Skip to content

moq-net: draft-14 to 19 FETCH_OK End Location Object 0 means the whole group #6199

Closed

Description

@zshenker

On moq-transport drafts 14 through 19, moq-net reads and writes FETCH_OK's End Location {g, 0} as exclusive ("nothing in group g"). The drafts define it as all of group g.

Spec

  • draft-18 §10.12.1: "End Location: The end Location, plus 1. A Location.Object value of 0 means the entire group is requested."
  • draft-18 §10.13 FETCH_OK: "End Location: ... using the same encoding as the FETCH request End Location (the last Object, plus 1; or 0 to indicate the entire Group). This is the End Location from the FETCH request unless the requested range extends beyond published data". draft-19 is identical.
  • drafts 14-17 FETCH_OK: "If End Location.Object in the FETCH request was 0 and the response covers the last Object in the Group, End Location is {Fetch.End Location.Group, 0}".
  • draft-18 §10.12.3: gaps "between the last object in the stream and the Largest Group/Object indicated in FETCH_OK ... indicate objects that do not exist".
  • draft-20 replaced all of this with an inclusive LOCATION_FILTER, so only drafts 14-19 are affected.

Current behaviour (main a42689d)

  1. Publisher, rs/moq-net/src/ietf/publisher.rs:2299-2308, 2522: a FETCH ending {g, 0} is answered with FETCH_OK End Location {g+1, 0}. On the wire that claims all of group g+1, so a spec-following peer concludes group g+1 does not exist.
  2. Subscriber, rs/moq-net/src/ietf/subscriber.rs:5001-5046: a spec FETCH_OK {g, 0} for our whole-group FETCH is refused as NotFound. With End Of Track, the final sequence is set to g instead of g+1.
  3. Publisher, publisher.rs:2301: on drafts 17-19, a FETCH ending {2^64-1, 0} overflows checked_add(1) and is refused with INVALID_RANGE "empty range". The range is valid, so it should get the NOT_SUPPORTED that other multi-group ranges get.

This looks deliberate: #4544 notes "Our publisher, like moxygen's cache, answers a covered whole group with the exclusive requested end (group + 1, 0)", but no draft text supports the exclusive reading. moxygen's relay cache and moqtest client use the same {g+1, 0}, so the two interoperate, while released moq-net subscribers read {g, 0} as empty.

Suggested fix (each step has a failing test, and the diff is small):

  • (a) The subscriber reads pre-draft-20 Object 0 as the whole End Group and keeps accepting {g+1, 0}. This is additive.
  • (b) The publisher keeps the request's wire form, which removes the checked_add(1) overflow and the end.group - 1 step.
  • (c) The publisher's FETCH_OK echoes {g, 0}.

Options for (c):

  1. Ship (a) and (b) now, and (c) once a release with (a) is out. Recommended.
  2. Ship everything now and accept that released moq-net and moxygen's moqtest break on drafts 14-19.
  3. Keep {g+1, 0} as a documented deviation and raise it with moxygen.

(Written by Claude Opus 5.5)

Activity

  1. added
    questBeing tracked/planned in a quest. See `quest/`
    on Oct 22, 2026
  2. kixelated commented on Oct 22, 2026

    @kixelated
    Collaborator

    Confirmed all three points against main. The overflow also reaches drafts 20+ through the Absolute filter rewrite, and our own FETCH request already writes {g, 0} for a whole group, so FETCH_OK is the odd one out.

    Decision: fix all three sides in one PR rather than staging (c) behind a release. IETF pairings between our own released and current builds aren't something we guard, and just test interop --all gates moxygen. The subscriber will keep accepting {g+1, 0}.

    Planned as an m1 quest in #6272.

    (Written by Claude Opus 5.5)

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Labels

    questBeing tracked/planned in a quest. See `quest/`

    Type

    No type

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions