Add a poll surface to web-transport-trait

[kixelated]

Summary

Make web-transport-trait poll-native, so consumers such as moq-net can drive the
transport from their own polling loop without duplicating logic between the poll and
async halves.

One rule throughout: poll_* are the required methods, the async ones are provided
helpers over them. A backend implements the poll surface once.

Both halves split the same way, and the poll traits carry no Send bound:

poll surface	async conveniences
PollSendStream	SendStream: PollSendStream + MaybeSend
PollRecvStream	RecvStream: PollRecvStream + MaybeSend
PollSession	Session: PollSession + MaybeSend + MaybeSync

The poll traits also hold the synchronous methods (finish, reset, stop,
set_priority, send_datagram, close, protocol, max_datagram_size, stats), so
a thread-pinned transport can implement them and be driven from a poll loop without
ever being Send. PollSession's associated types only require the poll halves, so
the Send requirement doesn't leak in through the stream types either — the whole
stack is expressible. Two tests pin that, each implementing the poll surface for a
type holding an Rc.

Backends implement the poll trait and opt in with an empty impl SendStream for MyStream {}. Overriding still works, and is how quinn, noq and iroh keep
write_chunk/read_chunk zero-copy — which is why this is a supertrait rather than a
blanket impl, since a blanket impl forbids those overrides (E0119).

&mut self on sessions

Session operations take &mut self, and Clone is not a supertrait.

&mut self is what makes a retained in-progress operation safe. A &self poll method
holding a retained future would have to either share one slot between concurrent
callers — the second polls the first's future with its own waker, so the first hangs
and the resolved stream goes to the wrong task — or clone the session on every call.
&mut self gives each handle one owner by construction.

Concurrency comes from cloning the concrete session; each clone gets independent poll
state. Dropping the Clone bound means a session that can't be duplicated is
expressible, and an application wanting a freely shared handle opts into wrapping it
rather than every implementation paying for it.

Where a backend already drives an operation from a poll loop it forwards natively:
quinn, quiche, noq and iroh accept through a FuturesUnordered, and qmux polls its
channels directly. The rest have no poll form anywhere, so SessionOps in the trait
crate holds those retained futures — one field per backend instead of six. Those
futures can't borrow the session, so each async body was split into an owned-args form
with the &self method delegating; qmux instead clones a handle, because its bodies
touch too many fields to thread through as arguments and its clone is refcount bumps
only.

Retention has a cost, paid five times

Retaining a future means holding what it owns for longer than the caller intends.
Every bug found in review was a variant of that, and each has a test that fails
without its fix.

• qmux leaked an outbound-queue slot on terminal transitions. reset, finish
  and a peer STOP_SENDING left the reservation behind. The queue has 8 slots.
• qmux held a slot while parked on flow control. The case with no terminal event
  at all: a caller cancels a write, the stream stays open, the slot is gone for its
  lifetime. The slot is now taken only across a synchronous credit claim and handed
  back before waiting on the peer's window, so the retained future holds nothing while
  parked — it retains the waiting, not a resource.
• qmux wedged clones on an abandoned accept. The accept and datagram receivers
  were shared behind a tokio::sync::Mutex, and the retained future held the guard
  across recv().await. Replaced with SharedRecv: a std::sync::Mutex locked for
  the poll itself, plus a waker list. The bad shape is now unrepresentable — a
  std::sync::Mutex guard isn't Send, so the compiler rejects it.
• iroh and quiche lost a concurrent accepter's waker. Every clone polls one shared
  accept struct, which keeps only the most recent waker. quinn and noq already had the
  fix; these never got it. It predates this PR, but telling callers to clone for
  concurrency is what makes it reachable.
• poll_read_buf fed an unchecked size to BufMut::advance_mut. That's unsafe
  while PollRecvStream is not, so a safe implementation could advance past the
  buffer. It now asserts — panic rather than clamp, since an implementation reporting
  more than the buffer it was handed is broken and quietly trusting a smaller number
  would hide it.

