(transform["writable"]) is true, throw a TypeError exception.
Let signal be options["signal"] if it exists, or undefined otherwise.
Let promise be ! ReadableStreamPipeTo(this, transform["writable"], options["preventClose"], options["preventAbort"], options["preventCancel"], signal).
Set promise.[[PromiseIsHandled]] to true.
Return transform["readable"].

A typical example of constructing pipe chain using

pipeThrough(transform,
  options)

would look like

httpResponseBody
  .pipeThrough(decompressorTransform)
  .pipeThrough(ignoreNonImageFilesTransform)
  .pipeTo(mediaGallery);

The pipeTo(destination, options) method steps are:

If ! IsReadableStreamLocked(this) is true, return a promise rejected with a TypeError exception.
If ! IsWritableStreamLocked(destination) is true, return a promise rejected with a TypeError exception.
Let signal be options["signal"] if it exists, or undefined otherwise.
Return ! ReadableStreamPipeTo(this, destination, options["preventClose"], options["preventAbort"], options["preventCancel"], signal).

An ongoing pipe operation can be stopped using an AbortSignal, as follows:

const controller = new AbortController();
readable.pipeTo(writable, { signal: controller.signal });

// ... some time later ...
controller.abort();

(The above omits error handling for the promise returned by pipeTo(). Additionally, the impact of the preventAbort and preventCancel options what happens when piping is stopped are worth considering.)

The above technique can be used to switch the ReadableStream being piped, while writing into the same WritableStream:

const controller = new AbortController();
const pipePromise = readable1.pipeTo(writable, { preventAbort: true, signal: controller.signal });

// ... some time later ...
controller.abort();

// Wait for the pipe to complete before starting a new one:
try {
 await pipePromise;
} catch (e) {
 // Swallow "AbortError" DOMExceptions as expected, but rethrow any unexpected failures.
 if (e.name !== "AbortError") {
  throw e;
 }
}

// Start the new pipe!
readable2.pipeTo(writable);

The tee() method steps are:

Return ? ReadableStreamTee(this, false).

Teeing a stream is most useful when you wish to let two independent consumers read from the stream in parallel, perhaps even at different speeds. For example, given a writable stream cacheEntry representing an on-disk file, and another writable stream httpRequestBody representing an upload to a remote server, you could pipe the same readable stream to both destinations at once:

const [forLocal, forRemote] = readableStream.tee();

Promise.all([
  forLocal.pipeTo(cacheEntry),
  forRemote.pipeTo(httpRequestBody)
])
.then(() => console.log("Saved the stream to the cache and also uploaded it!"))
.catch(e => console.error("Either caching or uploading failed: ", e));

4.2.5. Asynchronous iteration

for await (const chunk of stream) { ... }

for await (const chunk of stream.values({ preventCancel: true })) { ... }

Asynchronously iterates over the chunks in the stream’s internal queue.

Asynchronously iterating over the stream will lock it, preventing any other consumer from acquiring a reader. The lock will be released if the async iterator’s return() method is called, e.g. by breaking out of the loop.

By default, calling the async iterator’s return() method will also cancel the stream. To prevent this, use the stream’s values() method, passing true for the preventCancel option.

The asynchronous iterator initialization steps for a ReadableStream, given stream, iterator, and args, are:

Let reader be ? AcquireReadableStreamDefaultReader(stream).
Set iterator’s reader to reader.
Let preventCancel be args[0]["preventCancel"].
Set iterator’s prevent cancel to preventCancel.

The get the next iteration result steps for a ReadableStream, given stream and iterator, are:

Let reader be iterator’s reader.
Assert: reader.[[stream]] is not undefined.
Let promise be a new promise.
Let readRequest be a new read request with the following items:
chunk steps, given chunk
1. Resolve promise with chunk.
close steps
1. Perform ! ReadableStreamDefaultReaderRelease(reader).
2. Resolve promise with end of iteration.
error steps, given e
1. Perform ! ReadableStreamDefaultReaderRelease(reader).
2. Reject promise with e.
Perform ! ReadableStreamDefaultReaderRead(this, readRequest).
Return promise.

The asynchronous iterator return steps for a ReadableStream, given stream, iterator, and arg, are:

Let reader be iterator’s reader.
Assert: reader.[[stream]] is not undefined.
Assert: reader.[[readRequests]] is empty, as the async iterator machinery guarantees that any previous calls to next() have settled before this is called.
If iterator’s prevent cancel is false:
1. Let result be ! ReadableStreamReaderGenericCancel(reader, arg).
2. Perform ! ReadableStreamDefaultReaderRelease(reader).
3. Return result.
Perform ! ReadableStreamDefaultReaderRelease(reader).
Return a promise resolved with undefined.

4.2.6. Transfer via `postMessage()`

destination.postMessage(rs, { transfer: [rs] });

Sends a ReadableStream to another frame, window, or worker.

The transferred stream can be used exactly like the original. The original will become locked and no longer directly usable.

ReadableStream objects are transferable objects. Their transfer steps, given value and dataHolder, are:

If ! IsReadableStreamLocked(value) is true, throw a "DataCloneError" DOMException.
Let port1 be a new MessagePort in the current Realm.
Let port2 be a new MessagePort in the current Realm.
Entangle port1 and port2.
Let writable be a new WritableStream in the current Realm.
Perform ! SetUpCrossRealmTransformWritable(writable, port1).
Let promise be ! ReadableStreamPipeTo(value, writable, false, false, false).
Set promise.[[PromiseIsHandled]] to true.
Set dataHolder.[[port]] to ! StructuredSerializeWithTransfer(port2, « port2 »).

Their transfer-receiving steps, given dataHolder and value, are:

Let deserializedRecord be ! StructuredDeserializeWithTransfer(dataHolder.[[port]], the current Realm).
Let port be deserializedRecord.[[Deserialized]].
Perform ! SetUpCrossRealmTransformReadable(value, port).

4.3. The `ReadableStreamGenericReader` mixin

The ReadableStreamGenericReader mixin defines common internal slots, getters and methods that are shared between ReadableStreamDefaultReader and ReadableStreamBYOBReader objects.

4.3.1. Mixin definition

The Web IDL definition for the ReadableStreamGenericReader mixin is given as follows:

interface mixin ReadableStreamGenericReader {
  readonly attribute Promise<undefined> closed;

  Promise<undefined> cancel(optional any reason);
};

4.3.2. Internal slots

Instances of classes including the ReadableStreamGenericReader mixin are created with the internal slots described in the following table:

Internal Slot	Description (non-normative)
[[closedPromise]]	A promise returned by the reader’s `closed` getter
[[stream]]	A `ReadableStream` instance that owns this reader

4.3.3. Methods and properties

The closed getter steps are:

Return this.[[closedPromise]].

The cancel(reason) method steps are:

If this.[[stream]] is undefined, return a promise rejected with a TypeError exception.
Return ! ReadableStreamReaderGenericCancel(this, reason).

4.4. The `ReadableStreamDefaultReader` class

The ReadableStreamDefaultReader class represents a default reader designed to be vended by a ReadableStream instance.

4.4.1. Interface definition

The Web IDL definition for the ReadableStreamDefaultReader class is given as follows:

[Exposed=*]
interface ReadableStreamDefaultReader {
  constructor(ReadableStream stream);

  Promise<ReadableStreamReadResult> read();
  undefined releaseLock();
};
ReadableStreamDefaultReader includes ReadableStreamGenericReader;

dictionary ReadableStreamReadResult {
  any value;
  boolean done;
};

4.4.2. Internal slots

Instances of ReadableStreamDefaultReader are created with the internal slots defined by ReadableStreamGenericReader, and those described in the following table:

Internal Slot	Description (non-normative)
[[readRequests]]	A list of read requests, used when a consumer requests chunks sooner than they are available

A read request is a struct containing three algorithms to perform in reaction to filling the readable stream's internal queue or changing its state. It has the following items:

chunk steps: An algorithm taking a chunk, called when a chunk is available for reading
close steps: An algorithm taking no arguments, called when no chunks are available because the stream is closed
error steps: An algorithm taking a JavaScript value, called when no chunks are available because the stream is errored

4.4.3. Constructor, methods, and properties

reader = new ReadableStreamDefaultReader(stream)

This is equivalent to calling stream.getReader().

await reader.closed

Returns a promise that will be fulfilled when the stream becomes closed, or rejected if the stream ever errors or the reader’s lock is released before the stream finishes closing.

await reader.cancel([ reason ])

If the reader is active, behaves the same as stream.cancel(reason).

{ value, done } = await reader.read()

Returns a promise that allows access to the next chunk from the stream’s internal queue, if available.

If the chunk does become available, the promise will be fulfilled with an object of the form { value: theChunk, done: false }.
If the stream becomes closed, the promise will be fulfilled with an object of the form { value: undefined, done: true }.
If the stream becomes errored, the promise will be rejected with the relevant error.

If reading a chunk causes the queue to become empty, more data will be pulled from the underlying source.

reader.releaseLock()

Releases the reader’s lock on the corresponding stream. After the lock is released, the reader is no longer active. If the associated stream is errored when the lock is released, the reader will appear errored in the same way from now on; otherwise, the reader will appear closed.

If the reader’s lock is released while it still has pending read requests, then the promises returned by the reader’s read() method are immediately rejected with a TypeError. Any unread chunks remain in the stream’s internal queue and can be read later by acquiring a new reader.

The new ReadableStreamDefaultReader(stream) constructor steps are:

Perform ? SetUpReadableStreamDefaultReader(this, stream).

The read() method steps are:

If this.[[stream]] is undefined, return a promise rejected with a TypeError exception.
Let promise be a new promise.
Let readRequest be a new read request with the following items:
chunk steps, given chunk
1. Resolve promise with «[ "value" → chunk, "done" → false ]».
close steps
1. Resolve promise with «[ "value" → undefined, "done" → true ]».
error steps, given e
1. Reject promise with e.
Perform ! ReadableStreamDefaultReaderRead(this, readRequest).
Return promise.

The releaseLock() method steps are:

If this.[[stream]] is undefined, return.
Perform ! ReadableStreamDefaultReaderRelease(this).

4.5. The `ReadableStreamBYOBReader` class

The ReadableStreamBYOBReader class represents a BYOB reader designed to be vended by a ReadableStream instance.

4.5.1. Interface definition

The Web IDL definition for the ReadableStreamBYOBReader class is given as follows:

[Exposed=*]
interface ReadableStreamBYOBReader {
  constructor(ReadableStream stream);

  Promise<ReadableStreamReadResult> read(ArrayBufferView view, optional ReadableStreamBYOBReaderReadOptions options = {});
  undefined releaseLock();
};
ReadableStreamBYOBReader includes ReadableStreamGenericReader;

dictionary ReadableStreamBYOBReaderReadOptions {
  [EnforceRange] unsigned long long min = 1;
};

4.5.2. Internal slots

Instances of ReadableStreamBYOBReader are created with the internal slots defined by ReadableStreamGenericReader, and those described in the following table:

Internal Slot	Description (non-normative)
[[readIntoRequests]]	A list of read-into requests, used when a consumer requests chunks sooner than they are available

A read-into request is a struct containing three algorithms to perform in reaction to filling the readable byte stream's internal queue or changing its state. It has the following items:

chunk steps: An algorithm taking a chunk, called when a chunk is available for reading
close steps: An algorithm taking a chunk or undefined, called when no chunks are available because the stream is closed
error steps: An algorithm taking a JavaScript value, called when no chunks are available because the stream is errored

The close steps take a chunk so that it can return the backing memory to the caller if possible. For example, byobReader.read(chunk) will fulfill with { value: newViewOnSameMemory, done: true } for closed streams. If the stream is canceled, the backing memory is discarded and byobReader.read(chunk) fulfills with the more traditional { value: undefined, done: true } instead.

4.5.3. Constructor, methods, and properties

reader = new ReadableStreamBYOBReader(stream)

This is equivalent to calling stream.getReader({ mode: "byob" }).

await reader.closed

Returns a promise that will be fulfilled when the stream becomes closed, or rejected if the stream ever errors or the reader’s lock is released before the stream finishes closing.

await reader.cancel([ reason ])

If the reader is active, behaves the same stream.cancel(reason).

{ value, done } = await reader.read(view[, { min }])

Attempts to read bytes into view, and returns a promise resolved with the result:

If the chunk does become available, the promise will be fulfilled with an object of the form { value: newView, done: false }. In this case, view will be detached and no longer usable, but newView will be a new view (of the same type) onto the same backing memory region, with the chunk’s data written into it.
If the stream becomes closed, the promise will be fulfilled with an object of the form { value: newView, done: true }. In this case, view will be detached and no longer usable, but newView will be a new view (of the same type) onto the same backing memory region, with no modifications, to ensure the memory is returned to the caller.
If the reader is canceled, the promise will be fulfilled with an object of the form { value: undefined, done: true }. In this case, the backing memory region of view is discarded and not returned to the caller.
If the stream becomes errored, the promise will be rejected with the relevant error.

If reading a chunk causes the queue to become empty, more data will be pulled from the underlying source.

If min is given, then the promise will only be fulfilled as soon as the given minimum number of elements are available. Here, the "number of elements" is given by newView’s length (for typed arrays) or newView’s byteLength (for DataViews). If the stream becomes closed, then the promise is fulfilled with the remaining elements in the stream, which might be fewer than the initially requested amount. If not given, then the promise resolves when at least one element is available.

reader.releaseLock()

The new ReadableStreamBYOBReader(stream) constructor steps are:

Perform ? SetUpReadableStreamBYOBReader(this, stream).

The read(view, options) method steps are:

If view.[[ByteLength]] is 0, return a promise rejected with a TypeError exception.
If view.[[ViewedArrayBuffer]].[[ArrayBufferByteLength]] is 0, return a promise rejected with a TypeError exception.
If ! IsDetachedBuffer(view.[[ViewedArrayBuffer]]) is true, return a promise rejected with a TypeError exception.
If options["min"] is 0, return a promise rejected with a TypeError exception.
If view has a [[TypedArrayName]] internal slot,
1. If options["min"] > view.[[ArrayLength]], return a promise rejected with a RangeError exception.
Otherwise (i.e., it is a DataView),
1. If options["min"] > view.[[ByteLength]], return a promise rejected with a RangeError exception.
If this.[[stream]] is undefined, return a promise rejected with a TypeError exception.
Let promise be a new promise.
Let readIntoRequest be a new read-into request with the following items:
chunk steps, given chunk
1. Resolve promise with «[ "value" → chunk, "done" → false ]».
close steps, given chunk
1. Resolve promise with «[ "value" → chunk, "done" → true ]».
error steps, given e
1. Reject promise with e.
Perform ! ReadableStreamBYOBReaderRead(this, view, options["min"], readIntoRequest).
Return promise.

The releaseLock() method steps are:

If this.[[stream]] is undefined, return.
Perform ! ReadableStreamBYOBReaderRelease(this).

4.6. The `ReadableStreamDefaultController` class

The ReadableStreamDefaultController class has methods that allow control of a ReadableStream's state and internal queue. When constructing a ReadableStream that is not a readable byte stream, the underlying source is given a corresponding ReadableStreamDefaultController instance to manipulate.

4.6.1. Interface definition

The Web IDL definition for the ReadableStreamDefaultController class is given as follows:

[Exposed=*]
interface ReadableStreamDefaultController {
  readonly attribute unrestricted double? desiredSize;

  undefined close();
  undefined enqueue(optional any chunk);
  undefined error(optional any e);
};

4.6.2. Internal slots

Instances of ReadableStreamDefaultController are created with the internal slots described in the following table:

Internal Slot	Description (non-normative)
[[cancelAlgorithm]]	A promise-returning algorithm, taking one argument (the cancel reason), which communicates a requested cancelation to the underlying source
[[closeRequested]]	A boolean flag indicating whether the stream has been closed by its underlying source, but still has chunks in its internal queue that have not yet been read
[[pullAgain]]	A boolean flag set to true if the stream’s mechanisms requested a call to the underlying source's pull algorithm to pull more data, but the pull could not yet be done since a previous call is still executing
[[pullAlgorithm]]	A promise-returning algorithm that pulls data from the underlying source
[[pulling]]	A boolean flag set to true while the underlying source's pull algorithm is executing and the returned promise has not yet fulfilled, used to prevent reentrant calls
[[queue]]	A list representing the stream’s internal queue of chunks
[[queueTotalSize]]	The total size of all the chunks stored in [[queue]] (see § 8.1 Queue-with-sizes)
[[started]]	A boolean flag indicating whether the underlying source has finished starting
[[strategyHWM]]	A number supplied to the constructor as part of the stream’s queuing strategy, indicating the point at which the stream will apply backpressure to its underlying source
[[strategySizeAlgorithm]]	An algorithm to calculate the size of enqueued chunks, as part of the stream’s queuing strategy
[[stream]]	The `ReadableStream` instance controlled

4.6.3. Methods and properties

desiredSize = controller.desiredSize: Returns the desired size to fill the controlled stream’s internal queue. It can be negative, if the queue is over-full. An underlying source ought to use this information to determine when and how to apply backpressure.
controller.close(): Closes the controlled readable stream. Consumers will still be able to read any previously-enqueued chunks from the stream, but once those are read, the stream will become closed.
controller.enqueue(chunk): Enqueues the given chunk chunk in the controlled readable stream.
controller.error(e): Errors the controlled readable stream, making all future interactions with it fail with the given error e.

The desiredSize getter steps are:

Return ! ReadableStreamDefaultControllerGetDesiredSize(this).

The close() method steps are:

If ! ReadableStreamDefaultControllerCanCloseOrEnqueue(this) is false, throw a TypeError exception.
Perform ! ReadableStreamDefaultControllerClose(this).

The enqueue(chunk) method steps are:

If ! ReadableStreamDefaultControllerCanCloseOrEnqueue(this) is false, throw a TypeError exception.
Perform ? ReadableStreamDefaultControllerEnqueue(this, chunk).

The error(e) method steps are:

Perform ! ReadableStreamDefaultControllerError(this, e).

4.6.4. Internal methods

The following are internal methods implemented by each ReadableStreamDefaultController instance. The readable stream implementation will polymorphically call to either these, or to their counterparts for BYOB controllers, as discussed in § 4.9.2 Interfacing with controllers.

[[CancelSteps]](reason) implements the [[CancelSteps]] contract. It performs the following steps:

Perform ! ResetQueue(this).
Let result be the result of performing this.[[cancelAlgorithm]], passing reason.
Perform ! ReadableStreamDefaultControllerClearAlgorithms(this).
Return result.

[[PullSteps]](readRequest) implements the [[PullSteps]] contract. It performs the following steps:

Let stream be this.[[stream]].
If this.[[queue]] is not empty,
1. Let chunk be ! DequeueValue(this).
2. If this.[[closeRequested]] is true and this.[[queue]] is empty,
  1. Perform ! ReadableStreamDefaultControllerClearAlgorithms(this).
  2. Perform ! ReadableStreamClose(stream).
3. Otherwise, perform ! ReadableStreamDefaultControllerCallPullIfNeeded(this).
4. Perform readRequest’s chunk steps, given chunk.
Otherwise,
1. Perform ! ReadableStreamAddReadRequest(stream, readRequest).
2. Perform ! ReadableStreamDefaultControllerCallPullIfNeeded(this).

[[ReleaseSteps]]() implements the [[ReleaseSteps]] contract. It performs the following steps:

Return.

4.7. The `ReadableByteStreamController` class

The ReadableByteStreamController class has methods that allow control of a ReadableStream's state and internal queue. When constructing a ReadableStream that is a readable byte stream, the underlying source is given a corresponding ReadableByteStreamController instance to manipulate.

4.7.1. Interface definition

The Web IDL definition for the ReadableByteStreamController class is given as follows:

[Exposed=*]
interface ReadableByteStreamController {
  readonly attribute ReadableStreamBYOBRequest? byobRequest;
  readonly attribute unrestricted double? desiredSize;

  undefined close();
  undefined enqueue(ArrayBufferView chunk);
  undefined error(optional any e);
};

4.7.2. Internal slots

Instances of ReadableByteStreamController are created with the internal slots described in the following table:

Internal Slot	Description (non-normative)
[[autoAllocateChunkSize]]	A positive integer, when the automatic buffer allocation feature is enabled. In that case, this value specifies the size of buffer to allocate. It is undefined otherwise.
[[byobRequest]]	A `ReadableStreamBYOBRequest` instance representing the current BYOB pull request, or null if there are no pending requests
[[cancelAlgorithm]]	A promise-returning algorithm, taking one argument (the cancel reason), which communicates a requested cancelation to the underlying byte source
[[closeRequested]]	A boolean flag indicating whether the stream has been closed by its underlying byte source, but still has chunks in its internal queue that have not yet been read
[[pullAgain]]	A boolean flag set to true if the stream’s mechanisms requested a call to the underlying byte source's pull algorithm to pull more data, but the pull could not yet be done since a previous call is still executing
[[pullAlgorithm]]	A promise-returning algorithm that pulls data from the underlying byte source
[[pulling]]	A boolean flag set to true while the underlying byte source's pull algorithm is executing and the returned promise has not yet fulfilled, used to prevent reentrant calls
[[pendingPullIntos]]	A list of pull-into descriptors
[[queue]]	A list of readable byte stream queue entries representing the stream’s internal queue of chunks
[[queueTotalSize]]	The total size, in bytes, of all the chunks stored in [[queue]] (see § 8.1 Queue-with-sizes)
[[started]]	A boolean flag indicating whether the underlying byte source has finished starting
[[strategyHWM]]	A number supplied to the constructor as part of the stream’s queuing strategy, indicating the point at which the stream will apply backpressure to its underlying byte source
[[stream]]	The `ReadableStream` instance controlled

Although ReadableByteStreamController instances have [[queue]] and [[queueTotalSize]] slots, we do not use most of the abstract operations in § 8.1 Queue-with-sizes on them, as the way in which we manipulate this queue is rather different than the others in the spec. Instead, we update the two slots together manually.

This might be cleaned up in a future spec refactoring.

A readable byte stream queue entry is a struct encapsulating the important aspects of a chunk for the specific case of readable byte streams. It has the following items:

buffer: An ArrayBuffer, which will be a transferred version of the one originally supplied by the underlying byte source
byte offset: A nonnegative integer number giving the byte offset derived from the view originally supplied by the underlying byte source
byte length: A nonnegative integer number giving the byte length derived from the view originally supplied by the underlying byte source

A pull-into descriptor is a struct used to represent pending BYOB pull requests. It has the following items:

buffer: An ArrayBuffer
buffer byte length: A positive integer representing the initial byte length of buffer
byte offset: A nonnegative integer byte offset into the buffer where the underlying byte source will start writing
byte length: A positive integer number of bytes which can be written into the buffer
bytes filled: A nonnegative integer number of bytes that have been written into the buffer so far
minimum fill: A positive integer representing the minimum number of bytes that must be written into the buffer before the associated read() request may be fulfilled. By default, this equals the element size.
element size: A positive integer representing the number of bytes that can be written into the buffer at a time, using views of the type described by the view constructor
view constructor: A typed array constructor or %DataView%, which will be used for constructing a view with which to write into the buffer
reader type: Either "default" or "byob", indicating what type of readable stream reader initiated this request, or "none" if the initiating reader was released

4.7.3. Methods and properties

byobRequest = controller.byobRequest: Returns the current BYOB pull request, or null if there isn’t one.
desiredSize = controller.desiredSize: Returns the desired size to fill the controlled stream’s internal queue. It can be negative, if the queue is over-full. An underlying byte source ought to use this information to determine when and how to apply backpressure.
controller.close(): Closes the controlled readable stream. Consumers will still be able to read any previously-enqueued chunks from the stream, but once those are read, the stream will become closed.
controller.enqueue(chunk): Enqueues the given chunk chunk in the controlled readable stream. The chunk has to be an ArrayBufferView instance, or else a TypeError will be thrown.
controller.error(e): Errors the controlled readable stream, making all future interactions with it fail with the given error e.

The byobRequest getter steps are:

Return ! ReadableByteStreamControllerGetBYOBRequest(this).

The desiredSize getter steps are:

Return ! ReadableByteStreamControllerGetDesiredSize(this).

The close() method steps are:

If this.[[closeRequested]] is true, throw a TypeError exception.
If this.[[stream]].[[state]] is not "readable", throw a TypeError exception.
Perform ? ReadableByteStreamControllerClose(this).

The enqueue(chunk) method steps are:

If chunk.[[ByteLength]] is 0, throw a TypeError exception.
If chunk.[[ViewedArrayBuffer]].[[ArrayBufferByteLength]] is 0, throw a TypeError exception.
If this.[[closeRequested]] is true, throw a TypeError exception.
If this.[[stream]].[[state]] is not "readable", throw a TypeError exception.
Return ? ReadableByteStreamControllerEnqueue(this, chunk).

The error(e) method steps are:

Perform ! ReadableByteStreamControllerError(this, e).

4.7.4. Internal methods

The following are internal methods implemented by each ReadableByteStreamController instance. The readable stream implementation will polymorphically call to either these, or to their counterparts for default controllers, as discussed in § 4.9.2 Interfacing with controllers.

[[CancelSteps]](reason) implements the [[CancelSteps]] contract. It performs the following steps:

Perform ! ReadableByteStreamControllerClearPendingPullIntos(this).
Perform ! ResetQueue(this).
Let result be the result of performing this.[[cancelAlgorithm]], passing in reason.
Perform ! ReadableByteStreamControllerClearAlgorithms(this).
Return result.

[[PullSteps]](readRequest) implements the [[PullSteps]] contract. It performs the following steps:

Let stream be this.[[stream]].
Assert: ! ReadableStreamHasDefaultReader(stream) is true.
If this.[[queueTotalSize]] > 0,
1. Assert: ! ReadableStreamGetNumReadRequests(stream) is 0.
2. Perform ! ReadableByteStreamControllerFillReadRequestFromQueue(this, readRequest).
3. Return.
Let autoAllocateChunkSize be this.[[autoAllocateChunkSize]].
If autoAllocateChunkSize is not undefined,
1. Let buffer be Construct(%ArrayBuffer%, « autoAllocateChunkSize »).
2. If buffer is an abrupt completion,
  1. Perform readRequest’s error steps, given buffer.[[Value]].
  2. Return.
3. Let pullIntoDescriptor be a new pull-into descriptor with
  
  buffer
  buffer.[[Value]]
  buffer byte length
  autoAllocateChunkSize
  byte offset
  0
  byte length
  autoAllocateChunkSize
  bytes filled
  0
  minimum fill
  1
  element size
  1
  view constructor
  %Uint8Array%
  reader type
  "default"
4. Append pullIntoDescriptor to this.[[pendingPullIntos]].
Perform ! ReadableStreamAddReadRequest(stream, readRequest).
Perform ! ReadableByteStreamControllerCallPullIfNeeded(this).

[[ReleaseSteps]]() implements the [[ReleaseSteps]] contract. It performs the following steps:

If this.[[pendingPullIntos]] is not empty,
1. Let firstPendingPullInto be this.[[pendingPullIntos]][0].
2. Set firstPendingPullInto’s reader type to "none".
3. Set this.[[pendingPullIntos]] to the list « firstPendingPullInto ».

4.8. The `ReadableStreamBYOBRequest` class

The ReadableStreamBYOBRequest class represents a pull-into request in a ReadableByteStreamController.

4.8.1. Interface definition

The Web IDL definition for the ReadableStreamBYOBRequest class is given as follows:

[Exposed=*]
interface ReadableStreamBYOBRequest {
  readonly attribute ArrayBufferView? view;

  undefined respond([EnforceRange] unsigned long long bytesWritten);
  undefined respondWithNewView(ArrayBufferView view);
};

4.8.2. Internal slots

Instances of ReadableStreamBYOBRequest are created with the internal slots described in the following table:

Internal Slot	Description (non-normative)
[[controller]]	The parent `ReadableByteStreamController` instance
[[view]]	A typed array representing the destination region to which the controller can write generated data, or null after the BYOB request has been invalidated.

4.8.3. Methods and properties

view = byobRequest.view

Returns the view for writing in to, or null if the BYOB request has already been responded to.

byobRequest.respond(bytesWritten)

Indicates to the associated readable byte stream that bytesWritten bytes were written into view, causing the result be surfaced to the consumer.

After this method is called, view will be transferred and no longer modifiable.

byobRequest.respondWithNewView(view)

Indicates to the associated readable byte stream that instead of writing into view, the underlying byte source is providing a new ArrayBufferView, which will be given to the consumer of the readable byte stream.

The new view has to be a view onto the same backing memory region as view, i.e. its buffer has to equal (or be a transferred version of) view's buffer. Its byteOffset has to equal view's byteOffset, and its byteLength (representing the number of bytes written) has to be less than or equal to that of view.

After this method is called, view will be transferred and no longer modifiable.

The view getter steps are:

Return this.[[view]].

The respond(bytesWritten) method steps are:

If this.[[controller]] is undefined, throw a TypeError exception.
If ! IsDetachedBuffer(this.[[view]].[[ArrayBuffer]]) is true, throw a TypeError exception.
Assert: this.[[view]].[[ByteLength]] > 0.
Assert: this.[[view]].[[ViewedArrayBuffer]].[[ByteLength]] > 0.
Perform ? ReadableByteStreamControllerRespond(this.[[controller]], bytesWritten).

