Sunbelt Computer Software

JsString is a string type that can represent JavaScript strings containing lone surrogates (U+D800-U+DFFF), which Rust's String (UTF-8) cannot. Internal representation uses an enum with a fast path: - Utf8(String): for the 99.99% of strings with no lone surrogates - Wtf8(Vec<u8>): WTF-8 encoded bytes when lone surrogates are present Custom serde support decodes __SURROGATE_XXXX__ markers (from bridge.ts) into WTF-8 lone surrogate bytes on deserialization, and re-encodes them as markers on serialization. Provides: Eq, Hash, Clone, Debug, Display, From<String>, From<&str>, utf16_len() for JS .length semantics, push_js_string() for concat, as_str() for zero-cost UTF-8 access, to_utf8_lossy() for display. 21 unit tests covering: serde round-trip, encoding/decoding, equality, concatenation, utf16_len for BMP/supplementary/surrogate, sentinel collision resistance. This is Phase 1 of the WTF-8 migration. Subsequent phases will replace String with JsString in AST fields, HIR, and optimization passes.

Change 5 AST fields from String to JsString: - StringLiteral.value - DirectiveLiteral.value - JSXText.value - TemplateElementValue.raw - TemplateElementValue.cooked These are the entry/exit points for JavaScript string content in the compiler. Downstream crates (lowering, optimization, codegen) will need corresponding changes in Phase 3.

…codegen, and optimization Replace String with JsString across the entire compiler pipeline: HIR: PrimitiveValue::String(JsString), PropertyLiteral::String(JsString), InstructionValue::JSXText { value: JsString }, TemplateQuasi { raw/cooked: JsString }, HirFunction.directives: Vec<JsString> Lowering: All StringLiteral→Primitive, JSXText, TemplateQuasi, ObjectPropertyKey, and directive extraction sites now use JsString. Codegen: All Primitive→StringLiteral, JSXText, TemplateQuasi, ObjectPropertyKey, directive, and string literal construction sites emit JsString. Optimization: String concat uses push_js_string(), .length uses utf16_len() (correctly returns 1 for lone surrogates instead of 18 for marker text), identifier checks use as_str() fast path. Debug printer: format_js_string() iterates WTF-8 bytes directly and emits \uXXXX for lone surrogates, matching TS's JSON.stringify output exactly. All tests pass: cargo build: clean cargo test: 30/30 (AST crate) yarn snap --rust: 1804/1804 yarn snap: 1804/1804 test-rust-port.sh: 1803/1803

Replace the hardcoded `__SURROGATE_` marker with a dynamically chosen prefix that's guaranteed absent from the source JSON. bridge.ts's chooseSurrogateMarker() starts with `__SURROGATE_` and prepends `__ESC_` until a collision-free prefix is found. The chosen marker is passed to Rust via `__surrogateMarker` in the options JSON. Rust side: set_surrogate_marker() stores the prefix in a thread-local. JsString's decode_markers/encode_markers read it dynamically instead of using the hardcoded prefix. This eliminates the sentinel collision bug where source code containing the literal text `__SURROGATE_D83E__` would have been corrupted. Verified: a fixture with literal `__SURROGATE_D83E__` in a string attribute survives compilation unchanged. All tests pass: cargo build + test: clean, 30/30 yarn snap --rust: 1804/1804 test-rust-port.sh: 1803/1803

Mechanical type changes in the SWC and OXC frontend crates to work with JsString instead of String for AST string fields (StringLiteral.value, DirectiveLiteral.value, JSXText.value, TemplateElementValue.raw/cooked). - convert_ast.rs: wrap String values in JsString::from() when constructing AST nodes - convert_ast_reverse.rs: use .to_string()/.as_str_unwrap() when extracting JsString values 63 error sites fixed across 4 files.

…o surrogates present When bridge.ts doesn't encode any surrogates (99.99% of compilations), pass __surrogateMarker: null instead of a marker string. Rust's JsString deserialization checks the thread-local marker: if None, it skips all string scanning and stores every string as the Utf8 fast-path variant. This eliminates the overhead of searching every string for the marker prefix during deserialization for the common case where no lone surrogates exist in the source. Changes: - bridge.ts: sanitizeJsonSurrogates returns {json, hadSurrogates}; passes null marker when no surrogates were found - js_string.rs: SURROGATE_MARKER thread-local changed from String to Option<String>; decode_markers returns None immediately when marker is None - lib.rs (NAPI): passes Option<&str> to set_surrogate_marker All tests pass: cargo test 105/105, yarn snap --rust 1804/1804.

Sunbelt Computer Software

PL/B Language Development and Support

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

Comparing changes

Open a pull request

Uh oh!

Commits on Jun 11, 2026

This comparison is taking too long to generate.

Uh oh!