Encoding

1. Preface

The UTF-8 encoding is the most appropriate encoding for interchange of Unicode, the universal coded character set. Therefore for new protocols and formats, as well as existing formats deployed in new contexts, this specification requires (and defines) the UTF-8 encoding.

The other (legacy) encodings have been defined to some extent in the past. However, user agents have not always implemented them in the same way, have not always used the same labels, and often differ in dealing with undefined and former proprietary areas of encodings. This specification addresses those gaps so that new user agents do not have to reverse engineer encoding implementations and existing user agents can converge.

In particular, this specification defines all those encodings, their algorithms to go from bytes to scalar values and back, and their canonical names and identifying labels. This specification also defines an API to expose part of the encoding algorithms to JavaScript.

User agents have also significantly deviated from the labels listed in the IANA Character Sets registry. To stop spreading legacy encodings further, this specification is exhaustive about the aforementioned details and therefore has no need for the registry. In particular, this specification does not provide a mechanism for extending any aspect of encodings.

2. Security background

There is a set of encoding security issues when the producer and consumer do not agree on the encoding in use, or on the way a given encoding is to be implemented. For instance, an attack was reported in 2011 where a Shift_JIS lead byte 0x82 was used to “mask” a 0x22 trail byte in a JSON resource of which an attacker could control some field. The producer did not see the problem even though this is an illegal byte combination. The consumer decoded it as a single U+FFFD and therefore changed the overall interpretation as U+0022 is an important delimiter. Decoders of encodings that use multiple bytes for scalar values now require that in case of an illegal byte combination, a scalar value in the range U+0000 to U+007F, inclusive, cannot be “masked”. For the aforementioned sequence the output would be U+FFFD U+0022. (As an unfortunate exception to this, the gb18030 decoder will “mask” up to one such byte at end-of-queue.)

This is a larger issue for encodings that map anything that is an ASCII byte to something that is not an ASCII code point, when there is no lead byte present. These are “ASCII-incompatible” encodings and other than ISO-2022-JP and UTF-16BE/LE, which are unfortunately required due to deployed content, they are not supported. (Investigation is ongoing whether more labels of other such encodings can be mapped to the replacement encoding, rather than the unknown encoding fallback.) An example attack is injecting carefully crafted content into a resource and then encouraging the user to override the encoding, resulting in, e.g., script execution.

Encoders used by URLs found in HTML and HTML’s form feature can also result in slight information loss when an encoding is used that cannot represent all scalar values. E.g., when a resource uses the windows-1252 encoding a server will not be able to distinguish between an end user entering “💩” and “💩” into a form.

The problems outlined here go away when exclusively using UTF-8, which is one of the many reasons that is now the mandatory encoding for all things.

See also the Browser UI chapter.

3. Terminology

This specification depends on the Infra Standard. [INFRA]

Hexadecimal numbers are prefixed with "0x".

In equations, all numbers are integers, addition is represented by "+", subtraction by "−", multiplication by "×", integer division by "/" (returns the quotient), modulo by "%" (returns the remainder of an integer division), logical left shifts by "<<", logical right shifts by ">>", bitwise AND by "&", and bitwise OR by "|".

For logical right shifts operands must have at least twenty-one bits precision.

An I/O queue is a type of list with items of a particular type (i.e., bytes or scalar values). End-of-queue is a special item that can be present in I/O queues of any type and it signifies that there are no more items in the queue.

To read an item from an I/O queue ioQueue, run these steps:

1. If ioQueue is empty, then wait until its size is at least 1.

2. If ioQueue[0] is end-of-queue, then return end-of-queue.

3. Remove ioQueue[0] and return it.

To read a number number of items from ioQueue, run these steps:

1. Let readItems be « ».

2. Perform the following step number times:

   1. Append to readItems the result of reading an item from ioQueue.

3. Remove end-of-queue from readItems.

4. Return readItems.

To peek a number number of items from an I/O queue ioQueue, run these steps:

1. Wait until either ioQueue’s size is equal to or greater than number, or ioQueue contains end-of-queue, whichever comes first.

2. Let prefix be « ».

3. For each n in the range 1 to number, inclusive:

   1. If ioQueue[n] is end-of-queue, break.

   2. Otherwise, append ioQueue[n] to prefix.

4. Return prefix.

To push an item item to an I/O queue ioQueue, run these steps:

1. If the last item in ioQueue is end-of-queue, then:

   1. If item is end-of-queue, do nothing.

   2. Otherwise, insert item before the last item in ioQueue.

2. Otherwise, append item to ioQueue.

To push a sequence of items to an I/O queue ioQueue is to push each item in the sequence to ioQueue, in the given order.

To restore an item other than end-of-queue to an I/O queue, perform the list prepend operation. To restore a list of items excluding end-of-queue to an I/O queue, insert those items, in the given order, before the first item in the queue.

Inserting the bytes « 0xF0, 0x9F » in an I/O queue « 0x92 0xA9, end-of-queue », results in an I/O queue « 0xF0, 0x9F, 0x92 0xA9, end-of-queue ». The next item to be read would be 0xF0.

To convert an I/O queue ioQueue into a list, string, or byte sequence, return the result of reading an indefinite number of items from ioQueue.

To convert a list, string, or byte sequence input into an I/O queue, run these steps:

1. Assert: if input is a list, then it does not contain end-of-queue.

2. Return an I/O queue containing the items in input, in order, followed by end-of-queue.

The Infra standard is expected to define some infrastructure around type conversions. See whatwg/infra issue #319. [INFRA]

I/O queues are defined as lists, not queues, because they feature a restore operation. However, this restore operation is an internal detail of the algorithms in this specification, and is not to be used by other standards. Implementations are free to find alternative ways to implement such algorithms, as detailed in Implementation considerations.

To obtain a scalar value from surrogates, given a leading surrogate leading and a trailing surrogate trailing, return 0x10000 + ((leading − 0xD800) << 10) + (trailing − 0xDC00).

4. Encodings

An encoding defines a mapping from a scalar value sequence to a byte sequence (and vice versa). Each encoding has a name, and one or more labels.

This specification defines three encodings with the same names as encoding schemes defined in the Unicode standard: UTF-8, UTF-16LE, and UTF-16BE. The encodings differ from the encoding schemes by byte order mark (also known as BOM) handling not being part of the encodings themselves and instead being part of wrapper algorithms in this specification, whereas byte order mark handling is part of the definition of the encoding schemes in the Unicode Standard. UTF-8 used together with the UTF-8 decode algorithm matches the encoding scheme of the same name. This specification does not provide wrapper algorithms that would combine with UTF-16LE and UTF-16BE to match the similarly-named encoding schemes. [UNICODE]

4.1. Encoders and decoders

Each encoding has an associated decoder and most of them have an associated encoder. Instances of decoders and encoders have a handler algorithm and might also have state. A handler algorithm takes an input I/O queue and an item, and returns finished, one or more items, error optionally with a code point, or continue.

The replacement and UTF-16BE/LE encodings have no encoder.

An error mode as used below is "replacement" or "fatal" for a decoder and "fatal" or "html" for an encoder.

An XML processor would set error mode to "fatal". [XML]

"html" exists as error mode due to HTML forms requiring a non-terminating legacy encoder. The "html" error mode causes a sequence to be emitted that cannot be distinguished from legitimate input and can therefore lead to silent data loss. Developers are strongly encouraged to use the UTF-8 encoding to prevent this from happening. [HTML]

To process a queue given an encoding’s decoder or encoder instance encoderDecoder, I/O queue input, I/O queue output, and error mode mode:

1. While true:

   1. Let result be the result of processing an item with the result of reading from input, encoderDecoder, input, output, and mode.

   2. If result is not continue, then return result.

To process an item given an item item, encoding’s encoder or decoder instance encoderDecoder, I/O queue input, I/O queue output, and error mode mode:

1. Assert: if encoderDecoder is an encoder instance, mode is not "replacement".

2. Assert: if encoderDecoder is a decoder instance, mode is not "html".

3. Assert: if encoderDecoder is an encoder instance, item is not a surrogate.

4. Let result be the result of running encoderDecoder’s handler on input and item.

5. If result is finished:

   1. Push end-of-queue to output.

   2. Return result.

6. Otherwise, if result is one or more items:

   1. Assert: if encoderDecoder is a decoder instance, result does not contain any surrogates.

   2. Push result to output.

7. Otherwise, if result is an error, switch on mode and run the associated steps:

   "replacement"

   Push U+FFFD (�) to output.