The respondWithNewView(view) method steps are:

If this.[[controller]] is undefined, throw a TypeError exception.
If ! IsDetachedBuffer(view.[[ViewedArrayBuffer]]) is true, throw a TypeError exception.
Return ? ReadableByteStreamControllerRespondWithNewView(this.[[controller]], view).

4.9. Abstract operations

4.9.1. Working with readable streams

The following abstract operations operate on ReadableStream instances at a higher level.

AcquireReadableStreamBYOBReader(stream) performs the following steps:

Let reader be a new ReadableStreamBYOBReader.
Perform ? SetUpReadableStreamBYOBReader(reader, stream).
Return reader.

AcquireReadableStreamDefaultReader(stream) performs the following steps:

Let reader be a new ReadableStreamDefaultReader.
Perform ? SetUpReadableStreamDefaultReader(reader, stream).
Return reader.

CreateReadableStream(startAlgorithm, pullAlgorithm, cancelAlgorithm[, highWaterMark, [, sizeAlgorithm]]) performs the following steps:

If highWaterMark was not passed, set it to 1.
If sizeAlgorithm was not passed, set it to an algorithm that returns 1.
Assert: ! IsNonNegativeNumber(highWaterMark) is true.
Let stream be a new ReadableStream.
Perform ! InitializeReadableStream(stream).
Let controller be a new ReadableStreamDefaultController.
Perform ? SetUpReadableStreamDefaultController(stream, controller, startAlgorithm, pullAlgorithm, cancelAlgorithm, highWaterMark, sizeAlgorithm).
Return stream.

This abstract operation will throw an exception if and only if the supplied startAlgorithm throws.

CreateReadableByteStream(startAlgorithm, pullAlgorithm, cancelAlgorithm) performs the following steps:

Let stream be a new ReadableStream.
Perform ! InitializeReadableStream(stream).
Let controller be a new ReadableByteStreamController.
Perform ? SetUpReadableByteStreamController(stream, controller, startAlgorithm, pullAlgorithm, cancelAlgorithm, 0, undefined).
Return stream.

This abstract operation will throw an exception if and only if the supplied startAlgorithm throws.

InitializeReadableStream(stream) performs the following steps:

Set stream.[[state]] to "readable".
Set stream.[[reader]] and stream.[[storedError]] to undefined.
Set stream.[[disturbed]] to false.

IsReadableStreamLocked(stream) performs the following steps:

If stream.[[reader]] is undefined, return false.
Return true.

ReadableStreamFromIterable(asyncIterable) performs the following steps:

Let stream be undefined.
Let iteratorRecord be ? GetIterator(asyncIterable, async).
Let startAlgorithm be an algorithm that returns undefined.
Let pullAlgorithm be the following steps:
1. Let nextResult be IteratorNext(iteratorRecord).
2. If nextResult is an abrupt completion, return a promise rejected with nextResult.[[Value]].
3. Let nextPromise be a promise resolved with nextResult.[[Value]].
4. Return the result of reacting to nextPromise with the following fulfillment steps, given iterResult:
  1. If iterResult is not an Object, throw a TypeError.
  2. Let done be ? IteratorComplete(iterResult).
  3. If done is true:
    1. Perform ! ReadableStreamDefaultControllerClose(stream.[[controller]]).
  4. Otherwise:
    1. Let value be ? IteratorValue(iterResult).
    2. Perform ! ReadableStreamDefaultControllerEnqueue(stream.[[controller]], value).
Let cancelAlgorithm be the following steps, given reason:
1. Let iterator be iteratorRecord.[[Iterator]].
2. Let returnMethod be GetMethod(iterator, "return").
3. If returnMethod is an abrupt completion, return a promise rejected with returnMethod.[[Value]].
4. If returnMethod.[[Value]] is undefined, return a promise resolved with undefined.
5. Let returnResult be Call(returnMethod.[[Value]], iterator, « reason »).
6. If returnResult is an abrupt completion, return a promise rejected with returnResult.[[Value]].
7. Let returnPromise be a promise resolved with returnResult.[[Value]].
8. Return the result of reacting to returnPromise with the following fulfillment steps, given iterResult:
  1. If iterResult is not an Object, throw a TypeError.
  2. Return undefined.
Set stream to ! CreateReadableStream(startAlgorithm, pullAlgorithm, cancelAlgorithm, 0).
Return stream.

ReadableStreamPipeTo(source, dest, preventClose, preventAbort, preventCancel[, signal]) performs the following steps:

Assert: source implements ReadableStream.
Assert: dest implements WritableStream.
Assert: preventClose, preventAbort, and preventCancel are all booleans.
If signal was not given, let signal be undefined.
Assert: either signal is undefined, or signal implements AbortSignal.
Assert: ! IsReadableStreamLocked(source) is false.
Assert: ! IsWritableStreamLocked(dest) is false.
If source.[[controller]] implements ReadableByteStreamController, let reader be either ! AcquireReadableStreamBYOBReader(source) or ! AcquireReadableStreamDefaultReader(source), at the user agent’s discretion.
Otherwise, let reader be ! AcquireReadableStreamDefaultReader(source).
Let writer be ! AcquireWritableStreamDefaultWriter(dest).
Set source.[[disturbed]] to true.
Let shuttingDown be false.
Let promise be a new promise.
If signal is not undefined,
1. Let abortAlgorithm be the following steps:
  1. Let error be signal’s abort reason.
  2. Let actions be an empty ordered set.
  3. If preventAbort is false, append the following action to actions:
    1. If dest.[[state]] is "writable", return ! WritableStreamAbort(dest, error).
    2. Otherwise, return a promise resolved with undefined.
  4. If preventCancel is false, append the following action action to actions:
    1. If source.[[state]] is "readable", return ! ReadableStreamCancel(source, error).
    2. Otherwise, return a promise resolved with undefined.
  5. Shutdown with an action consisting of getting a promise to wait for all of the actions in actions, and with error.
2. If signal is aborted, perform abortAlgorithm and return promise.
3. Add abortAlgorithm to signal.
In parallel but not really; see #905, using reader and writer, read all chunks from source and write them to dest. Due to the locking provided by the reader and writer, the exact manner in which this happens is not observable to author code, and so there is flexibility in how this is done. The following constraints apply regardless of the exact algorithm used:
- Public API must not be used: while reading or writing, or performing any of the operations below, the JavaScript-modifiable reader, writer, and stream APIs (i.e. methods on the appropriate prototypes) must not be used. Instead, the streams must be manipulated directly.
- Backpressure must be enforced:
  - While WritableStreamDefaultWriterGetDesiredSize(writer) is ≤ 0 or is null, the user agent must not read from reader.
  - If reader is a BYOB reader, WritableStreamDefaultWriterGetDesiredSize(writer) should be used as a basis to determine the size of the chunks read from reader.
    
    It’s frequently inefficient to read chunks that are too small or too large. Other information might be factored in to determine the optimal chunk size.
  - Reads or writes should not be delayed for reasons other than these backpressure signals.
    
    An implementation that waits for each write to successfully complete before proceeding to the next read/write operation violates this recommendation. In doing so, such an implementation makes the internal queue of dest useless, as it ensures dest always contains at most one queued chunk.
- Shutdown must stop activity: if shuttingDown becomes true, the user agent must not initiate further reads from reader, and must only perform writes of already-read chunks, as described below. In particular, the user agent must check the below conditions before performing any reads or writes, since they might lead to immediate shutdown.
- Error and close states must be propagated: the following conditions must be applied in order.
  1. Errors must be propagated forward: if source.[[state]] is or becomes "errored", then
    1. If preventAbort is false, shutdown with an action of ! WritableStreamAbort(dest, source.[[storedError]]) and with source.[[storedError]].
    2. Otherwise, shutdown with source.[[storedError]].
  2. Errors must be propagated backward: if dest.[[state]] is or becomes "errored", then
    1. If preventCancel is false, shutdown with an action of ! ReadableStreamCancel(source, dest.[[storedError]]) and with dest.[[storedError]].
    2. Otherwise, shutdown with dest.[[storedError]].
  3. Closing must be propagated forward: if source.[[state]] is or becomes "closed", then
    1. If preventClose is false, shutdown with an action of ! WritableStreamDefaultWriterCloseWithErrorPropagation(writer).
    2. Otherwise, shutdown.
  4. Closing must be propagated backward: if ! WritableStreamCloseQueuedOrInFlight(dest) is true or dest.[[state]] is "closed", then
    1. Assert: no chunks have been read or written.
    2. Let destClosed be a new TypeError.
    3. If preventCancel is false, shutdown with an action of ! ReadableStreamCancel(source, destClosed) and with destClosed.
    4. Otherwise, shutdown with destClosed.
- Shutdown with an action: if any of the above requirements ask to shutdown with an action action, optionally with an error originalError, then:
  1. If shuttingDown is true, abort these substeps.
  2. Set shuttingDown to true.
  3. If dest.[[state]] is "writable" and ! WritableStreamCloseQueuedOrInFlight(dest) is false,
    1. If any chunks have been read but not yet written, write them to dest.
    2. Wait until every chunk that has been read has been written (i.e. the corresponding promises have settled).
  4. Let p be the result of performing action.
  5. Upon fulfillment of p, finalize, passing along originalError if it was given.
  6. Upon rejection of p with reason newError, finalize with newError.
- Shutdown: if any of the above requirements or steps ask to shutdown, optionally with an error error, then:
  1. If shuttingDown is true, abort these substeps.
  2. Set shuttingDown to true.
  3. If dest.[[state]] is "writable" and ! WritableStreamCloseQueuedOrInFlight(dest) is false,
    1. If any chunks have been read but not yet written, write them to dest.
    2. Wait until every chunk that has been read has been written (i.e. the corresponding promises have settled).
  4. Finalize, passing along error if it was given.
- Finalize: both forms of shutdown will eventually ask to finalize, optionally with an error error, which means to perform the following steps:
  1. Perform ! WritableStreamDefaultWriterRelease(writer).
  2. If reader implements ReadableStreamBYOBReader, perform ! ReadableStreamBYOBReaderRelease(reader).
  3. Otherwise, perform ! ReadableStreamDefaultReaderRelease(reader).
  4. If signal is not undefined, remove abortAlgorithm from signal.
  5. If error was given, reject promise with error.
  6. Otherwise, resolve promise with undefined.
Return promise.

Various abstract operations performed here include object creation (often of promises), which usually would require specifying a realm for the created object. However, because of the locking, none of these objects can be observed by author code. As such, the realm used to create them does not matter.

ReadableStreamTee(stream, cloneForBranch2) will tee a given readable stream.

The second argument, cloneForBranch2, governs whether or not the data from the original stream will be cloned (using HTML’s serializable objects framework) before appearing in the second of the returned branches. This is useful for scenarios where both branches are to be consumed in such a way that they might otherwise interfere with each other, such as by transferring their chunks. However, it does introduce a noticeable asymmetry between the two branches, and limits the possible chunks to serializable ones. [HTML]

If stream is a readable byte stream, then cloneForBranch2 is ignored and chunks are cloned unconditionally.

In this standard ReadableStreamTee is always called with cloneForBranch2 set to false; other specifications pass true via the tee wrapper algorithm.

It performs the following steps:

Assert: stream implements ReadableStream.
Assert: cloneForBranch2 is a boolean.
If stream.[[controller]] implements ReadableByteStreamController, return ? ReadableByteStreamTee(stream).
Return ? ReadableStreamDefaultTee(stream, cloneForBranch2).

ReadableStreamDefaultTee(stream, cloneForBranch2) performs the following steps:

Assert: stream implements ReadableStream.
Assert: cloneForBranch2 is a boolean.
Let reader be ? AcquireReadableStreamDefaultReader(stream).
Let reading be false.
Let readAgain be false.
Let canceled1 be false.
Let canceled2 be false.
Let reason1 be undefined.
Let reason2 be undefined.
Let branch1 be undefined.
Let branch2 be undefined.
Let cancelPromise be a new promise.
Let pullAlgorithm be the following steps:
1. If reading is true,
  1. Set readAgain to true.
  2. Return a promise resolved with undefined.
2. Set reading to true.
3. Let readRequest be a read request with the following items:
  chunk steps, given chunk
  1. Queue a microtask to perform the following steps:
    
    Set readAgain to false.
    
    Let chunk1 and chunk2 be chunk.
    
    If canceled2 is false and cloneForBranch2 is true,
    
    Let cloneResult be StructuredClone(chunk2).
    
    If cloneResult is an abrupt completion,
    
    Perform ! ReadableStreamDefaultControllerError(branch1.[[controller]], cloneResult.[[Value]]).
    
    Perform ! ReadableStreamDefaultControllerError(branch2.[[controller]], cloneResult.[[Value]]).
    
    Resolve cancelPromise with ! ReadableStreamCancel(stream, cloneResult.[[Value]]).
    
    Return.
    
    Otherwise, set chunk2 to cloneResult.[[Value]].
    
    If canceled1 is false, perform ! ReadableStreamDefaultControllerEnqueue(branch1.[[controller]], chunk1).
    
    If canceled2 is false, perform ! ReadableStreamDefaultControllerEnqueue(branch2.[[controller]], chunk2).
    
    Set reading to false.
    
    If readAgain is true, perform pullAlgorithm.
  The microtask delay here is necessary because it takes at least a microtask to detect errors, when we use reader.[[closedPromise]] below. We want errors in stream to error both branches immediately, so we cannot let successful synchronously-available reads happen ahead of asynchronously-available errors.
  close steps
  1. Set reading to false.
  2. If canceled1 is false, perform ! ReadableStreamDefaultControllerClose(branch1.[[controller]]).
  3. If canceled2 is false, perform ! ReadableStreamDefaultControllerClose(branch2.[[controller]]).
  4. If canceled1 is false or canceled2 is false, resolve cancelPromise with undefined.
  error steps
  1. Set reading to false.
4. Perform ! ReadableStreamDefaultReaderRead(reader, readRequest).
5. Return a promise resolved with undefined.
Let cancel1Algorithm be the following steps, taking a reason argument:
1. Set canceled1 to true.
2. Set reason1 to reason.
3. If canceled2 is true,
  1. Let compositeReason be ! CreateArrayFromList(« reason1, reason2 »).
  2. Let cancelResult be ! ReadableStreamCancel(stream, compositeReason).
  3. Resolve cancelPromise with cancelResult.
4. Return cancelPromise.
Let cancel2Algorithm be the following steps, taking a reason argument:
1. Set canceled2 to true.
2. Set reason2 to reason.
3. If canceled1 is true,
  1. Let compositeReason be ! CreateArrayFromList(« reason1, reason2 »).
  2. Let cancelResult be ! ReadableStreamCancel(stream, compositeReason).
  3. Resolve cancelPromise with cancelResult.
4. Return cancelPromise.
Let startAlgorithm be an algorithm that returns undefined.
Set branch1 to ! CreateReadableStream(startAlgorithm, pullAlgorithm, cancel1Algorithm).
Set branch2 to ! CreateReadableStream(startAlgorithm, pullAlgorithm, cancel2Algorithm).
Upon rejection of reader.[[closedPromise]] with reason r,
1. Perform ! ReadableStreamDefaultControllerError(branch1.[[controller]], r).
2. Perform ! ReadableStreamDefaultControllerError(branch2.[[controller]], r).
3. If canceled1 is false or canceled2 is false, resolve cancelPromise with undefined.
Return « branch1, branch2 ».

ReadableByteStreamTee(stream) performs the following steps:

Assert: stream implements ReadableStream.
Assert: stream.[[controller]] implements ReadableByteStreamController.
Let reader be ? AcquireReadableStreamDefaultReader(stream).
Let reading be false.
Let readAgainForBranch1 be false.
Let readAgainForBranch2 be false.
Let canceled1 be false.
Let canceled2 be false.
Let reason1 be undefined.
Let reason2 be undefined.
Let branch1 be undefined.
Let branch2 be undefined.
Let cancelPromise be a new promise.
Let forwardReaderError be the following steps, taking a thisReader argument:
1. Upon rejection of thisReader.[[closedPromise]] with reason r,
  1. If thisReader is not reader, return.
  2. Perform ! ReadableByteStreamControllerError(branch1.[[controller]], r).
  3. Perform ! ReadableByteStreamControllerError(branch2.[[controller]], r).
  4. If canceled1 is false or canceled2 is false, resolve cancelPromise with undefined.
Let pullWithDefaultReader be the following steps:
1. If reader implements ReadableStreamBYOBReader,
  1. Assert: reader.[[readIntoRequests]] is empty.
  2. Perform ! ReadableStreamBYOBReaderRelease(reader).
  3. Set reader to ! AcquireReadableStreamDefaultReader(stream).
  4. Perform forwardReaderError, given reader.
2. Let readRequest be a read request with the following items:
  chunk steps, given chunk
  1. Queue a microtask to perform the following steps:
    
    Set readAgainForBranch1 to false.
    
    Set readAgainForBranch2 to false.
    
    Let chunk1 and chunk2 be chunk.
    
    If canceled1 is false and canceled2 is false,
    
    Let cloneResult be CloneAsUint8Array(chunk).
    
    If cloneResult is an abrupt completion,
    
    Perform ! ReadableByteStreamControllerError(branch1.[[controller]], cloneResult.[[Value]]).
    
    Perform ! ReadableByteStreamControllerError(branch2.[[controller]], cloneResult.[[Value]]).
    
    Resolve cancelPromise with ! ReadableStreamCancel(stream, cloneResult.[[Value]]).
    
    Return.
    
    Otherwise, set chunk2 to cloneResult.[[Value]].
    
    If canceled1 is false, perform ! ReadableByteStreamControllerEnqueue(branch1.[[controller]], chunk1).
    
    If canceled2 is false, perform ! ReadableByteStreamControllerEnqueue(branch2.[[controller]], chunk2).
    
    Set reading to false.
    
    If readAgainForBranch1 is true, perform pull1Algorithm.
    
    Otherwise, if readAgainForBranch2 is true, perform pull2Algorithm.
  The microtask delay here is necessary because it takes at least a microtask to detect errors, when we use reader.[[closedPromise]] below. We want errors in stream to error both branches immediately, so we cannot let successful synchronously-available reads happen ahead of asynchronously-available errors.
  close steps
  1. Set reading to false.
  2. If canceled1 is false, perform ! ReadableByteStreamControllerClose(branch1.[[controller]]).
  3. If canceled2 is false, perform ! ReadableByteStreamControllerClose(branch2.[[controller]]).
  4. If branch1.[[controller]].[[pendingPullIntos]] is not empty, perform ! ReadableByteStreamControllerRespond(branch1.[[controller]], 0).
  5. If branch2.[[controller]].[[pendingPullIntos]] is not empty, perform ! ReadableByteStreamControllerRespond(branch2.[[controller]], 0).
  6. If canceled1 is false or canceled2 is false, resolve cancelPromise with undefined.
  error steps
  1. Set reading to false.
3. Perform ! ReadableStreamDefaultReaderRead(reader, readRequest).
Let pullWithBYOBReader be the following steps, given view and forBranch2:
1. If reader implements ReadableStreamDefaultReader,
  1. Assert: reader.[[readRequests]] is empty.
  2. Perform ! ReadableStreamDefaultReaderRelease(reader).
  3. Set reader to ! AcquireReadableStreamBYOBReader(stream).
  4. Perform forwardReaderError, given reader.
2. Let byobBranch be branch2 if forBranch2 is true, and branch1 otherwise.
3. Let otherBranch be branch2 if forBranch2 is false, and branch1 otherwise.
4. Let readIntoRequest be a read-into request with the following items:
  chunk steps, given chunk
  1. Queue a microtask to perform the following steps:
    
    Set readAgainForBranch1 to false.
    
    Set readAgainForBranch2 to false.
    
    Let byobCanceled be canceled2 if forBranch2 is true, and canceled1 otherwise.
    
    Let otherCanceled be canceled2 if forBranch2 is false, and canceled1 otherwise.
    
    If otherCanceled is false,
    
    Let cloneResult be CloneAsUint8Array(chunk).
    
    If cloneResult is an abrupt completion,
    
    Perform ! ReadableByteStreamControllerError(byobBranch.[[controller]], cloneResult.[[Value]]).
    
    Perform ! ReadableByteStreamControllerError(otherBranch.[[controller]], cloneResult.[[Value]]).
    
    Resolve cancelPromise with ! ReadableStreamCancel(stream, cloneResult.[[Value]]).
    
    Return.
    
    Otherwise, let clonedChunk be cloneResult.[[Value]].
    
    If byobCanceled is false, perform ! ReadableByteStreamControllerRespondWithNewView(byobBranch.[[controller]], chunk).
    
    Perform ! ReadableByteStreamControllerEnqueue(otherBranch.[[controller]], clonedChunk).
    
    Otherwise, if byobCanceled is false, perform ! ReadableByteStreamControllerRespondWithNewView(byobBranch.[[controller]], chunk).
    
    Set reading to false.
    
    If readAgainForBranch1 is true, perform pull1Algorithm.
    
    Otherwise, if readAgainForBranch2 is true, perform pull2Algorithm.
  The microtask delay here is necessary because it takes at least a microtask to detect errors, when we use reader.[[closedPromise]] below. We want errors in stream to error both branches immediately, so we cannot let successful synchronously-available reads happen ahead of asynchronously-available errors.
  close steps, given chunk
  1. Set reading to false.
  2. Let byobCanceled be canceled2 if forBranch2 is true, and canceled1 otherwise.
  3. Let otherCanceled be canceled2 if forBranch2 is false, and canceled1 otherwise.
  4. If byobCanceled is false, perform ! ReadableByteStreamControllerClose(byobBranch.[[controller]]).
  5. If otherCanceled is false, perform ! ReadableByteStreamControllerClose(otherBranch.[[controller]]).
  6. If chunk is not undefined,
    
    Assert: chunk.[[ByteLength]] is 0.
    
    If byobCanceled is false, perform ! ReadableByteStreamControllerRespondWithNewView(byobBranch.[[controller]], chunk).
    
    If otherCanceled is false and otherBranch.[[controller]].[[pendingPullIntos]] is not empty, perform ! ReadableByteStreamControllerRespond(otherBranch.[[controller]], 0).
  7. If byobCanceled is false or otherCanceled is false, resolve cancelPromise with undefined.
  error steps
  1. Set reading to false.
5. Perform ! ReadableStreamBYOBReaderRead(reader, view, 1, readIntoRequest).
Let pull1Algorithm be the following steps:
1. If reading is true,
  1. Set readAgainForBranch1 to true.
  2. Return a promise resolved with undefined.
2. Set reading to true.
3. Let byobRequest be ! ReadableByteStreamControllerGetBYOBRequest(branch1.[[controller]]).
4. If byobRequest is null, perform pullWithDefaultReader.
5. Otherwise, perform pullWithBYOBReader, given byobRequest.[[view]] and false.
6. Return a promise resolved with undefined.
Let pull2Algorithm be the following steps:
1. If reading is true,
  1. Set readAgainForBranch2 to true.
  2. Return a promise resolved with undefined.
2. Set reading to true.
3. Let byobRequest be ! ReadableByteStreamControllerGetBYOBRequest(branch2.[[controller]]).
4. If byobRequest is null, perform pullWithDefaultReader.
5. Otherwise, perform pullWithBYOBReader, given byobRequest.[[view]] and true.
6. Return a promise resolved with undefined.
Let cancel1Algorithm be the following steps, taking a reason argument:
1. Set canceled1 to true.
2. Set reason1 to reason.
3. If canceled2 is true,
  1. Let compositeReason be ! CreateArrayFromList(« reason1, reason2 »).
  2. Let cancelResult be ! ReadableStreamCancel(stream, compositeReason).
  3. Resolve cancelPromise with cancelResult.
4. Return cancelPromise.
Let cancel2Algorithm be the following steps, taking a reason argument:
1. Set canceled2 to true.
2. Set reason2 to reason.
3. If canceled1 is true,
  1. Let compositeReason be ! CreateArrayFromList(« reason1, reason2 »).
  2. Let cancelResult be ! ReadableStreamCancel(stream, compositeReason).
  3. Resolve cancelPromise with cancelResult.
4. Return cancelPromise.
Let startAlgorithm be an algorithm that returns undefined.
Set branch1 to ! CreateReadableByteStream(startAlgorithm, pull1Algorithm, cancel1Algorithm).
Set branch2 to ! CreateReadableByteStream(startAlgorithm, pull2Algorithm, cancel2Algorithm).
Perform forwardReaderError, given reader.
Return « branch1, branch2 ».

4.9.2. Interfacing with controllers

In terms of specification factoring, the way that the ReadableStream class encapsulates the behavior of both simple readable streams and readable byte streams into a single class is by centralizing most of the potentially-varying logic inside the two controller classes, ReadableStreamDefaultController and ReadableByteStreamController. Those classes define most of the stateful internal slots and abstract operations for how a stream’s internal queue is managed and how it interfaces with its underlying source or underlying byte source.

Each controller class defines three internal methods, which are called by the ReadableStream algorithms:

[[CancelSteps]](reason): The controller’s steps that run in reaction to the stream being canceled, used to clean up the state stored in the controller and inform the underlying source.
[[PullSteps]](readRequest): The controller’s steps that run when a default reader is read from, used to pull from the controller any queued chunks, or pull from the underlying source to get more chunks.
[[ReleaseSteps]](): The controller’s steps that run when a reader is released, used to clean up reader-specific resources stored in the controller.

(These are defined as internal methods, instead of as abstract operations, so that they can be called polymorphically by the ReadableStream algorithms, without having to branch on which type of controller is present.)

The rest of this section concerns abstract operations that go in the other direction: they are used by the controller implementations to affect their associated ReadableStream object. This translates internal state changes of the controller into developer-facing results visible through the ReadableStream's public API.

ReadableStreamAddReadIntoRequest(stream, readRequest) performs the following steps:

Assert: stream.[[reader]] implements ReadableStreamBYOBReader.
Assert: stream.[[state]] is "readable" or "closed".
Append readRequest to stream.[[reader]].[[readIntoRequests]].

ReadableStreamAddReadRequest(stream, readRequest performs the following steps:

Assert: stream.[[reader]] implements ReadableStreamDefaultReader.
Assert: stream.[[state]] is "readable".
Append readRequest to stream.[[reader]].[[readRequests]].

ReadableStreamCancel(stream, reason) performs the following steps:

Set stream.[[disturbed]] to true.
If stream.[[state]] is "closed", return a promise resolved with undefined.
If stream.[[state]] is "errored", return a promise rejected with stream.[[storedError]].
Perform ! ReadableStreamClose(stream).
Let reader be stream.[[reader]].
If reader is not undefined and reader implements ReadableStreamBYOBReader,
1. Let readIntoRequests be reader.[[readIntoRequests]].
2. Set reader.[[readIntoRequests]] to an empty list.
3. For each readIntoRequest of readIntoRequests,
  1. Perform readIntoRequest’s close steps, given undefined.
Let sourceCancelPromise be ! stream.[[controller]].[[CancelSteps]](reason).
Return the result of reacting to sourceCancelPromise with a fulfillment step that returns undefined.

ReadableStreamClose(stream) performs the following steps:

Assert: stream.[[state]] is "readable".
Set stream.[[state]] to "closed".
Let reader be stream.[[reader]].
If reader is undefined, return.
Resolve reader.[[closedPromise]] with undefined.
If reader implements ReadableStreamDefaultReader,
1. Let readRequests be reader.[[readRequests]].
2. Set reader.[[readRequests]] to an empty list.
3. For each readRequest of readRequests,
  1. Perform readRequest’s close steps.

ReadableStreamError(stream, e) performs the following steps:

Assert: stream.[[state]] is "readable".
Set stream.[[state]] to "errored".
Set stream.[[storedError]] to e.
Let reader be stream.[[reader]].
If reader is undefined, return.
Reject reader.[[closedPromise]] with e.
Set reader.[[closedPromise]].[[PromiseIsHandled]] to true.
If reader implements ReadableStreamDefaultReader,
1. Perform ! ReadableStreamDefaultReaderErrorReadRequests(reader, e).
Otherwise,
1. Assert: reader implements ReadableStreamBYOBReader.
2. Perform ! ReadableStreamBYOBReaderErrorReadIntoRequests(reader, e).

ReadableStreamFulfillReadIntoRequest(stream, chunk, done) performs the following steps:

Assert: ! ReadableStreamHasBYOBReader(stream) is true.
Let reader be stream.[[reader]].
Assert: reader.[[readIntoRequests]] is not empty.
Let readIntoRequest be reader.[[readIntoRequests]][0].
Remove readIntoRequest from reader.[[readIntoRequests]].
If done is true, perform readIntoRequest’s close steps, given chunk.
Otherwise, perform readIntoRequest’s chunk steps, given chunk.

ReadableStreamFulfillReadRequest(stream, chunk, done) performs the following steps:

Assert: ! ReadableStreamHasDefaultReader(stream) is true.
Let reader be stream.[[reader]].
Assert: reader.[[readRequests]] is not empty.
Let readRequest be reader.[[readRequests]][0].
Remove readRequest from reader.[[readRequests]].
If done is true, perform readRequest’s close steps.
Otherwise, perform readRequest’s chunk steps, given chunk.

