MOQ Wiki

Media over QUIC Transport (MOQT)

37 min read

2026-06-07: DTS consensus readout flips to “separate WG extension document” — Cullen Jennings contests; June-4 normative PRs merge; 3 new design issues open; London Day-1 agenda published. The chairs’ readout of the DTS/SWITCH consensus call (closed June 4) concludes there is “strong consensus for DTS to be adopted as an extension described by a WG document” — not integrated into base draft-18 — and invites Will Law to publish his wilaw/dts4moq repo as a new WG document draft-ietf-moq-dts4moq (door left open: “it is possible that the document later reaches a state where it could be merged into MOQT”). This reverses the trajectory the wiki tracked: through late May the on-list votes favored integration into MOQT and Cullen Jennings explicitly argued against a separate draft (May 27: “I disagree it should be done as a separate draft”), which prompted wilaw’s base-spec PR #1638 (still OPEN, +155/−0). Cullen now pushes back on the process June 6: “How did you reach the conclusion that there is consensus this should be an extension instead of in the MoQT draft?” — PR #1638’s base-spec integration path is now superseded-in-direction by the readout’s separate-document outcome (closer to Magnus Westerlund’s original proposal). draft-ietf-moq-dts4moq invited but not yet submitted as of June 7. Normative June-2/3 work lands: PR #1640 EXPIRES (martinduke) MERGED June 4; sharmafb PR #1645 (PUBLISH_OK = 0x1E) + #1648 (soften REQUEST_UPDATE response) + #1649 (PUBLISH_DONE wording) all MERGED June 4; PR #1476 “Delivery timeouts are both Track and Object Properties” (afrind, OPEN since Feb) MERGED June 5, closing Issue #606 (“Delivery timeout cannot be expressed per subgroup”). Issues #1641 (TimEvens PUBLISH_DONE) + #1639 (martinduke EXPIRES) auto-CLOSE; ianswett’s PR #1451 “Allow multiple Subscriptions to a Track” CLOSED June 6 (subsumed by the Concurrent Subscribe agenda item #1633). 3 new design issues OPEN: #1650 (afrind, “Forward Handling SHOULD set Forward=1 upstream is too strong”), #1651 (martinduke, “How does GROUP_ORDER interact with PUBLISH?”), #1652 (Kostya Vasilyev, “SETUP - extensions negotiation”). London Day-1 agenda (interim-08) now published — Martin Duke June 5 “shifted a bit more time to MOQT blockers”: Concurrent Subscribe (Ian/#1633), Object Range Filters (Mo/PR #1518), Joining FETCH (Alan+Mo/PR #1642, 60 min), Request Blocking (Alan/#1519), two hour-long “Other MOQT Issues” blocks (Alan+Ian), then SWITCH/DTS Implementation Lessons (Ali) demoted to 1600-1700. See interim-meetings + discussions-2026-06. quiche moqt draft-19/refactor continues: vasilvv 8a11928a+6b6a9b74 June 5 factor control-message queueing into a MoqtControlMessageQueue class; f3ce1892 June 6 “Support the ‘has first object in the subgroup’ bit in the subgroup header” (draft-18 FIRST_OBJECT bit decode).

2026-06-04: sharmafb (Suhas Sathyanarayana, Cisco) opens editorial sprint on draft-18 June 3 21:09-21:27 UTC — 5 contributions in ~18 minutes addressing newly-found wire/text inconsistencies: PR #1645 OPEN “Change PUBLISH_OK = 0x1E in table” (PUBLISH_OK message type was removed); Issue #1646 OPEN “Application-specific property codepoints conflict” (§2.5 reserves 0x38-0x3F + 0x3800-0x3FFF for application use, but §15.8 reserves 0x78-0x7F — direct codepoint collision); Issue #1647 OPEN “TIMESTAMP and SUBGROUP_DELIVERY_TIMEOUT have same value” (§15.8 codepoint collision); PR #1648 OPEN “Soften one mandatory response for a REQUEST_UPDATE” (§10.9.1 already allows coalesced failed updates to produce only one REQUEST_ERROR for multiple updates); PR #1649 OPEN “Modify text about PUBLISH_DONE on a control stream” (companion to Tim Evens’s Issue #1641 where afrind acknowledged “Request streams are considered ‘control’ streams in most of the text, but we can probably crisp this up”). Plus earlier same day Issue #1644 OPEN 15:55 UTC by kixelated “Duplicate Track Properties on repeated SUBSCRIBE_OK / FETCH_OK” — “Track properties are meant to be fixed for the lifetime of a track. They can be cached indefinitely, at least in theory. However, each SUBSCRIBE_OK and FETCH_OK transmits a fresh copy of the Track Properties even if the subscriber already has that information cached. The most egregious example would be a FETCH for each group; we keep transmitting redundant track properties.” — this is the same gap kixelated already addresses on the moq-lite-05 side via moq-dev/moq PR #1609 OPEN (TRACK_INFO Track Stream 0x6 removes immutable props from SUBSCRIBE_OK); now surfacing as a moq-transport spec issue too. Cumulative active draft-18 issue/PR inventory heading into June 4 = 6 normative PRs OPEN (Fill #1642, EXPIRES #1640, DTS #1638, PUBLISH_OK #1645, REQUEST_UPDATE response #1648, PUBLISH_DONE wording #1649) + 8 design issues OPEN (bidi-stream-credit #1637, request-blocking #1519, PUBLISH_DONE control #1641, bidi-stream-FIN #1643, duplicate track props #1644, codepoint conflict #1646, TIMESTAMP collision #1647, plus carry-overs #1633/#1634/#1635/#1636). draft-18 errata-style issues now outpace design-issue resolution: 5 of 5 sharmafb items are wire-table-vs-text/codepoint-collision-style, suggesting draft-18 needs an editorial -19 pass or a draft-18 errata document. Parallel — quiche moqt starts draft-19 wire prep in earnest June 3: 3 vasilvv commits June 3 03:08-10:58 UTC: 4096d2e3 “Switch MOQT to using new varint format” (draft-19 wire bump, follows the moq-WG-discussed varint compaction), 8427f949 “Use delta encoding for Subscribe Filters of type Absolute Range” (wire compaction), 96c9a9c0 “Merge CLIENT_SETUP and SERVER_SETUP into a single control message” (draft-19 architectural simplification reducing the handshake to one control message per side). Plus martinduke commits 52de014c June 2 22:45 UTC “Update MoqtResponseCallback to include MessageParameters” (REQUEST_OK includes this field) + c25d5258 June 3 15:30 UTC ASAN fix. First sustained draft-19 wire-prep work the wiki has tracked — until now quiche moqt commits were refactor work (May 14-30 stream-class factoring), now lifting concrete wire-shape changes. Carry-forward: sharmafb’s 5-item editorial sprint lifts the London Day-1 0900-1045 “MOQT Issues” 180-min slot agenda toward triage-heavy: 6 PRs + 8 issues in a 180-min slot ≈ 13 min per item, infeasible to ship all 6 PRs on the floor — chairs will likely defer pure-editorial items (PUBLISH_OK code change #1645 + PUBLISH_DONE wording #1649) to editor’s discretion, leaving DTS + EXPIRES + Fill as the three substantive merges-on-floor candidates.