   "html"

   Push 0x26 (&), 0x23 (#), followed by the shortest sequence of 0x30 (0) to 0x39 (9), inclusive, representing result’s code point’s value in base ten, followed by 0x3B (;) to output.

   "fatal"

   Return result.

8. Return continue.

4.2. Names and labels

The table below lists all encodings and their labels user agents must support. User agents must not support any other encodings or labels.

For each encoding, ASCII-lowercasing its name yields one of its labels.

Authors must use the UTF-8 encoding and must use its (ASCII case-insensitive) "utf-8" label to identify it.

New protocols and formats, as well as existing formats deployed in new contexts, must use the UTF-8 encoding exclusively. If these protocols and formats need to expose the encoding’s name or label, they must expose it as "utf-8".

To get an encoding from a string label, run these steps:

1. Remove any leading and trailing ASCII whitespace from label.

2. If label is an ASCII case-insensitive match for any of the labels listed in the table below, then return the corresponding encoding; otherwise return failure.

This is a more basic and restrictive algorithm of mapping labels to encodings than section 1.4 of Unicode Technical Standard #22 prescribes, as that is necessary to be compatible with deployed content.

All encodings and their labels are also available as non-normative encodings.json resource.

The set of supported encodings is primarily based on the intersection of the sets supported by major browser engines when the development of this standard started, while removing encodings that were rarely used legitimately but that could be used in attacks. The inclusion of some encodings is questionable in the light of anecdotal evidence of the level of use by existing Web content. That is, while they have been broadly supported by browsers, it is unclear if they are broadly used by Web content. However, an effort has not been made to eagerly remove single-byte encodings that were broadly supported by browsers or are part of the ISO 8859 series. In particular, the necessity of the inclusion of IBM866, macintosh, x-mac-cyrillic, ISO-8859-3, ISO-8859-10, ISO-8859-14, and ISO-8859-16 is doubtful for the purpose of supporting existing content, but there are no plans to remove these.

4.3. Output encodings

To get an output encoding from an encoding encoding, run these steps:

1. If encoding is replacement or UTF-16BE/LE, then return UTF-8.

2. Return encoding.

The get an output encoding algorithm is useful for URL parsing and HTML form submission, which both need exactly this.

5. Indexes

Most legacy encodings make use of an index. An index is an ordered list of entries, each entry consisting of a pointer and a corresponding code point. Within an index pointers are unique and code points can be duplicated.

An efficient implementation likely has two indexes per encoding. One optimized for its decoder and one for its encoder.

To find the pointers and their corresponding code points in an index, let lines be the result of splitting the resource’s contents on U+000A. Then remove each item in lines that is the empty string or starts with U+0023. Then the pointers and their corresponding code points are found by splitting each item in lines on U+0009. The first subitem is the pointer (as a decimal number) and the second is the corresponding code point (as a hexadecimal number). Other subitems are not relevant.

To signify changes an index includes an Identifier and a Date. If an Identifier has changed, so has the index.

The index code point for pointer in index is the code point corresponding to pointer in index, or null if pointer is not in index.

The index pointer for code point in index is the first pointer corresponding to code point in index, or null if code point is not in index.

These are the indexes defined by this specification, excluding index single-byte, which have their own table:

The index gb18030 ranges code point for pointer is the return value of these steps:

1. If pointer is greater than 39419 and less than 189000, or pointer is greater than 1237575, return null.

2. If pointer is 7457, return code point U+E7C7.

3. Let offset be the last pointer in index gb18030 ranges that is less than or equal to pointer and let code point offset be its corresponding code point.

4. Return a code point whose value is code point offset + pointer − offset.

The index gb18030 ranges pointer for code point is the return value of these steps:

1. If code point is U+E7C7, return pointer 7457.

2. Let offset be the last code point in index gb18030 ranges that is less than or equal to code point and let pointer offset be its corresponding pointer.

3. Return a pointer whose value is pointer offset + code point − offset.

The index Shift_JIS pointer for code point is the return value of these steps:

1. Let index be index jis0208 excluding all entries whose pointer is in the range 8272 to 8835, inclusive.

   The index jis0208 contains duplicate code points so the exclusion of these entries causes later code points to be used.

2. Return the index pointer for code point in index.

The index Big5 pointer for code point is the return value of these steps:

1. Let index be index Big5 excluding all entries whose pointer is less than (0xA1 - 0x81) × 157.

   Avoid returning Hong Kong Supplementary Character Set extensions literally.

2. If code point is U+2550, U+255E, U+2561, U+256A, U+5341, or U+5345, return the last pointer corresponding to code point in index.

   There are other duplicate code points, but for those the first pointer is to be used.

3. Return the index pointer for code point in index.

All indexes are also available as a non-normative indexes.json resource. (Index gb18030 ranges has a slightly different format here, to be able to represent ranges.)

6. Hooks for standards

To UTF-8 decode an I/O queue of bytes ioQueue given an optional I/O queue of scalar values output (default « »), run these steps:

1. Let buffer be the result of peeking three bytes from ioQueue, converted to a byte sequence.

2. If buffer is 0xEF 0xBB 0xBF, then read three bytes from ioQueue. (Do nothing with those bytes.)

3. Process a queue with an instance of UTF-8’s decoder, ioQueue, output, and "replacement".

4. Return output.

To UTF-8 decode without BOM an I/O queue of bytes ioQueue given an optional I/O queue of scalar values output (default « »), run these steps:

1. Process a queue with an instance of UTF-8’s decoder, ioQueue, output, and "replacement".

2. Return output.

To UTF-8 decode without BOM or fail an I/O queue of bytes ioQueue given an optional I/O queue of scalar values output (default « »), run these steps:

1. Let potentialError be the result of processing a queue with an instance of UTF-8’s decoder, ioQueue, output, and "fatal".

2. If potentialError is an error, then return failure.

3. Return output.

To UTF-8 encode an I/O queue of scalar values ioQueue given an optional I/O queue of bytes output (default « »), return the result of encoding ioQueue with encoding UTF-8 and output.

6.1. Legacy hooks for standards

To decode an I/O queue of bytes ioQueue given a fallback encoding encoding and an optional I/O queue of scalar values output (default « »), run these steps:

1. Let BOMEncoding be the result of BOM sniffing ioQueue.

2. If BOMEncoding is non-null:

   1. Set encoding to BOMEncoding.

   2. Read three bytes from ioQueue, if BOMEncoding is UTF-8; otherwise read two bytes. (Do nothing with those bytes.)

   For compatibility with deployed content, the byte order mark is more authoritative than anything else. In a context where HTTP is used this is in violation of the semantics of the `Content-Type` header.

3. Process a queue with an instance of encoding’s decoder, ioQueue, output, and "replacement".

4. Return output.

To BOM sniff an I/O queue of bytes ioQueue, run these steps:

1. Let BOM be the result of peeking 3 bytes from ioQueue, converted to a byte sequence.

2. For each of the rows in the table below, starting with the first one and going down, if BOM starts with the bytes given in the first column, then return the encoding given in the cell in the second column of that row. Otherwise, return null.

   Byte order mark	Encoding
   0xEF 0xBB 0xBF	UTF-8
   0xFE 0xFF	UTF-16BE
   0xFF 0xFE	UTF-16LE

This hook is a workaround for the fact that decode has no way to communicate back to the caller that it has found a byte order mark and is therefore not using the provided encoding. The hook is to be invoked before decode, and it will return an encoding corresponding to the byte order mark found, or null otherwise.