Other fixes

• write_chunk's default wrote a single partial write_buf despite promising the
  whole chunk. Now write_all_buf.
• poll_read_buf transmuted an UninitSlice into a &mut [u8] — UB regardless of
  whether the callee reads it. It now zeroes first (overridable to skip the memset),
  and poll_read_chunk reads into a BytesMut::zeroed, dropping its unsafe entirely.
• poll_read_buf/poll_read_chunk reported a full destination as end of stream.
  The backends deliberately distinguish the two (Some(0) for an empty slice); the
  defaults were throwing that away, turning "no room" into apparent truncation.
• Cached stream headers are now web_transport_proto::Header, an inline [u8; 16].
  Two varints is the most a header can be, so it's Copy and the open path no longer
  allocates — cloning a Vec per stream defeated the point of caching the bytes.
• qmux's write/write_buf looped until the buffer drained; they now do a partial
  write like every other backend, which is what the trait has always documented.
  Callers wanting the old behavior use write_all/write_all_buf.
• quinn's and noq's write_buf overrides are gone. They took bytes with
  copy_to_bytes before the write completed — the hazard quinn's own comment warns
  about. The write_chunk overrides, where the caller hands over ownership, stay.

Downstream impact

Built moq-net against this branch with [patch.crates-io]: 227 errors across 12
files. 25 are borrow-structural — sessions moved into spawned 'static tasks — and
the rest are impl-signature updates and cascade. That is a deliberate choice to get
the interface right rather than minimize churn in a consumer that is moving to a
poll-based design anyway.

Breaking

Implementors provide poll_* instead of the async methods, moving type Error and
the synchronous methods to the poll traits, then add an empty impl Session/impl SendStream/impl RecvStream. Session callers hold their session mutably and clone
it for concurrent operations. Callers of the async methods are otherwise unaffected
unless they use the synchronous ones, which need the poll trait imported.

Commits are marked ! with BREAKING CHANGE footers; release-plz owns the version
bump, so no Cargo.toml version is touched here.

Validation

• just fix, just check, just test — all green
• native and wasm32, all-features clippy with -D warnings
• Rust, TypeScript, and Rust↔TypeScript interop tests
• Each qmux fix was verified to fail its test when reverted; the abandoned-slot
  mutation hangs the suite, which is the failure mode it guards

Known gap: the iroh/quiche waker fix has no behavioural test. It mirrors quinn's
proven code and compiles, but a two-cloned-handles test needs a live endpoint per
backend and wasn't built. Of everything here, that change rests on review alone.

(written by Opus 5)

[sourcery-ai]

Sorry @kixelated, you have reached your weekly rate limit of 500000 diff characters.

Please try again later or upgrade to continue using Sourcery

[coderabbitai]

Note

Reviews paused

It looks like this branch is under active development. To avoid overwhelming you with review comments due to an influx of new commits, CodeRabbit has automatically paused this review. You can configure this behavior by changing the reviews.auto_review.auto_pause_after_reviewed_commits setting.

Use the following commands to manage reviews:

• @coderabbitai resume to resume automatic reviews.
• @coderabbitai review to trigger a single review.

Use the checkboxes below for quick actions:

• ▶️ Resume reviews
• 🔍 Trigger review

Walkthrough

The change introduces poll-based session, send-stream, and receive-stream interfaces with retained operation state. QMux stream reservations, reads, writes, terminal handling, and tests are adapted to polling. Iroh, Noq, Quiche, and Quinn integrations implement the new interfaces, while protocol headers use a fixed-size encoded representation. Examples and integration tests now use complete-buffer writes and mutable sessions where required.

🚥 Pre-merge checks | ✅ 5

✅ Passed checks (5 passed)

Check name	Status	Explanation
Docstring Coverage	✅ Passed	Docstring coverage is 100.00% which is sufficient. The required threshold is 80.00%.
Linked Issues check	✅ Passed	Check skipped because no linked issues were found for this pull request.
Out of Scope Changes check	✅ Passed	Check skipped because no linked issues were found for this pull request.
Title check	✅ Passed	The title clearly summarizes the main change: adding a poll-based surface to web-transport-trait.
Description check	✅ Passed	The description is directly aligned with the PR and accurately describes the poll-native trait refactor.

✨ Finishing Touches

✨ Simplify code

• Create PR with simplified code
• Commit simplified code in branch codex/web-transport-poll

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands.}

[sourcery-ai]

Sorry @kixelated, you have reached your weekly rate limit of 500000 diff characters.

Please try again later or upgrade to continue using Sourcery

[coderabbitai]

Actionable comments posted: 4

🤖 Prompt for all review comments with AI agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

Inline comments:
In `@rs/qmux/src/session.rs`:
- Around line 2228-2248: In the pending-write resolution flow, add an early
return after calculating and releasing credit when `to_send` is zero, before
constructing or sending the `Stream` frame. Return `Poll::Ready(Ok(0))` so the
permit is dropped and no empty non-FIN frame is queued; preserve the existing
send and offset-update path for positive lengths.
- Around line 2504-2512: Update poll_closed to preserve existing self.buffer
contents before assigning newly polled stream.data; only replace the buffer when
it is empty, while retaining the current closed-state and polling behavior.