2026-06-03: afrind opens PR #1642 June 2 17:01 UTC “Replace Joining Fetch with Subscription Fill, add Current Group” (+188/−148, 1f) — structural redesign of MOQT’s past-object retrieval: Joining Fetch types removed entirely (Joining Fetch types, Joining Fetches section, Joining Fetch Range Calculation, Joining Location, INVALID_JOINING_REQUEST_ID error code all deleted); Standalone Fetch fields inlined directly into FETCH wire format; new subscription filter types added — AbsoluteStartFill (0x5), AbsoluteRangeFill (0x6), CurrentGroup (0x7), RelativeStartFill (0x8); new Current Group Delivery section with publisher subgroup ordering + relay object accounting requirements (applies to any filter covering current group); new Fill Semantics section defining fill fetch stream behavior; new FillDescending (0x3) group order (descending for fill, ascending for subscribe); LARGEST_OBJECT serves as fill boundary in SUBSCRIBE_OK; FILL_TIMEOUT extended to apply in SUBSCRIBE messages with fill filter types; REQUEST_UPDATE with a fill filter type opens a new fill fetch stream using the REQUEST_UPDATE’s Request ID; fill filter types + CurrentGroup disallowed in PUBLISH_OK (stale LARGEST_OBJECT); recommends forward=0 + REQUEST_UPDATE pattern instead. Publisher opens a single fill fetch stream (FETCH_HEADER reusing the subscription’s Request ID) for past objects, while delivering current/future via subscribe subgroups/datagrams. Replaces ianswett’s PR #1627 “SUBSCRIBE with Joining Fetch” OPEN May 3 — ianswett auto-CLOSES #1627 June 3 00:16 UTC, 30 days OPEN. The architectural simplification (one filter dimension instead of two request types) and the FillDescending group order directly answer ianswett’s Issue #1614 “(JOINING) FETCH + SUBSCRIBE prioritization”. Plus 2 new draft-18 design issues: Issue #1641 OPEN June 2 06:19 UTC by Tim Evens (Cisco) “Publish DONE, control stream or request stream?” — surfaces text/wire inconsistency where the spec says “Because PUBLISH_DONE is sent on a control stream, it is likely to arrive at the receiver before late-arriving objects” but PUBLISH_DONE actually rides the request stream; afrind reply June 2 20:45 UTC: “Request streams are considered ‘control’ streams in most of the text, but we can probably crisp this up”. Issue #1643 OPEN June 2 18:49 UTC by afrind “Is it a request or session error to FIN a response bidi stream without a control message first?” — “eg: publisher responds to FETCH with FIN but not FETCH_OK/ERROR? If I read HTTP/3 correctly, this is just a stream error there” — companion to #1635 (“Should a subscriber treat FIN/RST on a bidi stream as equivalent to PUBLISH_DONE?”) and #1634 (“What semantics do FIN or RST on a request stream carry?”). London Day-1 0900-1045 “MOQT Issues” 180-min slot now has 3 normative-text deliverables (DTS PR #1638 + EXPIRES PR #1640 + Fill PR #1642) + 4 design-issue threads (bidi-stream-credit #1637, request-blocking #1519, PUBLISH_DONE wording #1641, bidi-stream-FIN #1643). With 3 hours and 3 PRs + 4 issues, the slot will require triage. Tim Evens cross-repo signal: same-author files transport Issue #1641 + openmoq/moqx PR #376 (Debian bookworm builds → CPM) + PR #377 (macOS Apple Clang 21 fmt fix) within ~18h June 2 06:19-23:42 UTC — Cisco engineer surfacing at both spec-design and build-system layers; 3rd Cisco contributor reaching mainline moq-wg artifacts in May-June 2026 alongside Suhas Nandakumar (cmsf/msf/secure-objects/privacy-pass editorial + cloudflare/moq-rs PR 171) and Mo Zanaty (loc draft author).

2026-06-02: wilaw opens PR #1638 June 1 11:06 UTC “Add Dynamic Track Switching (DTS)” (+147/−0, 1 file, fixes Issue #259) — DTS lands as a PR-against-moq-transport-18 base spec, the integration path Cullen Jennings May 27 explicitly preferred (“I disagree it should be done as a separate draft. I think it should be worked as a PR that can be discussed in context”) against Magnus Westerlund’s separate-Internet-Draft proposal. Consensus call cumulative tally heading into June 4 close = 4 YES + 0 NO (Will Law May 26 + Cullen Jennings May 27 + Gwendal Simon May 28 + Yu You-Nokia May 29 + Ali Begen May 29; 3 of 4 SWITCH co-authors + Nokia implementation report) = substantively decided 14 days before close. DTS PR #1638 body: “Adds the Dynamic Track Switching (DTS) feature, detailing subscriber operations, relay behavior, and bandwidth allocation algorithm. Adds a new SWITCHING-SET-ASSIGNMENT parameter.”. Plus same day: Issue #1639 OPEN June 1 19:50 UTC by martinduke “EXPIRES should be allowed in more REQUEST_OK sub-types” — “IIUC EXPIRES is basically saying ‘this is how long your auth token is good for.’ If that’s the case, this parameter is useful for any response for a request that contains an auth token, which is all of them.”, calls out FETCH_OK + TRACK_STATUS as exceptions; PR #1640 OPEN 22:14 UTC by martinduke “Add EXPIRES to most request types” (+2/−1, 1 file, fixes #1639). London Day-1 0900-1045 “MOQT Issues” 180-min slot now has 2 normative-text deliverables (DTS PR #1638 + EXPIRES PR #1640) + 2 design-issue threads (bidi-stream-credit Issue #1637 + request-blocking Issue #1519). Plus 2 google/quiche moqt commits June 1 17:28-17:34 UTC by martinduke: 3b9d5450 “Move parameter handling from session to RemoteTrack” + 0b92a8b4 “Get rid of MoqtUpstreamFetch::LocationIsValid because FETCH streams are diff-encoded; Malformed Tracks are no longer possible to encode in a FETCH stream” = breaks 5-day quiche moqt silence, incremental implementation refactor work (not draft-19 wire prep).

