TextEncoder: encodeInto() method
The TextEncoder.encodeInto()
method takes a
string to encode and a destination Uint8Array
to put resulting UTF-8 encoded text into, and returns a dictionary object indicating the
progress of the encoding.
This is potentially more performant than the older encode()
method — especially when the target buffer is a view into a Wasm heap.
Syntax
encodeInto(string, uint8Array)
Parameters
string
-
A string containing the text to encode.
uint8Array
-
A
Uint8Array
object instance to place the resulting UTF-8 encoded text into.
Return value
An object, which contains two members:
read
-
The number of UTF-16 units of code from the source that has been converted over to UTF-8. This may be less than
string.length
ifuint8Array
did not have enough space. written
-
The number of bytes modified in the destination
Uint8Array
. The bytes written are guaranteed to form complete UTF-8 byte sequences.
Encode into a specific position
encodeInto()
always puts its output at the start of the array.
However, it is sometimes useful to make the output start at a particular index.
The solution is TypedArray.prototype.subarray()
:
const encoder = new TextEncoder();
function encodeIntoAtPosition(string, u8array, position) {
return encoder.encodeInto(
string,
position ? u8array.subarray(position | 0) : u8array,
);
}
const u8array = new Uint8Array(8);
encodeIntoAtPosition("hello", u8array, 2);
console.log(u8array.join()); // 0,0,104,101,108,108,111,0
Buffer sizing
To convert a JavaScript string s
, the output space needed for full conversion is never less than s.length
bytes and never greater than s.length * 3
bytes.
The exact UTF-8-to-UTF-16 length ratio for your string depends on the language you are working with:
- For basic English text that uses mostly ASCII characters, the ratio is close to 1.
- For text in scripts using characters U+0080 to U+07FF, which includes Greek, Cyrillic, Hebrew, Arabic, etc., the ratio is about 2.
- For text in scripts using characters U+0800 to U+FFFF, which includes Chinese, Japanese, Korean, etc., the ratio is about 3.
- It's not common for entire scripts to be written in non-BMP characters (although they do exist). These characters are usually math symbols, emojis, historical scripts, etc. The ratio for these characters is 2, because they take 4 bytes in UTF-8 and 2 in UTF-16.
If the output allocation (typically within Wasm heap) is expected to be short-lived, it makes sense to allocate s.length * 3
bytes for the output, in which case the first conversion attempt is guaranteed to convert the whole string.
For example, if your text is primarily English, it is unlikely that long text will exceed s.length * 2
bytes in length.
Thus, a more optimistic approach might be to allocate s.length * 2 + 5
bytes, and perform reallocation in the rare circumstance that the optimistic prediction was wrong.
If the output is expected to be long-lived, it makes sense to compute minimum allocation roundUpToBucketSize(s.length)
, the maximum allocation size s.length * 3
, and to have a chosen (as a tradeoff between memory usage and speed) threshold t
such that if roundUpToBucketSize(s.length) + t >= s.length * 3
, you allocate for s.length * 3
.
Otherwise, first allocate for roundUpToBucketSize(s.length)
and convert.
If the read
item it the return dictionary is s.length
, the conversion is done.
If not, reallocate the target buffer to written + (s.length - read) * 3
and then convert the rest by taking a substring of s
starting from index read
and a subbuffer of the target buffer starting from index written
.
Above roundUpToBucketSize()
is a function that rounds up to the allocator bucket size.
For example, if your Wasm allocator is known to use power-of-two buckets, roundUpToBucketSize()
should return the argument if it is a power-of-two or the next power-of-two otherwise.
If the behavior of the Wasm allocator is unknown, roundUpToBucketSize()
should be an identity function.
If the behavior of your allocator is unknown, you might want to have up to two reallocation steps and make the first reallocation step multiply the remaining unconverted length by two instead of three.
However, in that case, it makes sense not to implement the usual multiplying by two of the already written buffer length, because in such a case if a second reallocation happened, it would always over-allocate compared to the original length times three.
The above advice assumes that you don't need to allocate space for a zero terminator.
That is, on the Wasm side you are working with Rust strings or a non-zero-terminating C++ class.
If you are working with C++ std::string
, even though the logical length is shown to you, you need to take the extra terminator byte into account when computing rounding up to allocator bucket size.
See the next section about C strings.
No Zero-termination
If the input string contains the character U+0000 in the input, encodeInto()
will write a 0x00 byte in the output.
encodeInto()
does not write a C-style 0x00 sentinel byte after the logical output.
If your Wasm program uses C strings, it's your responsibility to write the 0x00
sentinel and you can't prevent your Wasm program from seeing a logically truncated string if the JavaScript string contained U+0000
.
Observe:
const encoder = new TextEncoder();
function encodeIntoWithSentinel(string, u8array, position) {
const stats = encoder.encodeInto(
string,
position ? u8array.subarray(position | 0) : u8array,
);
if (stats.written < u8array.length) u8array[stats.written] = 0; // append null if room
return stats;
}
Examples
<p class="source">This is a sample paragraph.</p>
<p class="result"></p>
const sourcePara = document.querySelector(".source");
const resultPara = document.querySelector(".result");
const string = sourcePara.textContent;
const textEncoder = new TextEncoder();
const utf8 = new Uint8Array(string.length);
const encodedResults = textEncoder.encodeInto(string, utf8);
resultPara.textContent +=
`Bytes read: ${encodedResults.read}` +
` | Bytes written: ${encodedResults.written}` +
` | Encoded result: ${utf8}`;
Specifications
Specification |
---|
Encoding Standard # ref-for-dom-textencoder-encodeinto① |
Browser compatibility
BCD tables only load in the browser
See also
- The
TextEncoder
interface it belongs to. TextEncoder.encode()