In `@rs/web-transport-trait/README.md`:
- Line 18: Correct the spelling of “underyling” to “underlying” in the README
sentence, leaving the surrounding wording and link unchanged.

In `@rs/web-transport-trait/src/lib.rs`:
- Around line 371-382: Update poll_read_chunk to explicitly limit the writable
destination to max bytes before calling poll_read_buf, rather than relying on
BytesMut::with_capacity(max.min(8 * 1024)). Preserve the existing 8 KiB
allocation cap and return behavior while ensuring the produced Bytes never
exceeds max.

🪄 Autofix (Beta)

Fix all unresolved CodeRabbit comments on this PR:

• Push a commit to this branch (recommended)
• Create a new PR with the fixes

---

ℹ️ Review info

⚙️ Run configuration

Configuration used: Organization UI

Review profile: CHILL

Plan: Pro Plus

Run ID: 8cf104e6-bc5d-4be4-99ed-cccd7f54972a

📥 Commits

Reviewing files that changed from the base of the PR and between d47327f and f07c7b6.

📒 Files selected for processing (21)

• rs/qmux/examples/h3qx.rs
• rs/qmux/src/session.rs
• rs/qmux/tests/backpressure.rs
• rs/qmux/tests/priority.rs
• rs/qmux/tests/qmux01.rs
• rs/qmux/tests/qmux02.rs
• rs/web-transport-iroh/src/recv.rs
• rs/web-transport-iroh/src/send.rs
• rs/web-transport-noq/src/recv.rs
• rs/web-transport-noq/src/send.rs
• rs/web-transport-quiche/src/ez/recv.rs
• rs/web-transport-quiche/src/ez/send.rs
• rs/web-transport-quiche/src/recv.rs
• rs/web-transport-quiche/src/send.rs
• rs/web-transport-quinn/src/recv.rs
• rs/web-transport-quinn/src/send.rs
• rs/web-transport-trait/README.md
• rs/web-transport-trait/src/lib.rs
• rs/web-transport-trait/src/poll.rs
• rs/web-transport-trait/src/util.rs
• rs/web-transport-trait/tests/defaults.rs

[coderabbitai]

Actionable comments posted: 1

Caution

Some comments are outside the diff and can’t be posted inline due to platform limitations.

⚠️ Outside diff range comments (1)

rs/qmux/src/session.rs (1)

1824-1840: 🩺 Stability & Availability | 🟡 Minor | ⚡ Quick win

Avoid holding the accept/datagram mutex inside retained futures.

poll_accept_uni, poll_accept_bi, and poll_recv_datagram clone a Session and retain the resulting future in self.ops; the async body holds self.accept_*/recv_datagram.lock().await across later recv().await polls. If a caller abandons the retained future after acquiring the lock, another Session clone that later calls the same accept/datagram method is blocked for the lifetime of that retained operation. These futures should not own the mutex guard; poll the underlying receiver separately from the retained future path.

🤖 Prompt for AI Agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@rs/qmux/src/session.rs` around lines 1824 - 1840, Refactor accept_uni_inner,
accept_bi_inner, and the corresponding datagram receive helper so their retained
futures do not hold accept_* or recv_datagram mutex guards across recv().await.
Poll or receive from the underlying channels outside the retained Session future
path, ensuring abandoned poll_accept_uni, poll_accept_bi, and poll_recv_datagram
operations release the mutex before another Session clone accesses the same
receiver.

🧹 Nitpick comments (3)

rs/web-transport-quiche/src/connection.rs (1)

56-57: 📐 Maintainability & Code Quality | 🔵 Trivial | 💤 Low value

Stale #[allow(unused)].

header_datagram is read in send_datagram (Line 289) and max_datagram_size (Line 314), so the allow no longer suppresses anything and hides future dead-code signal.

♻️ Proposed change

     header_bi: Header,
-    #[allow(unused)]
     header_datagram: Header,

🤖 Prompt for AI Agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@rs/web-transport-quiche/src/connection.rs` around lines 56 - 57, Remove the
stale #[allow(unused)] attribute from the header_datagram field in the
connection struct. Keep header_datagram unchanged, since it is actively used by
send_datagram and max_datagram_size.

rs/web-transport-trait/src/util.rs (2)

236-245: 📐 Maintainability & Code Quality | 🔵 Trivial | ⚡ Quick win

cleared_future_restarts doesn't assert the restart.

