[kixelated]

Problem

Ten moq-tokio duration fields ignored values set in code after CLI parsing or a settings merge. Usage stored each parser default as if the flag had been given, so it beat the typed field, and a parse left the skipped typed field at Duration::ZERO, which could not be told apart from unset.

Approach

The typed fields become Option<Duration>. Each knob resolves the flag or environment value actually given, then the typed field, then the default. The fallback/resolve/explicit machinery in cli::Duration loses its callers and is deleted, leaving it the humantime parser. One test macro covers the precedence for all eleven duration knobs (the ten here plus #5350's backoff timeout): parsed, merged round-trip, zero and nonzero, a given flag still winning, and a config built in code.

Since the flags have no parser default, --help no longer printed [default: X] for them. Each flag now carries a one-line help that states its default, which also gives these flags a description they never had.

Public API

Breaking, Rust only (moq-tokio):

Field	Before	After
connect::Config::race	Duration	Option<Duration>
connect::Config::resolution_delay	Duration	Option<Duration>
connect::Config::timeout	Duration	Option<Duration>
listen::Config::timeout	Duration	Option<Duration>
quic::Config::idle_timeout	Duration	Option<Duration>
quic::Config::keep_alive	Duration	Option<Duration>
Backoff::initial	Duration	Option<Duration>
Backoff::max	Duration	Option<Duration>
connection::Goaway::handover	Duration	Option<Duration>
websocket::Config::delay	Duration	Option<Duration>

Default still holds Some(default) for each, matching #5350's Backoff::timeout. None uses the default; Some(Duration::ZERO) keeps each knob's zero behavior. The Resolved structs (connect, quic, websocket) and listen::Config::resolved_timeout keep their types.

Unchanged: CLI flag names, environment variables, TOML keys, the C ABI (moq.h), moq-ffi records, hand-written bindings, and the wire.

Rebased on #5484

#5484 removed the WebSocket-to-QUIC upgrade and reworked client.rs, connection.rs, and websocket.rs. Merged main in and re-checked each knob:

• No knob was removed: all ten still exist and are still read.
• race is the Happy Eyeballs stagger between resolved addresses; fix(tokio)!: keep the transport that won the connect race #5484 did not touch it.
• WebSocket delay is still QUIC's head start, skipped while a server is remembered as a WebSocket win. fix(tokio)!: keep the transport that won the connect race #5484 moved the forget on a QUIC win into the race itself; zero still means no head start.
• GOAWAY handover now caps only a peer GOAWAY handover. The client-initiated upgrade handover that also read it is gone.
• Connect timeout bounds the dial and handshake only, since no upgrade outlives the attempt anymore.

The only conflict was in upgrade tests #5484 deleted; its new quic_winning_cancels_the_websocket_dial test now sets websocket.delay = Some(Duration::ZERO). The doc/setup/upgrade.md entry still holds against main: #5350 is unreleased, so backoff.timeout stays in the list.

Decisions

Picked by the iterating agent, without a prompt:

• Help text for the eleven flags:
  • ✅ A one-line help per flag stating its default (recommended: restores the default in --help without a parser value).
  • Leave help without defaults, as fix(moq-tokio)!: a backoff timeout set in code wins over the parser default #5350 did for --backoff-timeout.
  • Restore default = "...": rejected, it reintroduces the bug.
• Merge main rather than rewrite the branch: ✅ the change re-derives cleanly; squash merge flattens it.

Validation

Final head e9cb3ceb294f7725c669218eee079253648a5cc9, merged with main at 20ca8a651.

• just check origin/main: every Rust test passed in at least one of two runs. Each run stopped on a tracked load flake, never in code this PR touches: run one on the uring teardown close (passed alone), run two on the auth-refresh clock tests (passed in run one).
• The modules just check runs after Rust passed when run directly: bench, quest, drill sensitivity, Python, Kotlin, Go, Dart, C++, shell, TOML, and GitHub lint.
• moq-relay --help shows each duration flag's description and default.
• CI passed on the final head, including TS validation.

Follow-ups

None from this PR.

(Written by Claude Opus 5.5)

🤖 Generated with Claude Code

[kixelated]

Duration implementation is complete at e7e3eb1c440073c5b43df1fbe2aaad534f4e2f46; no API or scope decisions remain. Leaving this PR draft because the final strict run hit the existing auth-refresh clock failures and final interop hit the existing browser refusal-close failure. Both are linked in the PR description. The identical native source previously passed all 3,975 Rust tests and full interop; that does not establish either existing failure as fixed. Recommended follow-up is the existing controlled-clock and browser close-code quest work, without widening bounds or retrying to obtain green. No merge or subsequent PR monitoring was performed.

(Written by GPT-6)

[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 2026-10-15T00:42:15.299384Z	b6e9a5b	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.

[coderabbitai]

No actionable comments were generated in the recent review. 🎉

ℹ️ Recent review info

⚙️ Run configuration

• Configuration used: Organization UI
• Review profile: CHILL
• Plan: Advanced
• Run ID: 45a8053b-20f5-4507-aa23-84dbc1301370

📥 Commits

Reviewing files that changed from the base of the PR and between b6e9a5b and b160a03.