ReadableStreamGetNumReadIntoRequests(stream) performs the following steps:

Assert: ! ReadableStreamHasBYOBReader(stream) is true.
Return stream.[[reader]].[[readIntoRequests]]'s size.

ReadableStreamGetNumReadRequests(stream) performs the following steps:

Assert: ! ReadableStreamHasDefaultReader(stream) is true.
Return stream.[[reader]].[[readRequests]]'s size.

ReadableStreamHasBYOBReader(stream) performs the following steps:

Let reader be stream.[[reader]].
If reader is undefined, return false.
If reader implements ReadableStreamBYOBReader, return true.
Return false.

ReadableStreamHasDefaultReader(stream) performs the following steps:

Let reader be stream.[[reader]].
If reader is undefined, return false.
If reader implements ReadableStreamDefaultReader, return true.
Return false.

4.9.3. Readers

The following abstract operations support the implementation and manipulation of ReadableStreamDefaultReader and ReadableStreamBYOBReader instances.

ReadableStreamReaderGenericCancel(reader, reason) performs the following steps:

Let stream be reader.[[stream]].
Assert: stream is not undefined.
Return ! ReadableStreamCancel(stream, reason).

ReadableStreamReaderGenericInitialize(reader, stream) performs the following steps:

Set reader.[[stream]] to stream.
Set stream.[[reader]] to reader.
If stream.[[state]] is "readable",
1. Set reader.[[closedPromise]] to a new promise.
Otherwise, if stream.[[state]] is "closed",
1. Set reader.[[closedPromise]] to a promise resolved with undefined.
Otherwise,
1. Assert: stream.[[state]] is "errored".
2. Set reader.[[closedPromise]] to a promise rejected with stream.[[storedError]].
3. Set reader.[[closedPromise]].[[PromiseIsHandled]] to true.

ReadableStreamReaderGenericRelease(reader) performs the following steps:

Let stream be reader.[[stream]].
Assert: stream is not undefined.
Assert: stream.[[reader]] is reader.
If stream.[[state]] is "readable", reject reader.[[closedPromise]] with a TypeError exception.
Otherwise, set reader.[[closedPromise]] to a promise rejected with a TypeError exception.
Set reader.[[closedPromise]].[[PromiseIsHandled]] to true.
Perform ! stream.[[controller]].[[ReleaseSteps]]().
Set stream.[[reader]] to undefined.
Set reader.[[stream]] to undefined.

ReadableStreamBYOBReaderErrorReadIntoRequests(reader, e) performs the following steps:

Let readIntoRequests be reader.[[readIntoRequests]].
Set reader.[[readIntoRequests]] to a new empty list.
For each readIntoRequest of readIntoRequests,
1. Perform readIntoRequest’s error steps, given e.

ReadableStreamBYOBReaderRead(reader, view, min, readIntoRequest) performs the following steps:

Let stream be reader.[[stream]].
Assert: stream is not undefined.
Set stream.[[disturbed]] to true.
If stream.[[state]] is "errored", perform readIntoRequest’s error steps given stream.[[storedError]].
Otherwise, perform ! ReadableByteStreamControllerPullInto(stream.[[controller]], view, min, readIntoRequest).

ReadableStreamBYOBReaderRelease(reader) performs the following steps:

Perform ! ReadableStreamReaderGenericRelease(reader).
Let e be a new TypeError exception.
Perform ! ReadableStreamBYOBReaderErrorReadIntoRequests(reader, e).

ReadableStreamDefaultReaderErrorReadRequests(reader, e) performs the following steps:

Let readRequests be reader.[[readRequests]].
Set reader.[[readRequests]] to a new empty list.
For each readRequest of readRequests,
1. Perform readRequest’s error steps, given e.

ReadableStreamDefaultReaderRead(reader, readRequest) performs the following steps:

Let stream be reader.[[stream]].
Assert: stream is not undefined.
Set stream.[[disturbed]] to true.
If stream.[[state]] is "closed", perform readRequest’s close steps.
Otherwise, if stream.[[state]] is "errored", perform readRequest’s error steps given stream.[[storedError]].
Otherwise,
1. Assert: stream.[[state]] is "readable".
2. Perform ! stream.[[controller]].[[PullSteps]](readRequest).

ReadableStreamDefaultReaderRelease(reader) performs the following steps:

Perform ! ReadableStreamReaderGenericRelease(reader).
Let e be a new TypeError exception.
Perform ! ReadableStreamDefaultReaderErrorReadRequests(reader, e).

SetUpReadableStreamBYOBReader(reader, stream) performs the following steps:

If ! IsReadableStreamLocked(stream) is true, throw a TypeError exception.
If stream.[[controller]] does not implement ReadableByteStreamController, throw a TypeError exception.
Perform ! ReadableStreamReaderGenericInitialize(reader, stream).
Set reader.[[readIntoRequests]] to a new empty list.

SetUpReadableStreamDefaultReader(reader, stream) performs the following steps:

If ! IsReadableStreamLocked(stream) is true, throw a TypeError exception.
Perform ! ReadableStreamReaderGenericInitialize(reader, stream).
Set reader.[[readRequests]] to a new empty list.

4.9.4. Default controllers

The following abstract operations support the implementation of the ReadableStreamDefaultController class.

ReadableStreamDefaultControllerCallPullIfNeeded(controller) performs the following steps:

Let shouldPull be ! ReadableStreamDefaultControllerShouldCallPull(controller).
If shouldPull is false, return.
If controller.[[pulling]] is true,
1. Set controller.[[pullAgain]] to true.
2. Return.
Assert: controller.[[pullAgain]] is false.
Set controller.[[pulling]] to true.
Let pullPromise be the result of performing controller.[[pullAlgorithm]].
Upon fulfillment of pullPromise,
1. Set controller.[[pulling]] to false.
2. If controller.[[pullAgain]] is true,
  1. Set controller.[[pullAgain]] to false.
  2. Perform ! ReadableStreamDefaultControllerCallPullIfNeeded(controller).
Upon rejection of pullPromise with reason e,
1. Perform ! ReadableStreamDefaultControllerError(controller, e).

ReadableStreamDefaultControllerShouldCallPull(controller) performs the following steps:

Let stream be controller.[[stream]].
If ! ReadableStreamDefaultControllerCanCloseOrEnqueue(controller) is false, return false.
If controller.[[started]] is false, return false.
If ! IsReadableStreamLocked(stream) is true and ! ReadableStreamGetNumReadRequests(stream) > 0, return true.
Let desiredSize be ! ReadableStreamDefaultControllerGetDesiredSize(controller).
Assert: desiredSize is not null.
If desiredSize > 0, return true.
Return false.

ReadableStreamDefaultControllerClearAlgorithms(controller) is called once the stream is closed or errored and the algorithms will not be executed any more. By removing the algorithm references it permits the underlying source object to be garbage collected even if the ReadableStream itself is still referenced.

This is observable using weak references. See tc39/proposal-weakrefs#31 for more detail.

It performs the following steps:

Set controller.[[pullAlgorithm]] to undefined.
Set controller.[[cancelAlgorithm]] to undefined.
Set controller.[[strategySizeAlgorithm]] to undefined.

ReadableStreamDefaultControllerClose(controller) performs the following steps:

If ! ReadableStreamDefaultControllerCanCloseOrEnqueue(controller) is false, return.
Let stream be controller.[[stream]].
Set controller.[[closeRequested]] to true.
If controller.[[queue]] is empty,
1. Perform ! ReadableStreamDefaultControllerClearAlgorithms(controller).
2. Perform ! ReadableStreamClose(stream).

ReadableStreamDefaultControllerEnqueue(controller, chunk) performs the following steps:

If ! ReadableStreamDefaultControllerCanCloseOrEnqueue(controller) is false, return.
Let stream be controller.[[stream]].
If ! IsReadableStreamLocked(stream) is true and ! ReadableStreamGetNumReadRequests(stream) > 0, perform ! ReadableStreamFulfillReadRequest(stream, chunk, false).
Otherwise,
1. Let result be the result of performing controller.[[strategySizeAlgorithm]], passing in chunk, and interpreting the result as a completion record.
2. If result is an abrupt completion,
  1. Perform ! ReadableStreamDefaultControllerError(controller, result.[[Value]]).
  2. Return result.
3. Let chunkSize be result.[[Value]].
4. Let enqueueResult be EnqueueValueWithSize(controller, chunk, chunkSize).
5. If enqueueResult is an abrupt completion,
  1. Perform ! ReadableStreamDefaultControllerError(controller, enqueueResult.[[Value]]).
  2. Return enqueueResult.
Perform ! ReadableStreamDefaultControllerCallPullIfNeeded(controller).

ReadableStreamDefaultControllerError(controller, e) performs the following steps:

Let stream be controller.[[stream]].
If stream.[[state]] is not "readable", return.
Perform ! ResetQueue(controller).
Perform ! ReadableStreamDefaultControllerClearAlgorithms(controller).
Perform ! ReadableStreamError(stream, e).

ReadableStreamDefaultControllerGetDesiredSize(controller) performs the following steps:

Let state be controller.[[stream]].[[state]].
If state is "errored", return null.
If state is "closed", return 0.
Return controller.[[strategyHWM]] − controller.[[queueTotalSize]].

ReadableStreamDefaultControllerHasBackpressure(controller) is used in the implementation of TransformStream. It performs the following steps:

If ! ReadableStreamDefaultControllerShouldCallPull(controller) is true, return false.
Otherwise, return true.

ReadableStreamDefaultControllerCanCloseOrEnqueue(controller) performs the following steps:

Let state be controller.[[stream]].[[state]].
If controller.[[closeRequested]] is false and state is "readable", return true.
Otherwise, return false.

The case where controller.[[closeRequested]] is false, but state is not "readable", happens when the stream is errored via controller.error(), or when it is closed without its controller’s controller.close() method ever being called: e.g., if the stream was closed by a call to stream.cancel().

SetUpReadableStreamDefaultController(stream, controller, startAlgorithm, pullAlgorithm, cancelAlgorithm, highWaterMark, sizeAlgorithm) performs the following steps:

Assert: stream.[[controller]] is undefined.
Set controller.[[stream]] to stream.
Perform ! ResetQueue(controller).
Set controller.[[started]], controller.[[closeRequested]], controller.[[pullAgain]], and controller.[[pulling]] to false.
Set controller.[[strategySizeAlgorithm]] to sizeAlgorithm and controller.[[strategyHWM]] to highWaterMark.
Set controller.[[pullAlgorithm]] to pullAlgorithm.
Set controller.[[cancelAlgorithm]] to cancelAlgorithm.
Set stream.[[controller]] to controller.
Let startResult be the result of performing startAlgorithm. (This might throw an exception.)
Let startPromise be a promise resolved with startResult.
Upon fulfillment of startPromise,
1. Set controller.[[started]] to true.
2. Assert: controller.[[pulling]] is false.
3. Assert: controller.[[pullAgain]] is false.
4. Perform ! ReadableStreamDefaultControllerCallPullIfNeeded(controller).
Upon rejection of startPromise with reason r,
1. Perform ! ReadableStreamDefaultControllerError(controller, r).

SetUpReadableStreamDefaultControllerFromUnderlyingSource(stream, underlyingSource, underlyingSourceDict, highWaterMark, sizeAlgorithm) performs the following steps:

Let controller be a new ReadableStreamDefaultController.
Let startAlgorithm be an algorithm that returns undefined.
Let pullAlgorithm be an algorithm that returns a promise resolved with undefined.
Let cancelAlgorithm be an algorithm that returns a promise resolved with undefined.
If underlyingSourceDict["start"] exists, then set startAlgorithm to an algorithm which returns the result of invoking underlyingSourceDict["start"] with argument list « controller » and callback this value underlyingSource.
If underlyingSourceDict["pull"] exists, then set pullAlgorithm to an algorithm which returns the result of invoking underlyingSourceDict["pull"] with argument list « controller » and callback this value underlyingSource.
If underlyingSourceDict["cancel"] exists, then set cancelAlgorithm to an algorithm which takes an argument reason and returns the result of invoking underlyingSourceDict["cancel"] with argument list « reason » and callback this value underlyingSource.
Perform ? SetUpReadableStreamDefaultController(stream, controller, startAlgorithm, pullAlgorithm, cancelAlgorithm, highWaterMark, sizeAlgorithm).

4.9.5. Byte stream controllers

ReadableByteStreamControllerCallPullIfNeeded(controller) performs the following steps:

Let shouldPull be ! ReadableByteStreamControllerShouldCallPull(controller).
If shouldPull is false, return.
If controller.[[pulling]] is true,
1. Set controller.[[pullAgain]] to true.
2. Return.
Assert: controller.[[pullAgain]] is false.
Set controller.[[pulling]] to true.
Let pullPromise be the result of performing controller.[[pullAlgorithm]].
Upon fulfillment of pullPromise,
1. Set controller.[[pulling]] to false.
2. If controller.[[pullAgain]] is true,
  1. Set controller.[[pullAgain]] to false.
  2. Perform ! ReadableByteStreamControllerCallPullIfNeeded(controller).
Upon rejection of pullPromise with reason e,
1. Perform ! ReadableByteStreamControllerError(controller, e).

ReadableByteStreamControllerClearAlgorithms(controller) is called once the stream is closed or errored and the algorithms will not be executed any more. By removing the algorithm references it permits the underlying byte source object to be garbage collected even if the ReadableStream itself is still referenced.

This is observable using weak references. See tc39/proposal-weakrefs#31 for more detail.

It performs the following steps:

Set controller.[[pullAlgorithm]] to undefined.
Set controller.[[cancelAlgorithm]] to undefined.

ReadableByteStreamControllerClearPendingPullIntos(controller) performs the following steps:

Perform ! ReadableByteStreamControllerInvalidateBYOBRequest(controller).
Set controller.[[pendingPullIntos]] to a new empty list.

ReadableByteStreamControllerClose(controller) performs the following steps:

Let stream be controller.[[stream]].
If controller.[[closeRequested]] is true or stream.[[state]] is not "readable", return.
If controller.[[queueTotalSize]] > 0,
1. Set controller.[[closeRequested]] to true.
2. Return.
If controller.[[pendingPullIntos]] is not empty,
1. Let firstPendingPullInto be controller.[[pendingPullIntos]][0].
2. If the remainder after dividing firstPendingPullInto’s bytes filled by firstPendingPullInto’s element size is not 0,
  1. Let e be a new TypeError exception.
  2. Perform ! ReadableByteStreamControllerError(controller, e).
  3. Throw e.
Perform ! ReadableByteStreamControllerClearAlgorithms(controller).
Perform ! ReadableStreamClose(stream).

ReadableByteStreamControllerCommitPullIntoDescriptor(stream, pullIntoDescriptor) performs the following steps:

Assert: stream.[[state]] is not "errored".
Assert: pullIntoDescriptor.reader type is not "none".
Let done be false.
If stream.[[state]] is "closed",
1. Assert: the remainder after dividing pullIntoDescriptor’s bytes filled by pullIntoDescriptor’s element size is 0.
2. Set done to true.
Let filledView be ! ReadableByteStreamControllerConvertPullIntoDescriptor(pullIntoDescriptor).
If pullIntoDescriptor’s reader type is "default",
1. Perform ! ReadableStreamFulfillReadRequest(stream, filledView, done).
Otherwise,
1. Assert: pullIntoDescriptor’s reader type is "byob".
2. Perform ! ReadableStreamFulfillReadIntoRequest(stream, filledView, done).

ReadableByteStreamControllerConvertPullIntoDescriptor(pullIntoDescriptor) performs the following steps:

Let bytesFilled be pullIntoDescriptor’s bytes filled.
Let elementSize be pullIntoDescriptor’s element size.
Assert: bytesFilled ≤ pullIntoDescriptor’s byte length.
Assert: the remainder after dividing bytesFilled by elementSize is 0.
Let buffer be ! TransferArrayBuffer(pullIntoDescriptor’s buffer).
Return ! Construct(pullIntoDescriptor’s view constructor, « buffer, pullIntoDescriptor’s byte offset, bytesFilled ÷ elementSize »).

ReadableByteStreamControllerEnqueue(controller, chunk) performs the following steps:

Let stream be controller.[[stream]].
If controller.[[closeRequested]] is true or stream.[[state]] is not "readable", return.
Let buffer be chunk.[[ViewedArrayBuffer]].
Let byteOffset be chunk.[[ByteOffset]].
Let byteLength be chunk.[[ByteLength]].
If ! IsDetachedBuffer(buffer) is true, throw a TypeError exception.
Let transferredBuffer be ? TransferArrayBuffer(buffer).
If controller.[[pendingPullIntos]] is not empty,
1. Let firstPendingPullInto be controller.[[pendingPullIntos]][0].
2. If ! IsDetachedBuffer(firstPendingPullInto’s buffer) is true, throw a TypeError exception.
3. Perform ! ReadableByteStreamControllerInvalidateBYOBRequest(controller).
4. Set firstPendingPullInto’s buffer to ! TransferArrayBuffer(firstPendingPullInto’s buffer).
5. If firstPendingPullInto’s reader type is "none", perform ? ReadableByteStreamControllerEnqueueDetachedPullIntoToQueue(controller, firstPendingPullInto).
If ! ReadableStreamHasDefaultReader(stream) is true,
1. Perform ! ReadableByteStreamControllerProcessReadRequestsUsingQueue(controller).
2. If ! ReadableStreamGetNumReadRequests(stream) is 0,
  1. Assert: controller.[[pendingPullIntos]] is empty.
  2. Perform ! ReadableByteStreamControllerEnqueueChunkToQueue(controller, transferredBuffer, byteOffset, byteLength).
3. Otherwise,
  1. Assert: controller.[[queue]] is empty.
  2. If controller.[[pendingPullIntos]] is not empty,
    1. Assert: controller.[[pendingPullIntos]][0]'s reader type is "default".
    2. Perform ! ReadableByteStreamControllerShiftPendingPullInto(controller).
  3. Let transferredView be ! Construct(%Uint8Array%, « transferredBuffer, byteOffset, byteLength »).
  4. Perform ! ReadableStreamFulfillReadRequest(stream, transferredView, false).
Otherwise, if ! ReadableStreamHasBYOBReader(stream) is true,
1. Perform ! ReadableByteStreamControllerEnqueueChunkToQueue(controller, transferredBuffer, byteOffset, byteLength).
2. Perform ! ReadableByteStreamControllerProcessPullIntoDescriptorsUsingQueue(controller).
Otherwise,
1. Assert: ! IsReadableStreamLocked(stream) is false.
2. Perform ! ReadableByteStreamControllerEnqueueChunkToQueue(controller, transferredBuffer, byteOffset, byteLength).
Perform ! ReadableByteStreamControllerCallPullIfNeeded(controller).

ReadableByteStreamControllerEnqueueChunkToQueue(controller, buffer, byteOffset, byteLength) performs the following steps:

Append a new readable byte stream queue entry with buffer buffer, byte offset byteOffset, and byte length byteLength to controller.[[queue]].
Set controller.[[queueTotalSize]] to controller.[[queueTotalSize]] + byteLength.

ReadableByteStreamControllerEnqueueClonedChunkToQueue(controller, buffer, byteOffset, byteLength) performs the following steps:

Let cloneResult be CloneArrayBuffer(buffer, byteOffset, byteLength, %ArrayBuffer%).
If cloneResult is an abrupt completion,
1. Perform ! ReadableByteStreamControllerError(controller, cloneResult.[[Value]]).
2. Return cloneResult.
Perform ! ReadableByteStreamControllerEnqueueChunkToQueue(controller, cloneResult.[[Value]], 0, byteLength).

ReadableByteStreamControllerEnqueueDetachedPullIntoToQueue(controller, pullIntoDescriptor) performs the following steps:

Assert: pullIntoDescriptor’s reader type is "none".
If pullIntoDescriptor’s bytes filled > 0, perform ? ReadableByteStreamControllerEnqueueClonedChunkToQueue(controller, pullIntoDescriptor’s buffer, pullIntoDescriptor’s byte offset, pullIntoDescriptor’s bytes filled).
Perform ! ReadableByteStreamControllerShiftPendingPullInto(controller).

ReadableByteStreamControllerError(controller, e) performs the following steps:

Let stream be controller.[[stream]].
If stream.[[state]] is not "readable", return.
Perform ! ReadableByteStreamControllerClearPendingPullIntos(controller).
Perform ! ResetQueue(controller).
Perform ! ReadableByteStreamControllerClearAlgorithms(controller).
Perform ! ReadableStreamError(stream, e).

ReadableByteStreamControllerFillHeadPullIntoDescriptor(controller, size, pullIntoDescriptor) performs the following steps:

Assert: either controller.[[pendingPullIntos]] is empty, or controller.[[pendingPullIntos]][0] is pullIntoDescriptor.
Assert: controller.[[byobRequest]] is null.
Set pullIntoDescriptor’s bytes filled to bytes filled + size.

ReadableByteStreamControllerFillPullIntoDescriptorFromQueue(controller, pullIntoDescriptor) performs the following steps:

Let maxBytesToCopy be min(controller.[[queueTotalSize]], pullIntoDescriptor’s byte length − pullIntoDescriptor’s bytes filled).
Let maxBytesFilled be pullIntoDescriptor’s bytes filled + maxBytesToCopy.
Let totalBytesToCopyRemaining be maxBytesToCopy.
Let ready be false.
Assert: pullIntoDescriptor’s bytes filled < pullIntoDescriptor’s minimum fill.
Let remainderBytes be the remainder after dividing maxBytesFilled by pullIntoDescriptor’s element size.
Let maxAlignedBytes be maxBytesFilled − remainderBytes.
If maxAlignedBytes ≥ pullIntoDescriptor’s minimum fill,
1. Set totalBytesToCopyRemaining to maxAlignedBytes − pullIntoDescriptor’s bytes filled.
2. Set ready to true.
  
  A descriptor for a read() request that is not yet filled up to its minimum length will stay at the head of the queue, so the underlying source can keep filling it.
Let queue be controller.[[queue]].
While totalBytesToCopyRemaining > 0,
1. Let headOfQueue be queue[0].
2. Let bytesToCopy be min(totalBytesToCopyRemaining, headOfQueue’s byte length).
3. Let destStart be pullIntoDescriptor’s byte offset + pullIntoDescriptor’s bytes filled.
4. Perform ! CopyDataBlockBytes(pullIntoDescriptor’s buffer.[[ArrayBufferData]], destStart, headOfQueue’s buffer.[[ArrayBufferData]], headOfQueue’s byte offset, bytesToCopy).
5. If headOfQueue’s byte length is bytesToCopy,
  1. Remove queue[0].
6. Otherwise,
  1. Set headOfQueue’s byte offset to headOfQueue’s byte offset + bytesToCopy.
  2. Set headOfQueue’s byte length to headOfQueue’s byte length − bytesToCopy.
7. Set controller.[[queueTotalSize]] to controller.[[queueTotalSize]] − bytesToCopy.
8. Perform ! ReadableByteStreamControllerFillHeadPullIntoDescriptor(controller, bytesToCopy, pullIntoDescriptor).
9. Set totalBytesToCopyRemaining to totalBytesToCopyRemaining − bytesToCopy.
If ready is false,
1. Assert: controller.[[queueTotalSize]] is 0.
2. Assert: pullIntoDescriptor’s bytes filled > 0.
3. Assert: pullIntoDescriptor’s bytes filled < pullIntoDescriptor’s minimum fill.
Return ready.

ReadableByteStreamControllerFillReadRequestFromQueue(controller, readRequest) performs the following steps:

Assert: controller.[[queueTotalSize]] > 0.
Let entry be controller.[[queue]][0].
Remove entry from controller.[[queue]].
Set controller.[[queueTotalSize]] to controller.[[queueTotalSize]] − entry’s byte length.
Perform ! ReadableByteStreamControllerHandleQueueDrain(controller).
Let view be ! Construct(%Uint8Array%, « entry’s buffer, entry’s byte offset, entry’s byte length »).
Perform readRequest’s chunk steps, given view.

ReadableByteStreamControllerGetBYOBRequest(controller) performs the following steps:

If controller.[[byobRequest]] is null and controller.[[pendingPullIntos]] is not empty,
1. Let firstDescriptor be controller.[[pendingPullIntos]][0].
2. Let view be ! Construct(%Uint8Array%, « firstDescriptor’s buffer, firstDescriptor’s byte offset + firstDescriptor’s bytes filled, firstDescriptor’s byte length − firstDescriptor’s bytes filled »).
3. Let byobRequest be a new ReadableStreamBYOBRequest.
4. Set byobRequest.[[controller]] to controller.
5. Set byobRequest.[[view]] to view.
6. Set controller.[[byobRequest]] to byobRequest.
Return controller.[[byobRequest]].

ReadableByteStreamControllerGetDesiredSize(controller) performs the following steps:

Let state be controller.[[stream]].[[state]].
If state is "errored", return null.
If state is "closed", return 0.
Return controller.[[strategyHWM]] − controller.[[queueTotalSize]].

ReadableByteStreamControllerHandleQueueDrain(controller) performs the following steps:

Assert: controller.[[stream]].[[state]] is "readable".
If controller.[[queueTotalSize]] is 0 and controller.[[closeRequested]] is true,
1. Perform ! ReadableByteStreamControllerClearAlgorithms(controller).
2. Perform ! ReadableStreamClose(controller.[[stream]]).
Otherwise,
1. Perform ! ReadableByteStreamControllerCallPullIfNeeded(controller).

ReadableByteStreamControllerInvalidateBYOBRequest(controller) performs the following steps:

If controller.[[byobRequest]] is null, return.
Set controller.[[byobRequest]].[[controller]] to undefined.
Set controller.[[byobRequest]].[[view]] to null.
Set controller.[[byobRequest]] to null.

ReadableByteStreamControllerProcessPullIntoDescriptorsUsingQueue(controller) performs the following steps:

Assert: controller.[[closeRequested]] is false.
While controller.[[pendingPullIntos]] is not empty,
1. If controller.[[queueTotalSize]] is 0, return.
2. Let pullIntoDescriptor be controller.[[pendingPullIntos]][0].
3. If ! ReadableByteStreamControllerFillPullIntoDescriptorFromQueue(controller, pullIntoDescriptor) is true,
  1. Perform ! ReadableByteStreamControllerShiftPendingPullInto(controller).
  2. Perform ! ReadableByteStreamControllerCommitPullIntoDescriptor(controller.[[stream]], pullIntoDescriptor).

ReadableByteStreamControllerProcessReadRequestsUsingQueue(controller) performs the following steps:

Let reader be controller.[[stream]].[[reader]].
Assert: reader implements ReadableStreamDefaultReader.
While reader.[[readRequests]] is not empty,
1. If controller.[[queueTotalSize]] is 0, return.
2. Let readRequest be reader.[[readRequests]][0].
3. Remove readRequest from reader.[[readRequests]].
4. Perform ! ReadableByteStreamControllerFillReadRequestFromQueue(controller, readRequest).

ReadableByteStreamControllerPullInto(controller, view, min, readIntoRequest) performs the following steps:

Let stream be controller.[[stream]].
Let elementSize be 1.
Let ctor be %DataView%.
If view has a [[TypedArrayName]] internal slot (i.e., it is not a DataView),
1. Set elementSize to the element size specified in the typed array constructors table for view.[[TypedArrayName]].
2. Set ctor to the constructor specified in the typed array constructors table for view.[[TypedArrayName]].
Let minimumFill be min × elementSize.
Assert: minimumFill ≥ 0 and minimumFill ≤ view.[[ByteLength]].
Assert: the remainder after dividing minimumFill by elementSize is 0.
Let byteOffset be view.[[ByteOffset]].
Let byteLength be view.[[ByteLength]].
Let bufferResult be TransferArrayBuffer(view.[[ViewedArrayBuffer]]).
If bufferResult is an abrupt completion,
1. Perform readIntoRequest’s error steps, given bufferResult.[[Value]].
2. Return.
Let buffer be bufferResult.[[Value]].
Let pullIntoDescriptor be a new pull-into descriptor with

buffer
buffer
buffer byte length
buffer.[[ArrayBufferByteLength]]
byte offset
byteOffset
byte length
byteLength
bytes filled
0
minimum fill
minimumFill
element size
elementSize
view constructor
ctor
reader type
"byob"
If controller.[[pendingPullIntos]] is not empty,
1. Append pullIntoDescriptor to controller.[[pendingPullIntos]].
2. Perform ! ReadableStreamAddReadIntoRequest(stream, readIntoRequest).
3. Return.
If stream.[[state]] is "closed",
1. Let emptyView be ! Construct(ctor, « pullIntoDescriptor’s buffer, pullIntoDescriptor’s byte offset, 0 »).
2. Perform readIntoRequest’s close steps, given emptyView.
3. Return.
If controller.[[queueTotalSize]] > 0,
1. If ! ReadableByteStreamControllerFillPullIntoDescriptorFromQueue(controller, pullIntoDescriptor) is true,
  1. Let filledView be ! ReadableByteStreamControllerConvertPullIntoDescriptor(pullIntoDescriptor).
  2. Perform ! ReadableByteStreamControllerHandleQueueDrain(controller).
  3. Perform readIntoRequest’s chunk steps, given filledView.
  4. Return.
2. If controller.[[closeRequested]] is true,
  1. Let e be a TypeError exception.
  2. Perform ! ReadableByteStreamControllerError(controller, e).
  3. Perform readIntoRequest’s error steps, given e.
  4. Return.
Append pullIntoDescriptor to controller.[[pendingPullIntos]].
Perform ! ReadableStreamAddReadIntoRequest(stream, readIntoRequest).
Perform ! ReadableByteStreamControllerCallPullIfNeeded(controller).

