From c6d4612ca2c9576694715745d05395228b5a7c46 Mon Sep 17 00:00:00 2001 From: Luke Curley Date: Sat, 24 Oct 2026 23:36:26 -0700 Subject: [PATCH 1/2] chore(quest): claim quest/m1/fmp4-edit-lists.md Quest-Claim: 4c096b214c9bafdc From c15d22f9ece10fdd0d699df6f70ec4312fc260aa Mon Sep 17 00:00:00 2001 From: Luke Curley Date: Sat, 24 Oct 2026 23:48:19 -0700 Subject: [PATCH 2/2] test(interop): hold the AAC tone to its phase The fMP4 importer now applies the edit list holding ffmpeg's 1024 AAC priming samples, so --phase covers the AAC rendition too. GStreamer's AAC stays unchecked: faad, which decodebin picks, drops the first frame's output and stamps the rest a frame early, a decoder delay moqsrc cannot know to compensate for. Co-Authored-By: Claude Opus 5.5 --- quest/m1/README.md | 1 - quest/m1/fmp4-edit-lists.md | 30 ------------------------------ test/interop/README.md | 11 +++++------ test/interop/interop.sh | 8 ++++---- 4 files changed, 9 insertions(+), 41 deletions(-) delete mode 100644 quest/m1/fmp4-edit-lists.md diff --git a/quest/m1/README.md b/quest/m1/README.md index e54cac761a4..4c4eb935aed 100644 --- a/quest/m1/README.md +++ b/quest/m1/README.md @@ -122,7 +122,6 @@ Catalog HLS field. - [Moxygen minimal interop](/quest/m1/smoke-moxygen.md) - waiting on Moxygen to send a registered filter type, then the smoke lane stops hiding its failure - [More tests under load](/quest/m1/test-flakes-2/README.md) - the second round of load-only failures, one quest per flake, fixed at the cause - [Broadcast lifecycle](/quest/m1/broadcast-lifecycle/README.md) - broadcasts are online or offline with no finished state in moq-net, finished-broadcast retention is a fixed moq-mux policy, and each linger timer has one meaning -- [fMP4 edit lists](/quest/m1/fmp4-edit-lists.md) - the fMP4 importer applies edit lists to frame timestamps, so the interop AAC tone passes the phase check - [fMP4 passthrough sync start](/quest/m1/fmp4-passthrough-sync.md) - a passthrough video import that starts mid-GOP publishes nothing before its first sync sample - [A smaller, safer API than release](/quest/m1/api-audit/README.md) - every surface added since release has a consumer or is gone, overlapping operations are collapsed, names mirror across Rust, JS, and the bindings, and every break says how to migrate - [moq-transport 18 and 22 only](/quest/m1/ietf-drop/README.md) - drafts 14-17 and 19-21 are deleted from Rust, JS, tests, interop lanes, and docs, removing the legacy control-stream adapter and its request limit diff --git a/quest/m1/fmp4-edit-lists.md b/quest/m1/fmp4-edit-lists.md deleted file mode 100644 index 17b33af0f89..00000000000 --- a/quest/m1/fmp4-edit-lists.md +++ /dev/null @@ -1,30 +0,0 @@ -# [S] Interop holds the AAC tone to its phase - -## Goal - -The interop AAC rendition passes `tone.py --phase` like Opus does, now that -the fMP4 importer applies the edit list holding ffmpeg's 1024 AAC priming -samples to the frame timestamps. - -## Plan - -Drop the `opus` condition from `phase_checked` in `test/interop/interop.sh`, -with the comment's AAC sentence, and the matching sentence in -`test/interop/README.md`. Both sit beside lines -[moq-dev/moq#6647](https://github.com/moq-dev/moq/pull/6647) rewrites, so -land this after it. - -With the condition dropped, `just test interop --all --publishers rust` -passes the AAC phase on every phase-checked subscriber except GStreamer, -which reads it 1024 samples early (`+878us`). The cause is its decoder, not -the timestamps: `faad`, which `decodebin` picks there, drops the first -frame's output and stamps the rest one frame early, while ffmpeg's decoder -keeps the priming at the input timestamps (an ffmpeg ADTS encode decodes -with its first audible sample at 2 through `faad` and at 1026 through -ffmpeg; `fdkaacdec` lands later still). Decide whether `moqsrc` compensates -for that decoder delay or the check skips GStreamer's AAC with the reason, -as it skips the native-JS subscriber. - -## Related - -- [MP4 export](/quest/m2/mp4-export.md) - its open "edit lists" item is the output side diff --git a/test/interop/README.md b/test/interop/README.md index 3e0ad28f65e..48f124b2de7 100644 --- a/test/interop/README.md +++ b/test/interop/README.md @@ -35,13 +35,12 @@ Go, and C++, `moq-c` for C, `decodebin` on each `audio_N` pad for GStreamer, and publisher's channel count, contiguous timestamps, and each channel's tone carrying most of its energy by a Goertzel filter. A garbled decode (wrong channel order or sample rate), silence, a timestamp gap, or a rendition left undecoded fails the cell. -Each publisher's tone starts at timestamp zero, so `--phase` also holds every Opus +Each publisher's tone starts at timestamp zero, so `--phase` also holds every rendition's tone to its timestamps within a sample, which fails a decoder that keeps -or drops the wrong pre-skip. It skips the browser publisher, which stamps its file -from its own page clock, and the native-JS subscriber, which leaves the pre-skip to the -player as WebCodecs does and stamps milliseconds. AAC is skipped until the fMP4 -importer applies the edit list holding ffmpeg's priming -([quest](/quest/m1/fmp4-edit-lists.md)). The browser plays each rendition in a page +or drops the wrong Opus pre-skip or AAC priming. It skips the browser publisher, which stamps its file +from its own page clock, the native-JS subscriber, which leaves the pre-skip to the +player as WebCodecs does and stamps milliseconds, and GStreamer's AAC, whose `faad` +decoder reads the priming a frame early. The browser plays each rendition in a page of its own, the player pinned to it by `controls.audio.target`, and runs the same filter on each channel of the player's audio graph. A negative control then has the Rust CLI swap its channels' tones, and every subscriber must refuse every rendition diff --git a/test/interop/interop.sh b/test/interop/interop.sh index b3d392c568a..d3cff0503d4 100755 --- a/test/interop/interop.sh +++ b/test/interop/interop.sh @@ -368,12 +368,12 @@ publisher_codecs() { # decode's timestamps, which fails a misapplied pre-skip. Every publisher's tone starts at # timestamp zero except the browser's, which stamps its file from its own page clock. The native-JS # subscriber hands WebCodecs raw Opus without the OpusHead, as the player does, which keeps the -# pre-skip for the player to trim, and its libav fallback stamps milliseconds. AAC waits on the -# fMP4 importer applying the edit list that carries ffmpeg's 1024 priming samples -# (quest/m1/fmp4-edit-lists.md). +# pre-skip for the player to trim, and its libav fallback stamps milliseconds. GStreamer's AAC is +# skipped for its decoder: `faad`, which decodebin picks, drops the first frame's output and stamps +# the rest a frame early, so it reads the priming 1024 samples early whatever the timestamps say. # shellcheck disable=SC2329 # reached from run_cell, which 'harness_spawn' invokes phase_checked() { - [[ "$1" != js* && "$2" != js-native-* && "$3" == opus ]] + [[ "$1" != js* && "$2" != js-native-* && ("$2" != gst || "$3" == opus) ]] } # Record a case for the CI report: `record case=C [field=value...]` (see