Discuss
  • String.fromCodePoint and surrogate pairs?

String.fromCodePoint and surrogate pairs?

# Erik Arvidsson (12 years ago)

It was suggested to me that we could probably extend String.fromCodePoint to be aware of UTF-16 code units too. It seems doable since the lead surrogate is not a valid code point.

The question is if it is worth it? It seems like we are going down a slippery slope if we start to do things like this. Should we also handle UTF-8 code units. Maybe it is better not to do this and try to get people to move away from UTF-16 code units and move them towards code points.

# Norbert Lindenberg (12 years ago)

Do you know what the people who talked to you mean by "aware of UTF-16 code units"?

As specified, String.fromCodePoint, accepts all UTF-16 code units because they use a subset of the integers allowed as code points (0 to 0xFFFF versus 0 to 0x10FFFF). For non-surrogate values, you get exactly what you expect. Surrogate values are interpreted as surrogate code points, which are valid code points in Unicode (their use makes a string ill-formed in Unicode terminology, but the proposed ECMAScript spec ignores issues of well-formedness for compatibility with ES5). Since in conversion to UTF-16 a surrogate code point just becomes the corresponding code unit, it can happen that two surrogate code points (an ill-formed sequence) become a well-formed surrogate pair: String.fromCodePoint(0xD83D, 0xDE04) => "\uD83D\uDE04" = "😄".

The story for UTF-8 is very different: Of course all UTF-8 code units would be accepted by String.fromCodePoint, but they would turn into a completely different character sequence. E.g., the UTF-8 byte sequence for 😄: String.fromCodePoint(0xF0, 0x9F, 0x98, 0x84) => "\u00F0\u009F\u0098\u0084" = "ð\u009F\u0098\u0084" (the last three are control characters).

Handling UTF-8 would require a way to identify the character encoding to convert from, which indicates the beginning of an encoding conversion API, and the internationalization ad-hoc decided not to work on one within ECMAScript. There is an API being defined as part of the encoding standard project at WhatWG.

Norbert

# Shawn Steele (12 years ago)

IMO String.fromCodePoint should disallow U+D800-U+DFFF.

There's already fromCharCode that does that, and a according to The Unicode Standard, isolated surrogates have no meaning

# Norbert Lindenberg (12 years ago)

The Unicode standard defines "code point" as any value in the range of integers from 0 to 0x10FFFF - see definitions D9 and D10 of chapter 3 [1].

Once you exclude surrogate code points, you have Unicode scalar values (definition D76), so you're basically proposing a String.fromScalarValue function. But then, why not also exclude code points that Unicode has defined as non-characters (chapter 16.7 [2])? It seems we're getting into policy-setting here, and so far ECMAScript has avoided setting policy for how you can use strings.

Norbert

[1] www.unicode.org/versions/Unicode6.2.0/ch03.pdf [2] www.unicode.org/versions/Unicode6.2.0/ch16.pdf

# Shawn Steele (12 years ago)

I was looking at D75 of 3.8 "Surrogates"

My point is that there's no "legal" scenario for converting basically a UTF-32 input to an isolated surrogate pair. No valid Unicode string could contain that. So why support it?

# Norbert Lindenberg (12 years ago)

I don't have a good scenario at hand either that would require support for surrogate code points, but in ECMAScript the question is often asked the other way around: Why reject it? And given that there are several ways already to construct strings that are ill-formed UTF-16 (e.g., "\uD800", String.fromCodeUnit(0xD800)), it's not clear why this particular path should be blocked.

(Sorry for letting this sit in my outbox for such a long time.)

Norbert

# Shawn Steele (12 years ago)

It doesn't make sense and is illegal unicode. Eg: it's corrupt data. So the only reason to accept it is to allow corrupt data, perhaps as a way of faking other non-Unicode data as a Unicode context. Which inevitably leads to problems, particularly on the web where people do whatever sneaky things the developer thinks works.

Assuming a use case for illegal unicode were ever found, it could be added later.

# Mark Davis ☕ (12 years ago)

There is a long discussion of this on the unicode list recently. A surrogate code point is not illegal Unicode. It is illegal in a UTF string, but is not illegal in a Unicode String ( www.unicode.org/glossary/#unicode_string)

I don't want to repeat that whole long discussion here.

Mark plus.google.com/114199149796022210033 * * — Il meglio è l’inimico del bene — **