ReadableByteStreamControllerRespond(controller, bytesWritten) performs the following steps:

Assert: controller.[[pendingPullIntos]] is not empty.
Let firstDescriptor be controller.[[pendingPullIntos]][0].
Let state be controller.[[stream]].[[state]].
If state is "closed",
1. If bytesWritten is not 0, throw a TypeError exception.
Otherwise,
1. Assert: state is "readable".
2. If bytesWritten is 0, throw a TypeError exception.
3. If firstDescriptor’s bytes filled + bytesWritten > firstDescriptor’s byte length, throw a RangeError exception.
Set firstDescriptor’s buffer to ! TransferArrayBuffer(firstDescriptor’s buffer).
Perform ? ReadableByteStreamControllerRespondInternal(controller, bytesWritten).

ReadableByteStreamControllerRespondInClosedState(controller, firstDescriptor) performs the following steps:

Assert: the remainder after dividing firstDescriptor’s bytes filled by firstDescriptor’s element size is 0.
If firstDescriptor’s reader type is "none", perform ! ReadableByteStreamControllerShiftPendingPullInto(controller).
Let stream be controller.[[stream]].
If ! ReadableStreamHasBYOBReader(stream) is true,
1. While ! ReadableStreamGetNumReadIntoRequests(stream) > 0,
  1. Let pullIntoDescriptor be ! ReadableByteStreamControllerShiftPendingPullInto(controller).
  2. Perform ! ReadableByteStreamControllerCommitPullIntoDescriptor(stream, pullIntoDescriptor).

ReadableByteStreamControllerRespondInReadableState(controller, bytesWritten, pullIntoDescriptor) performs the following steps:

Assert: pullIntoDescriptor’s bytes filled + bytesWritten ≤ pullIntoDescriptor’s byte length.
Perform ! ReadableByteStreamControllerFillHeadPullIntoDescriptor(controller, bytesWritten, pullIntoDescriptor).
If pullIntoDescriptor’s reader type is "none",
1. Perform ? ReadableByteStreamControllerEnqueueDetachedPullIntoToQueue(controller, pullIntoDescriptor).
2. Perform ! ReadableByteStreamControllerProcessPullIntoDescriptorsUsingQueue(controller).
3. Return.
If pullIntoDescriptor’s bytes filled < pullIntoDescriptor’s minimum fill, return.

A descriptor for a read() request that is not yet filled up to its minimum length will stay at the head of the queue, so the underlying source can keep filling it.
Perform ! ReadableByteStreamControllerShiftPendingPullInto(controller).
Let remainderSize be the remainder after dividing pullIntoDescriptor’s bytes filled by pullIntoDescriptor’s element size.
If remainderSize > 0,
1. Let end be pullIntoDescriptor’s byte offset + pullIntoDescriptor’s bytes filled.
2. Perform ? ReadableByteStreamControllerEnqueueClonedChunkToQueue(controller, pullIntoDescriptor’s buffer, end − remainderSize, remainderSize).
Set pullIntoDescriptor’s bytes filled to pullIntoDescriptor’s bytes filled − remainderSize.
Perform ! ReadableByteStreamControllerCommitPullIntoDescriptor(controller.[[stream]], pullIntoDescriptor).
Perform ! ReadableByteStreamControllerProcessPullIntoDescriptorsUsingQueue(controller).

ReadableByteStreamControllerRespondInternal(controller, bytesWritten) performs the following steps:

Let firstDescriptor be controller.[[pendingPullIntos]][0].
Assert: ! CanTransferArrayBuffer(firstDescriptor’s buffer) is true.
Perform ! ReadableByteStreamControllerInvalidateBYOBRequest(controller).
Let state be controller.[[stream]].[[state]].
If state is "closed",
1. Assert: bytesWritten is 0.
2. Perform ! ReadableByteStreamControllerRespondInClosedState(controller, firstDescriptor).
Otherwise,
1. Assert: state is "readable".
2. Assert: bytesWritten > 0.
3. Perform ? ReadableByteStreamControllerRespondInReadableState(controller, bytesWritten, firstDescriptor).
Perform ! ReadableByteStreamControllerCallPullIfNeeded(controller).

ReadableByteStreamControllerRespondWithNewView(controller, view) performs the following steps:

Assert: controller.[[pendingPullIntos]] is not empty.
Assert: ! IsDetachedBuffer(view.[[ViewedArrayBuffer]]) is false.
Let firstDescriptor be controller.[[pendingPullIntos]][0].
Let state be controller.[[stream]].[[state]].
If state is "closed",
1. If view.[[ByteLength]] is not 0, throw a TypeError exception.
Otherwise,
1. Assert: state is "readable".
2. If view.[[ByteLength]] is 0, throw a TypeError exception.
If firstDescriptor’s byte offset + firstDescriptor’ bytes filled is not view.[[ByteOffset]], throw a RangeError exception.
If firstDescriptor’s buffer byte length is not view.[[ViewedArrayBuffer]].[[ByteLength]], throw a RangeError exception.
If firstDescriptor’s bytes filled + view.[[ByteLength]] > firstDescriptor’s byte length, throw a RangeError exception.
Let viewByteLength be view.[[ByteLength]].
Set firstDescriptor’s buffer to ? TransferArrayBuffer(view.[[ViewedArrayBuffer]]).
Perform ? ReadableByteStreamControllerRespondInternal(controller, viewByteLength).

ReadableByteStreamControllerShiftPendingPullInto(controller) performs the following steps:

Assert: controller.[[byobRequest]] is null.
Let descriptor be controller.[[pendingPullIntos]][0].
Remove descriptor from controller.[[pendingPullIntos]].
Return descriptor.

ReadableByteStreamControllerShouldCallPull(controller) performs the following steps:

Let stream be controller.[[stream]].
If stream.[[state]] is not "readable", return false.
If controller.[[closeRequested]] is true, return false.
If controller.[[started]] is false, return false.
If ! ReadableStreamHasDefaultReader(stream) is true and ! ReadableStreamGetNumReadRequests(stream) > 0, return true.
If ! ReadableStreamHasBYOBReader(stream) is true and ! ReadableStreamGetNumReadIntoRequests(stream) > 0, return true.
Let desiredSize be ! ReadableByteStreamControllerGetDesiredSize(controller).
Assert: desiredSize is not null.
If desiredSize > 0, return true.
Return false.

SetUpReadableByteStreamController(stream, controller, startAlgorithm, pullAlgorithm, cancelAlgorithm, highWaterMark, autoAllocateChunkSize) performs the following steps:

Assert: stream.[[controller]] is undefined.
If autoAllocateChunkSize is not undefined,
1. Assert: ! IsInteger(autoAllocateChunkSize) is true.
2. Assert: autoAllocateChunkSize is positive.
Set controller.[[stream]] to stream.
Set controller.[[pullAgain]] and controller.[[pulling]] to false.
Set controller.[[byobRequest]] to null.
Perform ! ResetQueue(controller).
Set controller.[[closeRequested]] and controller.[[started]] to false.
Set controller.[[strategyHWM]] to highWaterMark.
Set controller.[[pullAlgorithm]] to pullAlgorithm.
Set controller.[[cancelAlgorithm]] to cancelAlgorithm.
Set controller.[[autoAllocateChunkSize]] to autoAllocateChunkSize.
Set controller.[[pendingPullIntos]] to a new empty list.
Set stream.[[controller]] to controller.
Let startResult be the result of performing startAlgorithm.
Let startPromise be a promise resolved with startResult.
Upon fulfillment of startPromise,
1. Set controller.[[started]] to true.
2. Assert: controller.[[pulling]] is false.
3. Assert: controller.[[pullAgain]] is false.
4. Perform ! ReadableByteStreamControllerCallPullIfNeeded(controller).
Upon rejection of startPromise with reason r,
1. Perform ! ReadableByteStreamControllerError(controller, r).

SetUpReadableByteStreamControllerFromUnderlyingSource(stream, underlyingSource, underlyingSourceDict, highWaterMark) performs the following steps:

Let controller be a new ReadableByteStreamController.
Let startAlgorithm be an algorithm that returns undefined.
Let pullAlgorithm be an algorithm that returns a promise resolved with undefined.
Let cancelAlgorithm be an algorithm that returns a promise resolved with undefined.
If underlyingSourceDict["start"] exists, then set startAlgorithm to an algorithm which returns the result of invoking underlyingSourceDict["start"] with argument list « controller » and callback this value underlyingSource.
If underlyingSourceDict["pull"] exists, then set pullAlgorithm to an algorithm which returns the result of invoking underlyingSourceDict["pull"] with argument list « controller » and callback this value underlyingSource.
If underlyingSourceDict["cancel"] exists, then set cancelAlgorithm to an algorithm which takes an argument reason and returns the result of invoking underlyingSourceDict["cancel"] with argument list « reason » and callback this value underlyingSource.
Let autoAllocateChunkSize be underlyingSourceDict["autoAllocateChunkSize"], if it exists, or undefined otherwise.
If autoAllocateChunkSize is 0, then throw a TypeError exception.
Perform ? SetUpReadableByteStreamController(stream, controller, startAlgorithm, pullAlgorithm, cancelAlgorithm, highWaterMark, autoAllocateChunkSize).

5. Writable streams

5.1. Using writable streams

The usual way to write to a writable stream is to simply pipe a readable stream to it. This ensures that backpressure is respected, so that if the writable stream’s underlying sink is not able to accept data as fast as the readable stream can produce it, the readable stream is informed of this and has a chance to slow down its data production.

readableStream.pipeTo(writableStream)
  .then(() => console.log("All data successfully written!"))
  .catch(e => console.error("Something went wrong!", e));

You can also write directly to writable streams by acquiring a writer and using its write() and close() methods. Since writable streams queue any incoming writes, and take care internally to forward them to the underlying sink in sequence, you can indiscriminately write to a writable stream without much ceremony:

function writeArrayToStream(array, writableStream) {
  const writer = writableStream.getWriter();
  array.forEach(chunk => writer.write(chunk).catch(() => {}));

  return writer.close();
}

writeArrayToStream([1, 2, 3, 4, 5], writableStream)
  .then(() => console.log("All done!"))
  .catch(e => console.error("Error with the stream: " + e));

Note how we use .catch(() => {}) to suppress any rejections from the write() method; we’ll be notified of any fatal errors via a rejection of the close() method, and leaving them un-caught would cause potential unhandledrejection events and console warnings.

In the previous example we only paid attention to the success or failure of the entire stream, by looking at the promise returned by the writer’s close() method. That promise will reject if anything goes wrong with the stream—initializing it, writing to it, or closing it. And it will fulfill once the stream is successfully closed. Often this is all you care about.

However, if you care about the success of writing a specific chunk, you can use the promise returned by the writer’s write() method:

writer.write("i am a chunk of data")
  .then(() => console.log("chunk successfully written!"))
  .catch(e => console.error(e));

What "success" means is up to a given stream instance (or more precisely, its underlying sink) to decide. For example, for a file stream it could simply mean that the OS has accepted the write, and not necessarily that the chunk has been flushed to disk. Some streams might not be able to give such a signal at all, in which case the returned promise will fulfill immediately.

The desiredSize and ready properties of writable stream writers allow producers to more precisely respond to flow control signals from the stream, to keep memory usage below the stream’s specified high water mark. The following example writes an infinite sequence of random bytes to a stream, using desiredSize to determine how many bytes to generate at a given time, and using ready to wait for the backpressure to subside.

async function writeRandomBytesForever(writableStream) {
  const writer = writableStream.getWriter();

  while (true) {
    await writer.ready;

    const bytes = new Uint8Array(writer.desiredSize);
    crypto.getRandomValues(bytes);

    // Purposefully don't await; awaiting writer.ready is enough.
    writer.write(bytes).catch(() => {});
  }
}

writeRandomBytesForever(myWritableStream).catch(e => console.error("Something broke", e));

Note how we don’t await the promise returned by write(); this would be redundant with awaiting the ready promise. Additionally, similar to a previous example, we use the .catch(() => {}) pattern on the promises returned by write(); in this case we’ll be notified about any failures awaiting the ready promise.

To further emphasize how it’s a bad idea to await the promise returned by write(), consider a modification of the above example, where we continue to use the WritableStreamDefaultWriter interface directly, but we don’t control how many bytes we have to write at a given time. In that case, the backpressure-respecting code looks the same:

async function writeSuppliedBytesForever(writableStream, getBytes) {
  const writer = writableStream.getWriter();

  while (true) {
    await writer.ready;

    const bytes = getBytes();
    writer.write(bytes).catch(() => {});
  }
}

Unlike the previous example, where—because we were always writing exactly writer.desiredSize bytes each time—the write() and ready promises were synchronized, in this case it’s quite possible that the ready promise fulfills before the one returned by write() does. Remember, the ready promise fulfills when the desired size becomes positive, which might be before the write succeeds (especially in cases with a larger high water mark).

In other words, awaiting the return value of write() means you never queue up writes in the stream’s internal queue, instead only executing a write after the previous one succeeds, which can result in low throughput.

5.2. The `WritableStream` class

The WritableStream represents a writable stream.

5.2.1. Interface definition

The Web IDL definition for the WritableStream class is given as follows:

[Exposed=*, Transferable]
interface WritableStream {
  constructor(optional object underlyingSink, optional QueuingStrategy strategy = {});

  readonly attribute boolean locked;

  Promise<undefined> abort(optional any reason);
  Promise<undefined> close();
  WritableStreamDefaultWriter getWriter();
};

5.2.2. Internal slots

Instances of WritableStream are created with the internal slots described in the following table:

Internal Slot	Description (non-normative)
[[backpressure]]	A boolean indicating the backpressure signal set by the controller
[[closeRequest]]	The promise returned from the writer’s `close()` method
[[controller]]	A `WritableStreamDefaultController` created with the ability to control the state and queue of this stream
[[Detached]]	A boolean flag set to true when the stream is transferred
[[inFlightWriteRequest]]	A slot set to the promise for the current in-flight write operation while the underlying sink's write algorithm is executing and has not yet fulfilled, used to prevent reentrant calls
[[inFlightCloseRequest]]	A slot set to the promise for the current in-flight close operation while the underlying sink's close algorithm is executing and has not yet fulfilled, used to prevent the `abort()` method from interrupting close
[[pendingAbortRequest]]	A pending abort request
[[state]]	A string containing the stream’s current state, used internally; one of "`writable`", "`closed`", "`erroring`", or "`errored`"
[[storedError]]	A value indicating how the stream failed, to be given as a failure reason or exception when trying to operate on the stream while in the "`errored`" state
[[writer]]	A `WritableStreamDefaultWriter` instance, if the stream is locked to a writer, or undefined if it is not
[[writeRequests]]	A list of promises representing the stream’s internal queue of write requests not yet processed by the underlying sink

The [[inFlightCloseRequest]] slot and [[closeRequest]] slot are mutually exclusive. Similarly, no element will be removed from [[writeRequests]] while [[inFlightWriteRequest]] is not undefined. Implementations can optimize storage for these slots based on these invariants.

A pending abort request is a struct used to track a request to abort the stream before that request is finally processed. It has the following items:

promise: A promise returned from WritableStreamAbort
reason: A JavaScript value that was passed as the abort reason to WritableStreamAbort
was already erroring: A boolean indicating whether or not the stream was in the "erroring" state when WritableStreamAbort was called, which impacts the outcome of the abort request

5.2.3. The underlying sink API

The WritableStream() constructor accepts as its first argument a JavaScript object representing the underlying sink. Such objects can contain any of the following properties:

dictionary UnderlyingSink {
  UnderlyingSinkStartCallback start;
  UnderlyingSinkWriteCallback write;
  UnderlyingSinkCloseCallback close;
  UnderlyingSinkAbortCallback abort;
  any type;
};

callback UnderlyingSinkStartCallback = any (WritableStreamDefaultController controller);
callback UnderlyingSinkWriteCallback = Promise<undefined> (any chunk, WritableStreamDefaultController controller);
callback UnderlyingSinkCloseCallback = Promise<undefined> ();
callback UnderlyingSinkAbortCallback = Promise<undefined> (optional any reason);

start(controller), of type UnderlyingSinkStartCallback

A function that is called immediately during creation of the WritableStream.

Typically this is used to acquire access to the underlying sink resource being represented.

write(chunk, controller), of type UnderlyingSinkWriteCallback

A function that is called when a new chunk of data is ready to be written to the underlying sink. The stream implementation guarantees that this function will be called only after previous writes have succeeded, and never before start() has succeeded or after close() or abort() have been called.

This function is used to actually send the data to the resource presented by the underlying sink, for example by calling a lower-level API.

If the process of writing data is asynchronous, and communicates success or failure signals back to its user, then this function can return a promise to signal success or failure. This promise return value will be communicated back to the caller of writer.write(), so they can monitor that individual write. Throwing an exception is treated the same as returning a rejected promise.

Note that such signals are not always available; compare e.g. § 10.6 A writable stream with no backpressure or success signals with § 10.7 A writable stream with backpressure and success signals. In such cases, it’s best to not return anything.

The promise potentially returned by this function also governs whether the given chunk counts as written for the purposes of computed the desired size to fill the stream’s internal queue. That is, during the time it takes the promise to settle, writer.desiredSize will stay at its previous value, only increasing to signal the desire for more chunks once the write succeeds.

Finally, the promise potentially returned by this function is used to ensure that well-behaved producers do not attempt to mutate the chunk before it has been fully processed. (This is not guaranteed by any specification machinery, but instead is an informal contract between producers and the underlying sink.)

close(), of type UnderlyingSinkCloseCallback

A function that is called after the producer signals, via writer.close(), that they are done writing chunks to the stream, and subsequently all queued-up writes have successfully completed.

This function can perform any actions necessary to finalize or flush writes to the underlying sink, and release access to any held resources.

If the shutdown process is asynchronous, the function can return a promise to signal success or failure; the result will be communicated via the return value of the called writer.close() method. Additionally, a rejected promise will error the stream, instead of letting it close successfully. Throwing an exception is treated the same as returning a rejected promise.

abort(reason), of type UnderlyingSinkAbortCallback

A function that is called after the producer signals, via stream.abort() or writer.abort(), that they wish to abort the stream. It takes as its argument the same value as was passed to those methods by the producer.

Writable streams can additionally be aborted under certain conditions during piping; see the definition of the pipeTo() method for more details.

This function can clean up any held resources, much like close(), but perhaps with some custom handling.

If the shutdown process is asynchronous, the function can return a promise to signal success or failure; the result will be communicated via the return value of the called writer.abort() method. Throwing an exception is treated the same as returning a rejected promise. Regardless, the stream will be errored with a new TypeError indicating that it was aborted.

type, of type any

This property is reserved for future use, so any attempts to supply a value will throw an exception.

The controller argument passed to start() and write() is an instance of WritableStreamDefaultController, and has the ability to error the stream. This is mainly used for bridging the gap with non-promise-based APIs, as seen for example in § 10.6 A writable stream with no backpressure or success signals.

5.2.4. Constructor, methods, and properties

stream = new WritableStream(underlyingSink[, strategy)

Creates a new WritableStream wrapping the provided underlying sink. See § 5.2.3 The underlying sink API for more details on the underlyingSink argument.

isLocked = stream.locked

Returns whether or not the writable stream is locked to a writer.

await stream.abort([ reason ])

Aborts the stream, signaling that the producer can no longer successfully write to the stream and it is to be immediately moved to an errored state, with any queued-up writes discarded. This will also execute any abort mechanism of the underlying sink.

The returned promise will fulfill if the stream shuts down successfully, or reject if the underlying sink signaled that there was an error doing so. Additionally, it will reject with a TypeError (without attempting to cancel the stream) if the stream is currently locked.

await stream.close()

Closes the stream. The underlying sink will finish processing any previously-written chunks, before invoking its close behavior. During this time any further attempts to write will fail (without erroring the stream).

The method returns a promise that will fulfill if all remaining chunks are successfully written and the stream successfully closes, or rejects if an error is encountered during this process. Additionally, it will reject with a TypeError (without attempting to cancel the stream) if the stream is currently locked.

writer = stream.getWriter()

Creates a writer (an instance of WritableStreamDefaultWriter) and locks the stream to the new writer. While the stream is locked, no other writer can be acquired until this one is released.

This functionality is especially useful for creating abstractions that desire the ability to write to a stream without interruption or interleaving. By getting a writer for the stream, you can ensure nobody else can write at the same time, which would cause the resulting written data to be unpredictable and probably useless.

The new WritableStream(underlyingSink, strategy) constructor steps are:

If underlyingSink is missing, set it to null.
Let underlyingSinkDict be underlyingSink, converted to an IDL value of type UnderlyingSink.

We cannot declare the underlyingSink argument as having the UnderlyingSink type directly, because doing so would lose the reference to the original object. We need to retain the object so we can invoke the various methods on it.
If underlyingSinkDict["type"] exists, throw a RangeError exception.

This is to allow us to add new potential types in the future, without backward-compatibility concerns.
Perform ! InitializeWritableStream(this).
Let sizeAlgorithm be ! ExtractSizeAlgorithm(strategy).
Let highWaterMark be ? ExtractHighWaterMark(strategy, 1).
Perform ? SetUpWritableStreamDefaultControllerFromUnderlyingSink(this, underlyingSink, underlyingSinkDict, highWaterMark, sizeAlgorithm).

The locked getter steps are:

Return ! IsWritableStreamLocked(this).

The abort(reason) method steps are:

If ! IsWritableStreamLocked(this) is true, return a promise rejected with a TypeError exception.
Return ! WritableStreamAbort(this, reason).

The close() method steps are:

If ! IsWritableStreamLocked(this) is true, return a promise rejected with a TypeError exception.
If ! WritableStreamCloseQueuedOrInFlight(this) is true, return a promise rejected with a TypeError exception.
Return ! WritableStreamClose(this).

The getWriter() method steps are:

Return ? AcquireWritableStreamDefaultWriter(this).

5.2.5. Transfer via `postMessage()`

destination.postMessage(ws, { transfer: [ws] });

Sends a WritableStream to another frame, window, or worker.

The transferred stream can be used exactly like the original. The original will become locked and no longer directly usable.

WritableStream objects are transferable objects. Their transfer steps, given value and dataHolder, are:

If ! IsWritableStreamLocked(value) is true, throw a "DataCloneError" DOMException.
Let port1 be a new MessagePort in the current Realm.
Let port2 be a new MessagePort in the current Realm.
Entangle port1 and port2.
Let readable be a new ReadableStream in the current Realm.
Perform ! SetUpCrossRealmTransformReadable(readable, port1).
Let promise be ! ReadableStreamPipeTo(readable, value, false, false, false).
Set promise.[[PromiseIsHandled]] to true.
Set dataHolder.[[port]] to ! StructuredSerializeWithTransfer(port2, « port2 »).

Their transfer-receiving steps, given dataHolder and value, are:

Let deserializedRecord be ! StructuredDeserializeWithTransfer(dataHolder.[[port]], the current Realm).
Let port be a deserializedRecord.[[Deserialized]].
Perform ! SetUpCrossRealmTransformWritable(value, port).

5.3. The `WritableStreamDefaultWriter` class

The WritableStreamDefaultWriter class represents a writable stream writer designed to be vended by a WritableStream instance.

5.3.1. Interface definition

The Web IDL definition for the WritableStreamDefaultWriter class is given as follows:

[Exposed=*]
interface WritableStreamDefaultWriter {
  constructor(WritableStream stream);

  readonly attribute Promise<undefined> closed;
  readonly attribute unrestricted double? desiredSize;
  readonly attribute Promise<undefined> ready;

  Promise<undefined> abort(optional any reason);
  Promise<undefined> close();
  undefined releaseLock();
  Promise<undefined> write(optional any chunk);
};

5.3.2. Internal slots

Instances of WritableStreamDefaultWriter are created with the internal slots described in the following table:

Internal Slot	Description (non-normative)
[[closedPromise]]	A promise returned by the writer’s `closed` getter
[[readyPromise]]	A promise returned by the writer’s `ready` getter
[[stream]]	A `WritableStream` instance that owns this reader

5.3.3. Constructor, methods, and properties

writer = new WritableStreamDefaultWriter(stream)

This is equivalent to calling stream.getWriter().

await writer.closed

Returns a promise that will be fulfilled when the stream becomes closed, or rejected if the stream ever errors or the writer’s lock is released before the stream finishes closing.

desiredSize = writer.desiredSize

Returns the desired size to fill the stream’s internal queue. It can be negative, if the queue is over-full. A producer can use this information to determine the right amount of data to write.

It will be null if the stream cannot be successfully written to (due to either being errored, or having an abort queued up). It will return zero if the stream is closed. And the getter will throw an exception if invoked when the writer’s lock is released.

await writer.ready

Returns a promise that will be fulfilled when the desired size to fill the stream’s internal queue transitions from non-positive to positive, signaling that it is no longer applying backpressure. Once the desired size dips back to zero or below, the getter will return a new promise that stays pending until the next transition.

If the stream becomes errored or aborted, or the writer’s lock is released, the returned promise will become rejected.

await writer.abort([ reason ])

If the reader is active, behaves the same as stream.abort(reason).

await writer.close()

If the reader is active, behaves the same as stream.close().

writer.releaseLock()

Releases the writer’s lock on the corresponding stream. After the lock is released, the writer is no longer active. If the associated stream is errored when the lock is released, the writer will appear errored in the same way from now on; otherwise, the writer will appear closed.

Note that the lock can still be released even if some ongoing writes have not yet finished (i.e. even if the promises returned from previous calls to write() have not yet settled). It’s not necessary to hold the lock on the writer for the duration of the write; the lock instead simply prevents other producers from writing in an interleaved manner.

await writer.write(chunk)

Writes the given chunk to the writable stream, by waiting until any previous writes have finished successfully, and then sending the chunk to the underlying sink's write() method. It will return a promise that fulfills with undefined upon a successful write, or rejects if the write fails or stream becomes errored before the writing process is initiated.

Note that what "success" means is up to the underlying sink; it might indicate simply that the chunk has been accepted, and not necessarily that it is safely saved to its ultimate destination.

If chunk is mutable, producers are advised to avoid mutating it after passing it to write(), until after the promise returned by write() settles. This ensures that the underlying sink receives and processes the same value that was passed in.

The new WritableStreamDefaultWriter(stream) constructor steps are:

Perform ? SetUpWritableStreamDefaultWriter(this, stream).

The closed getter steps are:

Return this.[[closedPromise]].

The desiredSize getter steps are:

If this.[[stream]] is undefined, throw a TypeError exception.
Return ! WritableStreamDefaultWriterGetDesiredSize(this).

The ready getter steps are:

Return this.[[readyPromise]].

The abort(reason) method steps are:

If this.[[stream]] is undefined, return a promise rejected with a TypeError exception.
Return ! WritableStreamDefaultWriterAbort(this, reason).

The close() method steps are:

Let stream be this.[[stream]].
If stream is undefined, return a promise rejected with a TypeError exception.
If ! WritableStreamCloseQueuedOrInFlight(stream) is true, return a promise rejected with a TypeError exception.
Return ! WritableStreamDefaultWriterClose(this).

The releaseLock() method steps are:

Let stream be this.[[stream]].
If stream is undefined, return.
Assert: stream.[[writer]] is not undefined.
Perform ! WritableStreamDefaultWriterRelease(this).

The write(chunk) method steps are:

If this.[[stream]] is undefined, return a promise rejected with a TypeError exception.
Return ! WritableStreamDefaultWriterWrite(this, chunk).

5.4. The `WritableStreamDefaultController` class

The WritableStreamDefaultController class has methods that allow control of a WritableStream's state. When constructing a WritableStream, the underlying sink is given a corresponding WritableStreamDefaultController instance to manipulate.

5.4.1. Interface definition

The Web IDL definition for the WritableStreamDefaultController class is given as follows:

[Exposed=*]
interface WritableStreamDefaultController {
  readonly attribute AbortSignal signal;
  undefined error(optional any e);
};

5.4.2. Internal slots

Instances of WritableStreamDefaultController are created with the internal slots described in the following table:

Internal Slot	Description (non-normative)
[[abortAlgorithm]]	A promise-returning algorithm, taking one argument (the abort reason), which communicates a requested abort to the underlying sink
[[abortController]]	An `AbortController` that can be used to abort the pending write or close operation when the stream is aborted.
[[closeAlgorithm]]	A promise-returning algorithm which communicates a requested close to the underlying sink
[[queue]]	A list representing the stream’s internal queue of chunks
[[queueTotalSize]]	The total size of all the chunks stored in [[queue]] (see § 8.1 Queue-with-sizes)
[[started]]	A boolean flag indicating whether the underlying sink has finished starting
[[strategyHWM]]	A number supplied by the creator of the stream as part of the stream’s queuing strategy, indicating the point at which the stream will apply backpressure to its underlying sink
[[strategySizeAlgorithm]]	An algorithm to calculate the size of enqueued chunks, as part of the stream’s queuing strategy
[[stream]]	The `WritableStream` instance controlled
[[writeAlgorithm]]	A promise-returning algorithm, taking one argument (the chunk to write), which writes data to the underlying sink

The close sentinel is a unique value enqueued into [[queue]], in lieu of a chunk, to signal that the stream is closed. It is only used internally, and is never exposed to web developers.