To encode an I/O queue of scalar values ioQueue given an encoding encoding and an optional I/O queue of bytes output (default « »), run these steps:

1. Let encoder be the result of getting an encoder from encoding.

2. Process a queue with encoder, ioQueue, output, and "html".

3. Return output.

This is a legacy hook for HTML forms. Layering UTF-8 encode on top is safe as it never triggers errors. [HTML]

To get an encoder from an encoding encoding:

1. Assert: encoding is not replacement or UTF-16BE/LE.

2. Return an instance of encoding’s encoder.

To encode or fail an I/O queue of scalar values ioQueue given an encoder instance encoder and an I/O queue of bytes output, run these steps:

1. Let potentialError be the result of processing a queue with encoder, ioQueue, output, and "fatal".

2. Push end-of-queue to output.

3. If potentialError is an error, then return error’s code point’s value.

4. Return null.

7. API

This section uses terminology from Web IDL. Browser user agents must support this API. JavaScript implementations should support this API. Other user agents or programming languages are encouraged to use an API suitable to their needs, which might not be this one. [WEBIDL]

function encodeArrayOfStrings(strings) {
  var encoder, encoded, len, bytes, view, offset;

  encoder = new TextEncoder();
  encoded = [];

  len = Uint32Array.BYTES_PER_ELEMENT;
  for (var i = 0; i < strings.length; i++) {
    len += Uint32Array.BYTES_PER_ELEMENT;
    encoded[i] = encoder.encode(strings[i]);
    len += encoded[i].byteLength;
  }

  bytes = new Uint8Array(len);
  view = new DataView(bytes.buffer);
  offset = 0;

  view.setUint32(offset, strings.length);
  offset += Uint32Array.BYTES_PER_ELEMENT;
  for (var i = 0; i < encoded.length; i += 1) {
    len = encoded[i].byteLength;
    view.setUint32(offset, len);
    offset += Uint32Array.BYTES_PER_ELEMENT;
    bytes.set(encoded[i], offset);
    offset += len;
  }
  return bytes.buffer;
}

function decodeArrayOfStrings(buffer, encoding) {
  var decoder, view, offset, num_strings, strings, len;

  decoder = new TextDecoder(encoding);
  view = new DataView(buffer);
  offset = 0;
  strings = [];

  num_strings = view.getUint32(offset);
  offset += Uint32Array.BYTES_PER_ELEMENT;
  for (var i = 0; i < num_strings; i++) {
    len = view.getUint32(offset);
    offset += Uint32Array.BYTES_PER_ELEMENT;
    strings[i] = decoder.decode(
      new DataView(view.buffer, offset, len));
    offset += len;
  }
  return strings;
}

7.1. Interface mixin TextDecoderCommon

interface mixin TextDecoderCommon {
  readonly attribute DOMString encoding;
  readonly attribute boolean fatal;
  readonly attribute boolean ignoreBOM;
};

The TextDecoderCommon interface mixin defines common getters that are shared between TextDecoder and TextDecoderStream objects. These objects have an associated:

encoding

An encoding.

decoder

A decoder instance.

I/O queue

An I/O queue of bytes.

ignore BOM

A boolean, initially false.

BOM seen

A boolean, initially false.

error mode

An error mode, initially "replacement".

The serialize I/O queue algorithm, given a TextDecoderCommon decoder and an I/O queue of scalar values ioQueue, runs these steps:

1. Let output be the empty string.

2. While true:

   1. Let item be the result of reading from ioQueue.

   2. If item is end-of-queue, then return output.

   3. If decoder’s encoding is UTF-8 or UTF-16BE/LE, and decoder’s ignore BOM and BOM seen are false, then:

      1. Set decoder’s BOM seen to true.

      2. If item is U+FEFF, then continue.

   4. Append item to output.

This algorithm is intentionally different with respect to BOM handling from the decode algorithm used by the rest of the platform to give API users more control.

The encoding getter steps are to return this’s encoding’s name, ASCII lowercased.

The fatal getter steps are to return true if this’s error mode is "fatal", otherwise false.

The ignoreBOM getter steps are to return this’s ignore BOM.

7.2. Interface TextDecoder

dictionary TextDecoderOptions {
  boolean fatal = false;
  boolean ignoreBOM = false;
};

dictionary TextDecodeOptions {
  boolean stream = false;
};

[Exposed=*]
interface TextDecoder {
  constructor(optional DOMString label = "utf-8", optional TextDecoderOptions options = {});

  USVString decode(optional AllowSharedBufferSource input, optional TextDecodeOptions options = {});
};
TextDecoder includes TextDecoderCommon;

A TextDecoder object has an associated do not flush, which is a boolean, initially false.

decoder = new TextDecoder([label = "utf-8" [, options]])

Returns a new TextDecoder object.

If label is either not a label or is a label for replacement, throws a RangeError.

decoder . encoding

Returns encoding’s name, lowercased.

decoder . fatal

Returns true if error mode is "fatal", otherwise false.

decoder . ignoreBOM

Returns the value of ignore BOM.

decoder . decode([input [, options]])

Returns the result of running encoding’s decoder. The method can be invoked zero or more times with options’s stream set to true, and then once without options’s stream (or set to false), to process a fragmented input. If the invocation without options’s stream (or set to false) has no input, it’s clearest to omit both arguments.

var string = "", decoder = new TextDecoder(encoding), buffer;
while(buffer = next_chunk()) {
  string += decoder.decode(buffer, {stream:true});
}
string += decoder.decode(); // end-of-queue

If the error mode is "fatal" and encoding’s decoder returns error, throws a TypeError.

The new TextDecoder(label, options) constructor steps are:

1. Let encoding be the result of getting an encoding from label.

2. If encoding is failure or replacement, then throw a RangeError.

3. Set this’s encoding to encoding.

4. If options["fatal"] is true, then set this’s error mode to "fatal".

5. Set this’s ignore BOM to options["ignoreBOM"].

The decode(input, options) method steps are:

1. If this’s do not flush is false, then set this’s decoder to a new instance of this’s encoding’s decoder, this’s I/O queue to the I/O queue of bytes « end-of-queue », and this’s BOM seen to false.

2. Set this’s do not flush to options["stream"].

3. If input is given, then push a copy of input to this’s I/O queue.

   Implementations are strongly encouraged to use an implementation strategy that avoids this copy. When doing so they will have to make sure that changes to input do not affect future calls to decode().

   The memory exposed by SharedArrayBuffer objects does not adhere to data race freedom properties required by the memory model of programming languages typically used for implementations. When implementing, take care to use the appropriate facilities when accessing memory exposed by SharedArrayBuffer objects.

4. Let output be the I/O queue of scalar values « end-of-queue ».

5. While true:

   1. Let item be the result of reading from this’s I/O queue.

   2. If item is end-of-queue and this’s do not flush is true, then return the result of running serialize I/O queue with this and output.

      The way streaming works is to not handle end-of-queue here when this’s do not flush is true and to not set it to false. That way in a subsequent invocation this’s decoder is not set anew in the first step of the algorithm and its state is preserved.

   3. Otherwise:

      1. Let result be the result of processing an item with item, this’s decoder, this’s I/O queue, output, and this’s error mode.

      2. If result is finished, then return the result of running serialize I/O queue with this and output.

      3. Otherwise, if result is error, throw a TypeError.

7.3. Interface mixin TextEncoderCommon

interface mixin TextEncoderCommon {
  readonly attribute DOMString encoding;
};

The TextEncoderCommon interface mixin defines common getters that are shared between TextEncoder and TextEncoderStream objects.

The encoding getter steps are to return "utf-8".

7.4. Interface TextEncoder

dictionary TextEncoderEncodeIntoResult {
  unsigned long long read;
  unsigned long long written;
};

[Exposed=*]
interface TextEncoder {
  constructor();

  [NewObject] Uint8Array encode(optional USVString input = "");
  TextEncoderEncodeIntoResult encodeInto(USVString source, [AllowShared] Uint8Array destination);
};
TextEncoder includes TextEncoderCommon;

A TextEncoder object offers no label argument as it only supports UTF-8. It also offers no stream option as no encoder requires buffering of scalar values.

encoder = new TextEncoder()

Returns a new TextEncoder object.

encoder . encoding

Returns "utf-8".

encoder . encode([input = ""])

Returns the result of running UTF-8’s encoder.

encoder . encodeInto(source, destination)

Runs the UTF-8 encoder on source, stores the result of that operation into destination, and returns the progress made as an object wherein read is the number of converted code units of source and written is the number of bytes modified in destination.

The new TextEncoder() constructor steps are to do nothing.

The encode(input) method steps are:

1. Convert input to an I/O queue of scalar values.

2. Let output be the I/O queue of bytes « end-of-queue ».

3. While true:

   1. Let item be the result of reading from input.

   2. Let result be the result of processing an item with item, an instance of the UTF-8 encoder, input, output, and "fatal".

   3. Assert: result is not an error.

      The UTF-8 encoder cannot return error.

   4. If result is finished, then convert output into a byte sequence and return a Uint8Array object wrapping an ArrayBuffer containing output.

The encodeInto(source, destination) method steps are:

1. Let read be 0.

2. Let written be 0.

3. Let encoder be an instance of the UTF-8 encoder.

4. Let unused be the I/O queue of scalar values « end-of-queue ».

   The handler algorithm invoked below requires this argument, but it is not used by the UTF-8 encoder.