It only checks !is_pending() after clear(). Poll again with a counting make to prove a fresh future is built.

💚 Proposed test

     fn cleared_future_restarts() {
+        let starts = Arc::new(AtomicUsize::new(0));
         let mut state = OpState::default();
         let waker = waker();
         let mut cx = Context::from_waker(&waker);
 
-        assert!(state.poll(&mut cx, pending::<()>).is_pending());
+        for _ in 0..2 {
+            let starts = starts.clone();
+            assert!(state
+                .poll(&mut cx, move || {
+                    starts.fetch_add(1, Ordering::Relaxed);
+                    pending::<()>()
+                })
+                .is_pending());
+            state.clear();
+        }
+
+        assert_eq!(starts.load(Ordering::Relaxed), 2);
         state.clear();
         assert!(!state.is_pending());
     }

🤖 Prompt for AI Agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@rs/web-transport-trait/src/util.rs` around lines 236 - 245, Extend the
cleared_future_restarts test to verify that clearing rebuilds the future: use a
counting make closure with state.poll, call state.clear(), then poll again and
assert the counter increases and the new future’s result is observed. Keep the
existing !state.is_pending() assertion.

---

103-118: 📐 Maintainability & Code Quality | 🔵 Trivial | 💤 Low value

Consider #[non_exhaustive] on SessionOps.

All fields are public and the struct is only ever built via Default. Marking it #[non_exhaustive] lets you add op slots later without a breaking change for out-of-tree backends that might construct it literally.

As per coding guidelines: "Treat public Rust API design as a compatibility commitment".

♻️ Proposed change

+#[non_exhaustive]
 pub struct SessionOps<S, R, E> {

🤖 Prompt for AI Agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@rs/web-transport-trait/src/util.rs` around lines 103 - 118, Mark the public
SessionOps struct as #[non_exhaustive] so future operation slots can be added
without breaking external implementations. Keep its existing public fields and
Default-based construction unchanged.

Source: Coding guidelines

🤖 Prompt for all review comments with AI agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

Inline comments:
In `@rs/web-transport-trait/src/lib.rs`:
- Around line 385-401: Update poll_read_buf and the poll_read API to accept
bytes::buf::UninitSlice directly, removing the transmute from UninitSlice to
[u8] and preserving the existing read-size and advance_mut behavior. Adjust all
implementations and call sites of poll_read to write into the uninitialized
destination without reading from it.

---

Outside diff comments:
In `@rs/qmux/src/session.rs`:
- Around line 1824-1840: Refactor accept_uni_inner, accept_bi_inner, and the
corresponding datagram receive helper so their retained futures do not hold
accept_* or recv_datagram mutex guards across recv().await. Poll or receive from
the underlying channels outside the retained Session future path, ensuring
abandoned poll_accept_uni, poll_accept_bi, and poll_recv_datagram operations
release the mutex before another Session clone accesses the same receiver.

---

Nitpick comments:
In `@rs/web-transport-quiche/src/connection.rs`:
- Around line 56-57: Remove the stale #[allow(unused)] attribute from the
header_datagram field in the connection struct. Keep header_datagram unchanged,
since it is actively used by send_datagram and max_datagram_size.

In `@rs/web-transport-trait/src/util.rs`:
- Around line 236-245: Extend the cleared_future_restarts test to verify that
clearing rebuilds the future: use a counting make closure with state.poll, call
state.clear(), then poll again and assert the counter increases and the new
future’s result is observed. Keep the existing !state.is_pending() assertion.
- Around line 103-118: Mark the public SessionOps struct as #[non_exhaustive] so
future operation slots can be added without breaking external implementations.
Keep its existing public fields and Default-based construction unchanged.

🪄 Autofix (Beta)

Fix all unresolved CodeRabbit comments on this PR:

• Push a commit to this branch (recommended)
• Create a new PR with the fixes

---

ℹ️ Review info

⚙️ Run configuration

Configuration used: Organization UI

Review profile: CHILL

Plan: Pro Plus

Run ID: 3e68ef2e-2d37-4599-8c3a-a215939f32bd

📥 Commits

Reviewing files that changed from the base of the PR and between bd70d43 and 688479e.