5.4.3. Methods and properties

controller.signal

An AbortSignal that can be used to abort the pending write or close operation when the stream is aborted.

controller.error(e)

Closes the controlled writable stream, making all future interactions with it fail with the given error e.

This method is rarely used, since usually it suffices to return a rejected promise from one of the underlying sink's methods. However, it can be useful for suddenly shutting down a stream in response to an event outside the normal lifecycle of interactions with the underlying sink.

The signal getter steps are:

Return this.[[abortController]]'s signal.

The error(e) method steps are:

Let state be this.[[stream]].[[state]].
If state is not "writable", return.
Perform ! WritableStreamDefaultControllerError(this, e).

5.4.4. Internal methods

The following are internal methods implemented by each WritableStreamDefaultController instance. The writable stream implementation will call into these.

The reason these are in method form, instead of as abstract operations, is to make it clear that the writable stream implementation is decoupled from the controller implementation, and could in the future be expanded with other controllers, as long as those controllers implemented such internal methods. A similar scenario is seen for readable streams (see § 4.9.2 Interfacing with controllers), where there actually are multiple controller types and as such the counterpart internal methods are used polymorphically.

[[AbortSteps]](reason) implements the [[AbortSteps]] contract. It performs the following steps:

Let result be the result of performing this.[[abortAlgorithm]], passing reason.
Perform ! WritableStreamDefaultControllerClearAlgorithms(this).
Return result.

[[ErrorSteps]]() implements the [[ErrorSteps]] contract. It performs the following steps:

Perform ! ResetQueue(this).

5.5. Abstract operations

5.5.1. Working with writable streams

The following abstract operations operate on WritableStream instances at a higher level.

AcquireWritableStreamDefaultWriter(stream) performs the following steps:

Let writer be a new WritableStreamDefaultWriter.
Perform ? SetUpWritableStreamDefaultWriter(writer, stream).
Return writer.

CreateWritableStream(startAlgorithm, writeAlgorithm, closeAlgorithm, abortAlgorithm, highWaterMark, sizeAlgorithm) performs the following steps:

Assert: ! IsNonNegativeNumber(highWaterMark) is true.
Let stream be a new WritableStream.
Perform ! InitializeWritableStream(stream).
Let controller be a new WritableStreamDefaultController.
Perform ? SetUpWritableStreamDefaultController(stream, controller, startAlgorithm, writeAlgorithm, closeAlgorithm, abortAlgorithm, highWaterMark, sizeAlgorithm).
Return stream.

This abstract operation will throw an exception if and only if the supplied startAlgorithm throws.

InitializeWritableStream(stream) performs the following steps:

Set stream.[[state]] to "writable".
Set stream.[[storedError]], stream.[[writer]], stream.[[controller]], stream.[[inFlightWriteRequest]], stream.[[closeRequest]], stream.[[inFlightCloseRequest]], and stream.[[pendingAbortRequest]] to undefined.
Set stream.[[writeRequests]] to a new empty list.
Set stream.[[backpressure]] to false.

IsWritableStreamLocked(stream) performs the following steps:

If stream.[[writer]] is undefined, return false.
Return true.

SetUpWritableStreamDefaultWriter(writer, stream) performs the following steps:

If ! IsWritableStreamLocked(stream) is true, throw a TypeError exception.
Set writer.[[stream]] to stream.
Set stream.[[writer]] to writer.
Let state be stream.[[state]].
If state is "writable",
1. If ! WritableStreamCloseQueuedOrInFlight(stream) is false and stream.[[backpressure]] is true, set writer.[[readyPromise]] to a new promise.
2. Otherwise, set writer.[[readyPromise]] to a promise resolved with undefined.
3. Set writer.[[closedPromise]] to a new promise.
Otherwise, if state is "erroring",
1. Set writer.[[readyPromise]] to a promise rejected with stream.[[storedError]].
2. Set writer.[[readyPromise]].[[PromiseIsHandled]] to true.
3. Set writer.[[closedPromise]] to a new promise.
Otherwise, if state is "closed",
1. Set writer.[[readyPromise]] to a promise resolved with undefined.
2. Set writer.[[closedPromise]] to a promise resolved with undefined.
Otherwise,
1. Assert: state is "errored".
2. Let storedError be stream.[[storedError]].
3. Set writer.[[readyPromise]] to a promise rejected with storedError.
4. Set writer.[[readyPromise]].[[PromiseIsHandled]] to true.
5. Set writer.[[closedPromise]] to a promise rejected with storedError.
6. Set writer.[[closedPromise]].[[PromiseIsHandled]] to true.

WritableStreamAbort(stream, reason) performs the following steps:

If stream.[[state]] is "closed" or "errored", return a promise resolved with undefined.
Signal abort on stream.[[controller]].[[abortController]] with reason.
Let state be stream.[[state]].
If state is "closed" or "errored", return a promise resolved with undefined.

We re-check the state because signaling abort runs author code and that might have changed the state.
If stream.[[pendingAbortRequest]] is not undefined, return stream.[[pendingAbortRequest]]'s promise.
Assert: state is "writable" or "erroring".
Let wasAlreadyErroring be false.
If state is "erroring",
1. Set wasAlreadyErroring to true.
2. Set reason to undefined.
Let promise be a new promise.
Set stream.[[pendingAbortRequest]] to a new pending abort request whose promise is promise, reason is reason, and was already erroring is wasAlreadyErroring.
If wasAlreadyErroring is false, perform ! WritableStreamStartErroring(stream, reason).
Return promise.

WritableStreamClose(stream) performs the following steps:

Let state be stream.[[state]].
If state is "closed" or "errored", return a promise rejected with a TypeError exception.
Assert: state is "writable" or "erroring".
Assert: ! WritableStreamCloseQueuedOrInFlight(stream) is false.
Let promise be a new promise.
Set stream.[[closeRequest]] to promise.
Let writer be stream.[[writer]].
If writer is not undefined, and stream.[[backpressure]] is true, and state is "writable", resolve writer.[[readyPromise]] with undefined.
Perform ! WritableStreamDefaultControllerClose(stream.[[controller]]).
Return promise.

5.5.2. Interfacing with controllers

To allow future flexibility to add different writable stream behaviors (similar to the distinction between default readable streams and readable byte streams), much of the internal state of a writable stream is encapsulated by the WritableStreamDefaultController class.

Each controller class defines two internal methods, which are called by the WritableStream algorithms:

[[AbortSteps]](reason): The controller’s steps that run in reaction to the stream being aborted, used to clean up the state stored in the controller and inform the underlying sink.
[[ErrorSteps]](): The controller’s steps that run in reaction to the stream being errored, used to clean up the state stored in the controller.

(These are defined as internal methods, instead of as abstract operations, so that they can be called polymorphically by the WritableStream algorithms, without having to branch on which type of controller is present. This is a bit theoretical for now, given that only WritableStreamDefaultController exists so far.)

The rest of this section concerns abstract operations that go in the other direction: they are used by the controller implementation to affect its associated WritableStream object. This translates internal state changes of the controllerinto developer-facing results visible through the WritableStream's public API.

WritableStreamAddWriteRequest(stream) performs the following steps:

Assert: ! IsWritableStreamLocked(stream) is true.
Assert: stream.[[state]] is "writable".
Let promise be a new promise.
Append promise to stream.[[writeRequests]].
Return promise.

WritableStreamCloseQueuedOrInFlight(stream) performs the following steps:

If stream.[[closeRequest]] is undefined and stream.[[inFlightCloseRequest]] is undefined, return false.
Return true.

WritableStreamDealWithRejection(stream, error) performs the following steps:

Let state be stream.[[state]].
If state is "writable",
1. Perform ! WritableStreamStartErroring(stream, error).
2. Return.
Assert: state is "erroring".
Perform ! WritableStreamFinishErroring(stream).

WritableStreamFinishErroring(stream) performs the following steps:

Assert: stream.[[state]] is "erroring".
Assert: ! WritableStreamHasOperationMarkedInFlight(stream) is false.
Set stream.[[state]] to "errored".
Perform ! stream.[[controller]].[[ErrorSteps]]().
Let storedError be stream.[[storedError]].
For each writeRequest of stream.[[writeRequests]]:
1. Reject writeRequest with storedError.
Set stream.[[writeRequests]] to an empty list.
If stream.[[pendingAbortRequest]] is undefined,
1. Perform ! WritableStreamRejectCloseAndClosedPromiseIfNeeded(stream).
2. Return.
Let abortRequest be stream.[[pendingAbortRequest]].
Set stream.[[pendingAbortRequest]] to undefined.
If abortRequest’s was already erroring is true,
1. Reject abortRequest’s promise with storedError.
2. Perform ! WritableStreamRejectCloseAndClosedPromiseIfNeeded(stream).
3. Return.
Let promise be ! stream.[[controller]].[[AbortSteps]](abortRequest’s reason).
Upon fulfillment of promise,
1. Resolve abortRequest’s promise with undefined.
2. Perform ! WritableStreamRejectCloseAndClosedPromiseIfNeeded(stream).
Upon rejection of promise with reason reason,
1. Reject abortRequest’s promise with reason.
2. Perform ! WritableStreamRejectCloseAndClosedPromiseIfNeeded(stream).

WritableStreamFinishInFlightClose(stream) performs the following steps:

Assert: stream.[[inFlightCloseRequest]] is not undefined.
Resolve stream.[[inFlightCloseRequest]] with undefined.
Set stream.[[inFlightCloseRequest]] to undefined.
Let state be stream.[[state]].
Assert: stream.[[state]] is "writable" or "erroring".
If state is "erroring",
1. Set stream.[[storedError]] to undefined.
2. If stream.[[pendingAbortRequest]] is not undefined,
  1. Resolve stream.[[pendingAbortRequest]]'s promise with undefined.
  2. Set stream.[[pendingAbortRequest]] to undefined.
Set stream.[[state]] to "closed".
Let writer be stream.[[writer]].
If writer is not undefined, resolve writer.[[closedPromise]] with undefined.
Assert: stream.[[pendingAbortRequest]] is undefined.
Assert: stream.[[storedError]] is undefined.

WritableStreamFinishInFlightCloseWithError(stream, error) performs the following steps:

Assert: stream.[[inFlightCloseRequest]] is not undefined.
Reject stream.[[inFlightCloseRequest]] with error.
Set stream.[[inFlightCloseRequest]] to undefined.
Assert: stream.[[state]] is "writable" or "erroring".
If stream.[[pendingAbortRequest]] is not undefined,
1. Reject stream.[[pendingAbortRequest]]'s promise with error.
2. Set stream.[[pendingAbortRequest]] to undefined.
Perform ! WritableStreamDealWithRejection(stream, error).

WritableStreamFinishInFlightWrite(stream) performs the following steps:

Assert: stream.[[inFlightWriteRequest]] is not undefined.
Resolve stream.[[inFlightWriteRequest]] with undefined.
Set stream.[[inFlightWriteRequest]] to undefined.

WritableStreamFinishInFlightWriteWithError(stream, error) performs the following steps:

Assert: stream.[[inFlightWriteRequest]] is not undefined.
Reject stream.[[inFlightWriteRequest]] with error.
Set stream.[[inFlightWriteRequest]] to undefined.
Assert: stream.[[state]] is "writable" or "erroring".
Perform ! WritableStreamDealWithRejection(stream, error).

WritableStreamHasOperationMarkedInFlight(stream) performs the following steps:

If stream.[[inFlightWriteRequest]] is undefined and stream.[[inFlightCloseRequest]] is undefined, return false.
Return true.

WritableStreamMarkCloseRequestInFlight(stream) performs the following steps:

Assert: stream.[[inFlightCloseRequest]] is undefined.
Assert: stream.[[closeRequest]] is not undefined.
Set stream.[[inFlightCloseRequest]] to stream.[[closeRequest]].
Set stream.[[closeRequest]] to undefined.

WritableStreamMarkFirstWriteRequestInFlight(stream) performs the following steps:

Assert: stream.[[inFlightWriteRequest]] is undefined.
Assert: stream.[[writeRequests]] is not empty.
Let writeRequest be stream.[[writeRequests]][0].
Remove writeRequest from stream.[[writeRequests]].
Set stream.[[inFlightWriteRequest]] to writeRequest.

WritableStreamRejectCloseAndClosedPromiseIfNeeded(stream) performs the following steps:

Assert: stream.[[state]] is "errored".
If stream.[[closeRequest]] is not undefined,
1. Assert: stream.[[inFlightCloseRequest]] is undefined.
2. Reject stream.[[closeRequest]] with stream.[[storedError]].
3. Set stream.[[closeRequest]] to undefined.
Let writer be stream.[[writer]].
If writer is not undefined,
1. Reject writer.[[closedPromise]] with stream.[[storedError]].
2. Set writer.[[closedPromise]].[[PromiseIsHandled]] to true.

WritableStreamStartErroring(stream, reason) performs the following steps:

Assert: stream.[[storedError]] is undefined.
Assert: stream.[[state]] is "writable".
Let controller be stream.[[controller]].
Assert: controller is not undefined.
Set stream.[[state]] to "erroring".
Set stream.[[storedError]] to reason.
Let writer be stream.[[writer]].
If writer is not undefined, perform ! WritableStreamDefaultWriterEnsureReadyPromiseRejected(writer, reason).
If ! WritableStreamHasOperationMarkedInFlight(stream) is false and controller.[[started]] is true, perform ! WritableStreamFinishErroring(stream).

WritableStreamUpdateBackpressure(stream, backpressure) performs the following steps:

Assert: stream.[[state]] is "writable".
Assert: ! WritableStreamCloseQueuedOrInFlight(stream) is false.
Let writer be stream.[[writer]].
If writer is not undefined and backpressure is not stream.[[backpressure]],
1. If backpressure is true, set writer.[[readyPromise]] to a new promise.
2. Otherwise,
  1. Assert: backpressure is false.
  2. Resolve writer.[[readyPromise]] with undefined.
Set stream.[[backpressure]] to backpressure.

5.5.3. Writers

The following abstract operations support the implementation and manipulation of WritableStreamDefaultWriter instances.

WritableStreamDefaultWriterAbort(writer, reason) performs the following steps:

Let stream be writer.[[stream]].
Assert: stream is not undefined.
Return ! WritableStreamAbort(stream, reason).

WritableStreamDefaultWriterClose(writer) performs the following steps:

Let stream be writer.[[stream]].
Assert: stream is not undefined.
Return ! WritableStreamClose(stream).

WritableStreamDefaultWriterCloseWithErrorPropagation(writer) performs the following steps:

Let stream be writer.[[stream]].
Assert: stream is not undefined.
Let state be stream.[[state]].
If ! WritableStreamCloseQueuedOrInFlight(stream) is true or state is "closed", return a promise resolved with undefined.
If state is "errored", return a promise rejected with stream.[[storedError]].
Assert: state is "writable" or "erroring".
Return ! WritableStreamDefaultWriterClose(writer).

This abstract operation helps implement the error propagation semantics of ReadableStream's pipeTo().

WritableStreamDefaultWriterEnsureClosedPromiseRejected(writer, error) performs the following steps:

If writer.[[closedPromise]].[[PromiseState]] is "pending", reject writer.[[closedPromise]] with error.
Otherwise, set writer.[[closedPromise]] to a promise rejected with error.
Set writer.[[closedPromise]].[[PromiseIsHandled]] to true.

WritableStreamDefaultWriterEnsureReadyPromiseRejected(writer, error) performs the following steps:

If writer.[[readyPromise]].[[PromiseState]] is "pending", reject writer.[[readyPromise]] with error.
Otherwise, set writer.[[readyPromise]] to a promise rejected with error.
Set writer.[[readyPromise]].[[PromiseIsHandled]] to true.

WritableStreamDefaultWriterGetDesiredSize(writer) performs the following steps:

Let stream be writer.[[stream]].
Let state be stream.[[state]].
If state is "errored" or "erroring", return null.
If state is "closed", return 0.
Return ! WritableStreamDefaultControllerGetDesiredSize(stream.[[controller]]).

WritableStreamDefaultWriterRelease(writer) performs the following steps:

Let stream be writer.[[stream]].
Assert: stream is not undefined.
Assert: stream.[[writer]] is writer.
Let releasedError be a new TypeError.
Perform ! WritableStreamDefaultWriterEnsureReadyPromiseRejected(writer, releasedError).
Perform ! WritableStreamDefaultWriterEnsureClosedPromiseRejected(writer, releasedError).
Set stream.[[writer]] to undefined.
Set writer.[[stream]] to undefined.

WritableStreamDefaultWriterWrite(writer, chunk) performs the following steps:

Let stream be writer.[[stream]].
Assert: stream is not undefined.
Let controller be stream.[[controller]].
Let chunkSize be ! WritableStreamDefaultControllerGetChunkSize(controller, chunk).
If stream is not equal to writer.[[stream]], return a promise rejected with a TypeError exception.
Let state be stream.[[state]].
If state is "errored", return a promise rejected with stream.[[storedError]].
If ! WritableStreamCloseQueuedOrInFlight(stream) is true or state is "closed", return a promise rejected with a TypeError exception indicating that the stream is closing or closed.
If state is "erroring", return a promise rejected with stream.[[storedError]].
Assert: state is "writable".
Let promise be ! WritableStreamAddWriteRequest(stream).
Perform ! WritableStreamDefaultControllerWrite(controller, chunk, chunkSize).
Return promise.

5.5.4. Default controllers

The following abstract operations support the implementation of the WritableStreamDefaultController class.

SetUpWritableStreamDefaultController(stream, controller, startAlgorithm, writeAlgorithm, closeAlgorithm, abortAlgorithm, highWaterMark, sizeAlgorithm) performs the following steps:

Assert: stream implements WritableStream.
Assert: stream.[[controller]] is undefined.
Set controller.[[stream]] to stream.
Set stream.[[controller]] to controller.
Perform ! ResetQueue(controller).
Set controller.[[abortController]] to a new AbortController.
Set controller.[[started]] to false.
Set controller.[[strategySizeAlgorithm]] to sizeAlgorithm.
Set controller.[[strategyHWM]] to highWaterMark.
Set controller.[[writeAlgorithm]] to writeAlgorithm.
Set controller.[[closeAlgorithm]] to closeAlgorithm.
Set controller.[[abortAlgorithm]] to abortAlgorithm.
Let backpressure be ! WritableStreamDefaultControllerGetBackpressure(controller).
Perform ! WritableStreamUpdateBackpressure(stream, backpressure).
Let startResult be the result of performing startAlgorithm. (This may throw an exception.)
Let startPromise be a promise resolved with startResult.
Upon fulfillment of startPromise,
1. Assert: stream.[[state]] is "writable" or "erroring".
2. Set controller.[[started]] to true.
3. Perform ! WritableStreamDefaultControllerAdvanceQueueIfNeeded(controller).
Upon rejection of startPromise with reason r,
1. Assert: stream.[[state]] is "writable" or "erroring".
2. Set controller.[[started]] to true.
3. Perform ! WritableStreamDealWithRejection(stream, r).

SetUpWritableStreamDefaultControllerFromUnderlyingSink(stream, underlyingSink, underlyingSinkDict, highWaterMark, sizeAlgorithm) performs the following steps:

Let controller be a new WritableStreamDefaultController.
Let startAlgorithm be an algorithm that returns undefined.
Let writeAlgorithm be an algorithm that returns a promise resolved with undefined.
Let closeAlgorithm be an algorithm that returns a promise resolved with undefined.
Let abortAlgorithm be an algorithm that returns a promise resolved with undefined.
If underlyingSinkDict["start"] exists, then set startAlgorithm to an algorithm which returns the result of invoking underlyingSinkDict["start"] with argument list « controller », exception behavior "rethrow", and callback this value underlyingSink.
If underlyingSinkDict["write"] exists, then set writeAlgorithm to an algorithm which takes an argument chunk and returns the result of invoking underlyingSinkDict["write"] with argument list « chunk, controller » and callback this value underlyingSink.
If underlyingSinkDict["close"] exists, then set closeAlgorithm to an algorithm which returns the result of invoking underlyingSinkDict["close"] with argument list «» and callback this value underlyingSink.
If underlyingSinkDict["abort"] exists, then set abortAlgorithm to an algorithm which takes an argument reason and returns the result of invoking underlyingSinkDict["abort"] with argument list « reason » and callback this value underlyingSink.
Perform ? SetUpWritableStreamDefaultController(stream, controller, startAlgorithm, writeAlgorithm, closeAlgorithm, abortAlgorithm, highWaterMark, sizeAlgorithm).

WritableStreamDefaultControllerAdvanceQueueIfNeeded(controller) performs the following steps:

Let stream be controller.[[stream]].
If controller.[[started]] is false, return.
If stream.[[inFlightWriteRequest]] is not undefined, return.
Let state be stream.[[state]].
Assert: state is not "closed" or "errored".
If state is "erroring",
1. Perform ! WritableStreamFinishErroring(stream).
2. Return.
If controller.[[queue]] is empty, return.
Let value be ! PeekQueueValue(controller).
If value is the close sentinel, perform ! WritableStreamDefaultControllerProcessClose(controller).
Otherwise, perform ! WritableStreamDefaultControllerProcessWrite(controller, value).

WritableStreamDefaultControllerClearAlgorithms(controller) is called once the stream is closed or errored and the algorithms will not be executed any more. By removing the algorithm references it permits the underlying sink object to be garbage collected even if the WritableStream itself is still referenced.

This is observable using weak references. See tc39/proposal-weakrefs#31 for more detail.

It performs the following steps:

Set controller.[[writeAlgorithm]] to undefined.
Set controller.[[closeAlgorithm]] to undefined.
Set controller.[[abortAlgorithm]] to undefined.
Set controller.[[strategySizeAlgorithm]] to undefined.

This algorithm will be performed multiple times in some edge cases. After the first time it will do nothing.

WritableStreamDefaultControllerClose(controller) performs the following steps:

Perform ! EnqueueValueWithSize(controller, close sentinel, 0).
Perform ! WritableStreamDefaultControllerAdvanceQueueIfNeeded(controller).

WritableStreamDefaultControllerError(controller, error) performs the following steps:

Let stream be controller.[[stream]].
Assert: stream.[[state]] is "writable".
Perform ! WritableStreamDefaultControllerClearAlgorithms(controller).
Perform ! WritableStreamStartErroring(stream, error).

WritableStreamDefaultControllerErrorIfNeeded(controller, error) performs the following steps:

If controller.[[stream]].[[state]] is "writable", perform ! WritableStreamDefaultControllerError(controller, error).

WritableStreamDefaultControllerGetBackpressure(controller) performs the following steps:

Let desiredSize be ! WritableStreamDefaultControllerGetDesiredSize(controller).
Return true if desiredSize ≤ 0, or false otherwise.

WritableStreamDefaultControllerGetChunkSize(controller, chunk) performs the following steps:

Let returnValue be the result of performing controller.[[strategySizeAlgorithm]], passing in chunk, and interpreting the result as a completion record.
If returnValue is an abrupt completion,
1. Perform ! WritableStreamDefaultControllerErrorIfNeeded(controller, returnValue.[[Value]]).
2. Return 1.
Return returnValue.[[Value]].

WritableStreamDefaultControllerGetDesiredSize(controller) performs the following steps:

Return controller.[[strategyHWM]] − controller.[[queueTotalSize]].

WritableStreamDefaultControllerProcessClose(controller) performs the following steps:

Let stream be controller.[[stream]].
Perform ! WritableStreamMarkCloseRequestInFlight(stream).
Perform ! DequeueValue(controller).
Assert: controller.[[queue]] is empty.
Let sinkClosePromise be the result of performing controller.[[closeAlgorithm]].
Perform ! WritableStreamDefaultControllerClearAlgorithms(controller).
Upon fulfillment of sinkClosePromise,
1. Perform ! WritableStreamFinishInFlightClose(stream).
Upon rejection of sinkClosePromise with reason reason,
1. Perform ! WritableStreamFinishInFlightCloseWithError(stream, reason).

WritableStreamDefaultControllerProcessWrite(controller, chunk) performs the following steps:

Let stream be controller.[[stream]].
Perform ! WritableStreamMarkFirstWriteRequestInFlight(stream).
Let sinkWritePromise be the result of performing controller.[[writeAlgorithm]], passing in chunk.
Upon fulfillment of sinkWritePromise,
1. Perform ! WritableStreamFinishInFlightWrite(stream).
2. Let state be stream.[[state]].
3. Assert: state is "writable" or "erroring".
4. Perform ! DequeueValue(controller).
5. If ! WritableStreamCloseQueuedOrInFlight(stream) is false and state is "writable",
  1. Let backpressure be ! WritableStreamDefaultControllerGetBackpressure(controller).
  2. Perform ! WritableStreamUpdateBackpressure(stream, backpressure).
6. Perform ! WritableStreamDefaultControllerAdvanceQueueIfNeeded(controller).
Upon rejection of sinkWritePromise with reason,
1. If stream.[[state]] is "writable", perform ! WritableStreamDefaultControllerClearAlgorithms(controller).
2. Perform ! WritableStreamFinishInFlightWriteWithError(stream, reason).

WritableStreamDefaultControllerWrite(controller, chunk, chunkSize) performs the following steps:

Let enqueueResult be EnqueueValueWithSize(controller, chunk, chunkSize).
If enqueueResult is an abrupt completion,
1. Perform ! WritableStreamDefaultControllerErrorIfNeeded(controller, enqueueResult.[[Value]]).
2. Return.
Let stream be controller.[[stream]].
If ! WritableStreamCloseQueuedOrInFlight(stream) is false and stream.[[state]] is "writable",
1. Let backpressure be ! WritableStreamDefaultControllerGetBackpressure(controller).
2. Perform ! WritableStreamUpdateBackpressure(stream, backpressure).
Perform ! WritableStreamDefaultControllerAdvanceQueueIfNeeded(controller).

6. Transform streams

6.1. Using transform streams

The natural way to use a transform stream is to place it in a pipe between a readable stream and a writable stream. Chunks that travel from the readable stream to the writable stream will be transformed as they pass through the transform stream. Backpressure is respected, so data will not be read faster than it can be transformed and consumed.

readableStream
  .pipeThrough(transformStream)
  .pipeTo(writableStream)
  .then(() => console.log("All data successfully transformed!"))
  .catch(e => console.error("Something went wrong!", e));

You can also use the readable and writable properties of a transform stream directly to access the usual interfaces of a readable stream and writable stream. In this example we supply data to the writable side of the stream using its writer interface. The readable side is then piped to anotherWritableStream.

const writer = transformStream.writable.getWriter();
writer.write("input chunk");
transformStream.readable.pipeTo(anotherWritableStream);

One use of identity transform streams is to easily convert between readable and writable streams. For example, the fetch() API accepts a readable stream request body, but it can be more convenient to write data for uploading via a writable stream interface. Using an identity transform stream addresses this:

const { writable, readable } = new TransformStream();
fetch("...", { body: readable }).then(response => /* ... */);

const writer = writable.getWriter();
writer.write(new Uint8Array([0x73, 0x74, 0x72, 0x65, 0x61, 0x6D, 0x73, 0x21]));
writer.close();

Another use of identity transform streams is to add additional buffering to a pipe. In this example we add extra buffering between readableStream and writableStream.

const writableStrategy = new ByteLengthQueuingStrategy({ highWaterMark: 1024 * 1024 });

readableStream
  .pipeThrough(new TransformStream(undefined, writableStrategy))
  .pipeTo(writableStream);

6.2. The `TransformStream` class

The TransformStream class is a concrete instance of the general transform stream concept.

6.2.1. Interface definition

The Web IDL definition for the TransformStream class is given as follows:

[Exposed=*, Transferable]
interface TransformStream {
  constructor(optional object transformer,
              optional QueuingStrategy writableStrategy = {},
              optional QueuingStrategy readableStrategy = {});

  readonly attribute ReadableStream readable;
  readonly attribute WritableStream writable;
};

6.2.2. Internal slots

Instances of TransformStream are created with the internal slots described in the following table:

Internal Slot	Description (non-normative)
[[backpressure]]	Whether there was backpressure on [[readable]] the last time it was observed
[[backpressureChangePromise]]	A promise which is fulfilled and replaced every time the value of [[backpressure]] changes
[[controller]]	A `TransformStreamDefaultController` created with the ability to control [[readable]] and [[writable]]
[[Detached]]	A boolean flag set to true when the stream is transferred
[[readable]]	The `ReadableStream` instance controlled by this object
[[writable]]	The `WritableStream` instance controlled by this object

6.2.3. The transformer API

The TransformStream() constructor accepts as its first argument a JavaScript object representing the transformer. Such objects can contain any of the following methods:

dictionary Transformer {
  TransformerStartCallback start;
  TransformerTransformCallback transform;
  TransformerFlushCallback flush;
  TransformerCancelCallback cancel;
  any readableType;
  any writableType;
};

callback TransformerStartCallback = any (TransformStreamDefaultController controller);
callback TransformerFlushCallback = Promise<undefined> (TransformStreamDefaultController controller);
callback TransformerTransformCallback = Promise<undefined> (any chunk, TransformStreamDefaultController controller);
callback TransformerCancelCallback = Promise<undefined> (any reason);

start(controller), of type TransformerStartCallback

A function that is called immediately during creation of the TransformStream.