📒 Files selected for processing: 4

• quest/m1/README.md
• rs/moq-relay/src/web.rs
• rs/moq-relay/tests/auth_lifetime.rs
• rs/moq-tokio/src/websocket.rs

💤 Files with no reviewable changes: 1

• quest/m1/README.md

Included review availability: This review used your included allowance. Your plan provides up to 4 included reviews per hour; 1 remain after this review.

Walkthrough

Selected moq-tokio duration configuration fields now use Option<Duration>. Resolution prefers parsed CLI or environment values, then configured values, then defaults. Optional serialization omits unset fields. C ABI, FFI, GStreamer, relay, and integration-test callers now use optional values. Tests cover precedence, zero values, defaults, validation, and resolved settings. Upgrade documentation records the behavior.

Priority: ⬇️ Low

Merge Risk: ⚪ Minimal · up to b160a

The reviewed changes update duration configuration and affected fixtures; no actionable merge-blocking risk is established.

Pre-merge checks: 4 · 1

Failed checks: 1 warning

Check name	Status	Explanation	Resolution
Docstring Coverage		Docstring coverage is 78.31% which is insufficient. The required threshold is 80.00%. Docstring coverage is scoped to functions touched by this diff. Analyzed 83 functions across 27 files.	Write docstrings for the functions missing them to satisfy the coverage threshold.

Passed checks: 4 passed

Check name	Status	Explanation
Title check		The title clearly identifies the main change: making parsed configuration durations honor values set in code.
Description check		The description directly explains the duration precedence issue, the Option-based solution, API impact, validation, and unchanged interfaces.
Linked Issues check		Check skipped because no linked issues were found for this pull request.
Out of Scope Changes check		Check skipped because no linked issues were found for this pull request.

• Fix all pre-merge checks with AI

✨ Finishing Touches

✨ Simplify code

• Commit to this branch
• Create a new PR

• Autofix · Keep fixing CodeRabbit findings and required CI, and resolving merge conflicts

Warning

Some tools did not complete. Review the errors below.

🔧 Clippy (1.98.1)

Clippy execution timed out

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.}

[kixelated]

Merge summary for head e9cb3ce.

• Merged main after fix(tokio)!: keep the transport that won the connect race #5484. All ten duration knobs survive it with the same zero semantics; handover now caps only a peer GOAWAY handover, and connect timeout no longer bounds a pending upgrade, since fix(tokio)!: keep the transport that won the connect race #5484 removed the upgrade. The only conflict was in upgrade tests fix(tokio)!: keep the transport that won the connect race #5484 deleted.
• Added a one-line --help per duration flag stating its default, which the flags lost along with their parser defaults (covers --backoff-timeout from fix(moq-tokio)!: a backoff timeout set in code wins over the parser default #5350 too).
• doc/setup/upgrade.md holds against main; fix(moq-tokio)!: a backoff timeout set in code wins over the parser default #5350 is unreleased, so backoff.timeout stays in the list.
• Reviews: Codex (no findings, 👍) and CodeRabbit (no actionable comments) on this head. Nothing to fix or decline.
• CI passed on this head. Local just check hit only the tracked uring-teardown and auth-refresh load flakes; each passed on rerun.

Public API: ten moq_tokio config fields go from Duration to Option<Duration> (table in the description). CLI, env, TOML, C ABI, bindings, and wire are unchanged.

(Written by Claude Opus 5.5)

[kixelated]

Automated review by review (OpenAI)

Reviewed commit: b6e9a5b.

No actionable bugs found. Option correctly separates unset from explicit zero; resolution keeps explicit CLI/environment values ahead of typed fields and defaults without manufacturing parser values. The shared eleven-knob regression coverage and documented Rust API migration are appropriate, and binding adaptations preserve their existing external representations.

Compared the current base-relative patch with the previously inspected e9cb3ce, accounting for main imports. Production changes are equivalent; the added drain/refused-dial fixture adaptations retain their intended zero-timeout behavior.

Verification: static review of configuration resolution, serialization, callers/bindings and tests. No repository tests or builds run; the PR description's validation refers to its earlier head.

[kixelated]

Re-enqueued at b6e9a5ba4083d5f9e199b653869ebbc1574b6194.

The queue dropped the previous head on a real semantic conflict, not a flake: the merge_group Test job failed to compile moq-relay's lib tests because main's new web drain fixture (rs/moq-relay/src/web.rs) sets config.listen.timeout = Duration::ZERO, which this PR makes Option<Duration>.

The new head merges main and adapts the two tests main added against the old types: that fixture now sets Some(Duration::ZERO), and a_refused_tcp_dial_keeps_retrying in moq-tokio's connection.rs takes main's Windows-safe version with Some(..) for backoff, timeout, and race. quest/m1/README.md keeps main's channel-count line and still drops this quest's entry. No other code changed.

Validation: cargo check --workspace --all-targets --all-features is clean, and cargo nextest run -p moq-tokio -p moq-relay --all-features passes (830 tests). Codex reviewed the new head with no findings.

(Written by Claude Opus 5.5)

[kixelated]

Re-enqueued at b160a03af5505691c23cb57987ec88a9da8a91e8. The queue dropped b6e9a5ba4 as unmergeable behind #5526, which also removed its own adjacent quest line from quest/m1/README.md. This merge of main drops both lines. No code changed since the head Codex reviewed, and the workspace still checks clean with all features.

(Written by Claude Opus 5.5)