• fix(transformer/async-to-generator): correct scope of inferred named FE in async-to-generator #21458 👈 (View in Graphite)
• main

How to use the Graphite Merge Queue

Add either label to this PR to merge it via the merge queue:

• 0-merge - adds this PR to the back of the merge queue
• hotfix - for urgent changes, fast-track this PR to the front of the merge queue

You must have a Graphite account in order to use the merge queue. Sign up using this link.

This stack of pull requests is managed by Graphite. Learn more about stacking.

Merging this PR will not alter performance

✅ 44 untouched benchmarks
⏩ 7 skipped benchmarks¹

_{Comparing fix/async-to-generator-inferred-name-scope (c5d2ef1) with main (b3ed467)²}

Verification against Rolldown's pipeline

I traced through rolldown/crates/rolldown/src/utils/pre_process_ecma_ast.rs (where Rolldown runs oxc's transformer) and reproduced the exact pipeline to confirm this fix propagates correctly downstream.

How Rolldown uses oxc

// Step 3: Transform (consumes initial scoping, drops the returned scoping)
let scoping = self.recreate_scoping(&mut scoping, program);
let _ret = Transformer::new(...).build_with_scoping(scoping, program);

// Step 5: Tree-shake (rebuilds scoping from AST before DCE)
let scoping = self.recreate_scoping(&mut scoping, program);  // ← SemanticBuilder::new().build(program)
Compressor::new(allocator).dead_code_elimination_with_scoping(program, scoping, options);

Between Step 3 and Step 5, Rolldown discards the transformer-returned scoping and calls SemanticBuilder::new().build(program) — rebuilding scoping from scratch by walking the post-transform AST.

The rebuild inspects func.r#type (via func.is_declaration() / func.is_expression() in oxc_semantic::builder::visit_function):

• is_declaration() → binds name in parent scope (before enter_scope).
• is_expression() → binds name in own scope (after enter_scope).

Pre-fix, oxc's transformer set r#type = FunctionDeclaration on the inner caller function even though it's wrapped in Expression::FunctionExpression(...). Rolldown's rebuild honors r#type, binding the name in the wrapper IIFE scope, causing DCE to (correctly per its view) inline the "unused" outer const.

End-to-end reproduction

I replicated Rolldown's exact pipeline (parse → initial SemanticBuilder → Transformer → fresh SemanticBuilder rebuild → Compressor with TreeShakeOptions) and ran the issue's input through both pre-fix and post-fix:

Pre-fix pipeline output:

import _asyncToGenerator from "@oxc-project/runtime/helpers/asyncToGenerator";
(function() {
    var _ref = _asyncToGenerator(function* () {
        yield new Promise((resolve) => setTimeout(resolve, 1e3));
        console.log("typeNext");
        typeNext();            // ReferenceError at runtime
    });
    return function typeNext() {
        return _ref.apply(this, arguments);
    };
})()();                         // outer const DCE'd away, IIFE double-called

Running in Node: ReferenceError: typeNext is not defined ← matches the Vite issue exactly.

Post-fix pipeline output:

import _asyncToGenerator from "@oxc-project/runtime/helpers/asyncToGenerator";
const typeNext = function() {  // outer const preserved
    var _ref = _asyncToGenerator(function* () {
        yield new Promise((resolve) => setTimeout(resolve, 1e3));
        console.log("typeNext");
        typeNext();             // correctly resolves to outer
    });
    return function typeNext() {
        return _ref.apply(this, arguments);
    };
}();
typeNext();

Running in Node: works correctly.

Summary

• No changes needed in Rolldown. The fix is purely in oxc's transformer emitting correct r#type and scope placement.
• Every tool that runs SemanticBuilder on oxc's post-transform AST benefits — including oxc's own minifier.
• The PR's transform_checker "Bindings mismatch" errors (pre-fix, now gone post-fix) are the exact symptom of what Rolldown's rebuild saw. Those tests are a direct proxy for Rolldown's behavior.

Finally, I figured out why no output changes could resolve the issue. An incorrect func.r#type causes semantic rebuilding in a different way, which leads to this bug.

Ah, sorry, I noticed that I incorrectly minimized the reproduction in #21125

I wonder if it makes sense to add debug mode in Rolldown that verifies whether generating semantic data from the intermediate code outputs the same semantic data.

Similar to what the transform_conformance test does.

I wonder if it makes sense to add debug mode in Rolldown that verifies whether generating semantic data from the intermediate code outputs the same semantic data.

I will take a look at how to achieve this

Merge activity

• Apr 15, 2:05 PM UTC: The merge label '0-merge' was detected. This PR will be added to the Graphite merge queue once it meets the requirements.
• Apr 15, 2:05 PM UTC: Dunqing added this pull request to the Graphite merge queue.
• Apr 15, 2:11 PM UTC: Merged by the Graphite merge queue.