Typically this is used to enqueue prefix chunks, using controller.enqueue(). Those chunks will be read from the readable side but don’t depend on any writes to the writable side.

If this initial process is asynchronous, for example because it takes some effort to acquire the prefix chunks, the function can return a promise to signal success or failure; a rejected promise will error the stream. Any thrown exceptions will be re-thrown by the TransformStream() constructor.

transform(chunk, controller), of type TransformerTransformCallback

A function called when a new chunk originally written to the writable side is ready to be transformed. The stream implementation guarantees that this function will be called only after previous transforms have succeeded, and never before start() has completed or after flush() has been called.

This function performs the actual transformation work of the transform stream. It can enqueue the results using controller.enqueue(). This permits a single chunk written to the writable side to result in zero or multiple chunks on the readable side, depending on how many times controller.enqueue() is called. § 10.9 A transform stream that replaces template tags demonstrates this by sometimes enqueuing zero chunks.

If the process of transforming is asynchronous, this function can return a promise to signal success or failure of the transformation. A rejected promise will error both the readable and writable sides of the transform stream.

The promise potentially returned by this function is used to ensure that well-behaved producers do not attempt to mutate the chunk before it has been fully transformed. (This is not guaranteed by any specification machinery, but instead is an informal contract between producers and the transformer.)

If no transform() method is supplied, the identity transform is used, which enqueues chunks unchanged from the writable side to the readable side.

flush(controller), of type TransformerFlushCallback

A function called after all chunks written to the writable side have been transformed by successfully passing through transform(), and the writable side is about to be closed.

Typically this is used to enqueue suffix chunks to the readable side, before that too becomes closed. An example can be seen in § 10.9 A transform stream that replaces template tags.

If the flushing process is asynchronous, the function can return a promise to signal success or failure; the result will be communicated to the caller of stream.writable.write(). Additionally, a rejected promise will error both the readable and writable sides of the stream. Throwing an exception is treated the same as returning a rejected promise.

(Note that there is no need to call controller.terminate() inside flush(); the stream is already in the process of successfully closing down, and terminating it would be counterproductive.)

cancel(reason), of type TransformerCancelCallback

A function called when the readable side is cancelled, or when the writable side is aborted.

Typically this is used to clean up underlying transformer resources when the stream is aborted or cancelled.

If the cancellation process is asynchronous, the function can return a promise to signal success or failure; the result will be communicated to the caller of stream.writable.abort() or stream.readable.cancel(). Throwing an exception is treated the same as returning a rejected promise.

(Note that there is no need to call controller.terminate() inside cancel(); the stream is already in the process of cancelling/aborting, and terminating it would be counterproductive.)

readableType, of type any

This property is reserved for future use, so any attempts to supply a value will throw an exception.

writableType, of type any

This property is reserved for future use, so any attempts to supply a value will throw an exception.

The controller object passed to start(), transform(), and flush() is an instance of TransformStreamDefaultController, and has the ability to enqueue chunks to the readable side, or to terminate or error the stream.

6.2.4. Constructor and properties

stream = new TransformStream([transformer[, writableStrategy[, readableStrategy]]])

Creates a new TransformStream wrapping the provided transformer. See § 6.2.3 The transformer API for more details on the transformer argument.

If no transformer argument is supplied, then the result will be an identity transform stream. See this example for some cases where that can be useful.

The writableStrategy and readableStrategy arguments are the queuing strategy objects for the writable and readable sides respectively. These are used in the construction of the WritableStream and ReadableStream objects and can be used to add buffering to a TransformStream, in order to smooth out variations in the speed of the transformation, or to increase the amount of buffering in a pipe. If they are not provided, the default behavior will be the same as a CountQueuingStrategy, with respective high water marks of 1 and 0.

readable = stream.readable

Returns a ReadableStream representing the readable side of this transform stream.

writable = stream.writable

Returns a WritableStream representing the writable side of this transform stream.

The new TransformStream(transformer, writableStrategy, readableStrategy) constructor steps are:

If transformer is missing, set it to null.
Let transformerDict be transformer, converted to an IDL value of type Transformer.

We cannot declare the transformer argument as having the Transformer type directly, because doing so would lose the reference to the original object. We need to retain the object so we can invoke the various methods on it.
If transformerDict["readableType"] exists, throw a RangeError exception.
If transformerDict["writableType"] exists, throw a RangeError exception.
Let readableHighWaterMark be ? ExtractHighWaterMark(readableStrategy, 0).
Let readableSizeAlgorithm be ! ExtractSizeAlgorithm(readableStrategy).
Let writableHighWaterMark be ? ExtractHighWaterMark(writableStrategy, 1).
Let writableSizeAlgorithm be ! ExtractSizeAlgorithm(writableStrategy).
Let startPromise be a new promise.
Perform ! InitializeTransformStream(this, startPromise, writableHighWaterMark, writableSizeAlgorithm, readableHighWaterMark, readableSizeAlgorithm).
Perform ? SetUpTransformStreamDefaultControllerFromTransformer(this, transformer, transformerDict).
If transformerDict["start"] exists, then resolve startPromise with the result of invoking transformerDict["start"] with argument list « this.[[controller]] » and callback this value transformer.
Otherwise, resolve startPromise with undefined.

The readable getter steps are:

Return this.[[readable]].

The writable getter steps are:

Return this.[[writable]].

6.2.5. Transfer via `postMessage()`

destination.postMessage(ts, { transfer: [ts] });

Sends a TransformStream to another frame, window, or worker.

The transferred stream can be used exactly like the original. Its readable and writable sides will become locked and no longer directly usable.

TransformStream objects are transferable objects. Their transfer steps, given value and dataHolder, are:

Let readable be value.[[readable]].
Let writable be value.[[writable]].
If ! IsReadableStreamLocked(readable) is true, throw a "DataCloneError" DOMException.
If ! IsWritableStreamLocked(writable) is true, throw a "DataCloneError" DOMException.
Set dataHolder.[[readable]] to ! StructuredSerializeWithTransfer(readable, « readable »).
Set dataHolder.[[writable]] to ! StructuredSerializeWithTransfer(writable, « writable »).

Their transfer-receiving steps, given dataHolder and value, are:

Let readableRecord be ! StructuredDeserializeWithTransfer(dataHolder.[[readable]], the current Realm).
Let writableRecord be ! StructuredDeserializeWithTransfer(dataHolder.[[writable]], the current Realm).
Set value.[[readable]] to readableRecord.[[Deserialized]].
Set value.[[writable]] to writableRecord.[[Deserialized]].
Set value.[[backpressure]], value.[[backpressureChangePromise]], and value.[[controller]] to undefined.

The [[backpressure]], [[backpressureChangePromise]], and [[controller]] slots are not used in a transferred TransformStream.

6.3. The `TransformStreamDefaultController` class

The TransformStreamDefaultController class has methods that allow manipulation of the associated ReadableStream and WritableStream. When constructing a TransformStream, the transformer object is given a corresponding TransformStreamDefaultController instance to manipulate.

6.3.1. Interface definition

The Web IDL definition for the TransformStreamDefaultController class is given as follows:

[Exposed=*]
interface TransformStreamDefaultController {
  readonly attribute unrestricted double? desiredSize;

  undefined enqueue(optional any chunk);
  undefined error(optional any reason);
  undefined terminate();
};

6.3.2. Internal slots

Instances of TransformStreamDefaultController are created with the internal slots described in the following table:

Internal Slot	Description (non-normative)
[[cancelAlgorithm]]	A promise-returning algorithm, taking one argument (the reason for cancellation), which communicates a requested cancellation to the transformer
[[finishPromise]]	A promise which resolves on completion of either the [[cancelAlgorithm]] or the [[flushAlgorithm]]. If this field is unpopulated (that is, undefined), then neither of those algorithms have been invoked yet
[[flushAlgorithm]]	A promise-returning algorithm which communicates a requested close to the transformer
[[stream]]	The `TransformStream` instance controlled
[[transformAlgorithm]]	A promise-returning algorithm, taking one argument (the chunk to transform), which requests the transformer perform its transformation

6.3.3. Methods and properties

desiredSize = controller.desiredSize: Returns the desired size to fill the readable side’s internal queue. It can be negative, if the queue is over-full.
controller.enqueue(chunk): Enqueues the given chunk chunk in the readable side of the controlled transform stream.
controller.error(e): Errors both the readable side and the writable side of the controlled transform stream, making all future interactions with it fail with the given error e. Any chunks queued for transformation will be discarded.
controller.terminate(): Closes the readable side and errors the writable side of the controlled transform stream. This is useful when the transformer only needs to consume a portion of the chunks written to the writable side.

The desiredSize getter steps are:

Let readableController be this.[[stream]].[[readable]].[[controller]].
Return ! ReadableStreamDefaultControllerGetDesiredSize(readableController).

The enqueue(chunk) method steps are:

Perform ? TransformStreamDefaultControllerEnqueue(this, chunk).

The error(e) method steps are:

Perform ? TransformStreamDefaultControllerError(this, e).

The terminate() method steps are:

Perform ? TransformStreamDefaultControllerTerminate(this).

6.4. Abstract operations

6.4.1. Working with transform streams

The following abstract operations operate on TransformStream instances at a higher level.

InitializeTransformStream(stream, startPromise, writableHighWaterMark, writableSizeAlgorithm, readableHighWaterMark, readableSizeAlgorithm) performs the following steps:

Let startAlgorithm be an algorithm that returns startPromise.
Let writeAlgorithm be the following steps, taking a chunk argument:
1. Return ! TransformStreamDefaultSinkWriteAlgorithm(stream, chunk).
Let abortAlgorithm be the following steps, taking a reason argument:
1. Return ! TransformStreamDefaultSinkAbortAlgorithm(stream, reason).
Let closeAlgorithm be the following steps:
1. Return ! TransformStreamDefaultSinkCloseAlgorithm(stream).
Set stream.[[writable]] to ! CreateWritableStream(startAlgorithm, writeAlgorithm, closeAlgorithm, abortAlgorithm, writableHighWaterMark, writableSizeAlgorithm).
Let pullAlgorithm be the following steps:
1. Return ! TransformStreamDefaultSourcePullAlgorithm(stream).
Let cancelAlgorithm be the following steps, taking a reason argument:
1. Return ! TransformStreamDefaultSourceCancelAlgorithm(stream, reason).
Set stream.[[readable]] to ! CreateReadableStream(startAlgorithm, pullAlgorithm, cancelAlgorithm, readableHighWaterMark, readableSizeAlgorithm).
Set stream.[[backpressure]] and stream.[[backpressureChangePromise]] to undefined.

The [[backpressure]] slot is set to undefined so that it can be initialized by TransformStreamSetBackpressure. Alternatively, implementations can use a strictly boolean value for [[backpressure]] and change the way it is initialized. This will not be visible to user code so long as the initialization is correctly completed before the transformer’s start() method is called.
Perform ! TransformStreamSetBackpressure(stream, true).
Set stream.[[controller]] to undefined.

TransformStreamError(stream, e) performs the following steps:

Perform ! ReadableStreamDefaultControllerError(stream.[[readable]].[[controller]], e).
Perform ! TransformStreamErrorWritableAndUnblockWrite(stream, e).

This operation works correctly when one or both sides are already errored. As a result, calling algorithms do not need to check stream states when responding to an error condition.

TransformStreamErrorWritableAndUnblockWrite(stream, e) performs the following steps:

Perform ! TransformStreamDefaultControllerClearAlgorithms(stream.[[controller]]).
Perform ! WritableStreamDefaultControllerErrorIfNeeded(stream.[[writable]].[[controller]], e).
Perform ! TransformStreamUnblockWrite(stream).

TransformStreamSetBackpressure(stream, backpressure) performs the following steps:

Assert: stream.[[backpressure]] is not backpressure.
If stream.[[backpressureChangePromise]] is not undefined, resolve stream.[[backpressureChangePromise]] with undefined.
Set stream.[[backpressureChangePromise]] to a new promise.
Set stream.[[backpressure]] to backpressure.

TransformStreamUnblockWrite(stream) performs the following steps:

If stream.[[backpressure]] is true, perform ! TransformStreamSetBackpressure(stream, false).

The TransformStreamDefaultSinkWriteAlgorithm abstract operation could be waiting for the promise stored in the [[backpressureChangePromise]] slot to resolve. The call to TransformStreamSetBackpressure ensures that the promise always resolves.

6.4.2. Default controllers

The following abstract operations support the implementaiton of the TransformStreamDefaultController class.

SetUpTransformStreamDefaultController(stream, controller, transformAlgorithm, flushAlgorithm, cancelAlgorithm) performs the following steps:

Assert: stream implements TransformStream.
Assert: stream.[[controller]] is undefined.
Set controller.[[stream]] to stream.
Set stream.[[controller]] to controller.
Set controller.[[transformAlgorithm]] to transformAlgorithm.
Set controller.[[flushAlgorithm]] to flushAlgorithm.
Set controller.[[cancelAlgorithm]] to cancelAlgorithm.

SetUpTransformStreamDefaultControllerFromTransformer(stream, transformer, transformerDict) performs the following steps:

Let controller be a new TransformStreamDefaultController.
Let transformAlgorithm be the following steps, taking a chunk argument:
1. Let result be TransformStreamDefaultControllerEnqueue(controller, chunk).
2. If result is an abrupt completion, return a promise rejected with result.[[Value]].
3. Otherwise, return a promise resolved with undefined.
Let flushAlgorithm be an algorithm which returns a promise resolved with undefined.
Let cancelAlgorithm be an algorithm which returns a promise resolved with undefined.
If transformerDict["transform"] exists, set transformAlgorithm to an algorithm which takes an argument chunk and returns the result of invoking transformerDict["transform"] with argument list « chunk, controller » and callback this value transformer.
If transformerDict["flush"] exists, set flushAlgorithm to an algorithm which returns the result of invoking transformerDict["flush"] with argument list « controller » and callback this value transformer.
If transformerDict["cancel"] exists, set cancelAlgorithm to an algorithm which takes an argument reason and returns the result of invoking transformerDict["cancel"] with argument list « reason » and callback this value transformer.
Perform ! SetUpTransformStreamDefaultController(stream, controller, transformAlgorithm, flushAlgorithm, cancelAlgorithm).

TransformStreamDefaultControllerClearAlgorithms(controller) is called once the stream is closed or errored and the algorithms will not be executed any more. By removing the algorithm references it permits the transformer object to be garbage collected even if the TransformStream itself is still referenced.

This is observable using weak references. See tc39/proposal-weakrefs#31 for more detail.

It performs the following steps:

Set controller.[[transformAlgorithm]] to undefined.
Set controller.[[flushAlgorithm]] to undefined.
Set controller.[[cancelAlgorithm]] to undefined.

TransformStreamDefaultControllerEnqueue(controller, chunk) performs the following steps:

Let stream be controller.[[stream]].
Let readableController be stream.[[readable]].[[controller]].
If ! ReadableStreamDefaultControllerCanCloseOrEnqueue(readableController) is false, throw a TypeError exception.
Let enqueueResult be ReadableStreamDefaultControllerEnqueue(readableController, chunk).
If enqueueResult is an abrupt completion,
1. Perform ! TransformStreamErrorWritableAndUnblockWrite(stream, enqueueResult.[[Value]]).
2. Throw stream.[[readable]].[[storedError]].
Let backpressure be ! ReadableStreamDefaultControllerHasBackpressure(readableController).
If backpressure is not stream.[[backpressure]],
1. Assert: backpressure is true.
2. Perform ! TransformStreamSetBackpressure(stream, true).

TransformStreamDefaultControllerError(controller, e) performs the following steps:

Perform ! TransformStreamError(controller.[[stream]], e).

TransformStreamDefaultControllerPerformTransform(controller, chunk) performs the following steps:

Let transformPromise be the result of performing controller.[[transformAlgorithm]], passing chunk.
Return the result of reacting to transformPromise with the following rejection steps given the argument r:
1. Perform ! TransformStreamError(controller.[[stream]], r).
2. Throw r.

TransformStreamDefaultControllerTerminate(controller) performs the following steps:

Let stream be controller.[[stream]].
Let readableController be stream.[[readable]].[[controller]].
Perform ! ReadableStreamDefaultControllerClose(readableController).
Let error be a TypeError exception indicating that the stream has been terminated.
Perform ! TransformStreamErrorWritableAndUnblockWrite(stream, error).

6.4.3. Default sinks

The following abstract operations are used to implement the underlying sink for the writable side of transform streams.

TransformStreamDefaultSinkWriteAlgorithm(stream, chunk) performs the following steps:

Assert: stream.[[writable]].[[state]] is "writable".
Let controller be stream.[[controller]].
If stream.[[backpressure]] is true,
1. Let backpressureChangePromise be stream.[[backpressureChangePromise]].
2. Assert: backpressureChangePromise is not undefined.
3. Return the result of reacting to backpressureChangePromise with the following fulfillment steps:
  1. Let writable be stream.[[writable]].
  2. Let state be writable.[[state]].
  3. If state is "erroring", throw writable.[[storedError]].
  4. Assert: state is "writable".
  5. Return ! TransformStreamDefaultControllerPerformTransform(controller, chunk).
Return ! TransformStreamDefaultControllerPerformTransform(controller, chunk).

TransformStreamDefaultSinkAbortAlgorithm(stream, reason) performs the following steps:

Let controller be stream.[[controller]].
If controller.[[finishPromise]] is not undefined, return controller.[[finishPromise]].
Let readable be stream.[[readable]].
Let controller.[[finishPromise]] be a new promise.
Let cancelPromise be the result of performing controller.[[cancelAlgorithm]], passing reason.
Perform ! TransformStreamDefaultControllerClearAlgorithms(controller).
React to cancelPromise:
1. If cancelPromise was fulfilled, then:
  1. If readable.[[state]] is "errored", reject controller.[[finishPromise]] with readable.[[storedError]].
  2. Otherwise:
    1. Perform ! ReadableStreamDefaultControllerError(readable.[[controller]], reason).
    2. Resolve controller.[[finishPromise]] with undefined.
2. If cancelPromise was rejected with reason r, then:
  1. Perform ! ReadableStreamDefaultControllerError(readable.[[controller]], r).
  2. Reject controller.[[finishPromise]] with r.
Return controller.[[finishPromise]].

TransformStreamDefaultSinkCloseAlgorithm(stream) performs the following steps:

Let controller be stream.[[controller]].
If controller.[[finishPromise]] is not undefined, return controller.[[finishPromise]].
Let readable be stream.[[readable]].
Let controller.[[finishPromise]] be a new promise.
Let flushPromise be the result of performing controller.[[flushAlgorithm]].
Perform ! TransformStreamDefaultControllerClearAlgorithms(controller).
React to flushPromise:
1. If flushPromise was fulfilled, then:
  1. If readable.[[state]] is "errored", reject controller.[[finishPromise]] with readable.[[storedError]].
  2. Otherwise:
    1. Perform ! ReadableStreamDefaultControllerClose(readable.[[controller]]).
    2. Resolve controller.[[finishPromise]] with undefined.
2. If flushPromise was rejected with reason r, then:
  1. Perform ! ReadableStreamDefaultControllerError(readable.[[controller]], r).
  2. Reject controller.[[finishPromise]] with r.
Return controller.[[finishPromise]].

6.4.4. Default sources

The following abstract operation is used to implement the underlying source for the readable side of transform streams.

TransformStreamDefaultSourceCancelAlgorithm(stream, reason) performs the following steps:

Let controller be stream.[[controller]].
If controller.[[finishPromise]] is not undefined, return controller.[[finishPromise]].
Let writable be stream.[[writable]].
Let controller.[[finishPromise]] be a new promise.
Let cancelPromise be the result of performing controller.[[cancelAlgorithm]], passing reason.
Perform ! TransformStreamDefaultControllerClearAlgorithms(controller).
React to cancelPromise:
1. If cancelPromise was fulfilled, then:
  1. If writable.[[state]] is "errored", reject controller.[[finishPromise]] with writable.[[storedError]].
  2. Otherwise:
    1. Perform ! WritableStreamDefaultControllerErrorIfNeeded(writable.[[controller]], reason).
    2. Perform ! TransformStreamUnblockWrite(stream).
    3. Resolve controller.[[finishPromise]] with undefined.
2. If cancelPromise was rejected with reason r, then:
  1. Perform ! WritableStreamDefaultControllerErrorIfNeeded(writable.[[controller]], r).
  2. Perform ! TransformStreamUnblockWrite(stream).
  3. Reject controller.[[finishPromise]] with r.
Return controller.[[finishPromise]].

TransformStreamDefaultSourcePullAlgorithm(stream) performs the following steps:

Assert: stream.[[backpressure]] is true.
Assert: stream.[[backpressureChangePromise]] is not undefined.
Perform ! TransformStreamSetBackpressure(stream, false).
Return stream.[[backpressureChangePromise]].

7. Queuing strategies

7.1. The queuing strategy API

The ReadableStream(), WritableStream(), and TransformStream() constructors all accept at least one argument representing an appropriate queuing strategy for the stream being created. Such objects contain the following properties:

dictionary QueuingStrategy {
  unrestricted double highWaterMark;
  QueuingStrategySize size;
};

callback QueuingStrategySize = unrestricted double (any chunk);

highWaterMark, of type unrestricted double

A non-negative number indicating the high water mark of the stream using this queuing strategy.

size(chunk) (non-byte streams only), of type QueuingStrategySize

A function that computes and returns the finite non-negative size of the given chunk value.

The result is used to determine backpressure, manifesting via the appropriate desiredSize property: either defaultController.desiredSize, byteController.desiredSize, or writer.desiredSize, depending on where the queuing strategy is being used. For readable streams, it also governs when the underlying source's pull() method is called.

This function has to be idempotent and not cause side effects; very strange results can occur otherwise.

For readable byte streams, this function is not used, as chunks are always measured in bytes.

Any object with these properties can be used when a queuing strategy object is expected. However, we provide two built-in queuing strategy classes that provide a common vocabulary for certain cases: ByteLengthQueuingStrategy and CountQueuingStrategy. They both make use of the following Web IDL fragment for their constructors:

dictionary QueuingStrategyInit {
  required unrestricted double highWaterMark;
};

7.2. The `ByteLengthQueuingStrategy` class

A common queuing strategy when dealing with bytes is to wait until the accumulated byteLength properties of the incoming chunks reaches a specified high-water mark. As such, this is provided as a built-in queuing strategy that can be used when constructing streams.

When creating a readable stream or writable stream, you can supply a byte-length queuing strategy directly:

const stream = new ReadableStream(
  { ... },
  new ByteLengthQueuingStrategy({ highWaterMark: 16 * 1024 })
);

In this case, 16 KiB worth of chunks can be enqueued by the readable stream’s underlying source before the readable stream implementation starts sending backpressure signals to the underlying source.

const stream = new WritableStream(
  { ... },
  new ByteLengthQueuingStrategy({ highWaterMark: 32 * 1024 })
);

In this case, 32 KiB worth of chunks can be accumulated in the writable stream’s internal queue, waiting for previous writes to the underlying sink to finish, before the writable stream starts sending backpressure signals to any producers.

It is not necessary to use ByteLengthQueuingStrategy with readable byte streams, as they always measure chunks in bytes. Attempting to construct a byte stream with a ByteLengthQueuingStrategy will fail.

7.2.1. Interface definition

The Web IDL definition for the ByteLengthQueuingStrategy class is given as follows:

[Exposed=*]
interface ByteLengthQueuingStrategy {
  constructor(QueuingStrategyInit init);

  readonly attribute unrestricted double highWaterMark;
  readonly attribute Function size;
};

7.2.2. Internal slots

Instances of ByteLengthQueuingStrategy have a [[highWaterMark]] internal slot, storing the value given in the constructor.

Additionally, every global object globalObject has an associated byte length queuing strategy size function, which is a Function whose value must be initialized as follows:

Let steps be the following steps, given chunk:
1. Return ? GetV(chunk, "byteLength").
Let F be ! CreateBuiltinFunction(steps, 1, "size", « », globalObject’s relevant Realm).
Set globalObject’s byte length queuing strategy size function to a Function that represents a reference to F, with callback context equal to globalObject’s relevant settings object.

This design is somewhat historical. It is motivated by the desire to ensure that size is a function, not a method, i.e. it does not check its this value. See whatwg/streams#1005 and heycam/webidl#819 for more background.

7.2.3. Constructor and properties

strategy = new ByteLengthQueuingStrategy({ highWaterMark })

Creates a new ByteLengthQueuingStrategy with the provided high water mark.

Note that the provided high water mark will not be validated ahead of time. Instead, if it is negative, NaN, or not a number, the resulting ByteLengthQueuingStrategy will cause the corresponding stream constructor to throw.

highWaterMark = strategy.highWaterMark

Returns the high water mark provided to the constructor.

strategy.size(chunk)

Measures the size of chunk by returning the value of its byteLength property.

The new ByteLengthQueuingStrategy(init) constructor steps are:

Set this.[[highWaterMark]] to init["highWaterMark"].

The highWaterMark getter steps are:

Return this.[[highWaterMark]].

The size getter steps are:

Return this's relevant global object's byte length queuing strategy size function.

7.3. The `CountQueuingStrategy` class

A common queuing strategy when dealing with streams of generic objects is to simply count the number of chunks that have been accumulated so far, waiting until this number reaches a specified high-water mark. As such, this strategy is also provided out of the box.

When creating a readable stream or writable stream, you can supply a count queuing strategy directly:

const stream = new ReadableStream(
  { ... },
  new CountQueuingStrategy({ highWaterMark: 10 })
);

In this case, 10 chunks (of any kind) can be enqueued by the readable stream’s underlying source before the readable stream implementation starts sending backpressure signals to the underlying source.

const stream = new WritableStream(
  { ... },
  new CountQueuingStrategy({ highWaterMark: 5 })
);

In this case, five chunks (of any kind) can be accumulated in the writable stream’s internal queue, waiting for previous writes to the underlying sink to finish, before the writable stream starts sending backpressure signals to any producers.

7.3.1. Interface definition

The Web IDL definition for the CountQueuingStrategy class is given as follows:

[Exposed=*]
interface CountQueuingStrategy {
  constructor(QueuingStrategyInit init);

  readonly attribute unrestricted double highWaterMark;
  readonly attribute Function size;
};

7.3.2. Internal slots

Instances of CountQueuingStrategy have a [[highWaterMark]] internal slot, storing the value given in the constructor.

Additionally, every global object globalObject has an associated count queuing strategy size function, which is a Function whose value must be initialized as follows:

Let steps be the following steps:
1. Return 1.
Let F be ! CreateBuiltinFunction(steps, 0, "size", « », globalObject’s relevant Realm).
Set globalObject’s count queuing strategy size function to a Function that represents a reference to F, with callback context equal to globalObject’s relevant settings object.

7.3.3. Constructor and properties

strategy = new CountQueuingStrategy({ highWaterMark })

Creates a new CountQueuingStrategy with the provided high water mark.

Note that the provided high water mark will not be validated ahead of time. Instead, if it is negative, NaN, or not a number, the resulting CountQueuingStrategy will cause the corresponding stream constructor to throw.

highWaterMark = strategy.highWaterMark

Returns the high water mark provided to the constructor.

strategy.size(chunk)

Measures the size of chunk by always returning 1. This ensures that the total queue size is a count of the number of chunks in the queue.

The new CountQueuingStrategy(init) constructor steps are:

Set this.[[highWaterMark]] to init["highWaterMark"].

The highWaterMark getter steps are:

Return this.[[highWaterMark]].

The size getter steps are:

Return this's relevant global object's count queuing strategy size function.

7.4. Abstract operations

The following algorithms are used by the stream constructors to extract the relevant pieces from a QueuingStrategy dictionary.

ExtractHighWaterMark(strategy, defaultHWM) performs the following steps:

If strategy["highWaterMark"] does not exist, return defaultHWM.
Let highWaterMark be strategy["highWaterMark"].
If highWaterMark is NaN or highWaterMark < 0, throw a RangeError exception.
Return highWaterMark.

+∞ is explicitly allowed as a valid high water mark. It causes backpressure to never be applied.

ExtractSizeAlgorithm(strategy) performs the following steps:

If strategy["size"] does not exist, return an algorithm that returns 1.
Return an algorithm that performs the following steps, taking a chunk argument:
1. Return the result of invoking strategy["size"] with argument list « chunk ».

8. Supporting abstract operations

The following abstract operations each support the implementation of more than one type of stream, and as such are not grouped under the major sections above.