2026-05-31: afrind May 30 16:00:09 UTC comment on Issue #1519 (vasilvv “Improve design of requests blocking on other requests”) replies to ianswett’s May 30 02:27 UTC ABR-approaches comment: “Most MOQT stacks don’t really offer direct application control of ‘send these two requests in the same QUIC packet’. I think 99% of the time that’s just going to happen — the control streams should have highest priority, and they are not large. Are we ok saying ‘this is mostly fine’ or do we need a MOQT mechanism to collect the related requests on the receiver side.” — surfaces a stack-level design constraint (no API to coordinate which requests share a QUIC packet) and asks whether the WG should add receiver-side request-coalescing or accept the existing implicit ordering. Combined with martinduke Issue #1637 (bidi-stream-credit, OPEN May 29) and Issue #1519 still OPEN since March 2 with active comments through May 30, MOQT’s request-message-on-bidi-stream design space is now actively under WG-list discussion ahead of London Day-1 0900-1045 “MOQT Issues” 180-min slot. No new commits or activity on google/quiche moqt May 30-31 beyond e05dcf91 ClangTidy fix May 30 06:20 UTC — substantive draft-19 prep work pauses while design discussion plays out in issues.

2026-05-30: Martin Duke Issue #1637 OPEN May 29 22:30:34 UTC “What does MOQT do without bidi stream credit?” — surfaces a draft-18 design gap: REQUEST_BLOCKED was removed but PUBLISH_BLOCKED retained asymmetrically; “if both directions are blocked, who is responsible?“. 3 numbered questions: (1) which request streams retry by MOQT vs. application? (2) when is peer expected to try SUBSCRIBE/PUBLISH inverse? (3) if both directions blocked, who opens? Cross-repo coordination: Issue #1637 is the design-discussion counterpart to martinduke’s May 28 google/quiche moqt commits c4503a21 (“Move IncomingDataStream … Other preparatory changes for requests on bidi streams”) + 997d6543 — implementation work + design issue publicly paired by same author within 24h, structural draft-19 direction. Ian Swett 4 comments through May 30 02:27 UTC on #1637 + #1519 (vasilvv “Improve design of requests blocking on other requests”, open since Mar 2): MoQT typically acts on PUBLISH_BLOCKED, “create a new MoQT session” for blocked-direction retry, 3 ABR approaches outlined (explicit-group SUBSCRIBE, bandwidth-check upswitch, use SWITCH or DTS). Carry-forward: London Day-1 0900-1045 “MOQT Issues” 180-min slot now structurally anchored around bidi-stream-credit + request-blocking + ABR-switching design, with #1637 + #1519 as the two open issues to resolve.

draft-ietf-moq-transport-18 | published 2026-05-12 | Datatracker

Draft-18 Implementation Status (as of 2026-05-20)

ImplementationStackMaintainerStatusPublic endpoint
moqRust + JSkixelatedShipped May 18 05:08 UTC (PR #1418, +1431/−477)cdn.moq.dev/anon (claims draft-18, untested)
imquicClorenzo-minieromoq-18 branch ready; merge gated on interop-runner PR #68. Lorenzo May 19 15:40 UTC: “finished the bulk of the work and the rest can wait”.lminiero.it:9000 (raw QUIC + WebTransport, served from branch)
mondain/moqxrC++Paul Gregoire (openmoq member)Shipped May 16–18; 8 fix-up commits May 19 14:17 → May 20 00:11 UTC after Lorenzo found a bidi-vs-uni SETUP-stream divergence in first interop attempt—
moq-rsRustmike-englishNot yet (Day +37 main-quiet)—
quiche moqtC++martin-duke / victor-vasilievNot yet (Day +8 quiet on draft-18 wire path; martinduke May 18-19 commits are refactors not wire-bumps)—
moqtailRustZafer GurelNot yet (Day +7 quiet)—
interop-runner matrix(target only)mike-englishTarget is still draft-16 (PR #68 Open, CI work in progress)—

6 days after publication, draft-18 has 3 distinct codebases shipped + 1 production deployment — fastest spec → first-impl turnaround the wiki has tracked (draft-17: ~3 weeks; draft-16: ~2 weeks). The accelerants: (1) the Claude-co-authored PR pattern on moq-dev/moq; (2) draft-18’s small-delta scope (7-byte varint, FIRST_OBJECT bit, RequestID removal, LARGEST_OBJECT) lets implementations stage changes incrementally on draft-17 plumbing.

Authors

Abstract

MOQT is a publish/subscribe protocol that runs over QUIC and WebTransport. It leverages transport capabilities including streams, datagrams, priorities, and partial reliability. MOQT enables content delivery both directly between endpoints and through intermediate relays, achieving scalable low-latency distribution. Despite its name referencing media, the specification emphasizes that MOQT remains content-agnostic and applicable across various use cases (abstract reframing in draft-18 — see discussions-2026-05 for Will Law’s parallel May 12 recharter proposal).

Recent Changes (draft-18, published 2026-05-12)

The draft-18 cut was assembled in two phases of merges to main:

Phase 1 — May 11 21:32–22:02 UTC editorial sprint (6 PRs in 30 min):

  • PR #1544 — Improve Startup Latency and 0-RTT (ianswett, fixes #420 #83). New sections on reducing startup latency / 0-RTT flow with WebTransport.
  • PR #1615 — Remove Required Request ID (ianswett, fixes #1603). Materialises the Apr 27 interim consensus. Required Request IDs are removed; Request IDs remain for individual requests (used by Joining FETCH and GOAWAY).
  • PR #1617 — Allow GOAWAY on request streams to migrate individual requests (afrind, fixes #1481). Per-request GOAWAY with zero-length URI causes endpoint to re-issue request on the specified URI session.
  • PR #1618 — Add FIRST_OBJECT bit to SUBGROUP_HEADER type (afrind). Bit 6 (0x40) signals subgroup contains the first object published by the original publisher; type byte expands 0b00X1XXXX → 0b0XX1XXXX (still 1-byte varint).
  • PR #1621 — Forbid relays from lying about LARGEST_OBJECT (afrind, fixes #1386).
  • PR #1629 — Clarify definition of scope (vasilvv, fixes #1432).

Phase 2 — May 12 20:02–23:07 UTC final merges (3 PRs):

  • PR #1625 — Improve Security Considerations section (suhasHere, +118/−7). Rebases / extends magnus-westerlund’s long-parked PR #1455. Closes Issue #783.
  • PR #1605 — Split DELIVERY_TIMEOUT into two types of timeout (vasilvv, +114/−76). OBJECT_DELIVERY_TIMEOUT (more precisely defined replacement for DELIVERY_TIMEOUT) + new SUBGROUP_DELIVERY_TIMEOUT (covers subgroups fully queued but not fully delivered). Closes Issue #667 (DELIVERY_TIMEOUT is unimplementable).
  • PR #1630 — Draft 18 release notes (afrind, +59/0). Body: “Behold”.

Datatracker upload followed at approximately May 12 23:30 UTC; afrind announcement on Slack #moq at May 13 01:15 CEST: “It’s heeeeere”.