5. Convert source to an I/O queue of scalar values.

6. While true:

   1. Let item be the result of reading from source.

   2. Let result be the result of running encoder’s handler on unused and item.

   3. If result is finished, then break.

   4. Otherwise:

      1. If destination’s byte length − written is greater than or equal to the number of bytes in result, then:

         1. If item is greater than U+FFFF, then increment read by 2.

         2. Otherwise, increment read by 1.

         3. Write the bytes in result into destination, with startingOffset set to written.

            See the warning for SharedArrayBuffer objects above.

         4. Increment written by the number of bytes in result.

      2. Otherwise, break.

7. Return «[ "read" → read, "written" → written ]».

function convertString(buffer, input, callback) {
  let bufferSize = 256,
      bufferStart = malloc(buffer, bufferSize),
      writeOffset = 0,
      readOffset = 0;
  while (true) {
    const view = new Uint8Array(buffer, bufferStart + writeOffset, bufferSize - writeOffset),
          {read, written} = cachedEncoder.encodeInto(input.substring(readOffset), view);
    readOffset += read;
    writeOffset += written;
    if (readOffset === input.length) {
      callback(bufferStart, writeOffset);
      free(buffer, bufferStart);
      return;
    }
    bufferSize *= 2;
    bufferStart = realloc(buffer, bufferStart, bufferSize);
  }
}

7.5. Interface TextDecoderStream

[Exposed=*]
interface TextDecoderStream {
  constructor(optional DOMString label = "utf-8", optional TextDecoderOptions options = {});
};
TextDecoderStream includes TextDecoderCommon;
TextDecoderStream includes GenericTransformStream;

decoder = new TextDecoderStream([label = "utf-8" [, options]])

Returns a new TextDecoderStream object.

If label is either not a label or is a label for replacement, throws a RangeError.

decoder . encoding

Returns encoding’s name, lowercased.

decoder . fatal

Returns true if error mode is "fatal", and false otherwise.

decoder . ignoreBOM

Returns the value of ignore BOM.

decoder . readable

Returns a readable stream whose chunks are strings resulting from running encoding’s decoder on the chunks written to writable.

decoder . writable

Returns a writable stream which accepts AllowSharedBufferSource chunks and runs them through encoding’s decoder before making them available to readable.

Typically this will be used via the pipeThrough() method on a ReadableStream source.

var decoder = new TextDecoderStream(encoding);
byteReadable
  .pipeThrough(decoder)
  .pipeTo(textWritable);

If the error mode is "fatal" and encoding’s decoder returns error, both readable and writable will be errored with a TypeError.

The new TextDecoderStream(label, options) constructor steps are:

1. Let encoding be the result of getting an encoding from label.

2. If encoding is failure or replacement, then throw a RangeError.

3. Set this’s encoding to encoding.

4. If options["fatal"] is true, then set this’s error mode to "fatal".

5. Set this’s ignore BOM to options["ignoreBOM"].

6. Set this’s decoder to a new instance of this’s encoding’s decoder, and set this’s I/O queue to a new I/O queue.

7. Let transformAlgorithm be an algorithm which takes a chunk argument and runs the decode and enqueue a chunk algorithm with this and chunk.

8. Let flushAlgorithm be an algorithm which takes no arguments and runs the flush and enqueue algorithm with this.

9. Let transformStream be a new TransformStream.

10. Set up transformStream with transformAlgorithm set to transformAlgorithm and flushAlgorithm set to flushAlgorithm.

11. Set this’s transform to transformStream.

The decode and enqueue a chunk algorithm, given a TextDecoderStream object decoder and a chunk, runs these steps:

1. Let bufferSource be the result of converting chunk to an AllowSharedBufferSource.

2. Push a copy of bufferSource to decoder’s I/O queue.

   See the warning for SharedArrayBuffer objects above.

3. Let output be the I/O queue of scalar values « end-of-queue ».

4. While true:

   1. Let item be the result of reading from decoder’s I/O queue.

   2. If item is end-of-queue, then:

      1. Let outputChunk be the result of running serialize I/O queue with decoder and output.

      2. If outputChunk is non-empty, then enqueue outputChunk in decoder’s transform.

      3. Return.

   3. Let result be the result of processing an item with item, decoder’s decoder, decoder’s I/O queue, output, and decoder’s error mode.

   4. If result is error, then throw a TypeError.

The flush and enqueue algorithm, which handles the end of data from the input ReadableStream object, given a TextDecoderStream object decoder, runs these steps:

1. Let output be the I/O queue of scalar values « end-of-queue ».

2. While true:

   1. Let item be the result of reading from decoder’s I/O queue.

   2. Let result be the result of processing an item with item, decoder’s decoder, decoder’s I/O queue, output, and decoder’s error mode.

   3. If result is finished, then:

      1. Let outputChunk be the result of running serialize I/O queue with decoder and output.

      2. If outputChunk is non-empty, then enqueue outputChunk in decoder’s transform.

      3. Return.

   4. Otherwise, if result is error, throw a TypeError.

7.6. Interface TextEncoderStream

[Exposed=*]
interface TextEncoderStream {
  constructor();
};
TextEncoderStream includes TextEncoderCommon;
TextEncoderStream includes GenericTransformStream;

A TextEncoderStream object has an associated:

encoder

An encoder instance.

leading surrogate

Null or a leading surrogate, initially null.

A TextEncoderStream object offers no label argument as it only supports UTF-8.

encoder = new TextEncoderStream()

Returns a new TextEncoderStream object.

encoder . encoding

Returns "utf-8".

encoder . readable

Returns a readable stream whose chunks are Uint8Arrays resulting from running UTF-8’s encoder on the chunks written to writable.

encoder . writable

Returns a writable stream which accepts string chunks and runs them through UTF-8’s encoder before making them available to readable.

Typically this will be used via the pipeThrough() method on a ReadableStream source.

textReadable
  .pipeThrough(new TextEncoderStream())
  .pipeTo(byteWritable);

The new TextEncoderStream() constructor steps are:

1. Set this’s encoder to an instance of the UTF-8 encoder.

2. Let transformAlgorithm be an algorithm which takes a chunk argument and runs the encode and enqueue a chunk algorithm with this and chunk.

3. Let flushAlgorithm be an algorithm which runs the encode and flush algorithm with this.

4. Let transformStream be a new TransformStream.

5. Set up transformStream with transformAlgorithm set to transformAlgorithm and flushAlgorithm set to flushAlgorithm.

6. Set this’s transform to transformStream.

The encode and enqueue a chunk algorithm, given a TextEncoderStream object encoder and chunk, runs these steps:

1. Let input be the result of converting chunk to a DOMString.

2. Convert input to an I/O queue of code units.

   DOMString, as well as an I/O queue of code units rather than scalar values, are used here so that a surrogate pair that is split between chunks can be reassembled into the appropriate scalar value. The behavior is otherwise identical to USVString. In particular, lone surrogates will be replaced with U+FFFD.

3. Let output be the I/O queue of bytes « end-of-queue ».

4. While true:

   1. Let item be the result of reading from input.

   2. If item is end-of-queue, then:

      1. Convert output into a byte sequence.

      2. If output is non-empty, then:

         1. Let chunk be a Uint8Array object wrapping an ArrayBuffer containing output.

         2. Enqueue chunk into encoder’s transform.

      3. Return.

   3. Let result be the result of executing the convert code unit to scalar value algorithm with encoder, item and input.

   4. If result is not continue, then process an item with result, encoder’s encoder, input, output, and "fatal".

The convert code unit to scalar value algorithm, given a TextEncoderStream object encoder, a code unit item, and an I/O queue of code units input, runs these steps:

1. If encoder’s leading surrogate is non-null, then:

   1. Let leadingSurrogate be encoder’s leading surrogate.

   2. Set encoder’s leading surrogate to null.

   3. If item is a trailing surrogate, then return a scalar value from surrogates given leadingSurrogate and item.

   4. Restore item to input.

   5. Return U+FFFD.

2. If item is a leading surrogate, then set encoder’s leading surrogate to item and return continue.

3. If item is a trailing surrogate, then return U+FFFD.

4. Return item.

This is equivalent to the "convert a string into a scalar value string" algorithm from the Infra Standard, but allows for surrogate pairs that are split between strings. [INFRA]

The encode and flush algorithm, given a TextEncoderStream object encoder, runs these steps:

1. If encoder’s leading surrogate is non-null, then:

   1. Let chunk be a Uint8Array object wrapping an ArrayBuffer containing 0xEF 0xBF 0xBD.

      This is U+FFFD (�) in UTF-8 bytes.