8.1. Queue-with-sizes

The streams in this specification use a "queue-with-sizes" data structure to store queued up values, along with their determined sizes. Various specification objects contain a queue-with-sizes, represented by the object having two paired internal slots, always named [[queue]] and [[queueTotalSize]]. [[queue]] is a list of value-with-sizes, and [[queueTotalSize]] is a JavaScript Number, i.e. a double-precision floating point number.

The following abstract operations are used when operating on objects that contain queues-with-sizes, in order to ensure that the two internal slots stay synchronized.

Due to the limited precision of floating-point arithmetic, the framework specified here, of keeping a running total in the [[queueTotalSize]] slot, is not equivalent to adding up the size of all chunks in [[queue]]. (However, this only makes a difference when there is a huge (~10¹⁵) variance in size between chunks, or when trillions of chunks are enqueued.)

In what follows, a value-with-size is a struct with the two items value and size.

DequeueValue(container) performs the following steps:

Assert: container has [[queue]] and [[queueTotalSize]] internal slots.
Assert: container.[[queue]] is not empty.
Let valueWithSize be container.[[queue]][0].
Remove valueWithSize from container.[[queue]].
Set container.[[queueTotalSize]] to container.[[queueTotalSize]] − valueWithSize’s size.
If container.[[queueTotalSize]] < 0, set container.[[queueTotalSize]] to 0. (This can occur due to rounding errors.)
Return valueWithSize’s value.

EnqueueValueWithSize(container, value, size) performs the following steps:

Assert: container has [[queue]] and [[queueTotalSize]] internal slots.
If ! IsNonNegativeNumber(size) is false, throw a RangeError exception.
If size is +∞, throw a RangeError exception.
Append a new value-with-size with value value and size size to container.[[queue]].
Set container.[[queueTotalSize]] to container.[[queueTotalSize]] + size.

PeekQueueValue(container) performs the following steps:

Assert: container has [[queue]] and [[queueTotalSize]] internal slots.
Assert: container.[[queue]] is not empty.
Let valueWithSize be container.[[queue]][0].
Return valueWithSize’s value.

ResetQueue(container) performs the following steps:

Assert: container has [[queue]] and [[queueTotalSize]] internal slots.
Set container.[[queue]] to a new empty list.
Set container.[[queueTotalSize]] to 0.

8.2. Transferable streams

Transferable streams are implemented using a special kind of identity transform which has the writable side in one realm and the readable side in another realm. The following abstract operations are used to implement these "cross-realm transforms".

CrossRealmTransformSendError(port, error) performs the following steps:

Perform PackAndPostMessage(port, "error", error), discarding the result.

As we are already in an errored state when this abstract operation is performed, we cannot handle further errors, so we just discard them.

PackAndPostMessage(port, type, value) performs the following steps:

Let message be OrdinaryObjectCreate(null).
Perform ! CreateDataProperty(message, "type", type).
Perform ! CreateDataProperty(message, "value", value).
Let targetPort be the port with which port is entangled, if any; otherwise let it be null.
Let options be «[ "transfer" → « » ]».
Run the message port post message steps providing targetPort, message, and options.

A JavaScript object is used for transfer to avoid having to duplicate the message port post message steps. The prototype of the object is set to null to avoid interference from %Object.prototype%.

PackAndPostMessageHandlingError(port, type, value) performs the following steps:

Let result be PackAndPostMessage(port, type, value).
If result is an abrupt completion,
1. Perform ! CrossRealmTransformSendError(port, result.[[Value]]).
Return result as a completion record.

SetUpCrossRealmTransformReadable(stream, port) performs the following steps:

Perform ! InitializeReadableStream(stream).
Let controller be a new ReadableStreamDefaultController.
Add a handler for port’s message event with the following steps:
1. Let data be the data of the message.
2. Assert: data is an Object.
3. Let type be ! Get(data, "type").
4. Let value be ! Get(data, "value").
5. Assert: type is a String.
6. If type is "chunk",
  1. Perform ! ReadableStreamDefaultControllerEnqueue(controller, value).
7. Otherwise, if type is "close",
  1. Perform ! ReadableStreamDefaultControllerClose(controller).
  2. Disentangle port.
8. Otherwise, if type is "error",
  1. Perform ! ReadableStreamDefaultControllerError(controller, value).
  2. Disentangle port.
Add a handler for port’s messageerror event with the following steps:
1. Let error be a new "DataCloneError" DOMException.
2. Perform ! CrossRealmTransformSendError(port, error).
3. Perform ! ReadableStreamDefaultControllerError(controller, error).
4. Disentangle port.
Enable port’s port message queue.
Let startAlgorithm be an algorithm that returns undefined.
Let pullAlgorithm be the following steps:
1. Perform ! PackAndPostMessage(port, "pull", undefined).
2. Return a promise resolved with undefined.
Let cancelAlgorithm be the following steps, taking a reason argument:
1. Let result be PackAndPostMessageHandlingError(port, "error", reason).
2. Disentangle port.
3. If result is an abrupt completion, return a promise rejected with result.[[Value]].
4. Otherwise, return a promise resolved with undefined.
Let sizeAlgorithm be an algorithm that returns 1.
Perform ! SetUpReadableStreamDefaultController(stream, controller, startAlgorithm, pullAlgorithm, cancelAlgorithm, 0, sizeAlgorithm).

Implementations are encouraged to explicitly handle failures from the asserts in this algorithm, as the input might come from an untrusted context. Failure to do so could lead to security issues.

SetUpCrossRealmTransformWritable(stream, port) performs the following steps:

Perform ! InitializeWritableStream(stream).
Let controller be a new WritableStreamDefaultController.
Let backpressurePromise be a new promise.
Add a handler for port’s message event with the following steps:
1. Let data be the data of the message.
2. Assert: data is an Object.
3. Let type be ! Get(data, "type").
4. Let value be ! Get(data, "value").
5. Assert: type is a String.
6. If type is "pull",
  1. If backpressurePromise is not undefined,
    1. Resolve backpressurePromise with undefined.
    2. Set backpressurePromise to undefined.
7. Otherwise, if type is "error",
  1. Perform ! WritableStreamDefaultControllerErrorIfNeeded(controller, value).
  2. If backpressurePromise is not undefined,
    1. Resolve backpressurePromise with undefined.
    2. Set backpressurePromise to undefined.
Add a handler for port’s messageerror event with the following steps:
1. Let error be a new "DataCloneError" DOMException.
2. Perform ! CrossRealmTransformSendError(port, error).
3. Perform ! WritableStreamDefaultControllerErrorIfNeeded(controller, error).
4. Disentangle port.
Enable port’s port message queue.
Let startAlgorithm be an algorithm that returns undefined.
Let writeAlgorithm be the following steps, taking a chunk argument:
1. If backpressurePromise is undefined, set backpressurePromise to a promise resolved with undefined.
2. Return the result of reacting to backpressurePromise with the following fulfillment steps:
  1. Set backpressurePromise to a new promise.
  2. Let result be PackAndPostMessageHandlingError(port, "chunk", chunk).
  3. If result is an abrupt completion,
    1. Disentangle port.
    2. Return a promise rejected with result.[[Value]].
  4. Otherwise, return a promise resolved with undefined.
Let closeAlgorithm be the folowing steps:
1. Perform ! PackAndPostMessage(port, "close", undefined).
2. Disentangle port.
3. Return a promise resolved with undefined.
Let abortAlgorithm be the following steps, taking a reason argument:
1. Let result be PackAndPostMessageHandlingError(port, "error", reason).
2. Disentangle port.
3. If result is an abrupt completion, return a promise rejected with result.[[Value]].
4. Otherwise, return a promise resolved with undefined.
Let sizeAlgorithm be an algorithm that returns 1.
Perform ! SetUpWritableStreamDefaultController(stream, controller, startAlgorithm, writeAlgorithm, closeAlgorithm, abortAlgorithm, 1, sizeAlgorithm).

Implementations are encouraged to explicitly handle failures from the asserts in this algorithm, as the input might come from an untrusted context. Failure to do so could lead to security issues.

8.3. Miscellaneous

The following abstract operations are a grab-bag of utilities.

CanTransferArrayBuffer(O) performs the following steps:

Assert: O is an Object.
Assert: O has an [[ArrayBufferData]] internal slot.
If ! IsDetachedBuffer(O) is true, return false.
If SameValue(O.[[ArrayBufferDetachKey]], undefined) is false, return false.
Return true.

IsNonNegativeNumber(v) performs the following steps:

If v is not a Number, return false.
If v is NaN, return false.
If v < 0, return false.
Return true.

TransferArrayBuffer(O) performs the following steps:

Assert: ! IsDetachedBuffer(O) is false.
Let arrayBufferData be O.[[ArrayBufferData]].
Let arrayBufferByteLength be O.[[ArrayBufferByteLength]].
Perform ? DetachArrayBuffer(O).

This will throw an exception if O has an [[ArrayBufferDetachKey]] that is not undefined, such as a WebAssembly.Memory's buffer. [WASM-JS-API-1]
Return a new ArrayBuffer object, created in the current Realm, whose [[ArrayBufferData]] internal slot value is arrayBufferData and whose [[ArrayBufferByteLength]] internal slot value is arrayBufferByteLength.

CloneAsUint8Array(O) performs the following steps:

Assert: O is an Object.
Assert: O has an [[ViewedArrayBuffer]] internal slot.
Assert: ! IsDetachedBuffer(O.[[ViewedArrayBuffer]]) is false.
Let buffer be ? CloneArrayBuffer(O.[[ViewedArrayBuffer]], O.[[ByteOffset]], O.[[ByteLength]], %ArrayBuffer%).
Let array be ! Construct(%Uint8Array%, « buffer »).
Return array.

StructuredClone(v) performs the following steps:

Let serialized be ? StructuredSerialize(v).
Return ? StructuredDeserialize(serialized, the current Realm).

9. Using streams in other specifications

Much of this standard concerns itself with the internal machinery of streams. Other specifications generally do not need to worry about these details. Instead, they should interface with this standard via the various IDL types it defines, along with the following definitions.

Specifications should not directly inspect or manipulate the various internal slots defined in this standard. Similarly, they should not use the abstract operations defined here. Such direct usage can break invariants that this standard otherwise maintains.

If your specification wants to interface with streams in a way not supported here, file an issue. This section is intended to grow organically as needed.

9.1. Readable streams

9.1.1. Creation and manipulation

To set up a newly-created-via-Web IDL ReadableStream object stream, given an optional algorithm pullAlgorithm, an optional algorithm cancelAlgorithm, an optional number highWaterMark (default 1), and an optional algorithm sizeAlgorithm, perform the following steps. If given, pullAlgorithm and cancelAlgorithm may return a promise. If given, sizeAlgorithm must be an algorithm accepting chunk objects and returning a number; and if given, highWaterMark must be a non-negative, non-NaN number.

Let startAlgorithm be an algorithm that returns undefined.
Let pullAlgorithmWrapper be an algorithm that runs these steps:
1. Let result be the result of running pullAlgorithm, if pullAlgorithm was given, or null otherwise. If this throws an exception e, return a promise rejected with e.
2. If result is a Promise, then return result.
3. Return a promise resolved with undefined.
Let cancelAlgorithmWrapper be an algorithm that runs these steps given reason:
1. Let result be the result of running cancelAlgorithm given reason, if cancelAlgorithm was given, or null otherwise. If this throws an exception e, return a promise rejected with e.
2. If result is a Promise, then return result.
3. Return a promise resolved with undefined.
If sizeAlgorithm was not given, then set it to an algorithm that returns 1.
Perform ! InitializeReadableStream(stream).
Let controller be a new ReadableStreamDefaultController.
Perform ! SetUpReadableStreamDefaultController(stream, controller, startAlgorithm, pullAlgorithmWrapper, cancelAlgorithmWrapper, highWaterMark, sizeAlgorithm).

To set up with byte reading support a newly-created-via-Web IDL ReadableStream object stream, given an optional algorithm pullAlgorithm, an optional algorithm cancelAlgorithm, and an optional number highWaterMark (default 0), perform the following steps. If given, pullAlgorithm and cancelAlgorithm may return a promise. If given, highWaterMark must be a non-negative, non-NaN number.

Let startAlgorithm be an algorithm that returns undefined.
Let pullAlgorithmWrapper be an algorithm that runs these steps:
1. Let result be the result of running pullAlgorithm, if pullAlgorithm was given, or null otherwise. If this throws an exception e, return a promise rejected with e.
2. If result is a Promise, then return result.
3. Return a promise resolved with undefined.
Let cancelAlgorithmWrapper be an algorithm that runs these steps:
1. Let result be the result of running cancelAlgorithm, if cancelAlgorithm was given, or null otherwise. If this throws an exception e, return a promise rejected with e.
2. If result is a Promise, then return result.
3. Return a promise resolved with undefined.
Perform ! InitializeReadableStream(stream).
Let controller be a new ReadableByteStreamController.
Perform ! SetUpReadableByteStreamController(stream, controller, startAlgorithm, pullAlgorithmWrapper, cancelAlgorithmWrapper, highWaterMark, undefined).

Creating a ReadableStream from other specifications is thus a two-step process, like so:

Let readableStream be a new ReadableStream.
Set up readableStream given….

Subclasses of ReadableStream will use the set up or set up with byte reading support operations directly on the this value inside their constructor steps.

The following algorithms must only be used on ReadableStream instances initialized via the above set up or set up with byte reading support algorithms (not, e.g., on web-developer-created instances):

A ReadableStream stream’s desired size to fill up to the high water mark is the result of running the following steps:

If stream is not readable, then return 0.
If stream.[[controller]] implements ReadableByteStreamController, then return ! ReadableByteStreamControllerGetDesiredSize(stream.[[controller]]).
Return ! ReadableStreamDefaultControllerGetDesiredSize(stream.[[controller]]).

A ReadableStream needs more data if its desired size to fill up to the high water mark is greater than zero.

To close a ReadableStream stream:

If stream.[[controller]] implements ReadableByteStreamController,
1. Perform ! ReadableByteStreamControllerClose(stream.[[controller]]).
2. If stream.[[controller]].[[pendingPullIntos]] is not empty, perform ! ReadableByteStreamControllerRespond(stream.[[controller]], 0).
Otherwise, perform ! ReadableStreamDefaultControllerClose(stream.[[controller]]).

To error a ReadableStream stream given a JavaScript value e:

If stream.[[controller]] implements ReadableByteStreamController, then perform ! ReadableByteStreamControllerError(stream.[[controller]], e).
Otherwise, perform ! ReadableStreamDefaultControllerError(stream.[[controller]], e).

To enqueue the JavaScript value chunk into a ReadableStream stream:

If stream.[[controller]] implements ReadableStreamDefaultController,
1. Perform ! ReadableStreamDefaultControllerEnqueue(stream.[[controller]], chunk).
Otherwise,
1. Assert: stream.[[controller]] implements ReadableByteStreamController.
2. Assert: chunk is an ArrayBufferView.
3. Let byobView be the current BYOB request view for stream.
4. If byobView is non-null, and chunk.[[ViewedArrayBuffer]] is byobView.[[ViewedArrayBuffer]], then:
  1. Assert: chunk.[[ByteOffset]] is byobView.[[ByteOffset]].
  2. Assert: chunk.[[ByteLength]] ≤ byobView.[[ByteLength]].
    
    These asserts ensure that the caller does not write outside the requested range in the current BYOB request view.
  3. Perform ? ReadableByteStreamControllerRespond(stream.[[controller]], chunk.[[ByteLength]]).
5. Otherwise, perform ? ReadableByteStreamControllerEnqueue(stream.[[controller]], chunk).

The following algorithms must only be used on ReadableStream instances initialized via the above set up with byte reading support algorithm:

The current BYOB request view for a ReadableStream stream is either an ArrayBufferView or null, determined by the following steps:

Assert: stream.[[controller]] implements ReadableByteStreamController.
Let byobRequest be ! ReadableByteStreamControllerGetBYOBRequest(stream.[[controller]]).
If byobRequest is null, then return null.
Return byobRequest.[[view]].

Specifications must not transfer or detach the underlying buffer of the current BYOB request view.

Implementations could do something equivalent to transferring, e.g. if they want to write into the memory from another thread. But they would need to make a few adjustments to how they implement the enqueue and close algorithms to keep the same observable consequences. In specification-land, transferring and detaching is just disallowed.

Specifications should, when possible, write into the current BYOB request view when it is non-null, and then call enqueue with that view. They should only create a new ArrayBufferView to pass to enqueue when the current BYOB request view is null, or when they have more bytes on hand than the current BYOB request view's byte length. This avoids unnecessary copies and better respects the wishes of the stream’s consumer.

The following pull from bytes algorithm implements these requirements, for the common case where bytes are derived from a byte sequence that serves as the specification-level representation of an underlying byte source. Note that it is conservative and leaves bytes in the byte sequence, instead of aggressively enqueueing them, so callers of this algorithm might want to use the number of remaining bytes as a backpressure signal.

To pull from bytes with a byte sequence bytes into a ReadableStream stream:

Assert: stream.[[controller]] implements ReadableByteStreamController.
Let available be bytes’s length.
Let desiredSize be available.
If stream’s current BYOB request view is non-null, then set desiredSize to stream’s current BYOB request view's byte length.
Let pullSize be the smaller value of available and desiredSize.
Let pulled be the first pullSize bytes of bytes.
Remove the first pullSize bytes from bytes.
If stream’s current BYOB request view is non-null, then:
1. Write pulled into stream’s current BYOB request view.
2. Perform ? ReadableByteStreamControllerRespond(stream.[[controller]], pullSize).
Otherwise,
1. Set view to the result of creating a Uint8Array from pulled in stream’s relevant Realm.
2. Perform ? ReadableByteStreamControllerEnqueue(stream.[[controller]], view).

Specifications must not write into the current BYOB request view or pull from bytes after closing the corresponding ReadableStream.

9.1.2. Reading

The following algorithms can be used on arbitrary ReadableStream instances, including ones that are created by web developers. They can all fail in various operation-specific ways, and these failures should be handled by the calling specification.

To get a reader for a ReadableStream stream, return ? AcquireReadableStreamDefaultReader(stream). The result will be a ReadableStreamDefaultReader.

This will throw an exception if stream is already locked.

To read a chunk from a ReadableStreamDefaultReader reader, given a read request readRequest, perform ! ReadableStreamDefaultReaderRead(reader, readRequest).

To read all bytes from a ReadableStreamDefaultReader reader, given successSteps, which is an algorithm accepting a byte sequence, and failureSteps, which is an algorithm accepting a JavaScript value: read-loop given reader, a new byte sequence, successSteps, and failureSteps.

For the purposes of the above algorithm, to read-loop given reader, bytes, successSteps, and failureSteps:

Let readRequest be a new read request with the following items:
chunk steps, given chunk
1. If chunk is not a Uint8Array object, call failureSteps with a TypeError and abort these steps.
2. Append the bytes represented by chunk to bytes.
3. Read-loop given reader, bytes, successSteps, and failureSteps.
  
  This recursion could potentially cause a stack overflow if implemented directly. Implementations will need to mitigate this, e.g. by using a non-recursive variant of this algorithm, or queuing a microtask, or using a more direct method of byte-reading as noted below.
close steps
1. Call successSteps with bytes.
error steps, given e
1. Call failureSteps with e.
Perform ! ReadableStreamDefaultReaderRead(reader, readRequest).

Because reader grants exclusive access to its corresponding ReadableStream, the actual mechanism of how to read cannot be observed. Implementations could use a more direct mechanism if convenient, such as acquiring and using a ReadableStreamBYOBReader instead of a ReadableStreamDefaultReader, or accessing the chunks directly.

To release a ReadableStreamDefaultReader reader, perform ! ReadableStreamDefaultReaderRelease(reader).

To cancel a ReadableStreamDefaultReader reader with reason, perform ! ReadableStreamReaderGenericCancel(reader, reason). The return value will be a promise that either fulfills with undefined, or rejects with a failure reason.

To cancel a ReadableStream stream with reason, return ! ReadableStreamCancel(stream, reason). The return value will be a promise that either fulfills with undefined, or rejects with a failure reason.

To tee a ReadableStream stream, return ? ReadableStreamTee(stream, true).

Because we pass true as the second argument to ReadableStreamTee, the second branch returned will have its chunks cloned (using HTML’s serializable objects framework) from those of the first branch. This prevents consumption of one of the branches from interfering with the other.

9.1.3. Introspection

The following predicates can be used on arbitrary ReadableStream objects. However, note that apart from checking whether or not the stream is locked, this direct introspection is not possible via the public JavaScript API, and so specifications should instead use the algorithms in § 9.1.2 Reading. (For example, instead of testing if the stream is readable, attempt to get a reader and handle any exception.)

A ReadableStream stream is readable if stream.[[state]] is "readable".

A ReadableStream stream is closed if stream.[[state]] is "closed".

A ReadableStream stream is errored if stream.[[state]] is "errored".

A ReadableStream stream is locked if ! IsReadableStreamLocked(stream) returns true.

A ReadableStream stream is disturbed if stream.[[disturbed]] is true.

This indicates whether the stream has ever been read from or canceled. Even more so than other predicates in this section, it is best consulted sparingly, since this is not information web developers have access to even indirectly. As such, branching platform behavior on it is undesirable.

9.2. Writable streams

9.2.1. Creation and manipulation

To set up a newly-created-via-Web IDL WritableStream object stream, given an algorithm writeAlgorithm, an optional algorithm closeAlgorithm, an optional algorithm abortAlgorithm, an optional number highWaterMark (default 1), an optional algorithm sizeAlgorithm, perform the following steps. writeAlgorithm must be an algorithm that accepts a chunk object and returns a promise. If given, closeAlgorithm and abortAlgorithm may return a promise. If given, sizeAlgorithm must be an algorithm accepting chunk objects and returning a number; and if given, highWaterMark must be a non-negative, non-NaN number.

Let startAlgorithm be an algorithm that returns undefined.
Let closeAlgorithmWrapper be an algorithm that runs these steps:
1. Let result be the result of running closeAlgorithm, if closeAlgorithm was given, or null otherwise. If this throws an exception e, return a promise rejected with e.
2. If result is a Promise, then return result.
3. Return a promise resolved with undefined.
Let abortAlgorithmWrapper be an algorithm that runs these steps given reason:
1. Let result be the result of running abortAlgorithm given reason, if abortAlgorithm was given, or null otherwise. If this throws an exception e, return a promise rejected with e.
2. If result is a Promise, then return result.
3. Return a promise resolved with undefined.
If sizeAlgorithm was not given, then set it to an algorithm that returns 1.
Perform ! InitializeWritableStream(stream).
Let controller be a new WritableStreamDefaultController.
Perform ! SetUpWritableStreamDefaultController(stream, controller, startAlgorithm, writeAlgorithm, closeAlgorithmWrapper, abortAlgorithmWrapper, highWaterMark, sizeAlgorithm).

Other specifications should be careful when constructing their writeAlgorithm to avoid in parallel reads from the given chunk, as such reads can violate the run-to-completion semantics of JavaScript. To avoid this, they can make a synchronous copy or transfer of the given value, using operations such as StructuredSerializeWithTransfer, get a copy of the bytes held by the buffer source, or transferring an ArrayBuffer. An exception is when the chunk is a SharedArrayBuffer, for which it is understood that parallel mutations are a fact of life.

Creating a WritableStream from other specifications is thus a two-step process, like so:

Let writableStream be a new WritableStream.
Set up writableStream given….

Subclasses of WritableStream will use the set up operation directly on the this value inside their constructor steps.

The following definitions must only be used on WritableStream instances initialized via the above set up algorithm:

To error a WritableStream stream given a JavaScript value e, perform ! WritableStreamDefaultControllerErrorIfNeeded(stream.[[controller]], e).

The signal of a WritableStream stream is stream.[[controller]].[[abortController]]'s signal. Specifications can add or remove algorithms to this AbortSignal, or consult whether it is aborted and its abort reason.

The usual usage is, after setting up the WritableStream, add an algorithm to its signal, which aborts any ongoing write operation to the underlying sink. Then, inside the writeAlgorithm, once the underlying sink has responded, check if the signal is aborted, and reject the returned promise with the signal’s abort reason if so.

9.2.2. Writing

The following algorithms can be used on arbitrary WritableStream instances, including ones that are created by web developers. They can all fail in various operation-specific ways, and these failures should be handled by the calling specification.

To get a writer for a WritableStream stream, return ? AcquireWritableStreamDefaultWriter(stream). The result will be a WritableStreamDefaultWriter.

This will throw an exception if stream is already locked.

To write a chunk to a WritableStreamDefaultWriter writer, given a value chunk, return ! WritableStreamDefaultWriterWrite(writer, chunk).

To release a WritableStreamDefaultWriter writer, perform ! WritableStreamDefaultWriterRelease(writer).

To close a WritableStream stream, return ! WritableStreamClose(stream). The return value will be a promise that either fulfills with undefined, or rejects with a failure reason.

To abort a WritableStream stream with reason, return ! WritableStreamAbort(stream, reason). The return value will be a promise that either fulfills with undefined, or rejects with a failure reason.

9.3. Transform streams

9.3.1. Creation and manipulation

To set up a newly-created-via-Web IDL TransformStream stream given an algorithm transformAlgorithm, an optional algorithm flushAlgorithm, and an optional algorithm cancelAlgorithm, perform the following steps. transformAlgorithm and, if given, flushAlgorithm and cancelAlgorithm, may return a promise.

Let writableHighWaterMark be 1.
Let writableSizeAlgorithm be an algorithm that returns 1.
Let readableHighWaterMark be 0.
Let readableSizeAlgorithm be an algorithm that returns 1.
Let transformAlgorithmWrapper be an algorithm that runs these steps given a value chunk:
1. Let result be the result of running transformAlgorithm given chunk. If this throws an exception e, return a promise rejected with e.
2. If result is a Promise, then return result.
3. Return a promise resolved with undefined.
Let flushAlgorithmWrapper be an algorithm that runs these steps:
1. Let result be the result of running flushAlgorithm, if flushAlgorithm was given, or null otherwise. If this throws an exception e, return a promise rejected with e.
2. If result is a Promise, then return result.
3. Return a promise resolved with undefined.
Let cancelAlgorithmWrapper be an algorithm that runs these steps given a value reason:
1. Let result be the result of running cancelAlgorithm given reason, if cancelAlgorithm was given, or null otherwise. If this throws an exception e, return a promise rejected with e.
2. If result is a Promise, then return result.
3. Return a promise resolved with undefined.
Let startPromise be a promise resolved with undefined.
Perform ! InitializeTransformStream(stream, startPromise, writableHighWaterMark, writableSizeAlgorithm, readableHighWaterMark, readableSizeAlgorithm).
Let controller be a new TransformStreamDefaultController.
Perform ! SetUpTransformStreamDefaultController(stream, controller, transformAlgorithmWrapper, flushAlgorithmWrapper, cancelAlgorithmWrapper).

Other specifications should be careful when constructing their transformAlgorithm to avoid in parallel reads from the given chunk, as such reads can violate the run-to-completion semantics of JavaScript. To avoid this, they can make a synchronous copy or transfer of the given value, using operations such as StructuredSerializeWithTransfer, get a copy of the bytes held by the buffer source, or transferring an ArrayBuffer. An exception is when the chunk is a SharedArrayBuffer, for which it is understood that parallel mutations are a fact of life.

Creating a TransformStream from other specifications is thus a two-step process, like so:

Let transformStream be a new TransformStream.
Set up transformStream given….

Subclasses of TransformStream will use the set up operation directly on the this value inside their constructor steps.

To create an identity TransformStream:

Let transformStream be a new TransformStream.
Set up transformStream with transformAlgorithm set to an algorithm which, given chunk, enqueues chunk in transformStream.
Return transformStream.

The following algorithms must only be used on TransformStream instances initialized via the above set up algorithm. Usually they are called as part of transformAlgorithm or flushAlgorithm.

To enqueue the JavaScript value chunk into a TransformStream stream, perform ! TransformStreamDefaultControllerEnqueue(stream.[[controller]], chunk).

To terminate a TransformStream stream, perform ! TransformStreamDefaultControllerTerminate(stream.[[controller]]).

To error a TransformStream stream given a JavaScript value e, perform ! TransformStreamDefaultControllerError(stream.[[controller]], e).

9.3.2. Wrapping into a custom class

Other specifications which mean to define custom transform streams might not want to subclass from the TransformStream interface directly. Instead, if they need a new class, they can create their own independent Web IDL interfaces, and use the following mixin:

interface mixin GenericTransformStream {
  readonly attribute ReadableStream readable;
  readonly attribute WritableStream writable;
};

Any platform object that includes the GenericTransformStream mixin has an associated transform, which is an actual TransformStream.

The readable getter steps are to return this's transform.[[readable]].

The writable getter steps are to return this's transform.[[writable]].

Including the GenericTransformStream mixin will give an IDL interface the appropriate readable and writable properties. To customize the behavior of the resulting interface, its constructor (or other initialization code) must set each instance’s transform to a new TransformStream, and then set it up with appropriate customizations via the transformAlgorithm and optionally flushAlgorithm arguments.

Note: Existing examples of this pattern on the web platform include CompressionStream and TextDecoderStream. [COMPRESSION] [ENCODING]

There’s no need to create a wrapper class if you don’t need any API beyond what the base TransformStream class provides. The most common driver for such a wrapper is needing custom constructor steps, but if your conceptual transform stream isn’t meant to be constructed, then using TransformStream directly is fine.