Key Concepts

  • publish-subscribe: Core messaging pattern - PUBLISH, SUBSCRIBE, FETCH, REQUEST_OK/REQUEST_ERROR, PUBLISH_DONE
  • subgroups-and-objects: Data hierarchy: Track > Group > Subgroup > Object
  • relays: First-class relay support for CDN-style distribution
  • track-properties: Metadata attached to tracks and objects via Key-Value Pairs
  • joining-fetch: Mechanism to fetch historical data when joining a live session
  • qmux: TCP fallback via QUIC multiplexing over TLS+TCP
  • URI Scheme: moqt:// URI scheme with fragment identifier support (PR #1571)

Recent Changes (draft-17, published 2026-03-02)

Draft-17 brought significant changes from draft-16:

  • Unidirectional control streams for 0-RTT subscribe capability
  • SUBSCRIBE_NAMESPACE split into two messages: SUBSCRIBE_NAMESPACE (namespace info) and SUBSCRIBE_TRACKS (PUBLISH notifications)
  • Subscription filters moved to be a Param (PR #1590)
  • Track Properties added to REQUEST_OK (PR #1576)
  • Allow 7-byte varint and non-minimal encodings (PR #1595)
  • Editorial: consistent use of “MOQT” for protocol references (PR #1597)
  • Editorial: use “message” instead of “frame” (PR #1587)

Active Issues (as of 2026-05-20)