   2. Enqueue chunk into encoder’s transform.

8. The encoding

8.1. UTF-8

8.1.1. UTF-8 decoder

A byte order mark has priority over a label as it has been found to be more accurate in deployed content. Therefore it is not part of the UTF-8 decoder algorithm, but rather the decode and UTF-8 decode algorithms.

UTF-8’s decoder has an associated UTF-8 code point, UTF-8 bytes seen, and UTF-8 bytes needed (all initially 0), a UTF-8 lower boundary (initially 0x80), and a UTF-8 upper boundary (initially 0xBF).

UTF-8’s decoder’s handler, given ioQueue and byte, runs these steps:

1. If byte is end-of-queue and UTF-8 bytes needed is not 0, set UTF-8 bytes needed to 0 and return error.

2. If byte is end-of-queue, return finished.

3. If UTF-8 bytes needed is 0, based on byte:

   0x00 to 0x7F

   Return a code point whose value is byte.

   0xC2 to 0xDF

   1. Set UTF-8 bytes needed to 1.

   2. Set UTF-8 code point to byte & 0x1F.

      The five least significant bits of byte.

   0xE0 to 0xEF

   1. If byte is 0xE0, set UTF-8 lower boundary to 0xA0.

   2. If byte is 0xED, set UTF-8 upper boundary to 0x9F.

   3. Set UTF-8 bytes needed to 2.

   4. Set UTF-8 code point to byte & 0xF.

      The four least significant bits of byte.

   0xF0 to 0xF4

   1. If byte is 0xF0, set UTF-8 lower boundary to 0x90.

   2. If byte is 0xF4, set UTF-8 upper boundary to 0x8F.

   3. Set UTF-8 bytes needed to 3.

   4. Set UTF-8 code point to byte & 0x7.

      The three least significant bits of byte.

   Otherwise

   Return error.

   Return continue.

4. If byte is not in the range UTF-8 lower boundary to UTF-8 upper boundary, inclusive, then:

   1. Set UTF-8 code point, UTF-8 bytes needed, and UTF-8 bytes seen to 0, set UTF-8 lower boundary to 0x80, and set UTF-8 upper boundary to 0xBF.

   2. Restore byte to ioQueue.

   3. Return error.

5. Set UTF-8 lower boundary to 0x80 and UTF-8 upper boundary to 0xBF.

6. Set UTF-8 code point to (UTF-8 code point << 6) | (byte & 0x3F)

   Shift the existing bits of UTF-8 code point left by six places and set the newly-vacated six least significant bits to the six least significant bits of byte.

7. Increase UTF-8 bytes seen by one.

8. If UTF-8 bytes seen is not equal to UTF-8 bytes needed, return continue.

9. Let code point be UTF-8 code point.

10. Set UTF-8 code point, UTF-8 bytes needed, and UTF-8 bytes seen to 0.

11. Return a code point whose value is code point.

The constraints in the UTF-8 decoder above match “Best Practices for Using U+FFFD” from the Unicode standard. No other behavior is permitted per the Encoding Standard (other algorithms that achieve the same result are fine, even encouraged). [UNICODE]

8.1.2. UTF-8 encoder

UTF-8’s encoder’s handler, given ioQueue and code point, runs these steps:

1. If code point is end-of-queue, return finished.

2. If code point is an ASCII code point, return a byte whose value is code point.

3. Set count and offset based on the range code point is in:

   U+0080 to U+07FF, inclusive

   1 and 0xC0

   U+0800 to U+FFFF, inclusive

   2 and 0xE0

   U+10000 to U+10FFFF, inclusive

   3 and 0xF0

4. Let bytes be a byte sequence whose first byte is (code point >> (6 × count)) + offset.

5. While count is greater than 0:

   1. Set temp to code point >> (6 × (count − 1)).

   2. Append to bytes 0x80 | (temp & 0x3F).

   3. Decrease count by one.

6. Return bytes bytes, in order.

This algorithm has identical results to the one described in the Unicode standard. It is included here for completeness. [UNICODE]

9. Legacy single-byte encodings

An encoding where each byte is either a single code point or nothing, is a single-byte encoding. Single-byte encodings share the decoder and encoder. Index single-byte, as referenced by the single-byte decoder and single-byte encoder, is defined by the following table, and depends on the single-byte encoding in use. All but two single-byte encodings have a unique index.

ISO-8859-8 and ISO-8859-8-I are distinct encoding names, because ISO-8859-8 has influence on the layout direction. And although historically this might have been the case for ISO-8859-6 and "ISO-8859-6-I" as well, that is no longer true.

9.1. single-byte decoder

Single-byte encodings’s decoder’s handler, given ioQueue and byte, runs these steps:

1. If byte is end-of-queue, return finished.

2. If byte is an ASCII byte, return a code point whose value is byte.

3. Let code point be the index code point for byte − 0x80 in index single-byte.

4. If code point is null, return error.

5. Return a code point whose value is code point.

9.2. single-byte encoder

Single-byte encodings’s encoder’s handler, given ioQueue and code point, runs these steps:

1. If code point is end-of-queue, return finished.

2. If code point is an ASCII code point, return a byte whose value is code point.

3. Let pointer be the index pointer for code point in index single-byte.

4. If pointer is null, return error with code point.

5. Return a byte whose value is pointer + 0x80.

10. Legacy multi-byte Chinese (simplified) encodings

10.1. GBK

10.1.1. GBK decoder

GBK’s decoder is gb18030’s decoder.

10.1.2. GBK encoder

GBK’s encoder is gb18030’s encoder with its is GBK set to true.

Not fully aliasing GBK with gb18030 is a conservative move to decrease the chances of breaking legacy servers and other consumers of content generated with GBK’s encoder.

10.2. gb18030

10.2.1. gb18030 decoder

gb18030’s decoder has an associated gb18030 first, gb18030 second, and gb18030 third (all initially 0x00).

gb18030’s decoder’s handler, given ioQueue and byte, runs these steps:

1. If byte is end-of-queue and gb18030 first, gb18030 second, and gb18030 third are 0x00, return finished.

2. If byte is end-of-queue, and gb18030 first, gb18030 second, or gb18030 third is not 0x00, set gb18030 first, gb18030 second, and gb18030 third to 0x00, and return error.

3. If gb18030 third is not 0x00, then:

   1. If byte is not in the range 0x30 to 0x39, inclusive, then:

      1. Restore « gb18030 second, gb18030 third, byte » to ioQueue.

      2. Set gb18030 first, gb18030 second, and gb18030 third to 0x00.

      3. Return error.

   2. Let code point be the index gb18030 ranges code point for ((gb18030 first − 0x81) × (10 × 126 × 10)) + ((gb18030 second − 0x30) × (10 × 126)) + ((gb18030 third − 0x81) × 10) + byte − 0x30.

   3. Set gb18030 first, gb18030 second, and gb18030 third to 0x00.

   4. If code point is null, return error.

   5. Return a code point whose value is code point.

4. If gb18030 second is not 0x00, then:

   1. If byte is in the range 0x81 to 0xFE, inclusive, set gb18030 third to byte and return continue.

   2. Restore « gb18030 second, byte » to ioQueue, set gb18030 first and gb18030 second to 0x00, and return error.

5. If gb18030 first is not 0x00, then:

   1. If byte is in the range 0x30 to 0x39, inclusive, set gb18030 second to byte and return continue.

   2. Let lead be gb18030 first, let pointer be null, and set gb18030 first to 0x00.

   3. Let offset be 0x40 if byte is less than 0x7F, otherwise 0x41.

   4. If byte is in the range 0x40 to 0x7E, inclusive, or 0x80 to 0xFE, inclusive, set pointer to (lead − 0x81) × 190 + (byte − offset).

