Add --strictObjectIterables (bikeshed): exclude primitive (string) from Iterable types

[petamoriken]

🔍 Search Terms

"iterable object", "string"

✅ Viability Checklist

• This wouldn't be a breaking change in existing TypeScript/JavaScript code
• This wouldn't change the runtime behavior of existing JavaScript code
• This could be implemented without emitting different JS based on the types of the expressions
• This isn't a runtime feature (e.g. library functionality, non-ECMAScript syntax with JavaScript output, new syntax sugar for JS, etc.)
• This isn't a request to add a new utility type: https://github.com/microsoft/TypeScript/wiki/No-New-Utility-Types
• This feature would agree with the rest of our Design Goals: https://github.com/Microsoft/TypeScript/wiki/TypeScript-Design-Goals

⭐ Suggestion

At the TC39 meeting in 2024-07, it was decided that Iterable expects objects.

Reject primitives in iterable-taking positions

Any time an iterable or async-iterable value (a value that has a Symbol.iterator or Symbol.asyncIterator method) is expected, primitives should be treated as if they were not iterable. Usually, this will mean throwing a TypeError. If the user provides a primitive wrapper Object such as a String Object, however, it should be treated like any other Object.

Although primitive Strings are default iterable (String.prototype has a Symbol.iterator method which enumerates code points), it is now considered a mistake to iterate a String without specifying whether the String is providing an abstraction over code units, code points, grapheme clusters, or something else.

NB: This convention is new as of 2024, and most earlier parts of the language do not follow it. In particular, positional destructuring (both binding and assignment), array spread, argument spread, for-of loops, yield *, the Set and AggregateError constructors, Object.groupBy, Map.groupBy, Promise.all, Promise.allSettled, Promise.any, Promise.race, Array.from, the static from methods on typed array constructors, and Iterator.from (Stage 3 at time of writing) all accept primitives where iterables are expected.

https://github.com/tc39/how-we-work/blob/main/normative-conventions.md#reject-primitives-in-iterable-taking-positions

To follow this decision, the Iterable and string types should be separated, and APIs that accept both types should be required to explicitly specify Iterable<any> | string. Since this would be a breaking change, how about adding a new --strictObjectIterables (bikeshed) option?

In practice, the upcoming ReadableStream.from will accept Iterable<any> and AsyncIterable<any>, but will be restricted to objects. whatwg/streams#1310

📃 Motivating Example

Enabling the --strictObjectIterables option raises a type error in the following example:

const iterable: Iterable<any> = "string";

💻 Use Cases

1. What do you want to use this for?

   Used for ReadableStream.from and other APIs to be added in the future.

2. What shortcomings exist with current approaches?

   Iterable object and string are not distinguished by default.

3. What workarounds are you using in the meantime?

   Maybe Iterable<any> & object can reject string.

[nmain]

"Old" functions such as Array.from will still accept primitive iterables at runtime, and still do something meaningful with them. IMO, those should be typed differently than any new function that rejects primitive iterables.

[RyanCavanaugh]

The results at #59683 outline the difficulty of this. Looking at the code, we see a lot of places where people write code that takes an implicit dependency on the iterability of strings, e.g. this kind of code is idiomatic JS and looks to be 100% correct:

// https://github.com/refined-github/refined-github/blob/6e167b9d881bd05db8cb92a5e140144cda7fd873/source/features/quick-review.tsx#L18C1-L18C54
const emojis = [...'🚀🐿️⚡️🤌🥳🥰🤩🥸😎🤯🚢🛫🏳️🏁'];

There's just a ton of code out there that takes inputs like URLs, hashes, file paths, programmer-authored strings, etc, and iterates it in a way that can't cause problems.

We also see places where people forget that strings are iterable and end up passing e.g. "foo" into a context where it will be interpreted as "f", "o", "o", which is a clear error:

// https://github.com/mui/material-ui/blob/8ab2137281267ad4fa766e4bd64d03baa5d44820/docs/data/material/getting-started/templates/dashboard/theme/themePrimitives.ts#L227
  shadows: [
    'none',
    ...(mode === 'dark'
      ? 'hsla(220, 30%, 5%, 0.7) 0px 4px 16px 0px, hsla(220, 25%, 10%, 0.8) 0px 8px 16px -5px'
      : 'hsla(220, 30%, 5%, 0.07) 0px 4px 16px 0px, hsla(220, 25%, 10%, 0.07) 0px 8px 16px -5px'),
  ] as Shadows

But both of these are just the same thing: spreading a string. The same thing happened for implicit Symbol.iterable calls. There doesn't seem to be a good way to tell, as a machine, if someone is iterating a string on purpose or not.

Worse, the iterator capability of string is the only way to do certain "correct" Unicode operations, e.g.:

console.log("👪".length); // 2, probably expected 1
console.log([..."👪"].length); // single codepoint, ok-ish

and we did see code that intentionally uses this behavior.

We'd need some way to tell apart intentional from non-intentional iteration to proceed here, and one doesn't seem to exist in the type system. Open to ideas, though.

[MichaelMitchell-at]

Worse, the iterator capability of string is the only way to do certain "correct" Unicode operations, e.g.:

Similar to what @nmain mentioned above perhaps Array.from could be given an overload to explicitly accept a string as an argument and it becomes the only blessed way to do things like Array.from("👪") rather than [..."👪"].

[Ginden]

Maybe add builtin type:

// Like `Iterable<T, TReturn, TNext>`, but excludes strings
type StrictIterable<T, TReturn = any, TNext = any> = intrinsic;