Design Issues

  • #1636 — Empty namespace clarification. OPENED May 21 10:50 UTC by RichLogan (Cisco). Body: “I was just reading the draft-18 text regarding namespace prefixes and full track names and wanted to clarify the rules for our implementation. I understand the use-case for zero element namespace prefixes, but wanted to check that publishing a track with a zero element namespace is allowed, and in theory, given the text as it is today, a track with a zero element namespace AND an empty name is also allowed?” — 3-comment AI-hallucination incident May 22: afrind 18:24 UTC initial reply (prefixed “With apologies for answering your question with AI”) hallucinated “between 1 and 32 Track Namespace Fields” + non-existent line numbers (511 / 544-546 / 3107-3108) claiming 0-element namespaces are a PROTOCOL_VIOLATION; RichLogan 21:15 UTC catches the hallucination by quoting the actual draft-18 §2.4.1 text (“between 0 and 32 Track Namespace Fields”); afrind 21:45 UTC explicit apology — “Wow, that is pretty embarrassing. I apologize. It spit out that sentence about 1 and 32 and that seemed familiar I didn’t double check it. Ouch.” — first explicit on-record acknowledged-AI-hallucination on moq-wg/moq-transport by an editor in spec discussion. Substantive question (allow vs. forbid 0-element namespace + empty name) remains open; afrind tentatively in favour of allowing (“sort of like path=/ in HTTP”). Belongs in the “draft-18 ambiguity backlog” alongside 1635 (afrind May 19 tranche) and #1632 (yuanchao-chris May 14 Properties Type collision).
  • #1635 — [draft-18] Should a subscriber treat FIN/RST on a bidi stream as equivalent to PUBLISH_DONE? OPENED May 19 18:37 UTC by afrind. Body: “PUBLISH_DONE conveys error information as well as stream count. Seems like FIN without PUBLISH_DONE is entirely avoidable and could be a PROTOCOL_VIOLATION?” First subscriber-side post-publication control-stream-lifecycle question.
  • #1634 — [draft-18] What semantics do FIN or RST on a request stream carry? OPENED May 19 18:34 UTC by afrind. Quotes §4.5 + §3.6 (STOP_SENDING is the canonical cancellation) and asks what action a receiver should take on a bare FIN or RST_STREAM. Self-comment May 19 21:45 UTC: “FIN almost always isn’t a cancellation… For TRACK_STATUS an early FIN is perfectly normal. For requests like FETCH, PUB_NS, SUB_NS, SUB_T and SUBSCRIBE, an early FIN can mean ‘I don’t plan to REQUEST_UPDATE’…” Sibling of #1635 on the publisher side.
  • #1633 — Should we allow more than one concurrent subscription per Track. OPENED May 19 16:33 UTC by afrind. Operationalises Luke Curley’s May 18 SVC-thread procedural gap discovery (SUBSCRIBE-same-track-twice illegal so SGPL-with-per-layer-priority needs multi-sub). Body lifts text from Editors’ IETF-125 slides with 4-proposal slate: 1a (Editors’ rec) = unique Track Aliases per sub, no dedup; 1b = same Track Alias, publisher coalesces (Ian Swett + Martin Duke prefer); 2 = single-sub + union filters; 3 = status quo (illegal). Same-day comments ian-swett May 19 17:08 UTC and afrind May 19 17:31 UTC re-argue 1b conflict-handling (afrind: “different DELIVERY_TIMEOUT, PRIORITY, GROUP_ORDER, AUTH… if publisher is deduplicating by single track alias, it’s unclear how to apply these”). May 21 substantive make-before-break thread (4 comments 18:43-21:21 UTC): martin-duke “Yes, 1b. The non-silly use case for this is make-before-break or just avoiding a session_error because of a lost RESET_STREAM”; ian-swett “With Subgroup filters, one clear usecase is starting the enhancement layer Subgroup at the next group boundary… end the current Subscription at the end of the current Group and start a new one at the next Group with the updated Subgroup filter”; martinduke “yeah, that’s a make-before-break. It also highlights that the group-as-join-point thinking in the design is still pretty muddled. Why can’t a REQUEST_UPDATE be triggered to start on the next group?”; ianswett “a simpler option might be to start a new Subscription with the new filters you want”. First explicit chair-side endorsement of 1b + new “REQUEST_UPDATE next-group offset” design alternative surfaced. Cites long-parked PR #1451 (open since April 13). Slated for London — 10-min Day-1 slot Jun 11 1300-1310 in the May 21 final agenda.
  • #1632 — MOQ-18: Properties Type collision with LOC-02. OPENED May 14 03:24 UTC by yuanchao-chris (new contributor — 2nd issue in 2 days). Reports that draft-18 §15.8-2 assigns Property Type IDs that diverge from draft-ietf-moq-loc-02’s commit-history values: MOQ-18 has TIMESTAMP=0x06 / TIMESCALE=0x08 / AUDIO_LEVEL=0x0C / VIDEO_FRAME_MARKING=0x0A / VIDEO_CONFIG=0x0D, while LOC-02 records TIMESTAMP=0x02 / AUDIO_LEVEL=0x06 / VIDEO_FRAME_MARKING=0x04. Twin issue filed simultaneously on the LOC side (moq-wg/loc Issue #20). PR #1624 (April 30, “provisional IANA registry for LOC properties”) was supposed to resolve issue-1550 but the assignment cuts in draft-18 §15.8-2 did not adopt the registry values. First publicly-flagged post-draft-18 cross-spec coordination failure — see discussions-2026-05.
  • #1631 — Track-level codec switching semantics. OPENED May 13 02:23 UTC by yuanchao-chris. Asks whether MoQ supports in-band codec migration (H265→H264, AV1→H264) within an existing Track, analogous to WebRTC PT change inside the same SSRC. alan-frindell May 13 05:11 UTC sketches “publisher make a new group in an ongoing track, and include codec information on properties communicated on Object 0 in the new group - or the Object 0 payload”. yuanchao-chris May 13 09:23 UTC confirms works in stream mode but in datagram mode needs property-stamped frames + REQUEST_UPDATE-based “ACK” semantics; also notes moq-msf §5.1.24 catalog track information needs alignment. First substantive post-draft-18 design conversation.
  • #1626 - Version negotiation for QMUX. Opened May 1 23:50 UTC by sharmafb (Suhas Sathyanarayana): “We have an idea of how version negotiation works for MoQ-over-HTTP/3 and how it works for MoQ-over-QUIC, but do we know how it’s going to work for MoQ-over-QMUX?” alan-frindell reply May 2 02:19 UTC: “We discussed quite a bit last IETF. The plan is to say something like TLS ALPN moqt-18 implies qmux-01” — first explicit statement of the QMUX/transport ALPN coupling for draft-18. See qmux for context.
  • #1622 - Request ID in GOAWAY isn’t useful. Opened Apr 30 by ian-swett (label Handshake and Session). Body: “After more thought (yes I approved #1559), I don’t think the Request ID in GOAWAY is actionable in MoQ. … Now that we’re removing Required Request ID (#1615) and we’ve already removed Request ID flow control, GOAWAY is one of the two remaining uses of Request ID (the other is Joining Fetch).” PR #1623 (revert) opened Apr 30 01:38 UTC. alan-frindell counter Apr 30 18:31:57 UTC: “My counter is - it’s trivial to put the request ID in goaway, and might be useful. If nothing else it can speed up retry when a new request is racing a GOAWAY.” — afrind’s first explicit pushback against the walk-back. PR #1623 now contested.
  • #1616 - Both PUBLISH_NAMESPACE and NAMESPACE are responses to SUBSCRIBE_NAMESPACE. CLOSED Apr 29 20:44 UTC via PR #1619 merge.
  • #1614 - (JOINING) FETCH + SUBSCRIBE prioritization. Opened Apr 27 by luke-curley (split from #1358). Concrete TTV math: at 1.5s into a 2s GoP with 3 Mb/s media on 4.5 Mb/s network, JOINING FETCH delivers TTV=1.33s while a hypothetical SUBSCRIBE filter=LargestGroup order=DESC delivers TTV=0.5s. Concludes: “we need order=DESC support for JOINING FETCH. Either some way of prioritizing between the SUBSCRIBE + JOINING FETCH, or cancelling the JOINING FETCH if the next group starts (kinda gross), or add back the LargestGroup filter (pls).” Renews pressure on PR #1607.
  • #1612 - What happens to Joining FETCH if fwd changes to 0? Opened Apr 23 by martin-duke. Apr 23 21:02 UTC: alan-frindell replied: “Changing the subscription from 1 to 0 after joining fetch has no effect on the FETCH. We can update the spec. Though now it seems like requiring fwd=1 is causing a lot of problems. I wonder if we should just allow fwd=0.” Addressed by PR #1620 (Apr 28 23:25 UTC).
  • #1603 - What is the use case for required-request-id. Apr 27 18:42 UTC interim outcome (ian-swett): “Conclusion was to remove required-request-id from draft 18 and fix Joining Fetch (if necessary?). Those who believe some functionality in this space is useful, such as for make-before-break, should explore those use cases in more detail and further describe what, if any, dependency structure between requests is needed in MoQ. Tentative plan is to discuss these at the London hybrid interim in June.” PR #1615 implements the removal. PRs #1604 (Martin’s structural fix) and #1613 (Alan’s flow control) lose their headline justification.
  • #1602 - Joining Fetch should be on the SUBSCRIBE/PUBLISH stream — CLOSED Apr 28 23:31 UTC by alan-frindell as duplicate of #1313 (“Joining FETCH as a separate control message creates edge cases and feature gaps”, ianswett, Oct 15).
  • #1601 - Joining FETCH session errors are subject to a race condition — CLOSED Apr 29 00:03 UTC via PR #1609 merge.
  • #1598 - Why PUBLISH_OK not REQUEST_OK? — CLOSED Apr 29 00:04 UTC via PR #1611 merge.
  • #1593 - RFC: Allow framing single Objects without Subgroup ID — CLOSED unmerged Apr 29 17:29 UTC by ian-swett. The PR proposed an alternative OBJECT_STREAM type. Made OBE by Apr 27 interim disposition on PR #1608 + PR #1618 (FIRST_OBJECT bit) replacing it.
  • #1582 - Caching and propagation of REQUEST_ERRORs (Design)
  • #1581 - Request cancellation should be able to specify an error code (resolved — PR #1606 merged Apr 23)
  • #1578 - Bikeshed: Largest Object should be Next Object. Apr 23: ian-swett agrees with the rename.
  • #1550 - Properties Type collision between moq-16 and loc-01 (LOC issue #10; loc-02 also has collisions per alan-frindell Apr 16). CLOSED Apr 30 18:10:19 UTC via PR #1624 merge (alan-frindell) — provisional IANA registry for LOC properties now established.
  • #1534 - REDIRECT. Apr 23 editor call: Remove REDIRECT message from PR. Use GOAWAY on a bidi stream to mean what REDIRECT did. PR #1534 MERGED May 1 01:11:59 UTC by alan-frindell (+50/−1) — both REDIRECT-as-REQUEST_ERROR and standalone REDIRECT message landed.
  • #1507 - (Referenced by Session-Level Tracks extension point)
  • #1481 - Do we need a way to move / goaway for individual track (fluffy, Feb 9). Addressed by PR #1617 (Apr 28 16:21 UTC) and PR #1534 (merged May 1 01:11:59 UTC) — REDIRECT message now provides the per-track move mechanism.
  • #1476 - DELIVERY_TIMEOUT is both Track and Object extension. Apr 23: Victor asks whether zero→non-zero transition is permitted.
  • #1405 - Single Object Subgroups don’t need a Subgroup ID. PR #1608 CLOSED unmerged Apr 28 21:19 UTC (alan-frindell: “WG didn’t think this was the right approach, but agreed we need a way to know if a subgroup contains the beginning”). Replaced by PR #1618 (FIRST_OBJECT bit in SUBGROUP_HEADER type — adopts the alternative yuyou suggested on PR #1608 Apr 28 07:29 UTC). PR #1608 formally CLOSED May 1 18:35 UTC (afrind: “Closing in favor of 1618”).
  • #1386 - Can a publisher ‘lie’ about what Largest Object is? (ian-swett, Dec 7 2025). Addressed by PR #1621 (Apr 28 23:50 UTC).