9.4. Other stream pairs

Apart from transform streams, discussed above, specifications often create pairs of readable and writable streams. This section gives some guidance for such situations.

In all such cases, specifications should use the names readable and writable for the two properties exposing the streams in question. They should not use other names (such as input/output or readableStream/writableStream), and they should not use methods or other non-property means of access to the streams.

9.4.1. Duplex streams

The most common readable/writable pair is a duplex stream, where the readable and writable streams represent two sides of a single shared resource, such as a socket, connection, or device.

The trickiest thing to consider when specifying duplex streams is how to handle operations like canceling the readable side, or closing or aborting the writable side. It might make sense to leave duplex streams "half open", with such operations one one side not impacting the other side. Or it might be best to carry over their effects to the other side, e.g. by specifying that your readable side’s cancelAlgorithm will close the writable side.

A basic example of a duplex stream, created through JavaScript instead of through specification prose, is found in § 10.8 A { readable, writable } stream pair wrapping the same underlying resource. It illustrates this carry-over behavior.

Another consideration is how to handle the creation of duplex streams which need to be acquired asynchronously, e.g. via establishing a connection. The preferred pattern here is to have a constructible class with a promise-returning property that fulfills with the actual duplex stream object. That duplex stream object can also then expose any information that is only available asynchronously, e.g. connection data. The container class can then provide convenience APIs, such as a function to close the entire connection instead of only closing individual sides.

An example of this more complex type of duplex stream is the still-being-specified WebSocketStream. See its explainer and design notes.

Because duplex streams obey the readable/writable property contract, they can be used with pipeThrough(). This doesn’t always make sense, but it could in cases where the underlying resource is in fact performing some sort of transformation.

For an arbitrary WebSocket, piping through a WebSocket-derived duplex stream doesn’t make sense. However, if the WebSocket server is specifically written so that it responds to incoming messages by sending the same data back in some transformed form, then this could be useful and convenient.

9.4.2. Endpoint pairs

Another type of readable/writable pair is an endpoint pair. In these cases the readable and writable streams represent the two ends of a longer pipeline, with the intention that web developer code insert transform streams into the middle of them.

Assuming we had a web-platform-provided function createEndpointPair(), web developers would write code like so:

const { readable, writable } = createEndpointPair();
await readable.pipeThrough(new TransformStream(...)).pipeTo(writable);

WebRTC Encoded Transform is an example of this technique, with its RTCRtpScriptTransformer interface which has both readable and writable attributes.

Despite such endpoint pairs obeying the readable/writable property contract, it never makes sense to pass them to pipeThrough().

9.5. Piping

The result of a ReadableStream readable piped to a WritableStream writable, given an optional boolean preventClose (default false), an optional boolean preventAbort (default false), an optional boolean preventCancel (default false), and an optional AbortSignal signal, is given by performing the following steps. They will return a Promise that fulfills when the pipe completes, or rejects with an exception if it fails.

Assert: ! IsReadableStreamLocked(readable) is false.
Assert: ! IsWritableStreamLocked(writable) is false.
Let signalArg be signal if signal was given, or undefined otherwise.
Return ! ReadableStreamPipeTo(readable, writable, preventClose, preventAbort, preventCancel, signalArg).

If one doesn’t care about the promise returned, referencing this concept can be a bit awkward. The best we can suggest is "pipe readable to writable".

The result of a ReadableStream readable piped through a TransformStream transform, given an optional boolean preventClose (default false), an optional boolean preventAbort (default false), an optional boolean preventCancel (default false), and an optional AbortSignal signal, is given by performing the following steps. The result will be the readable side of transform.

Assert: ! IsReadableStreamLocked(readable) is false.
Assert: ! IsWritableStreamLocked(transform.[[writable]]) is false.
Let signalArg be signal if signal was given, or undefined otherwise.
Let promise be ! ReadableStreamPipeTo(readable, transform.[[writable]], preventClose, preventAbort, preventCancel, signalArg).
Set promise.[[PromiseIsHandled]] to true.
Return transform.[[readable]].

To create a proxy for a ReadableStream stream, perform the following steps. The result will be a new ReadableStream object which pulls its data from stream, while stream itself becomes immediately locked and disturbed.

Let identityTransform be the result of creating an identity TransformStream.
Return the result of stream piped through identityTransform.

10. Examples of creating streams

This section, and all its subsections, are non-normative.

The previous examples throughout the standard have focused on how to use streams. Here we show how to create a stream, using the ReadableStream, WritableStream, and TransformStream constructors.

10.1. A readable stream with an underlying push source (no backpressure support)

The following function creates readable streams that wrap WebSocket instances [WEBSOCKETS], which are push sources that do not support backpressure signals. It illustrates how, when adapting a push source, usually most of the work happens in the start() method.

function makeReadableWebSocketStream(url, protocols) {
  const ws = new WebSocket(url, protocols);
  ws.binaryType = "arraybuffer";

  return new ReadableStream({
    start(controller) {
      ws.onmessage = event => controller.enqueue(event.data);
      ws.onclose = () => controller.close();
      ws.onerror = () => controller.error(new Error("The WebSocket errored!"));
    },

    cancel() {
      ws.close();
    }
  });
}

We can then use this function to create readable streams for a web socket, and pipe that stream to an arbitrary writable stream:

const webSocketStream = makeReadableWebSocketStream("wss://example.com:443/", "protocol");

webSocketStream.pipeTo(writableStream)
  .then(() => console.log("All data successfully written!"))
  .catch(e => console.error("Something went wrong!", e));

This specific style of wrapping a web socket interprets web socket messages directly as chunks. This can be a convenient abstraction, for example when piping to a writable stream or transform stream for which each web socket message makes sense as a chunk to consume or transform.

However, often when people talk about "adding streams support to web sockets", they are hoping instead for a new capability to send an individual web socket message in a streaming fashion, so that e.g. a file could be transferred in a single message without holding all of its contents in memory on the client side. To accomplish this goal, we’d instead want to allow individual web socket messages to themselves be ReadableStream instances. That isn’t what we show in the above example.

For more background, see this discussion.

10.2. A readable stream with an underlying push source and backpressure support

The following function returns readable streams that wrap "backpressure sockets," which are hypothetical objects that have the same API as web sockets, but also provide the ability to pause and resume the flow of data with their readStop and readStart methods. In doing so, this example shows how to apply backpressure to underlying sources that support it.

function makeReadableBackpressureSocketStream(host, port) {
  const socket = createBackpressureSocket(host, port);

  return new ReadableStream({
    start(controller) {
      socket.ondata = event => {
        controller.enqueue(event.data);

        if (controller.desiredSize <= 0) {
          // The internal queue is full, so propagate
          // the backpressure signal to the underlying source.
          socket.readStop();
        }
      };

      socket.onend = () => controller.close();
      socket.onerror = () => controller.error(new Error("The socket errored!"));
    },

    pull() {
      // This is called if the internal queue has been emptied, but the
      // stream's consumer still wants more data. In that case, restart
      // the flow of data if we have previously paused it.
      socket.readStart();
    },

    cancel() {
      socket.close();
    }
  });
}

We can then use this function to create readable streams for such "backpressure sockets" in the same way we do for web sockets. This time, however, when we pipe to a destination that cannot accept data as fast as the socket is producing it, or if we leave the stream alone without reading from it for some time, a backpressure signal will be sent to the socket.

10.3. A readable byte stream with an underlying push source (no backpressure support)

The following function returns readable byte streams that wraps a hypothetical UDP socket API, including a promise-returning select2() method that is meant to be evocative of the POSIX select(2) system call.

Since the UDP protocol does not have any built-in backpressure support, the backpressure signal given by desiredSize is ignored, and the stream ensures that when data is available from the socket but not yet requested by the developer, it is enqueued in the stream’s internal queue, to avoid overflow of the kernel-space queue and a consequent loss of data.

This has some interesting consequences for how consumers interact with the stream. If the consumer does not read data as fast as the socket produces it, the chunks will remain in the stream’s internal queue indefinitely. In this case, using a BYOB reader will cause an extra copy, to move the data from the stream’s internal queue to the developer-supplied buffer. However, if the consumer consumes the data quickly enough, a BYOB reader will allow zero-copy reading directly into developer-supplied buffers.

(You can imagine a more complex version of this example which uses desiredSize to inform an out-of-band backpressure signaling mechanism, for example by sending a message down the socket to adjust the rate of data being sent. That is left as an exercise for the reader.)

const DEFAULT_CHUNK_SIZE = 65536;

function makeUDPSocketStream(host, port) {
  const socket = createUDPSocket(host, port);

  return new ReadableStream({
    type: "bytes",

    start(controller) {
      readRepeatedly().catch(e => controller.error(e));

      function readRepeatedly() {
        return socket.select2().then(() => {
          // Since the socket can become readable even when there’s
          // no pending BYOB requests, we need to handle both cases.
          let bytesRead;
          if (controller.byobRequest) {
            const v = controller.byobRequest.view;
            bytesRead = socket.readInto(v.buffer, v.byteOffset, v.byteLength);
            if (bytesRead === 0) {
              controller.close();
            }
            controller.byobRequest.respond(bytesRead);
          } else {
            const buffer = new ArrayBuffer(DEFAULT_CHUNK_SIZE);
            bytesRead = socket.readInto(buffer, 0, DEFAULT_CHUNK_SIZE);
            if (bytesRead === 0) {
              controller.close();
            } else {
              controller.enqueue(new Uint8Array(buffer, 0, bytesRead));
            }
          }

          if (bytesRead === 0) {
            return;
          }

          return readRepeatedly();
        });
      }
    },

    cancel() {
      socket.close();
    }
  });
}

ReadableStream instances returned from this function can now vend BYOB readers, with all of the aforementioned benefits and caveats.

10.4. A readable stream with an underlying pull source

The following function returns readable streams that wrap portions of the Node.js file system API (which themselves map fairly directly to C’s fopen, fread, and fclose trio). Files are a typical example of pull sources. Note how in contrast to the examples with push sources, most of the work here happens on-demand in the pull() function, and not at startup time in the start() function.

const fs = require("fs").promises;
const CHUNK_SIZE = 1024;

function makeReadableFileStream(filename) {
  let fileHandle;
  let position = 0;

  return new ReadableStream({
    async start() {
      fileHandle = await fs.open(filename, "r");
    },

    async pull(controller) {
      const buffer = new Uint8Array(CHUNK_SIZE);

      const { bytesRead } = await fileHandle.read(buffer, 0, CHUNK_SIZE, position);
      if (bytesRead === 0) {
        await fileHandle.close();
        controller.close();
      } else {
        position += bytesRead;
        controller.enqueue(buffer.subarray(0, bytesRead));
      }
    },

    cancel() {
      return fileHandle.close();
    }
  });
}

We can then create and use readable streams for files just as we could before for sockets.

10.5. A readable byte stream with an underlying pull source

The following function returns readable byte streams that allow efficient zero-copy reading of files, again using the Node.js file system API. Instead of using a predetermined chunk size of 1024, it attempts to fill the developer-supplied buffer, allowing full control.

const fs = require("fs").promises;
const DEFAULT_CHUNK_SIZE = 1024;

function makeReadableByteFileStream(filename) {
 let fileHandle;
 let position = 0;

  return new ReadableStream({
    type: "bytes",

    async start() {
      fileHandle = await fs.open(filename, "r");
    },

    async pull(controller) {
      // Even when the consumer is using the default reader, the auto-allocation
      // feature allocates a buffer and passes it to us via byobRequest.
      const v = controller.byobRequest.view;

      const { bytesRead } = await fileHandle.read(v, 0, v.byteLength, position);
      if (bytesRead === 0) {
        await fileHandle.close();
        controller.close();
        controller.byobRequest.respond(0);
      } else {
        position += bytesRead;
        controller.byobRequest.respond(bytesRead);
      }
    },

    cancel() {
      return fileHandle.close();
    },

    autoAllocateChunkSize: DEFAULT_CHUNK_SIZE
  });
}

With this in hand, we can create and use BYOB readers for the returned ReadableStream. But we can also create default readers, using them in the same simple and generic manner as usual. The adaptation between the low-level byte tracking of the underlying byte source shown here, and the higher-level chunk-based consumption of a default reader, is all taken care of automatically by the streams implementation. The auto-allocation feature, via the autoAllocateChunkSize option, even allows us to write less code, compared to the manual branching in § 10.3 A readable byte stream with an underlying push source (no backpressure support).

10.6. A writable stream with no backpressure or success signals

The following function returns a writable stream that wraps a WebSocket [WEBSOCKETS]. Web sockets do not provide any way to tell when a given chunk of data has been successfully sent (without awkward polling of bufferedAmount, which we leave as an exercise to the reader). As such, this writable stream has no ability to communicate accurate backpressure signals or write success/failure to its producers. That is, the promises returned by its writer's write() method and ready getter will always fulfill immediately.

function makeWritableWebSocketStream(url, protocols) {
  const ws = new WebSocket(url, protocols);

  return new WritableStream({
    start(controller) {
      ws.onerror = () => {
        controller.error(new Error("The WebSocket errored!"));
        ws.onclose = null;
      };
      ws.onclose = () => controller.error(new Error("The server closed the connection unexpectedly!"));
      return new Promise(resolve => ws.onopen = resolve);
    },

    write(chunk) {
      ws.send(chunk);
      // Return immediately, since the web socket gives us no easy way to tell
      // when the write completes.
    },

    close() {
      return closeWS(1000);
    },

    abort(reason) {
      return closeWS(4000, reason && reason.message);
    },
  });

  function closeWS(code, reasonString) {
    return new Promise((resolve, reject) => {
      ws.onclose = e => {
        if (e.wasClean) {
          resolve();
        } else {
          reject(new Error("The connection was not closed cleanly"));
        }
      };
      ws.close(code, reasonString);
    });
  }
}

We can then use this function to create writable streams for a web socket, and pipe an arbitrary readable stream to it:

const webSocketStream = makeWritableWebSocketStream("wss://example.com:443/", "protocol");

readableStream.pipeTo(webSocketStream)
  .then(() => console.log("All data successfully written!"))
  .catch(e => console.error("Something went wrong!", e));

See the earlier note about this style of wrapping web sockets into streams.

10.7. A writable stream with backpressure and success signals

The following function returns writable streams that wrap portions of the Node.js file system API (which themselves map fairly directly to C’s fopen, fwrite, and fclose trio). Since the API we are wrapping provides a way to tell when a given write succeeds, this stream will be able to communicate backpressure signals as well as whether an individual write succeeded or failed.

const fs = require("fs").promises;

function makeWritableFileStream(filename) {
  let fileHandle;

  return new WritableStream({
    async start() {
      fileHandle = await fs.open(filename, "w");
    },

    write(chunk) {
      return fileHandle.write(chunk, 0, chunk.length);
    },

    close() {
      return fileHandle.close();
    },

    abort() {
      return fileHandle.close();
    }
  });
}

We can then use this function to create a writable stream for a file, and write individual chunks of data to it:

const fileStream = makeWritableFileStream("/example/path/on/fs.txt");
const writer = fileStream.getWriter();

writer.write("To stream, or not to stream\n");
writer.write("That is the question\n");

writer.close()
  .then(() => console.log("chunks written and stream closed successfully!"))
  .catch(e => console.error(e));

Note that if a particular call to fileHandle.write takes a longer time, the returned promise will fulfill later. In the meantime, additional writes can be queued up, which are stored in the stream’s internal queue. The accumulation of chunks in this queue can change the stream to return a pending promise from the ready getter, which is a signal to producers that they would benefit from backing off and stopping writing, if possible.

The way in which the writable stream queues up writes is especially important in this case, since as stated in the documentation for fileHandle.write, "it is unsafe to use filehandle.write multiple times on the same file without waiting for the promise." But we don’t have to worry about that when writing the makeWritableFileStream function, since the stream implementation guarantees that the underlying sink's write() method will not be called until any promises returned by previous calls have fulfilled!

10.8. A { readable, writable } stream pair wrapping the same underlying resource

The following function returns an object of the form { readable, writable }, with the readable property containing a readable stream and the writable property containing a writable stream, where both streams wrap the same underlying web socket resource. In essence, this combines § 10.1 A readable stream with an underlying push source (no backpressure support) and § 10.6 A writable stream with no backpressure or success signals.

While doing so, it illustrates how you can use JavaScript classes to create reusable underlying sink and underlying source abstractions.

function streamifyWebSocket(url, protocol) {
  const ws = new WebSocket(url, protocols);
  ws.binaryType = "arraybuffer";

  return {
    readable: new ReadableStream(new WebSocketSource(ws)),
    writable: new WritableStream(new WebSocketSink(ws))
  };
}

class WebSocketSource {
  constructor(ws) {
    this._ws = ws;
  }

  start(controller) {
    this._ws.onmessage = event => controller.enqueue(event.data);
    this._ws.onclose = () => controller.close();

    this._ws.addEventListener("error", () => {
      controller.error(new Error("The WebSocket errored!"));
    });
  }

  cancel() {
    this._ws.close();
  }
}

class WebSocketSink {
  constructor(ws) {
    this._ws = ws;
  }

  start(controller) {
    this._ws.onclose = () => controller.error(new Error("The server closed the connection unexpectedly!"));
    this._ws.addEventListener("error", () => {
      controller.error(new Error("The WebSocket errored!"));
      this._ws.onclose = null;
    });

    return new Promise(resolve => this._ws.onopen = resolve);
  }

  write(chunk) {
    this._ws.send(chunk);
  }

  close() {
    return this._closeWS(1000);
  }

  abort(reason) {
    return this._closeWS(4000, reason && reason.message);
  }

  _closeWS(code, reasonString) {
    return new Promise((resolve, reject) => {
      this._ws.onclose = e => {
        if (e.wasClean) {
          resolve();
        } else {
          reject(new Error("The connection was not closed cleanly"));
        }
      };
      this._ws.close(code, reasonString);
    });
  }
}

We can then use the objects created by this function to communicate with a remote web socket, using the standard stream APIs:

const streamyWS = streamifyWebSocket("wss://example.com:443/", "protocol");
const writer = streamyWS.writable.getWriter();
const reader = streamyWS.readable.getReader();

writer.write("Hello");
writer.write("web socket!");

reader.read().then(({ value, done }) => {
  console.log("The web socket says: ", value);
});

Note how in this setup canceling the readable side will implicitly close the writable side, and similarly, closing or aborting the writable side will implicitly close the readable side.

See the earlier note about this style of wrapping web sockets into streams.

10.9. A transform stream that replaces template tags

It’s often useful to substitute tags with variables on a stream of data, where the parts that need to be replaced are small compared to the overall data size. This example presents a simple way to do that. It maps strings to strings, transforming a template like "Time: {{time}} Message: {{message}}" to "Time: 15:36 Message: hello" assuming that { time: "15:36", message: "hello" } was passed in the substitutions parameter to LipFuzzTransformer.

This example also demonstrates one way to deal with a situation where a chunk contains partial data that cannot be transformed until more data is received. In this case, a partial template tag will be accumulated in the partialChunk property until either the end of the tag is found or the end of the stream is reached.

class LipFuzzTransformer {
  constructor(substitutions) {
    this.substitutions = substitutions;
    this.partialChunk = "";
    this.lastIndex = undefined;
  }

  transform(chunk, controller) {
    chunk = this.partialChunk + chunk;
    this.partialChunk = "";
    // lastIndex is the index of the first character after the last substitution.
    this.lastIndex = 0;
    chunk = chunk.replace(/\{\{([a-zA-Z0-9_-]+)\}\}/g, this.replaceTag.bind(this));
    // Regular expression for an incomplete template at the end of a string.
    const partialAtEndRegexp = /\{(\{([a-zA-Z0-9_-]+(\})?)?)?$/g;
    // Avoid looking at any characters that have already been substituted.
    partialAtEndRegexp.lastIndex = this.lastIndex;
    this.lastIndex = undefined;
    const match = partialAtEndRegexp.exec(chunk);
    if (match) {
      this.partialChunk = chunk.substring(match.index);
      chunk = chunk.substring(0, match.index);
    }
    controller.enqueue(chunk);
  }

  flush(controller) {
    if (this.partialChunk.length > 0) {
      controller.enqueue(this.partialChunk);
    }
  }

  replaceTag(match, p1, offset) {
    let replacement = this.substitutions[p1];
    if (replacement === undefined) {
      replacement = "";
    }
    this.lastIndex = offset + replacement.length;
    return replacement;
  }
}

In this case we define the transformer to be passed to the TransformStream constructor as a class. This is useful when there is instance data to track.

The class would be used in code like:

const data = { userName, displayName, icon, date };
const ts = new TransformStream(new LipFuzzTransformer(data));

fetchEvent.respondWith(
  fetch(fetchEvent.request.url).then(response => {
    const transformedBody = response.body
      // Decode the binary-encoded response to string
      .pipeThrough(new TextDecoderStream())
      // Apply the LipFuzzTransformer
      .pipeThrough(ts)
      // Encode the transformed string
      .pipeThrough(new TextEncoderStream());
    return new Response(transformedBody);
  })
);

For simplicity, LipFuzzTransformer performs unescaped text substitutions. In real applications, a template system that performs context-aware escaping is good practice for security and robustness.

10.10. A transform stream created from a sync mapper function

The following function allows creating new TransformStream instances from synchronous "mapper" functions, of the type you would normally pass to Array.prototype.map. It demonstrates that the API is concise even for trivial transforms.

function mapperTransformStream(mapperFunction) {
  return new TransformStream({
    transform(chunk, controller) {
      controller.enqueue(mapperFunction(chunk));
    }
  });
}

This function can then be used to create a TransformStream that uppercases all its inputs:

const ts = mapperTransformStream(chunk => chunk.toUpperCase());
const writer = ts.writable.getWriter();
const reader = ts.readable.getReader();

writer.write("No need to shout");

// Logs "NO NEED TO SHOUT":
reader.read().then(({ value }) => console.log(value));

Although a synchronous transform never causes backpressure itself, it will only transform chunks as long as there is no backpressure, so resources will not be wasted.

Exceptions error the stream in a natural way:

const ts = mapperTransformStream(chunk => JSON.parse(chunk));
const writer = ts.writable.getWriter();
const reader = ts.readable.getReader();

writer.write("[1, ");

// Logs a SyntaxError, twice:
reader.read().catch(e => console.error(e));
writer.write("{}").catch(e => console.error(e));

10.11. Using an identity transform stream as a primitive to create new readable streams

Combining an identity transform stream with pipeTo() is a powerful way to manipulate streams. This section contains a couple of examples of this general technique.

It’s sometimes natural to treat a promise for a readable stream as if it were a readable stream. A simple adapter function is all that’s needed:

function promiseToReadable(promiseForReadable) {
  const ts = new TransformStream();

  promiseForReadable
      .then(readable => readable.pipeTo(ts.writable))
      .catch(reason => ts.writable.abort(reason))
      .catch(() => {});

  return ts.readable;
}

Here, we pipe the data to the writable side and return the readable side. If the pipe errors, we abort the writable side, which automatically propagates the error to the returned readable side. If the writable side had already been errored by pipeTo(), then the abort() call will return a rejection, which we can safely ignore.

A more complex extension of this is concatenating multiple readable streams into one:

function concatenateReadables(readables) {
  const ts = new TransformStream();
  let promise = Promise.resolve();

  for (const readable of readables) {
    promise = promise.then(
     () => readable.pipeTo(ts.writable, { preventClose: true }),
     reason => {
       return Promise.all([
         ts.writable.abort(reason),
         readable.cancel(reason)
       ]);
     }
   );
  }

  promise.then(() => ts.writable.close(),
               reason => ts.writable.abort(reason))
         .catch(() => {});

  return ts.readable;
}

The error handling here is subtle because canceling the concatenated stream has to cancel all the input streams. However, the success case is simple enough. We just pipe each stream in the readables iterable one at a time to the identity transform stream's writable side, and then close it when we are done. The readable side is then a concatenation of all the chunks from all of of the streams. We return it from the function. Backpressure is applied as usual.

Acknowledgments

The editors would like to thank Anne van Kesteren, AnthumChris, Arthur Langereis, Ben Kelly, Bert Belder, Brian di Palma, Calvin Metcalf, Dominic Tarr, Ed Hager, Eric Skoglund, Forbes Lindesay, Forrest Norvell, Gary Blackwood, Gorgi Kosev, Gus Caplan, 贺师俊 (hax), Isaac Schlueter, isonmad, Jake Archibald, Jake Verbaten, James Pryor, Janessa Det, Jason Orendorff, Jeffrey Yasskin, Jeremy Roman, Jens Nockert, Lennart Grahl, Luca Casonato, Mangala Sadhu Sangeet Singh Khalsa, Marcos Caceres, Marvin Hagemeister, Mattias Buelens, Michael Mior, Mihai Potra, Nidhi Jaju, Romain Bellessort, Simon Menke, Stephen Sugden, Surma, Tab Atkins, Tanguy Krotoff, Thorsten Lorenz, Till Schneidereit, Tim Caswell, Trevor Norris, tzik, Will Chan, Youenn Fablet, 平野裕 (Yutaka Hirano), and Xabier Rodríguez for their contributions to this specification. Community involvement in this specification has been above and beyond; we couldn’t have done it without you.

This standard is written by Adam Rice (Google, [email protected]), Domenic Denicola (Google, [email protected]), Mattias Buelens, and 吉野剛史 (Takeshi Yoshino, [email protected]).

Intellectual property rights

Copyright © WHATWG (Apple, Google, Mozilla, Microsoft). This work is licensed under a Creative Commons Attribution 4.0 International License. To the extent portions of it are incorporated into source code, such portions in the source code are licensed under the BSD 3-Clause License instead.

This is the Living Standard. Those interested in the patent-review version should view the Living Standard Review Draft.

Streams

Abstract

1. Introduction

2. Model

2.1. Readable streams

2.2. Writable streams

2.3. Transform streams

2.4. Pipe chains and backpressure

2.5. Internal queues and queuing strategies

2.6. Locking

3. Conventions

4. Readable streams

4.1. Using readable streams

4.2. The ReadableStream class

4.2.1. Interface definition

4.2.2. Internal slots

4.2.3. The underlying source API

4.2.4. Constructor, methods, and properties

4.2.5. Asynchronous iteration

4.2.6. Transfer via postMessage()

4.3. The ReadableStreamGenericReader mixin

4.3.1. Mixin definition

4.3.2. Internal slots

4.3.3. Methods and properties

4.4. The ReadableStreamDefaultReader class

4.4.1. Interface definition

4.4.2. Internal slots

4.4.3. Constructor, methods, and properties

4.5. The ReadableStreamBYOBReader class

4.5.1. Interface definition

4.5.2. Internal slots

4.5.3. Constructor, methods, and properties

4.6. The ReadableStreamDefaultController class

4.6.1. Interface definition

4.6.2. Internal slots

4.6.3. Methods and properties

4.6.4. Internal methods

4.7. The ReadableByteStreamController class

4.7.1. Interface definition

4.7.2. Internal slots

4.7.3. Methods and properties

4.7.4. Internal methods

4.8. The ReadableStreamBYOBRequest class

4.8.1. Interface definition

4.8.2. Internal slots

4.8.3. Methods and properties

4.9. Abstract operations

4.9.1. Working with readable streams

4.9.2. Interfacing with controllers

4.9.3. Readers

4.9.4. Default controllers

4.9.5. Byte stream controllers

5. Writable streams

5.1. Using writable streams

5.2. The WritableStream class

5.2.1. Interface definition

5.2.2. Internal slots

5.2.3. The underlying sink API

5.2.4. Constructor, methods, and properties

5.2.5. Transfer via postMessage()

5.3. The WritableStreamDefaultWriter class

5.3.1. Interface definition

5.3.2. Internal slots

5.3.3. Constructor, methods, and properties

5.4. The WritableStreamDefaultController class

5.4.1. Interface definition

5.4.2. Internal slots

5.4.3. Methods and properties

5.4.4. Internal methods

5.5. Abstract operations

5.5.1. Working with writable streams

5.5.2. Interfacing with controllers

5.5.3. Writers

5.5.4. Default controllers

6. Transform streams

6.1. Using transform streams

6.2. The TransformStream class

6.2.1. Interface definition

6.2.2. Internal slots

6.2.3. The transformer API

4.2. The `ReadableStream` class

4.2.6. Transfer via `postMessage()`

4.3. The `ReadableStreamGenericReader` mixin

4.4. The `ReadableStreamDefaultReader` class

4.5. The `ReadableStreamBYOBReader` class

4.6. The `ReadableStreamDefaultController` class

4.7. The `ReadableByteStreamController` class

4.8. The `ReadableStreamBYOBRequest` class

5.2. The `WritableStream` class

5.2.5. Transfer via `postMessage()`

5.3. The `WritableStreamDefaultWriter` class

5.4. The `WritableStreamDefaultController` class

6.2. The `TransformStream` class

6.2.5. Transfer via `postMessage()`

6.3. The `TransformStreamDefaultController` class

7.2. The `ByteLengthQueuingStrategy` class

7.3. The `CountQueuingStrategy` class