📒 Files selected for processing (29)

• rs/qmux/examples/h3qx.rs
• rs/qmux/examples/interop-client.rs
• rs/qmux/examples/interop-server.rs
• rs/qmux/src/session.rs
• rs/qmux/tests/backpressure.rs
• rs/qmux/tests/datagram.rs
• rs/qmux/tests/priority.rs
• rs/qmux/tests/qmux01.rs
• rs/qmux/tests/qmux02.rs
• rs/web-transport-iroh/src/recv.rs
• rs/web-transport-iroh/src/send.rs
• rs/web-transport-iroh/src/session.rs
• rs/web-transport-noq/src/recv.rs
• rs/web-transport-noq/src/send.rs
• rs/web-transport-noq/src/session.rs
• rs/web-transport-proto/src/header.rs
• rs/web-transport-proto/src/lib.rs
• rs/web-transport-quiche/src/connection.rs
• rs/web-transport-quiche/src/ez/recv.rs
• rs/web-transport-quiche/src/ez/send.rs
• rs/web-transport-quiche/src/recv.rs
• rs/web-transport-quiche/src/send.rs
• rs/web-transport-quinn/src/recv.rs
• rs/web-transport-quinn/src/send.rs
• rs/web-transport-quinn/src/session.rs
• rs/web-transport-trait/README.md
• rs/web-transport-trait/src/lib.rs
• rs/web-transport-trait/src/util.rs
• rs/web-transport-trait/tests/defaults.rs

🚧 Files skipped from review as they are similar to previous changes (19)

• rs/qmux/examples/interop-client.rs
• rs/qmux/tests/qmux02.rs
• rs/qmux/examples/h3qx.rs
• rs/qmux/tests/priority.rs
• rs/web-transport-quiche/src/ez/recv.rs
• rs/web-transport-quiche/src/ez/send.rs
• rs/web-transport-trait/README.md
• rs/qmux/tests/qmux01.rs
• rs/qmux/examples/interop-server.rs
• rs/web-transport-trait/tests/defaults.rs
• rs/web-transport-quinn/src/recv.rs
• rs/web-transport-quiche/src/recv.rs
• rs/web-transport-iroh/src/recv.rs
• rs/web-transport-quinn/src/send.rs
• rs/qmux/tests/backpressure.rs
• rs/web-transport-noq/src/recv.rs
• rs/web-transport-iroh/src/send.rs
• rs/web-transport-quiche/src/send.rs
• rs/web-transport-noq/src/send.rs

[kixelated]

Re the outside-diff finding on rs/qmux/src/session.rs (accept/datagram mutex held inside retained futures) — confirmed and fixed in 97b7036.

The diagnosis was right, including that it only became reachable once the future was retained rather than owned by the caller. A handle that started an accept and stopped polling kept the tokio::sync::MutexGuard alive inside ops, and every other clone blocked on the mutex until that handle was polled again or dropped. That's a hang, not a slowdown.

Rather than poll the receiver outside the retained path, qmux now shares it through SharedRecv: a std::sync::Mutex locked only for the poll itself, plus a waker list so concurrent callers don't clobber each other's registration (mpsc::Receiver::poll_recv stores one waker — the same reason quinn's SessionAccept keeps a Vec<Waker>). Accept and datagrams are now poll-native for qmux too, so three of its six retained operations are gone.

The bad shape is also no longer expressible: a std::sync::Mutex guard isn't Send, so holding one across an await in a Send future fails to compile.

Covered by abandoned_accept_does_not_wedge_a_clone in rs/qmux/tests/qmux01.rs, verified to hang against the pre-fix code and pass after.

🤖 Addressed by Claude Code

[coderabbitai]

Actionable comments posted: 2

🧹 Nitpick comments (2)

rs/web-transport-quiche/src/connection.rs (1)

592-602: 📐 Maintainability & Code Quality | 🔵 Trivial | 💤 Low value

Stray comment now attached to the wrong item.

The "This is poll-based because we accept and decode streams in parallel…" note explains poll_accept_uni, but it now sits directly above remember's doc comment. Move it back down to poll_accept_uni.

🤖 Prompt for AI Agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@rs/web-transport-quiche/src/connection.rs` around lines 592 - 602, Move the
three-line poll-based/FuturesUnordered comment from above remember to the
poll_accept_uni method it describes, keeping remember’s documentation limited to
its waker-registration behavior.

rs/web-transport-trait/src/lib.rs (1)

78-81: 📐 Maintainability & Code Quality | 🔵 Trivial | 💤 Low value

Add a doc comment to the exported BiStreams alias.

