super.<prop> value-form lowering silently returns child override instead of parent #774

Background

PR #754 lowered value-form super.<prop> (introduced for the NestJS smoke fixture's rxjs OperatorSubscriber pattern) by rewriting it to this.<prop> / this[expr] in crates/perry-hir/src/lower/expr_misc.rs. The inline comment is honest about the scope:

The substitution is correct when the child does not override the property (the dominant rxjs / NestJS pattern). When the child does override, this approximation will resolve to the override rather than the parent — a TODO for a future explicit super-vtable path in codegen.

This is a silent semantic regression for any user code that overrides a parent property and reaches up via super.<prop> to read the parent's value. There is no compile error, no runtime warning — the program just returns the wrong value.

Repro shape

class A {
    foo = "A";
}
class B extends A {
    foo = "B";
    parentFoo() {
        return super.foo; // strict JS: "A"; perry today: "B"
    }
}
console.log(new B().parentFoo());

Expected (node --experimental-strip-types): A
Actual (perry HEAD): B

Why this matters

The PR #754 review accepted the lowering for the merge because the rxjs/NestJS pattern (no override) is the dominant case and the alternative was a hard lower_bail!. That tradeoff is fine, but it leaves user code that does override silently broken.

Suggested fix

Carry an explicit super-vtable lookup through codegen so super.<prop> resolves through the parent's prototype regardless of any child override. The HIR variant probably wants to be an explicit Expr::SuperPropertyGet { property } rather than the current this.<prop> rewrite, so codegen can route it to the parent class's vtable.

Regression net

A parity test that pins the wrong behavior would let a future fix flip it to passing. Suggested location: test-files/test_super_property_override.ts with the snippet above.

Related

• PR Add legacy TypeScript decorator metadata support #754 audit
• crates/perry-hir/src/lower/expr_misc.rs (substitution site + the explanatory comment)