[t0ms]

Stacked on #5306, which it targets until that lands.

Problem

A padded export ts could not carry a CBR contribution feed at the feed's own multiplex rate when
the feed fills that rate with no null packets (#5362). The export spent packets the source does not:

• Video filling its level's leak was admitted below Rbx. The schedule charged it whole 188-byte
  packets at the lesser of Rx and Rbx, plus the clock packet, and MB held only a transport buffer's
  worth ahead of the leak.
• Every 25 ms slot spent a packet on a clock of its own, where the source carries its PCR in the
  video's adaptation field.
• Every audio frame took a PES of its own, where the source packs several.

At the source's rate the export fell behind until its lead ran out, and it stopped on missed a
decode deadline on the video PID.

Approach

One commit per step, each retiring its quest:

• fix(ts): admit leak-bound video at its full Rbx (quest/m1/tstd/video-leak.md). TB drains
  whole packets at Rx, the clock packet among them, and MB passes elementary bytes on into EB at
  Rbx. An unpadded slot carries the lesser of the two, with fractions carried from slot to slot and
  no one-packet slack.
• fix(ts): let video's MB run BSmux ahead of its leak. MB may hold BSmux ahead of the leak
  (2.14.3.1), so audio taking a position between video packets no longer idles it.
• fix(ts): count audio PES headers in their decoder buffer. Only video's MB discards the PES
  header before EB. For audio and system PIDs, B holds it until the access unit after it is removed
  (ISO 13818-1 2.4.2.3). This reverses a5640abee's accounting for PIDs without a leak and keeps
  video's. A 2,592-byte frame no longer fits a 2,592-byte B, so that test now pushes the largest
  frame that fits with its 14-byte header.
• feat(ts): carry the PCR in the clock PID and pack audio frames per PES
  (quest/m1/tstd/mux-overhead.md).
  • Each slot's PCR rides the adaptation field of the PCR PID's first packet in the slot, reflowing
    that PES's payload. A slot without one keeps its clock packet.
  • With a --delay, consecutive frames of every audio codec share a PES while they run on without
    a gap and fit B with their header. The codecs are AAC, MP2, AC-3, E-AC-3, and AC-3 in a private
    stream. Each frame leaves B at its own decode time.
  • A PES starts wherever a frame's place on the media timeline is a multiple of the frames B
    holds, so exporters that joined at different frames cut the same PESs.

Tested with:

• just test ts --leak: 1080p with no HRD at level 4.0's 24 Mb/s leak round-trips the whole 30 s
  clip. Before, the export stopped 12.7 s in.
• just test ts --full, new, run in the interop workflow. The --leak video beside MP2 and DVB
  AC-3, muxed with its PCR in the video PID and several audio frames per PES, with the nulls
  dropped, exports at its own 24.75 Mb/s. It covers the whole 30 s and the strict T-STD check
  passes. The new overhead.py counts 8,026 packets spent on clocks and audio framing, against
  the source's 8,165.
• just test ts, --hrd, and --headroom pass.
• A 280 s replay of a 25 Mb/s CBR contribution feed with no nulls. On main at b581c27ed, the
  export stops 31 s in on the video PID. With the first two commits it runs through at 26 Mb/s,
  but stops 185 s in at 25 Mb/s. With all four it runs through at the source's own 25 Mb/s. The
  strict T-STD check, compliance.py and PCR accuracy all pass, and every frame of every track
  is the source's.
• just check on macOS with host tools passes, apart from the moq-tokio flake below and the steps
  whose tools are missing here: the js module (no wasm-bindgen), nix flake check, and the Swift
  step (no Xcode).

Impact

• No public API change. Schedule is private to moq-mux.
• Wire: padded export ts output carries each slot's PCR in the PCR PID's adaptation field
  whenever that PID has a packet in the slot.
• Wire: with a --delay, an exported audio PES carries several frames, up to its B. With no
  delay, it carries one frame, as before.
• An audio frame that fills B with no room for its PES header is now refused.
• Adds just test ts --full and its CI step, and test/ts/overhead.py.

Alternatives

• Keep audio PES headers out of B, as a5640abee does. Then PESs packed up to B overflow it by
  their header, which compliance.py's strict check reports on --full.
• Put the PCR in a stuffed adaptation field of a packet of its own. That costs as much as the clock
  packet. Reflowing a PES the slot already carries costs at most one more packet, and only when its
  last packet has no room left.
• Pack a fixed number of frames per PES. The number that fits depends on the codec and bitrate,
  so the PES size follows B instead.

Follow-ups

• A unit may go out its window, which is --delay, plus a slot and its drain ahead of its decode.
  So at --delay 1s or more, the first audio PES after startup can sit in B for longer than the
  T-STD's 1 s. --full runs at 900 ms for this reason. This belongs with the send-ahead quest.
• With per-frame removal, a PES could exceed B, provided its frames never overflow B as they
  arrive. On the 25 Mb/s feed above, the source packs about 9.3 AC-3 frames per PES, while B caps
  the export at seven. So the export spends 2.5 % more packets than the source on AC-3, and 1.0 %
  more on MP2. It still fits, since it repeats PAT and PMT less often than the source.
• moq-tokio's a_refused_tcp_dial_keeps_retrying flakes under load. Within its 200 ms budget it
  sometimes sees timed out connecting instead of Connection refused. It is unrelated to this PR.

(Written by Claude Opus 5.5)

---

^{Need help on this PR? Tag @codesmith-bot with what you need.}