Open PRs

  • PR #1625 (Apr 30 05:59 UTC, suhas-nandakumar, +132/−1) — Rebased and Update Security Considerations PR from Magnus Westerlund. Rebases and extends Magnus Westerlund’s long-parked Security Considerations PR (gloinul’s #1455). suhasHere requests gloinul review.
  • PR #1623 (Apr 30 01:38 UTC, ian-swett, +0/−10, label Handshake and Session) — Remove Request ID from GOAWAY (reverts #1559, fixes #1622). Pure-removal patch. Pairs with PR #1615 (RRID removal). Apr 30 18:31:57 UTC: alan-frindell pushed back on the parent issue: “trivial to put the request ID in goaway, and might be useful. If nothing else it can speed up retry when a new request is racing a GOAWAY.” PR now contested.
  • PR #1621 (Apr 28 23:50 UTC, alan-frindell, +8/−1) — Forbid relays from lying about LARGEST_OBJECT (fixes #1386). Body: “If we want to serve cached objects in response to SUBSCRIBE, lying is not the correct approach.” Closes a long-running 2025 design issue.
  • PR #1620 (Apr 28 23:25 UTC, alan-frindell, +2/0) — Clarify Joining FETCH is unaffected by fwd changing to 0 (fixes #1612). ian-swett APPROVED Apr 29 20:40:08 UTC. Two approvals → ready to merge.
  • PR #1618 (Apr 28 21:33 UTC, alan-frindell, +20/−10) — Add FIRST_OBJECT bit to SUBGROUP_HEADER type. The replacement for PR #1608’s approach. suhas-nandakumar APPROVED Apr 29 23:45:55 UTC: “Looks fine to me”. ianswett added Apr 29 20:48 UTC inline comment: “Ideally, I’d like this to be required, which was a perk of my other proposal to force the Subgroup ID==First Object ID”. Long Apr 29 20:54-22:04 UTC ianswett comments still re-litigating the closed PR #1608 approach — the Apr 29 list thread “Knowing the start of a Subgroup” (ianswett, afrind, Cullen) continues that debate. PR ready to merge despite parallel list dispute.
  • PR #1617 (Apr 28 16:21 UTC, alan-frindell, +85/−73) — Allow GOAWAY on request streams to migrate individual requests (fixes #1481). Body: “GOAWAY can now appear on a request stream using the same wire format as the control-stream form, but without Request ID (optional, present only on control stream). A client MUST send a zero-length URI in any GOAWAY.” ianswett comment Apr 29 18:10 UTC: “I think we should remove Request ID from GOAWAY entirely, since I don’t think it has much practical value.” — directly motivated PR #1623.
  • PR #1615 (Apr 27 19:48 UTC, ian-swett, +3/−52, label Control Messages) — Remove Required Request ID (fixes #1603). Body: “Removes ‘Required Request ID’. Does not remove Request ID, because it is used by Joining Fetch and GOAWAY.” Implements the Apr 27 interim decision. victor-vasiliev APPROVED. Unblocked by PR #1609 merge.
  • PR #1613 (Apr 23 23:10 UTC, alan-frindell, +30/0, label: Design) — Add MAX_REQUEST_UPDATES setup option and TOO_MANY_REQUEST_UPDATES error (references #1063). Adds per-stream flow control for REQUEST_UPDATE messages. Now superseded by the Apr 27 interim decision to remove RRID outright (PR #1615). PR #1613 was the alternative-frame to PR #1604’s structural reshuffling; both lose headline justification post-interim.
  • PR #1607 (Apr 18) - [Draft/RFC] Largest Available Group filter (Victor Vasiliev). Simpler alternative to REWIND: current group only, always serves complete group, no relay-side backfill. Coalescing point of the Apr 17–18 mailing-list convergence on LargestGroup/CurrentGroup. Apr 19: luke-curley left the first review — pushes back on the “MUST serve a complete group” / “full cache only” framing. Apr 23: suhas-nandakumar marked the PR CHANGES_REQUESTED (15:07 UTC, the first hard blocker). Apr 24 23:10 UTC: luke-curley returned with concrete defense — catalogs MUST use LargestGroup (NextGroup never resolves on dormant tracks), and Twitch’s TTV math shows 333 ms startup-time savings at the median (1s into 2s GoP, 1.5 Mb/s media on 3 Mb/s network → 0.66s vs 1s wait). Proposes the combined CurrentGroup + NGR race idiom. Separately, ian-swett’s Apr 23 inline “force Subgroup ID = first Object ID” suggestion has now been split into standalone PR #1608. See joining-fetch-dissent.
  • PR #1605 - Split DELIVERY_TIMEOUT into two types of timeout (Victor Vasiliev, Apr 14). OBJECT_DELIVERY_TIMEOUT replaces existing DELIVERY_TIMEOUT; new SUBGROUP_DELIVERY_TIMEOUT covers subgroups that have been fully queued but not yet fully delivered. Fixes #667 and #606. Apr 23 morning: First review by ian-swett — “this looks reasonable, but I don’t intuitively understand why two timeouts are necessary”. Apr 23 afternoon: alan-frindell added three suggestions — explicitly permit cancellation of retransmissions after delivery timeout, and evaluate delivery timeout “as late as possible” after internal queuing. On the Apr 27 interim agenda.
  • PR #1604 - Joining FETCH with subscription (implements #1602) — active review; Gwendal Simon (Apr 16) notes this is the subscriber-initiated sibling of the relay-initiated PUBLISH+catch-up pattern in SWITCH #1378. Apr 23: alan-frindell asked martin-duke for thoughts on Gwendal’s FETCH+PUBLISH_DONE race proposal; Martin replies that REQUEST_UPDATE ordering is a general problem and orthogonal to this PR. Apr 23 20:55 UTC: Martin added text that “killing SUBSCRIBE also kills the FETCH. I’m not sure how else to do it; there’s no other way to turn off the SUBSCRIBE.” Apr 23 20:57 UTC: Martin noted the PR “Now fixes #1612 as well”. Now mostly superseded by the Apr 27 interim decision to remove RRID outright (PR #1615) — Martin’s structural fix for RRID multiplication is no longer needed if RRID is gone. Branch is dirty (merge conflict with main).
  • PR #1591 - RFC: Add flow control for Subscriptions
  • PR #1588 - Add internationalization statement for moqt URI scheme
  • PR #1544 - Improve Startup Latency and 0-RTT (ian-swett, opened Mar 8, fixes #420 + #83). Apr 28 01:46 UTC: martin-thomson (former QUIC WG chair, IAB member) joined the review with a substantive rewrite of the introductory sentences. Apr 30 02:12-02:29 UTC: ian-swett posted 6 inline reply/suggestion comments working through Thomson’s rewrite — defends gRPC-style 0-RTT WT path, two suggestion patches around cache-expiry semantics for replayed objects, and a rewrite for resource-exhaustion mitigation (“Relays MAY defer initiating upstream subscriptions until the handshake is complete or reject 0-RTT entirely to mitigate resource exhaustion from replayed packets.”). Apr 30 03:25:32 UTC: Thomson asks “Do you have a specific response code that a relay could use so that the client can know that this was something that can be retried?” The 0-RTT review is now in serious dialogue.
  • PR #1378 - SWITCH for Client-Side ABR — relay-initiated PUBLISH + inline catch-up design; Apr 16 polish pass by Gwendal Simon. See switch-abr.

