[Brooooooklyn]

Summary

Closes #505.

Since oxc 0.155 (#589), string values are JSStr, which has no &str form for a lone surrogate. So on main:

@Directive({ selector: '[d]' })
export class D {
  @Input() 'x-\uD800': any;          // dropped from `inputs`, no error
  @Input('z-\uDC00') z: any;         // error: "must resolve to a string ..."
  b = input('', { alias: 'b-\uD800' }); // error: "Alias needs to be a string ..."
}

ngtsc compiles all of them. I checked this by running ngc 22.2.3 (full and partial) and the linker on these cases. It prints each lone code unit in a generated string as an uppercase \uXXXX:

inputs: { "x-\uD800": "x-\uD800", z: [0, "z-\uDC00", "z"], b: [1, "b-\uD800", "b"] }

This PR emits the same, in full, partial and JIT output, in setClassMetadata and in the .d.ts.

How

A Rust &str can't hold a lone surrogate, so util/lone_surrogate.rs stores each lone code unit 0xD800 + k as two characters: the noncharacter U+FDD0 (which Unicode reserves for a program's internal use), then U+10F800 + k. Only that pair is a lone surrogate. Either character on its own, as real text, is printed as itself.

read   'x-\uD800'  ->  "x-" U+FDD0 U+10F800          member names, aliases, signal API
                                                     aliases, evaluated metadata strings
emit   U+FDD0 U+10F800+k  ->  \uD800+k               emitter escape_string, dts
                                                     escape_dts_string, dts quote,
                                                     JIT escape_js_string

Piece	Change
util/lone_surrogate.rs	Stores and reads back lone surrogates.
directive/evaluator.rs	A lone-surrogate string evaluates to its string (was Dynamic). Transform types and the ngAcceptInputType_ clash check find such a member.
directive/property_decorators.rs, class_metadata/builders.rs	Member names, @Input/input()/model()/output() aliases and propDecorators keys keep it. A string query predicate takes the same path as any string, so lone_surrogate_selectors goes.
directive/decorator.rs	Drops the alias errors that existed only because the value couldn't be read. A selector holding a lone surrogate keeps main's error, because the selector parser can't read it.
component/transform.rs	JIT member names and aliases keep it. The error plumbing #589 added for this case goes. The __decorate member name is now escaped.
output/emitter.rs, dts.rs, directive/dts_type.rs	Print it as \uXXXX.

Where this differs from ngtsc

ngtsc writes invalid output in two places. This PR writes valid output there instead:

• @ViewChild('r') 'q-\uD800': ngtsc writes ctx.q-<U+FFFD> = .... This PR writes ctx["q-\uD800"] = ....
• An ngAcceptInputType_ member with a lone surrogate: ngtsc writes it unquoted, as U+FFFD. This PR quotes it.

Side effects of reading these strings

• styles and host attribute values holding a lone surrogate now compile like ngtsc. On main they were errors.
• A decorator query predicate is pooled into a const like any other (const _c0 = ["\uD800"]), as ngtsc does.

Not in this PR (same on main)

• @HostBinding/@HostListener arguments, host keys, hostDirectives mappings, exportAs, @Pipe({name}) and @Attribute(...) holding a lone surrogate.
• An inline template holding a lone surrogate. The component is silently left uncompiled.
• Separate crash: the CSS selector parser panics on any non-ASCII selector (selector: '[ké]' panics at pipeline/selector.rs:240).
• Diagnostics that quote such a name print the stored characters rather than \uXXXX.

Tests

tests/lone_surrogate_names_test.rs has 12 cases, with expected fragments taken from ngc's output. All 12 fail on main, except the two that pin main behaviour (the selector error, and private-use characters printed as themselves). The cases:

• decorator inputs and outputs with .d.ts;
• partial output;
• signal APIs;
• the inputs:/outputs: arrays;
• lowercase escapes, and a lone unit at the start or end of a name;
• JIT;
• host values;
• the selector error;
• real U+10F800 and U+FDD0 characters printed as themselves;
• quoted ngAcceptInputType_;
• keyed query writes;
• the clash error.

Three existing tests asserted the old behaviour (an error, or an inline predicate). They now assert ngtsc's output.

cargo fmt --all -- --check, cargo test --all-features (3340 passed) and cargo run -p oxc_angular_conformance (1264/1264, no snapshot diff) pass.

🤖 Generated with Claude Code

[chatgpt-codex-connector]

Codex Review Summary

This comment shows the latest Codex review activity on this pull request.

Review	Status	Commit	Review trigger
📝 Code Review	✅ Completed 2026-10-21T05:58:16.212859Z	c6da2db	New commits

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review" or "@codex security review".

Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings.

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: e5869b8826

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".