Champions: Guy Bedford, Bradley Farias, Michael Ficarra
Status: stage 4
ECMAScript string values are a finite ordered sequence of zero or more 16-bit unsigned integer values. However, ECMAScript does not place any restrictions or requirements on the integer values except that they must be 16-bit unsigned integers. In well-formed strings, each integer value in the sequence represents a single 16-bit code unit of UTF-16-encoded Unicode text. However, not all sequences of UTF-16 code units represent UTF-16-encoded Unicode text. In well-formed strings, code units in the range 0xD800..0xDBFF (leading surrogates) and 0xDC00..0xDFFF (trailing surrogates) must appear paired and in order. Strings with unpaired or out-of-order surrogates are ill-formed.
In WebIDL, possibly-ill-formed Strings are referred to using the DOMString type. But for interfaces which operate on Unicode text, WebIDL also defines the USVString type as the set of all possible sequences of Unicode Scalar Values, which are all of the Unicode code points apart from the surrogate code points (U+0000..U+D7FF and U+E000..U+10FFFF), representing Unicode text.
The WebAssembly Component Model requires well-formed strings, as do some compile-to-JS programming languages, many data encodings, network interfaces, filesystem interfaces, etc. Interfacing JavaScript strings with such APIs is a common use case that therefore suffers from conversion burdens. In particular because conversion from DOMString to USVString is lossy (common options are to replace unpaired surrogates or to throw an error) there is a regular need for string validation both within the platform and for certain userland use case scenarios.
The proposal is to define in ECMA-262 a method to verify if a given ECMAScript string is well-formed or not. As a highly common scenario for interfaces between JavaScript/web APIs and those that operate on Unicode text, this test should be as efficient as possible, ideally scaling independently from the length of the string. In addition to improved performance, this method will also increase the clarity for readers of code where this test is being performed — especially those without extensive Unicode or regular expression knowledge.
The proposal also adds a method to ensure that a String is well-formed by replacing any lone or out-of-order surrogates, if any are present, with U+FFFD (REPLACEMENT CHARACTER). This operation mimics the operation already performed within various web and non-ECMAScript APIs, and reduces the chance that a consumer who would otherwise have to write this method does so in an incompatible or incorrect way.
These methods are String.prototype.isWellFormed and String.prototype.toWellFormed, usable for example as in if (!someString.isWellFormed()) { someString = someString.toWellFormed(); }.
The validation algorithm is effectively the standard UTF-16 validation algorithm, iterating through the string and pairing UTF-16 surrogates, failing validation for any unpaired surrogates.
In JavaScript this can be achieved with the regular expression test:
!/\p{Surrogate}/u.test(str);Or more explicitly using an algorithm along the lines of:
function isWellFormed(str) {
for (let i = 0; i < str.length; ++i) {
const isSurrogate = (str.charCodeAt(i) & 0xF800) == 0xD800;
if (!isSurrogate) {
continue;
}
const isLeadingSurrogate = str.charCodeAt(i) < 0xDC00;
if (!isLeadingSurrogate) {
return false; // unpaired trailing surrogate
}
const isFollowedByTrailingSurrogate = i + 1 < str.length && (str.charCodeAt(i + 1) & 0xFC00) == 0xDC00;
if (!isFollowedByTrailingSurrogate) {
return false; // unpaired leading surrogate
}
++i;
}
return true;
}Unfortunately, these user-land implementations require iterating the entire string.
- WebIDL USVString
- Lone surrogate handling discussion on WhatWG/encoding
- Node.js
util.toUSVString
It is possible to answer the question, but it is not possible to do it in sub-linear time.
Performance optimizations are up to implementations and are not guaranteed by this specification, yet enabling a builtin method that is faster than validation in userland is desirable. Perhaps well-formedness state could be cached per string and propagated through string operations to avoid the need for a linear scan where possible, say where strings are already lists of Unicode Scalar Values, while propagating state in operations involving strings containing inner unpaired surrogates will likely still require (re)scans.
Are consumers going to do anything other than convert when they encounter ill-formed strings? If so, why not provide only a conversion method with a fast path for well-formed strings?
Consumers may want to throw/error when encountering ill-formed strings. Also, consumers may want to defer the conversion or the error until later when the String is actually interpreted as Unicode text. These use cases justify the test-only method.