Recently Merged

  • PR #1542 - Split SUBSCRIBE_NAMESPACE into SUBSCRIBE_NAMESPACE and SUBSCRIBE_TRACKS (merged May 1 22:59:13 UTC by alan-frindell, +215/−135, fixes #1458). Replaces single SUBSCRIBE_NAMESPACE (0x11) with SUBSCRIBE_NAMESPACE (0x50) for namespace discovery + SUBSCRIBE_TRACKS (0x51) for track subscriptions. Removes SUBSCRIBE_NAMESPACE_OPTIONS + BOTH mode; adds TRACK_NAMESPACE_PREFIX (0x34) allowing REQUEST_UPDATE to change the prefix. Approvals: ianswett (Mar 9), vasilvv (Apr 27), suhas-nandakumar (final May 1 18:32:48 UTC on commit 4aa849a). Closes one of the longest-standing draft-17 design splits. Anticipated by moqtail PR #180 (zafergurel) which had already opened against the post-Apr-29 split design.
  • PR #1534 - Add REDIRECT for request errors and established subscriptions (merged May 1 01:11:59 UTC by alan-frindell, +50/−1). Lands both: (1) Redirect structure (Connect URI + Full Track Name) + REDIRECT error code on REQUEST_ERROR for new requests; (2) standalone REDIRECT message for established subscriptions/publications. Servers receiving a non-empty Connect URI MUST close with PROTOCOL_VIOLATION; non-empty Track Namespace + empty Connect URI is an internal track redirect. Closes Issue #1481 (per-track move). Two approvals: Vasilvv (Apr 27 23:01 UTC) + suhasHere (Apr 29 17:56:52 UTC).
  • PR #1624 - Add provisional registry for LOC properties (merged Apr 30 18:10:18 UTC by alan-frindell, +11/0, fixes #1550). suhas-nandakumar’s provisional IANA registry resolving the cross-draft 0x02/0x04 collision between MOQT and LOC. Closes the cross-draft #1550 saga without renumbering existing codepoints.
  • PR #1619 - Fix SUBSCRIBE_NAMESPACE response message name (merged Apr 29 20:44 UTC by ian-swett, +1/−1, fixes #1616). Editorial: response to SUBSCRIBE_NAMESPACE is NAMESPACE, not PUBLISH_NAMESPACE.
  • PR #1611 - Remove PUBLISH_OK message type, make it a REQUEST_OK alias (merged Apr 29 00:04 UTC by alan-frindell, +11/−30, fixes #1598). Wire-format change: removes the PUBLISH_OK code point; PUBLISH_OK becomes a textual shorthand consistent with the other REQUEST_OK aliases introduced in PR #1610. Approvals from ian-swett, @sharmafb (Suhas Sathyanarayana), @sandarsh.
  • PR #1609 - Joining Fetch forward state mismatch is a request error (merged Apr 29 00:03 UTC by alan-frindell, +3/−2, fixes #1601). Downgrades session-fatal forward-state mismatch (race between REQUEST_UPDATE fwd=1 and Joining FETCH on different streams) to a request error. Approvals from ian-swett, @sharmafb, @sandarsh. Unblocks PR #1615 (RRID removal) per afrind’s Apr 28 21:17 UTC comment.
  • PR #1586 - Make Object ID and Group ID delta encoded in Fetch responses (merged Apr 27 05:24 UTC by alan-frindell, +32/−23). Replaces inline raw IDs with delta encoding for FETCH responses. Final wording: “If the Group ID Delta field is present, the Object ID is the value of Object ID Delta if present. When the Group ID Delta field is not present, the Object ID is the prior Object’s ID plus the Object ID Delta if present.” — directly resolves the Apr 23 ambiguity afrind flagged. afrind’s last suggested-text patch landed Apr 27 05:23 UTC immediately before merging. Closes martin-duke’s long-running Issue #877 (“Pack the bits”) and Issue #1345 (“Separate the list of reasons for malformed tracks into two lists”, yekuiwang).
  • PR #1610 - Define textual aliases for REQUEST_OK by request type (merged Apr 23 21:03 UTC by alan-frindell, +22/−17). Editorial: introduces REQUEST_UPDATE_OK, TRACK_STATUS_OK, SUBSCRIBE_NAMESPACE_OK, PUBLISH_NAMESPACE_OK as shorthand for REQUEST_OK (in response to X). Unblocks PR #1611 (PUBLISH_OK removal).
  • PR #1606 - Generalize stream reset codes to all request streams (merged Apr 23 18:32 UTC by alan-frindell, fixes #1581). Adds GOING_AWAY (0x4), EXPIRED_AUTH_TOKEN (0x7), SESSION_CLOSED; aligns TOO_FAR_BEHIND / EXPIRED codes between stream-reset and PUBLISH_DONE registries. First merge to main since draft-17 publication.
  • PR #1596 - Exclude your own tracks from SUBSCRIBE_NAMESPACE (Apr 16, fixes #1585)
  • PR #1562 - RFC: Add Session-Level Tracks reserved namespace (merged Apr 16 by alan-frindell) — reserves .session namespace tuple[0] for transport-internal tracks; relays MUST NOT forward; unknown session tracks MUST be rejected with NOT_SUPPORTED; IANA registry established under Specification Required policy
  • PR #1490 - FILL_TIMEOUT parameter (Apr 14; subscriber’s max wait to fill a FETCH gap before Unknown; addresses part of #1023)

Previously Merged (Apr 9-10)

  • PR #1599 - Move normative text on Track Alias
  • PR #1597 - Consistently use MOQT for protocol references
  • PR #1595 - Allow 7-byte varint and non-minimal encodings
  • PR #1590 - Subscription filters are a Param
  • PR #1583 - Allow publisher to reopen subgroup after REQUEST_UPDATE fwd 0→1
  • PR #1577 - Clarify Joining Fetch ordering with Forward State transitions
  • PR #1576 - Add Track Properties to REQUEST_OK
  • PR #1540 - Allow coalescing REQUEST_UPDATE processing

ALPN Negotiation

Draft-17 uses ALPN strings for version negotiation:

  • qmux-00.moqt-17 - QUIC multiplexed, draft-17
  • qmux-00.moqt-16 - QUIC multiplexed, draft-16
  • qmux-00 with no suffix implies draft-14 (legacy)

Related Drafts

Mailing List

REWIND Consensus Call (Apr 16, 2026, deadline May 1)

Chair Magnus Westerlund opened a consensus call on how to proceed with draft-duke-moq-subscribe-rewind following interim-13. Three options on the ballot: (1) no action until MOQT is published, (2) adopt as an MOQT extension, (3) use as the basis for a PR to merge when editors decide. Deadline: May 1, 2026. The draft targets issues #861, #1039, #1358, #1362, and #1386 plus Boulder-meeting concerns about head-of-line blocking in Joining Fetch.

