class X extends PassThrough never installs the subclass's _transform override, for any heritage shape — including a plain import { PassThrough } from "node:stream". This is a deeper gap than the bound-export heritage problem PR #10649 fixes for Readable/Writable/Duplex/Transform, and is deliberately out of scope there.
Why it is different from the sibling shapes
canonical_native_parent_name (crates/perry-hir/src/lower_decl/class_decl.rs) lists Readable, Writable, Duplex and Transform as node:stream native parents. It does not list PassThrough. So HIR never recognises a PassThrough heritage as a native stream parent at all, and the hidden _transform field that the runtime shim (js_node_stream_transform_subclass_init) reads is never pre-seeded — for any spelling of the heritage, static or dynamic.
That means the fix shape used for the other four does not transfer. This was established empirically while writing #10649, not inferred: adding a PassThrough arm to js_fetch_or_value_super alone was confirmed to change nothing. Recording that here is the main reason this issue exists — it is exactly the finding that saves the next person an afternoon, and it currently lives only in a code comment inside #10649's diff, where it will be met as a surprise rather than as a known gap.
Expected vs actual
import { PassThrough } from "node:stream";
class Tap extends PassThrough {
_transform(chunk, _enc, cb) { cb(null, String(chunk).toUpperCase()); }
}
const t = new Tap();
t.on("data", (c) => console.log("got:", String(c)));
t.write("abc");
- Node 26.5.1:
got: ABC
- Perry: the override is not installed;
write() does not route through _transform.
Fix direction
The gap is at the HIR level, so the fix belongs with canonical_native_parent_name and whatever pre-seeds the hidden field — not with an additional runtime dispatch arm. Worth checking at the same time whether PassThrough needs treating as its own base or as a Transform subclass, since it is one in Node, and whether the pre-seeding can be derived from that relationship rather than enumerated a fifth time.
Related
class X extends PassThroughnever installs the subclass's_transformoverride, for any heritage shape — including a plainimport { PassThrough } from "node:stream". This is a deeper gap than the bound-export heritage problem PR #10649 fixes forReadable/Writable/Duplex/Transform, and is deliberately out of scope there.Why it is different from the sibling shapes
canonical_native_parent_name(crates/perry-hir/src/lower_decl/class_decl.rs) listsReadable,Writable,DuplexandTransformas node:stream native parents. It does not listPassThrough. So HIR never recognises aPassThroughheritage as a native stream parent at all, and the hidden_transformfield that the runtime shim (js_node_stream_transform_subclass_init) reads is never pre-seeded — for any spelling of the heritage, static or dynamic.That means the fix shape used for the other four does not transfer. This was established empirically while writing #10649, not inferred: adding a
PassThrougharm tojs_fetch_or_value_superalone was confirmed to change nothing. Recording that here is the main reason this issue exists — it is exactly the finding that saves the next person an afternoon, and it currently lives only in a code comment inside #10649's diff, where it will be met as a surprise rather than as a known gap.Expected vs actual
got: ABCwrite()does not route through_transform.Fix direction
The gap is at the HIR level, so the fix belongs with
canonical_native_parent_nameand whatever pre-seeds the hidden field — not with an additional runtime dispatch arm. Worth checking at the same time whetherPassThroughneeds treating as its own base or as aTransformsubclass, since it is one in Node, and whether the pre-seeding can be derived from that relationship rather than enumerated a fifth time.Related
extends stream.Stream(the bareStreambase, not a subclass name) is also outside fix(runtime): dispatch node:stream super() through any bound-export heritage shape #10649's match list and may want checking as part of the same pass.