   5. Let code point be null if pointer is null, otherwise the index code point for pointer in index gb18030.

   6. If code point is non-null, return a code point whose value is code point.

   7. If byte is an ASCII byte, restore byte to ioQueue.

   8. Return error.

6. If byte is an ASCII byte, return a code point whose value is byte.

7. If byte is 0x80, return code point U+20AC.

8. If byte is in the range 0x81 to 0xFE, inclusive, set gb18030 first to byte and return continue.

9. Return error.

10.2.2. gb18030 encoder

gb18030’s encoder has an associated is GBK (initially false).

gb18030’s encoder’s handler, given ioQueue and code point, runs these steps:

1. If code point is end-of-queue, return finished.

2. If code point is an ASCII code point, return a byte whose value is code point.

3. If code point is U+E5E5, return error with code point.

   Index gb18030 maps 0xA3 0xA0 to U+3000 rather than U+E5E5 for compatibility with deployed content. Therefore it cannot roundtrip.

4. If is GBK is true and code point is U+20AC, return byte 0x80.

5. If there is a row in the table below whose first column is code point, then return the two bytes on the same row listed in the second column:

   Code point	Bytes
   U+E78D	0xA6 0xD9
   U+E78E	0xA6 0xDA
   U+E78F	0xA6 0xDB
   U+E790	0xA6 0xDC
   U+E791	0xA6 0xDD
   U+E792	0xA6 0xDE
   U+E793	0xA6 0xDF
   U+E794	0xA6 0xEC
   U+E795	0xA6 0xED
   U+E796	0xA6 0xF3
   U+E81E	0xFE 0x59
   U+E826	0xFE 0x61
   U+E82B	0xFE 0x66
   U+E82C	0xFE 0x67
   U+E832	0xFE 0x6D
   U+E843	0xFE 0x7E
   U+E854	0xFE 0x90
   U+E864	0xFE 0xA0

   This asymmetric encoder table preserves compatibility with the GB18030-2005 standard. See also the explanation at index gb18030 ranges.

6. Let pointer be the index pointer for code point in index gb18030.

7. If pointer is non-null, then:

   1. Let lead be pointer / 190 + 0x81.

   2. Let trail be pointer % 190.

   3. Let offset be 0x40 if trail is less than 0x3F, otherwise 0x41.

   4. Return two bytes whose values are lead and trail + offset.

8. If is GBK is true, return error with code point.

9. Set pointer to the index gb18030 ranges pointer for code point.

10. Let byte1 be pointer / (10 × 126 × 10).

11. Set pointer to pointer % (10 × 126 × 10).

12. Let byte2 be pointer / (10 × 126).

13. Set pointer to pointer % (10 × 126).

14. Let byte3 be pointer / 10.

15. Let byte4 be pointer % 10.

16. Return four bytes whose values are byte1 + 0x81, byte2 + 0x30, byte3 + 0x81, byte4 + 0x30.

11. Legacy multi-byte Chinese (traditional) encodings

11.1. Big5

11.1.1. Big5 decoder

Big5’s decoder has an associated Big5 lead (initially 0x00).

Big5’s decoder’s handler, given ioQueue and byte, runs these steps:

1. If byte is end-of-queue and Big5 lead is not 0x00, set Big5 lead to 0x00 and return error.

2. If byte is end-of-queue and Big5 lead is 0x00, return finished.

3. If Big5 lead is not 0x00, let lead be Big5 lead, let pointer be null, set Big5 lead to 0x00, and then:

   1. Let offset be 0x40 if byte is less than 0x7F, otherwise 0x62.

   2. If byte is in the range 0x40 to 0x7E, inclusive, or 0xA1 to 0xFE, inclusive, set pointer to (lead − 0x81) × 157 + (byte − offset).

   3. If there is a row in the table below whose first column is pointer, return the two code points listed in its second column (the third column is irrelevant):

      Pointer	Code points	Notes
      1133	U+00CA U+0304	Ê̄ (LATIN CAPITAL LETTER E WITH CIRCUMFLEX AND MACRON)
      1135	U+00CA U+030C	Ê̌ (LATIN CAPITAL LETTER E WITH CIRCUMFLEX AND CARON)
      1164	U+00EA U+0304	ê̄ (LATIN SMALL LETTER E WITH CIRCUMFLEX AND MACRON)
      1166	U+00EA U+030C	ê̌ (LATIN SMALL LETTER E WITH CIRCUMFLEX AND CARON)

      Since indexes are limited to single code points this table is used for these pointers.

   4. Let code point be null if pointer is null, otherwise the index code point for pointer in index Big5.

   5. If code point is non-null, return a code point whose value is code point.

   6. If byte is an ASCII byte, restore byte to ioQueue.

   7. Return error.

4. If byte is an ASCII byte, return a code point whose value is byte.

5. If byte is in the range 0x81 to 0xFE, inclusive, set Big5 lead to byte and return continue.

6. Return error.

11.1.2. Big5 encoder

Big5’s encoder’s handler, given ioQueue and code point, runs these steps:

1. If code point is end-of-queue, return finished.

2. If code point is an ASCII code point, return a byte whose value is code point.

3. Let pointer be the index Big5 pointer for code point.

4. If pointer is null, return error with code point.

5. Let lead be pointer / 157 + 0x81.

6. Let trail be pointer % 157.

7. Let offset be 0x40 if trail is less than 0x3F, otherwise 0x62.

8. Return two bytes whose values are lead and trail + offset.

12. Legacy multi-byte Japanese encodings

12.1. EUC-JP

12.1.1. EUC-JP decoder

EUC-JP’s decoder has an associated EUC-JP jis0212 (initially false) and EUC-JP lead (initially 0x00).

EUC-JP’s decoder’s handler, given ioQueue and byte, runs these steps:

1. If byte is end-of-queue and EUC-JP lead is not 0x00, set EUC-JP lead to 0x00, and return error.

2. If byte is end-of-queue and EUC-JP lead is 0x00, return finished.

3. If EUC-JP lead is 0x8E and byte is in the range 0xA1 to 0xDF, inclusive, set EUC-JP lead to 0x00 and return a code point whose value is 0xFF61 − 0xA1 + byte.

4. If EUC-JP lead is 0x8F and byte is in the range 0xA1 to 0xFE, inclusive, set EUC-JP jis0212 to true, set EUC-JP lead to byte, and return continue.

5. If EUC-JP lead is not 0x00, let lead be EUC-JP lead, set EUC-JP lead to 0x00, and then:

   1. Let code point be null.

   2. If lead and byte are both in the range 0xA1 to 0xFE, inclusive, then set code point to the index code point for (lead − 0xA1) × 94 + byte − 0xA1 in index jis0208 if EUC-JP jis0212 is false and in index jis0212 otherwise.

   3. Set EUC-JP jis0212 to false.

   4. If code point is non-null, return a code point whose value is code point.

   5. If byte is an ASCII byte, restore byte to ioQueue.

   6. Return error.

6. If byte is an ASCII byte, return a code point whose value is byte.

7. If byte is 0x8E, 0x8F, or in the range 0xA1 to 0xFE, inclusive, set EUC-JP lead to byte and return continue.

8. Return error.

12.1.2. EUC-JP encoder

EUC-JP’s encoder’s handler, given ioQueue and code point, runs these steps:

1. If code point is end-of-queue, return finished.

2. If code point is an ASCII code point, return a byte whose value is code point.

3. If code point is U+00A5, return byte 0x5C.

4. If code point is U+203E, return byte 0x7E.

5. If code point is in the range U+FF61 to U+FF9F, inclusive, return two bytes whose values are 0x8E and code point − 0xFF61 + 0xA1.

6. If code point is U+2212, set it to U+FF0D.

7. Let pointer be the index pointer for code point in index jis0208.

   If pointer is non-null, it is less than 8836 due to the nature of index jis0208 and the index pointer operation.

8. If pointer is null, return error with code point.

9. Let lead be pointer / 94 + 0xA1.

10. Let trail be pointer % 94 + 0xA1.

11. Return two bytes whose values are lead and trail.

12.2. ISO-2022-JP

12.2.1. ISO-2022-JP decoder

ISO-2022-JP’s decoder has an associated ISO-2022-JP decoder state (initially ASCII), ISO-2022-JP decoder output state (initially ASCII), ISO-2022-JP lead (initially 0x00), and ISO-2022-JP output (initially false).

ISO-2022-JP’s decoder’s handler, given ioQueue and byte, runs these steps, switching on ISO-2022-JP decoder state:

ASCII

Based on byte:

0x1B

Set ISO-2022-JP decoder state to escape start and return continue.

0x00 to 0x7F, excluding 0x0E, 0x0F, and 0x1B

Set ISO-2022-JP output to false and return a code point whose value is byte.

end-of-queue

Return finished.