Over Apr 17–18 the thread pivoted toward a simpler LargestGroup / CurrentGroup / CurrentGroupFill filter (alan-frindell, luke-curley, victor-vasiliev all aligned). Alan backed Option 1 Apr 17 (FILL_TIMEOUT=0 already solves core HOL-blocking scenarios). Vasiliev captured the emerging direction as PR #1607 (Draft/RFC) — current group only, always complete group, no relay backfill.

Apr 27–28 thread re-eruption (9 messages over 19 hours during/after interim-14):

  • suhas-nandakumar Apr 27 06:55 UTC: REWIND can’t handle gapped caches.
  • luke-curley Apr 27 08:33 UTC: HTTP analogy — “imagine if HTTP operated based on the cache state… allowed to return a partial response with byte range 68-419”. Would only support REWIND if relays must attempt upstream retrieval.
  • Gwendal Simon (Synamedia) Apr 27 16:12 UTC: Acknowledges Luke’s PR #1378 feedback, commits to updating SWITCH with explicit cache-continuity condition.
  • martin-duke Apr 27 12:23 UTC: “The ‘best-effortness’ of REWIND is critical to the design, and is consistent with what I briefed in Boulder.”
  • luke-curley Apr 27 12:46 UTC: Reiterates HTTP analogy + supports PR #1607 instead.
  • martin-duke Apr 27 12:52 UTC compromise: “would you accept something that is still best-effort… but does not preclude the relay doing something more aggressive”.
  • luke-curley Apr 27 13:18 UTC agrees: “A relay MUST deliver objects within a sub-group in order (SUBSCRIBE semantics). Otherwise, the relay MUST skip the remainder of the sub-group.”
  • ian-swett Apr 28 02:03 UTC: “I’m open to some variant of REWIND, but not very optimistic that we’ll get consensus on anything more complex than CurrentGroupFill.” Endorses CurrentGroupFill (Alan’s sketched alternative); main rationale for more complex REWIND would be removing Joining Fetch entirely.
  • Gwendal Simon Apr 28 10:38 UTC: Pushes back on Joining FETCH removal as a live-streaming-deployment concern. Proposes proactive past-object delivery via SUBSCRIBE_OK parameter [Start_Group, Live_Edge).
  • Magnus Westerlund (chair) Apr 28 10:49 UTC: Asks for explicit ballot positions; design discussion has eclipsed the ballot itself.
  • ian-swett Apr 29 17:27 UTC: Clarifies his Apr 28 endorsement is “his individual position” as an editor, not an editorial mandate. Endorses “some variant of option 3” — afrind’s CurrentGroupFill specifically.
  • Cullen Fluffy Jennings Apr 29: Explicit option-#1 vote — “I support option #1. I do not think we should not take on Rewind until we have MoQT wrapped up. … when the working group eventually addresses this topic, we need to start with the use case we are trying solve.” First explicit option-#1 ballot vote on the list.

The May 1 deadline is in 1 day. With at least one explicit option-#1 vote (Cullen), strong option-3-with-CurrentGroupFill positions (Luke, Ian Swett individually), Martin Duke’s compromise-floor framing, and Gwendal’s live-streaming pushback on Joining FETCH removal, chair Magnus will need to interpret a split outcome. See joining-fetch-dissent and discussions-2026-04.

”Knowing the start of a Subgroup” Thread (Apr 29-30) — now 10+ messages

Re-litigates the Apr 27 interim disposition (PR #1608 closed, #1618 adopted) on the list. Apr 29 had 3 messages; Apr 30 added 7 more.

  • ian-swett Apr 29 (msg): Still prefers his closed PR #1608 (Subgroup ID == first Object ID) over PR #1618 (FIRST_OBJECT bit). Asks for community feedback.
  • alan-frindell Apr 29 (msg): Supports #1608. Asks Mo, Cullen, Magnus, Suhas to “provide examples of what would be broken by #1608”.
  • Cullen Fluffy Jennings Apr 30 ~14:00 MDT (msg): Three critiques of #1608 — favors explicit signaling (PR #1618 approach) over “pinning to very weird implicit signaling”.
  • Magnus Westerlund Apr 30 (msg): replies on examples broken by #1608.
  • ian-swett Apr 30 (msg): follow-up.
  • luke-curley Apr 30 ~10:41 PDT (msg): Critical of both proposals. Against #1608: “information is already implicit via FETCH or SUBSCRIBE object arrival order”. Against #1618: the bit “only helps REWIND (if there’s a fragmented cache) for the first object… not expensive to add, but you still need a plan to handle the rest of the gaps”. Proposes a third option: an 0-indexed counter per object within a subgroup so relays can detect cache gaps on a subgroup basis.
  • alan-frindell Apr 30 (msg).
  • suhas-nandakumar Apr 30 (msg).
  • luke-curley Apr 30 follow-up (msg).
  • Mo Zanaty Apr 30 22:06 UTC (msg): Endorses #1618 over #1608. Concrete AV1 example: temporal-only AV1 layering produces frame numbers (0-5) ≠ layer numbers (0,2,1,2,0,2), violating any rule tying Subgroup ID to first Object ID. Calls #1608 “a footgun for devs to screw up” — “even the working group fell into this trap”. On subgroup-vs-datagram tie-breaking: “subgroup wins”; datagrams at matching priority are “disposable and therefore lower priority”.

Net: list debate is now a three-way design discussion (#1608 / #1618 / Luke’s per-subgroup counter) on top of an already-APPROVED PR #1618. Cullen + Mo aligned with #1618; Magnus engaged but unclear; Luke proposing a fourth path. PR #1618 has not yet been merged despite approval.

Pre-interim mailing list activity now surfaced: Mo Zanaty Apr 27 14:43 UTC (msg): “1608 is a major change to the core data model that makes subgroups semantically meaningless”. afrind Apr 27 16:04 UTC reply asks for application examples that would break — establishing the burden-of-proof shift that ultimately led to PR #1608’s closure.

Consensus Call on draft-17 (March 2026)

martin-duke initiated a consensus call on the mailing list (2026-03-24) for draft-17 changes. The thread received responses through April 10, 2026.

7-Byte Varint Encoding Debate

Multi-week mailing list thread (March 19 - April 7) about the new varint format introduced in draft-17. The 7-byte encoding (6 leading ones) was initially marked invalid. Discussion among alan-frindell, martin-duke, suhas-nandakumar, Ian Swett, Mo Zanaty, and Christian Huitema led to PR #1595 allowing 7-byte varint and non-minimal encodings.

External Links

Graph View

DraftsConceptsPeopleImplDiscInteropOtherTags
DraftsConceptsPeopleImplDiscInteropOtherTags

Backlinks