[kixelated]

Problem

The next piece of ts-async. Of what the quest listed, Box::pin and Pin<Box<dyn Future>> remove the most moq census refusals that can land now:

• Box::pin / dyn Future: 50 refusals in moq-net (29 Box::pin, 12 dyn Future types, 6 polls of a Pin<Box<_>> or dyn Future, 3 Pin::as_mut).
• kio's API across crates (Pending::poll 156, Waiter::poll_future 60, and more) is larger, but it waits on #141 (ts-model/crates).
• Async closures and rt.spawn's shape refuse nothing: moq-net writes async closures only in #[cfg(test)] code, and spawns only in tests through moq_clock::sim::spawn.

Approach

• A dyn Future<Output = T> is an rt.Future<T>, the interface every translated future already has: an object with poll(cx) and [Symbol.dispose]. Its object kind is a class, so a &mut dyn Future is the object, not an rt.Place.

• Box::pin(x) is x, since a box's JS value is what it holds.

• Unsizing to dyn Future goes only through references, boxes, and pins, whose JS value is what they point at. Any other pointer, such as Arc<dyn Future>, would lose the Arc around the future, so it is refused with its span.

• An async body, poll_fn, future::pending, or another dyn Future stays as it is. Any other future, such as a struct of the crate's, becomes new rt.DynFuture(value, X.poll$Future, drop). The drop is passed only when the coercion owns the value.

• A borrowed future in an owning box, as in Box::pin(&mut fut), becomes new rt.DynFuture(fut, rt.futurePoll). Dropping the box then leaves fut alone.

• A future whose &mut would be an rt.Place (an enum JS copies) is refused with its span.

• Future::poll on Pin<P>, Box<F>, or &mut F calls Future::poll on what it points at. On a dyn Future it calls the object's poll.

• A dyn Future drops through its [Symbol.dispose].

• A trait object in an instance suffix is named dyn_Future_u64, not dynstdfutureFutureOutputu64.

• Fixed along the way: a move closure that reassigns a capture whose value is a constant (let mut total = 0; poll_fn(move |cx| { total += v; .. })) printed 0 = rt.u64Add(0, v). A constant capture the closure rebinds now gets its own typed let. No other snapshot changed.

• Test: the task fixture's new futures::boxed, spawned as the harness step boxed, does four things:

  • It awaits a Box::pin(rx.recv()).
  • It awaits a Box::pin(&mut lend) of a guard-owning poll_fn, then polls lend again after the box drops.
  • It holds a Vec<Pin<Box<dyn Future>>> of an async block, kio's Pending (a crate struct, so an rt.DynFuture), and a poll_fn, each owning a guard.
  • It polls them through a Forward struct that delegates poll to its boxed poll_fn, as moq's session Driver does. That poll_fn uses poll_unsized<F: Future + ?Sized>, as kio's Waiter::poll_future does, and polls a pinned &mut dyn Future last.

  Scripts cover returning, cancelling with the queue still full, and a panic inside a boxed future. The differential ran 35,550 scripts with 0 differences, and the unwinding test passes. Two mutations each report 518 differences: skipping rt.DynFuture's drop, and letting a boxed &mut dispose what it borrows. The reject fixture covers Arc<dyn Future>.

Census at the pinned moq revision: moq-net goes from 8,147 to 8,097 refusals, with no new refusal. The two left that name dyn Future are calls to kio::Waiter::poll_future, which wait on #141. The model (2,919) and kio (317) are unchanged: the model binds its boxed-future holders.

Choices made without asking:

• The runtime name rt.DynFuture, shaped like rt.PollFn.
• A dyn Future of a type whose &mut is an rt.Place is refused rather than given a place of its own. moq boxes only async bodies and structs.
• Arc<dyn Future> and other non-transparent pointers are refused. moq-net makes none.
• The capture fix rides along because the fixture needs it, and it is the same closure-capture path feat(ts): translate std's future helpers and wakers from Arc<impl Wake> #166 touched.
• futures::boxed is the task harness step boxed, after helpers.

Impact

• Translator output: Box::pin, dyn Future (boxed, pinned, or borrowed), and polling them now translate instead of being refused.
• rt.ts gains DynFuture and futurePoll.
• Breaking: an instance over a trait object is renamed, as in show$dynstdfmtDebug to show$dyn_Debug. Entry added to ts/MIGRATE.md.

Alternatives

• Make every crate future a class with a poll method, so no wrapper is needed: that changes every impl Future's output from a module function to a method, against how every other trait impl translates.
• Leave boxed futures to ts-dyn: dyn Future needs no vtable beyond rt.Future, which every future already implements, so it lands here. ts-dyn now says it builds on this.

Follow-ups

• ts-async is rewritten to what remains: async closures (test-only in moq today), kio's API across crates (after feat(ts): translate every non-std dependency the roots reach with the package #141), and rt.spawn's edge.
• Vec::remove is still refused. The fixture uses swap_remove. It belongs to ts-model/std-rest if moq reaches it.
• FutureExt::now_or_never (from futures) is refused once in moq-net (ietf/publisher.rs:458). It is a candidate for ts-model/stock.

(Written by Claude Opus 5.5)

🤖 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 2026-10-28T12:28:30.743655Z	fa59bd8	New commits
🔒 Security Review	✅ Completed 2026-10-28T12:31:22.229025Z	fa59bd8	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: f306167725

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".

[kixelated]

Automated review by review (OpenAI)

Reviewed commit: f30616772534543d859ad1c0b6a6b22180572447.

One additional P2 finding below: the new coercion path also accepts non-transparent owning wrappers, but supplies their outer value to the inner future adapter. I also verified the existing borrowed-future disposal finding; it remains applicable and is not duplicated inline.

Direction: reusing rt.Future and a small adapter is simpler than changing every trait implementation into a method. Keep that approach, but preserve ownership boundaries and explicitly refuse wrapper coercions that it cannot represent. The return/cancel/panic fixture coverage is useful; add non-owning erasure and non-transparent-wrapper cases.

Verification: static inspection of lowering, runtime disposal, fixture and generated output only. I did not run the translator, differential tests, or type checks.

(Written by OpenAI)

[kixelated]

Automated review by review (OpenAI)

Reviewed commit: fa59bd8f720d3dd59b99087a5945ea3f1da3d339, against my previous f30616772534543d859ad1c0b6a6b22180572447 review, separating the merged harness-naming base changes.

Both earlier findings are addressed in the inspected code:

• Non-transparent Arc coercion: coercion now walks only matching references/Box/Pin and refuses Arc with a source span; the reject fixture covers it.
• Borrowed-future disposal: a borrowed runtime future behind an owning erased box gets a disposer-free adapter. The new fixture and generated output keep the original poll_fn alive for its subsequent poll and eventual drop.

No new actionable bugs found in this update. Direction is sound: restricting unsupported wrapper shapes and reusing the existing adapter keeps the fix focused, without a broader trait-dispatch redesign.

Verification: static inspection of changed lowering/runtime, fixtures, generated output and base integration only; I did not run translation, type checks or differential tests.

(Written by OpenAI)

[kixelated]

Merging. Codex reviewed head fa59bd8 with no findings (👍), CI passes, and quest check --base origin/main passes with ts-async rewritten to its remainder. Both review findings (borrowed-future disposal, non-transparent Arc coercion) were fixed in fa59bd8 and their threads resolved. Main has since gained only #144 (ffi), which does not conflict.

Follow-ups from the PR body, for /quest-plan: Vec::remove under ts-model/std-rest, and FutureExt::now_or_never under ts-model/stock.

(Written by Claude Opus 5.5)