Otherwise

Set ISO-2022-JP output to false and return error.

Roman

Based on byte:

0x1B

Set ISO-2022-JP decoder state to escape start and return continue.

0x5C

Set ISO-2022-JP output to false and return code point U+00A5.

0x7E

Set ISO-2022-JP output to false and return code point U+203E.

0x00 to 0x7F, excluding 0x0E, 0x0F, 0x1B, 0x5C, and 0x7E

Set ISO-2022-JP output to false and return a code point whose value is byte.

end-of-queue

Return finished.

Otherwise

Set ISO-2022-JP output to false and return error.

katakana

Based on byte:

0x1B

Set ISO-2022-JP decoder state to escape start and return continue.

0x21 to 0x5F

Set ISO-2022-JP output to false and return a code point whose value is 0xFF61 − 0x21 + byte.

end-of-queue

Return finished.

Otherwise

Set ISO-2022-JP output to false and return error.

Lead byte

Based on byte:

0x1B

Set ISO-2022-JP decoder state to escape start and return continue.

0x21 to 0x7E

Set ISO-2022-JP output to false, ISO-2022-JP lead to byte, ISO-2022-JP decoder state to trail byte, and return continue.

end-of-queue

Return finished.

Otherwise

Set ISO-2022-JP output to false and return error.

Trail byte

Based on byte:

0x1B

Set ISO-2022-JP decoder state to escape start and return error.

0x21 to 0x7E

1. Set the ISO-2022-JP decoder state to lead byte.

2. Let pointer be (ISO-2022-JP lead − 0x21) × 94 + byte − 0x21.

3. Let code point be the index code point for pointer in index jis0208.

4. If code point is null, return error.

5. Return a code point whose value is code point.

end-of-queue

Set the ISO-2022-JP decoder state to lead byte and return error.

Otherwise

Set ISO-2022-JP decoder state to lead byte and return error.

Escape start

1. If byte is either 0x24 or 0x28, set ISO-2022-JP lead to byte, ISO-2022-JP decoder state to escape, and return continue.

2. If byte is not end-of-queue, then restore byte to ioQueue.

3. Set ISO-2022-JP output to false, ISO-2022-JP decoder state to ISO-2022-JP decoder output state, and return error.

Escape

1. Let lead be ISO-2022-JP lead and set ISO-2022-JP lead to 0x00.

2. Let state be null.

3. If lead is 0x28 and byte is 0x42, set state to ASCII.

4. If lead is 0x28 and byte is 0x4A, set state to Roman.

5. If lead is 0x28 and byte is 0x49, set state to katakana.

6. If lead is 0x24 and byte is either 0x40 or 0x42, set state to lead byte.

7. If state is non-null, then:

   1. Set ISO-2022-JP decoder state and ISO-2022-JP decoder output state to state.

   2. Let output be the value of ISO-2022-JP output.

   3. Set ISO-2022-JP output to true.

   4. Return continue, if output is false, and error otherwise.

8. If byte is end-of-queue, then restore lead to ioQueue; otherwise, restore « lead, byte » to ioQueue.

9. Set ISO-2022-JP output to false, ISO-2022-JP decoder state to ISO-2022-JP decoder output state and return error.

12.2.2. ISO-2022-JP encoder

ISO-2022-JP’s encoder has an associated ISO-2022-JP encoder state which is ASCII, Roman, or jis0208 (initially ASCII).

ISO-2022-JP’s encoder’s handler, given ioQueue and code point, runs these steps:

1. If code point is end-of-queue and ISO-2022-JP encoder state is not ASCII, set ISO-2022-JP encoder state to ASCII, and return three bytes 0x1B 0x28 0x42.

2. If code point is end-of-queue and ISO-2022-JP encoder state is ASCII, return finished.

3. If ISO-2022-JP encoder state is ASCII or Roman, and code point is U+000E, U+000F, or U+001B, return error with U+FFFD.

   This returns U+FFFD rather than code point to prevent attacks.

4. If ISO-2022-JP encoder state is ASCII and code point is an ASCII code point, return a byte whose value is code point.

5. If ISO-2022-JP encoder state is Roman and code point is an ASCII code point, excluding U+005C and U+007E, or is U+00A5 or U+203E, then:

   1. If code point is an ASCII code point, return a byte whose value is code point.

   2. If code point is U+00A5, return byte 0x5C.

   3. If code point is U+203E, return byte 0x7E.

6. If code point is an ASCII code point, and ISO-2022-JP encoder state is not ASCII, restore code point to ioQueue, set ISO-2022-JP encoder state to ASCII, and return three bytes 0x1B 0x28 0x42.

7. If code point is either U+00A5 or U+203E, and ISO-2022-JP encoder state is not Roman, restore code point to ioQueue, set ISO-2022-JP encoder state to Roman, and return three bytes 0x1B 0x28 0x4A.

8. If code point is U+2212, set it to U+FF0D.

9. If code point is in the range U+FF61 to U+FF9F, inclusive, set it to the index code point for code point − 0xFF61 in index ISO-2022-JP katakana.

10. Let pointer be the index pointer for code point in index jis0208.

    If pointer is non-null, it is less than 8836 due to the nature of index jis0208 and the index pointer operation.

11. If pointer is null, then:

    1. If ISO-2022-JP encoder state is jis0208, then restore code point to ioQueue, set ISO-2022-JP encoder state to ASCII, and return three bytes 0x1B 0x28 0x42.

    2. Return error with code point.

12. If ISO-2022-JP encoder state is not jis0208, restore code point to ioQueue, set ISO-2022-JP encoder state to jis0208, and return three bytes 0x1B 0x24 0x42.

13. Let lead be pointer / 94 + 0x21.

14. Let trail be pointer % 94 + 0x21.

15. Return two bytes whose values are lead and trail.

12.3. Shift_JIS

12.3.1. Shift_JIS decoder

Shift_JIS’s decoder has an associated Shift_JIS lead (initially 0x00).

Shift_JIS’s decoder’s handler, given ioQueue and byte, runs these steps:

1. If byte is end-of-queue and Shift_JIS lead is not 0x00, set Shift_JIS lead to 0x00 and return error.

2. If byte is end-of-queue and Shift_JIS lead is 0x00, return finished.

3. If Shift_JIS lead is not 0x00, let lead be Shift_JIS lead, let pointer be null, set Shift_JIS lead to 0x00, and then:

   1. Let offset be 0x40 if byte is less than 0x7F, otherwise 0x41.

   2. Let lead offset be 0x81 if lead is less than 0xA0, otherwise 0xC1.

   3. If byte is in the range 0x40 to 0x7E, inclusive, or 0x80 to 0xFC, inclusive, set pointer to (lead − lead offset) × 188 + byte − offset.

   4. If pointer is in the range 8836 to 10715, inclusive, return a code point whose value is 0xE000 − 8836 + pointer.

      This is interoperable legacy from Windows known as EUDC.

   5. Let code point be null if pointer is null, otherwise the index code point for pointer in index jis0208.

   6. If code point is non-null, return a code point whose value is code point.

   7. If byte is an ASCII byte, restore byte to ioQueue.

   8. Return error.

4. If byte is an ASCII byte or 0x80, return a code point whose value is byte.

5. If byte is in the range 0xA1 to 0xDF, inclusive, return a code point whose value is 0xFF61 − 0xA1 + byte.

6. If byte is in the range 0x81 to 0x9F, inclusive, or 0xE0 to 0xFC, inclusive, set Shift_JIS lead to byte and return continue.

7. Return error.

12.3.2. Shift_JIS encoder

Shift_JIS’s encoder’s handler, given ioQueue and code point, runs these steps:

1. If code point is end-of-queue, return finished.

2. If code point is an ASCII code point or U+0080, return a byte whose value is code point.

3. If code point is U+00A5, return byte 0x5C.

4. If code point is U+203E, return byte 0x7E.

5. If code point is in the range U+FF61 to U+FF9F, inclusive, return a byte whose value is code point − 0xFF61 + 0xA1.

6. If code point is U+2212, set it to U+FF0D.

7. Let pointer be the index Shift_JIS pointer for code point.

8. If pointer is null, return error with code point.

9. Let lead be pointer / 188.

10. Let lead offset be 0x81 if lead is less than 0x1F, otherwise 0xC1.

11. Let trail be pointer % 188.

12. Let offset be 0x40 if trail is less than 0x3F, otherwise 0x41.

13. Return two bytes whose values are lead + lead offset and trail + offset.

13. Legacy multi-byte Korean encodings

13.1. EUC-KR

13.1.1. EUC-KR decoder

EUC-KR’s decoder has an associated EUC-KR lead (initially 0x00).