Every other public item in this module is documented; this alias is part of the public surface (it appears in poll_accept_bi/poll_open_bi signatures).

📝 Proposed doc

+/// The `(send, recv)` pair returned when opening or accepting a bidirectional
+/// stream on a [`PollSession`].
 pub type BiStreams<S> = (
     <S as PollSession>::SendStream,
     <S as PollSession>::RecvStream,
 );

As per coding guidelines: "Document exported Rust and TypeScript APIs".

🤖 Prompt for AI Agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@rs/web-transport-trait/src/lib.rs` around lines 78 - 81, Add a Rust doc
comment directly above the exported BiStreams type alias, describing that it
represents the send and receive streams returned by bidirectional session
operations such as poll_accept_bi and poll_open_bi.

Source: Coding guidelines

🤖 Prompt for all review comments with AI agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

Inline comments:
In `@rs/qmux/tests/qmux01.rs`:
- Around line 326-328: Replace the fixed sleep in the qmux test with explicit
synchronization: after the server’s timeout assertion and once the abandoned
accept is parked, send a readiness signal, then await that signal before the
client calls open_uni. Use the test’s existing task/channel coordination
mechanisms and preserve the test’s focus on the other handle’s behavior.

In `@rs/web-transport-trait/src/lib.rs`:
- Around line 419-438: Update the read flow around the visible buf.chunk_mut()
and len handling to detect len == 0 before invoking poll_read, and return
Poll::Ready(Ok(Some(0))) (or the established full-buffer assertion) instead of
allowing the result to be collapsed into EOF. Preserve the existing behavior for
non-empty buffers.

---

Nitpick comments:
In `@rs/web-transport-quiche/src/connection.rs`:
- Around line 592-602: Move the three-line poll-based/FuturesUnordered comment
from above remember to the poll_accept_uni method it describes, keeping
remember’s documentation limited to its waker-registration behavior.

In `@rs/web-transport-trait/src/lib.rs`:
- Around line 78-81: Add a Rust doc comment directly above the exported
BiStreams type alias, describing that it represents the send and receive streams
returned by bidirectional session operations such as poll_accept_bi and
poll_open_bi.

🪄 Autofix (Beta)

Fix all unresolved CodeRabbit comments on this PR:

• Push a commit to this branch (recommended)
• Create a new PR with the fixes

---

ℹ️ Review info

⚙️ Run configuration

Configuration used: Organization UI

Review profile: CHILL

Plan: Pro Plus

Run ID: 889172f9-a807-4b71-a498-8296a6b365ad

📥 Commits

Reviewing files that changed from the base of the PR and between 688479e and 4b81a5e.

📒 Files selected for processing (18)

• rs/qmux/examples/h3qx.rs
• rs/qmux/examples/interop-client.rs
• rs/qmux/examples/interop-server.rs
• rs/qmux/src/session.rs
• rs/qmux/src/tcp.rs
• rs/qmux/tests/backpressure.rs
• rs/qmux/tests/datagram.rs
• rs/qmux/tests/negotiation.rs
• rs/qmux/tests/qmux01.rs
• rs/qmux/tests/qmux02.rs
• rs/web-transport-iroh/src/session.rs
• rs/web-transport-noq/src/session.rs
• rs/web-transport-quiche/src/connection.rs
• rs/web-transport-quinn/src/session.rs
• rs/web-transport-trait/README.md
• rs/web-transport-trait/src/lib.rs
• rs/web-transport-trait/src/util.rs
• rs/web-transport-trait/tests/defaults.rs

🚧 Files skipped from review as they are similar to previous changes (12)

• rs/qmux/examples/interop-server.rs
• rs/qmux/examples/interop-client.rs
• rs/qmux/tests/backpressure.rs
• rs/qmux/examples/h3qx.rs
• rs/web-transport-trait/tests/defaults.rs
• rs/web-transport-noq/src/session.rs
• rs/web-transport-trait/README.md
• rs/qmux/tests/qmux02.rs
• rs/web-transport-trait/src/util.rs
• rs/web-transport-quinn/src/session.rs
• rs/web-transport-iroh/src/session.rs
• rs/qmux/src/session.rs

[kixelated]

Also fixed the stray-comment nitpick in rs/web-transport-quiche/src/connection.rs — the poll-based/FuturesUnordered note was displaced above remember when I inserted the waker helpers, and is back on poll_accept_uni.

🤖 Addressed by Claude Code