test(moq-net): cover the announce loop's demand/linger state machine

[kixelated]

What

Adds direct regression coverage for the demand/linger state machine in run_announce (rs/moq-net/src/lite/publisher.rs), introduced by #2424 and now on main. Previously only the lower-level pieces were tested (broadcast::Demand edges, the origin reselect gate, the RouteCost codec); the announce loop's hysteresis itself (Op::Idle, Op::Linger, COST_LINGER, the fused linger sleep, idle_at cleared on Op::Route) had none. Flagged by CodeRabbit on #2424.

Tests only, no behavior change, so this targets main.

How

A minimal in-memory harness: run_announce runs against a sink session whose send stream captures bytes (decoded back into AnnounceOk / AnnounceBroadcast messages via a cursor, so each step asserts exactly the wire messages it caused) and whose recv stream stays pending forever. Tokio time is paused, so the 5s linger is deterministic; under start_paused the 1ms settle() is a true quiescence barrier, since auto-advance only fires once the runtime is idle.

Six tests:

• drain_defers_the_cold_restore: draining demand while zero is advertised sends nothing, and stays quiet inside the window.
• demand_return_cancels_the_restore: demand returning cancels the restore, and the next drain starts a fresh window rather than inheriting the old deadline.
• linger_expiry_restores_the_cold_cost: an expired linger sends exactly one ANNOUNCE_RESTART with the cold cost, then settles.
• route_change_supersedes_the_linger: a route change mid-linger restarts with the new chain and clears the pending restore.
• staggered_lingers_restore_independently: two broadcasts with staggered drains each restore on their own deadline.
• demand_reprices_warm_immediately: the mirror edge, a viewer arriving on a cold advertisement re-prices to zero with no hysteresis.

Verification

Every test was mutation-checked so none passes for the wrong reason:

Mutation	Caught by
COST_LINGER = 0	drain / cancel / route-change
deadline scan .min() → .max()	staggered (only)
cancel arm stops clearing idle_at	cancel (only)
outgoing_cost ignores demand	livelocks the loop, caught as a paused-time hang

Two of these came out of reviewing the first pass, which had real holes worth calling out:

• The cancellation test didn't test cancellation. Asserting only silence can't distinguish a cleared deadline from one that fires and re-prices to the same zero cost, taking the unchanged-route arm without writing — both are quiet. It now drains a second time inside the window, so an uncancelled deadline sends the restart four seconds early. Confirmed: the old assertion passed under that mutation, the new one fails.
• Two tests could pass on a panicked announce task, since a dead task also writes nothing and the trailing assert_quiet() would swallow it (tokio::spawn parks the panic in the handle). Harness::assert_idle now checks the JoinHandle alongside the wire.

Cross-package sync

Test-only change in rs/moq-net; no wire, API, or doc surface touched, so no other rows apply.

🤖 Generated with Claude Code

(written by Fable 5)

[sourcery-ai]

Hey - I've reviewed your changes and they look great!

---

Sourcery is free for open source - if you like our reviews please consider sharing them ✨

• X
• Mastodon
• LinkedIn
• Facebook

_{Help me be more useful! Please click 👍 or 👎 on each comment and I'll use the feedback to improve your reviews.}

[coderabbitai]

No actionable comments were generated in the recent review. 🎉

ℹ️ Recent review info

⚙️ Run configuration

Configuration used: Organization UI

Review profile: CHILL

Plan: Pro

Run ID: c5f26c71-836b-4335-9375-80dcea743d9f

📥 Commits

Reviewing files that changed from the base of the PR and between c74e583 and 034b120.

📒 Files selected for processing (1)

• rs/moq-net/src/lite/publisher.rs

---

Walkthrough

Adds a test module for the publisher announce loop using fake WebTransport sessions and streams that capture and decode emitted messages. A harness creates broadcasts with configurable viewer demand and runs the loop with paused Tokio time. Tests cover deferred and canceled cold-cost restoration, independent linger deadlines, one-shot expiry behavior, route-change supersession, and immediate repricing when demand arrives.

🚥 Pre-merge checks | ✅ 5

✅ Passed checks (5 passed)

Check name	Status	Explanation
Title check	✅ Passed	The title clearly and concisely summarizes the main change: new tests for the announce loop's demand/linger state machine.
Description check	✅ Passed	The description directly matches the changeset and explains the new test-only coverage in detail.
Docstring Coverage	✅ Passed	No functions found in the changed files to evaluate docstring coverage. Skipping docstring coverage check.
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.

✨ Finishing Touches

✨ Simplify code

• Create PR with simplified code
• Commit simplified code in branch claude/lite-announce-linger-tests

---

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.}