EUC-KR’s decoder’s handler, given ioQueue and byte, runs these steps:

1. If byte is end-of-queue and EUC-KR lead is not 0x00, set EUC-KR lead to 0x00 and return error.

2. If byte is end-of-queue and EUC-KR lead is 0x00, return finished.

3. If EUC-KR lead is not 0x00, let lead be EUC-KR lead, let pointer be null, set EUC-KR lead to 0x00, and then:

   1. If byte is in the range 0x41 to 0xFE, inclusive, set pointer to (lead − 0x81) × 190 + (byte − 0x41).

   2. Let code point be null if pointer is null, otherwise the index code point for pointer in index EUC-KR.

   3. If code point is non-null, return a code point whose value is code point.

   4. If byte is an ASCII byte, restore byte to ioQueue.

   5. Return error.

4. If byte is an ASCII byte, return a code point whose value is byte.

5. If byte is in the range 0x81 to 0xFE, inclusive, set EUC-KR lead to byte and return continue.

6. Return error.

13.1.2. EUC-KR encoder

EUC-KR’s encoder’s handler, given ioQueue and code point, runs these steps:

1. If code point is end-of-queue, return finished.

2. If code point is an ASCII code point, return a byte whose value is code point.

3. Let pointer be the index pointer for code point in index EUC-KR.

4. If pointer is null, return error with code point.

5. Let lead be pointer / 190 + 0x81.

6. Let trail be pointer % 190 + 0x41.

7. Return two bytes whose values are lead and trail.

14. Legacy miscellaneous encodings

14.1. replacement

The replacement encoding exists to prevent certain attacks that abuse a mismatch between encodings supported on the server and the client.

14.1.1. replacement decoder

replacement’s decoder has an associated replacement error returned (initially false).

replacement’s decoder’s handler, given ioQueue and byte, runs these steps:

1. If byte is end-of-queue, return finished.

2. If replacement error returned is false, set replacement error returned to true and return error.

3. Return finished.

14.2. Common infrastructure for UTF-16BE/LE

UTF-16BE/LE is UTF-16BE or UTF-16LE.

14.2.1. shared UTF-16 decoder

A byte order mark has priority over a label as it has been found to be more accurate in deployed content. Therefore it is not part of the shared UTF-16 decoder algorithm, but rather the decode algorithm.

shared UTF-16 decoder has an associated UTF-16 lead byte and UTF-16 leading surrogate (both initially null), and is UTF-16BE decoder (initially false).

shared UTF-16 decoder’s handler, given ioQueue and byte, runs these steps:

1. If byte is end-of-queue and either UTF-16 lead byte or UTF-16 leading surrogate is non-null, set UTF-16 lead byte and UTF-16 leading surrogate to null, and return error.

2. If byte is end-of-queue and UTF-16 lead byte and UTF-16 leading surrogate are null, return finished.

3. If UTF-16 lead byte is null, set UTF-16 lead byte to byte and return continue.

4. Let code unit be the result of:

   is UTF-16BE decoder is true

   (UTF-16 lead byte << 8) + byte.

   is UTF-16BE decoder is false

   (byte << 8) + UTF-16 lead byte.

   Then set UTF-16 lead byte to null.

5. If UTF-16 leading surrogate is non-null:

   1. Let leadingSurrogate be UTF-16 leading surrogate.

   2. Set UTF-16 leading surrogate to null.

   3. If code unit is a trailing surrogate, then return a scalar value from surrogates given leadingSurrogate and code unit.

   4. Let byte1 be code unit >> 8.

   5. Let byte2 be code unit & 0x00FF.

   6. Let bytes be a list of two bytes whose values are byte1 and byte2, if is UTF-16BE decoder is true; otherwise byte2 and byte1.

   7. Restore bytes to ioQueue and return error.

6. If code unit is a leading surrogate, then set UTF-16 leading surrogate to code unit and return continue.

7. If code unit is a trailing surrogate, then return error.

8. Return code point code unit.

14.3. UTF-16BE

14.3.1. UTF-16BE decoder

UTF-16BE’s decoder is shared UTF-16 decoder with its is UTF-16BE decoder set to true.

14.4. UTF-16LE

"utf-16" is a label for UTF-16LE to deal with deployed content.

14.4.1. UTF-16LE decoder

UTF-16LE’s decoder is shared UTF-16 decoder.

14.5. x-user-defined

While technically this is a single-byte encoding, it is defined separately as it can be implemented algorithmically.

14.5.1. x-user-defined decoder

x-user-defined’s decoder’s handler, given ioQueue and byte, runs these steps:

1. If byte is end-of-queue, return finished.

2. If byte is an ASCII byte, return a code point whose value is byte.

3. Return a code point whose value is 0xF780 + byte − 0x80.

14.5.2. x-user-defined encoder

x-user-defined’s encoder’s handler, given ioQueue and code point, runs these steps:

1. If code point is end-of-queue, return finished.

2. If code point is an ASCII code point, return a byte whose value is code point.

3. If code point is in the range U+F780 to U+F7FF, inclusive, return a byte whose value is code point − 0xF780 + 0x80.

4. Return error with code point.

15. Browser UI

Browsers are encouraged to not enable overriding the encoding of a resource. If such a feature is nonetheless present, browsers should not offer UTF-16BE/LE as an option, due to the aforementioned security issues. Browsers should also disable this feature if the resource was decoded using UTF-16BE/LE.

Implementation considerations

Instead of supporting I/O queues with arbitrary restore, the decoders for encodings in this standard could be implemented with:

1. The ability to unread the current byte.

2. A single-byte buffer for gb18030 (an ASCII byte) and ISO-2022-JP (0x24 or 0x28).

   For gb18030 when hitting a bogus byte while gb18030 third is not 0x00, gb18030 second could be moved into the single-byte buffer to be returned next, and gb18030 third would be the new gb18030 first, checked for not being 0x00 after the single-byte buffer was returned and emptied. This is possible as the range for the first and third byte in gb18030 is identical.

The ISO-2022-JP encoder needs ISO-2022-JP encoder state as additional state, but other than that, none of the encoders for encodings in this standard require additional state or buffers.

Acknowledgments

There have been a lot of people that have helped make encodings more interoperable over the years and thereby furthered the goals of this standard. Likewise many people have helped making this standard what it is today.

With that, many thanks to Adam Rice, Alan Chaney, Alexander Shtuchkin, Allen Wirfs-Brock, Andreu Botella, Aneesh Agrawal, Arkadiusz Michalski, Asmus Freytag, Ben Noordhuis, Bnaya Peretz, Boris Zbarsky, Bruno Haible, Cameron McCormack, Charles McCathieNeville, Christopher Foo, CodifierNL, David Carlisle, Domenic Denicola, Dominique Hazaël-Massieux, Doug Ewell, Erik van der Poel, 譚永鋒 (Frank Yung-Fong Tang), Glenn Maynard, Gordon P. Hemsley, Henri Sivonen, Ian Hickson, J. King, James Graham, Jeffrey Yasskin, John Tamplin, Joshua Bell, 村井純 (Jun Murai), 신정식 (Jungshik Shin), Jxck, 강 성훈 (Kang Seonghoon), 川幡太一 (Kawabata Taichi), Ken Lunde, Ken Whistler, Kenneth Russell, 田村健人 (Kent Tamura), Leif Halvard Silli, Luke Wagner, Maciej Hirsz, Makoto Kato, Mark Callow, Mark Crispin, Mark Davis, Martin Dürst, Masatoshi Kimura, Mattias Buelens, Ms2ger, Nigel Megitt, Nigel Tao, Norbert Lindenberg, Øistein E. Andersen, Peter Krefting, Philip Jägenstedt, Philip Taylor, Richard Ishida, Robbert Broersma, Robert Mustacchi, Ryan Dahl, Sam Sneddon, Shawn Steele, Simon Montagu, Simon Pieters, Simon Sapin, Stephen Checkoway, 寺田健 (Takeshi Terada), Vyacheslav Matva, Wolf Lammen, and 成瀬ゆい (Yui Naruse) for being awesome.

This standard is written by Anne van Kesteren (Apple, annevk@annevk.nl). The API chapter was initially written by Joshua Bell (Google).

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.