Repository navigation
macOS Plex Original: retained Dolby Vision P7 configuration fails decoder initialization #316
Description
Scope
Native macOS 27 beta: a Plex Original/video-copy stream carrying Dolby Vision Profile 7 configuration fails decoder initialization. This is not a blanket HDR failure and does not establish failure on other Apple platforms.
Reproduction and controls
- Source: 4K HEVC Main10, MKV, Dolby Vision P7 with HDR10-compatible base layer; TrueHD audio.
- Plex decision: video copy, audio transcode; initialization/first-byte prewarm succeeds.
- Delivered fMP4:
hvc1, Main10hvcC,colr/nclxprimaries 9 / transfer 16 / matrix 9, retaineddvcCprofile 7. - Decoder log:
kVTCouldNotFindVideoDecoderErr(-12906), thenkVTVideoDecoderUnsupportedDataFormatErr(-12910), surfaced as AVFoundation undecodable media (-11855). - Reproduces with HDR enabled through the master and HDR disabled through the existing bounded single-variant child selector.
- A distinct-source HEVC/PQ/BT.2020 HDR10 control (without DV configuration) plays on the same native host with both display settings; decoded frames were inspected.
- Explicit capped video transcode of the failing source visibly plays. This does not prove HDR output or preservation of DV metadata.
- Original correctly requests encoding consent on failure; declining does not start an encoder.
Candidate scope / open questions
Investigate compatible base-layer delivery/normalization rather than weakening the P5 guard, removing color metadata blindly, or silently transcoding. No broad P7 policy change is justified by this one source/packaging result.
Stats currently predicts HDR10 fallback from source/copy intent before any runtime signal is observed. A narrow candidate separates unverified requests from observed stream signaling; neither is measured HDR light output.
Physical iPad/iPhone deployment is currently blocked by device preparation/availability; Vision Pro unavailable; no physical tvOS device available. HDR luminance/colorimetry and other DV profiles remain open. Private identifiers, media, logs and frames are not included here.
Activity
jlipworth commentedon Sep 14, 2026
Additional bounded native-Mac controls (same macOS beta host):
Candidate diagnostic wording now distinguishes observed stream signaling from unverified copy/encode intent. Physical-device coverage remains gated; this issue is not resolved.
jlipworth commentedon Sep 14, 2026
Evidence correction: the native HDR10 control used the
originalprobe scenario, which verifies playback progression/hold but does not seek even if a seek-time argument is present. Its decoded PQ/BT.2020 result stands, but the earlier native-control “seek passed” wording should be read as “hold passed.” The explicitly encoded DV control used the actualseekscenario. Initial/postseek capture directory names alone are not proof of a seek.Separate iPhone/iPad simulator HDR10 controls now pass the actual zero-bitrate
seekscenario with position snapshots and inspected post-seek frames. These do not establish native-Mac seek coverage or physical HDR output. DV decoder work remains separate.jlipworth commentedon Sep 14, 2026
Investigation checkpoint (local, not published as a playback fix):
dvcCentry from the extension-atom dictionary permits decoder creation, preserving subtype, dimensions,hvcCand other extensions. No compressed samples are submitted by this experiment; it does not prove safe metadata removal, accurate decoding or normalized HLS playback.sustainedOfflineInitialWait, two expectations); its focused 10-test suite passed. No all-green hosted claim is made.Temporary acquisition code was reverted and the isolated staged app removed; private evidence and the isolated app session are preserved. P5 safety, Generic profile and explicit encoding consent remain unchanged. Investigation paused at the unlock/sample-acquisition gate; issue remains open.
jlipworth commentedon Sep 14, 2026
Exact-source sample and local-HLS follow-up (still no shipping transformation):
dvcCfrom the format description. A second run additionally removed only RPU/EL NAL types 62/63, retaining base VCL bytes, other NALs and timing. Both runs produced identical timestamp-matched raw 10-bit pixel hashes for every frame (193 unique), with PQ/BT.2020 attachments; visible late frames were inspected.dvcCreplaced by a same-sizefreeatom played the unchanged segment and sought from 5 seconds to beyond 7 seconds with an inspected visible frame. An explicit base-only experimental fragment also played/sought; reopening the configuration-only HLS repeated success.This establishes an HDR10-base delivery path for this short segment on this native host, not full Dolby Vision or enhancement-layer reconstruction, measured HDR output, general P7 support, or safe production normalization. MEL/FEL characterization, later content, live Plex multi-segment/range/seek/restart integration and hardened malformed-input regression coverage remain open. The local one-segment seek is not a server-seek test.
Temporary acquisition changes were reverted; owned app/window/server resources cleaned up and private evidence preserved. No P5/Generic/consent changes, server configuration edits, source-library rewrites or release changes. Fresh hermetic/tooling/docs checks pass; the previously reported full native hosted timing-test failure remains a non-green gate. Issue remains open.
jlipworth commentedon Sep 14, 2026
HDR diagnostics/evidence PR #317 is merged and reachable on
origin/mainat7ecdf6b1.This landed honest stream-versus-display diagnostics, bounded probe evidence and the documented Plex HDR10/HDR10+-source simulator results—not a DV P7 normalization fix. Final PR CI passed docs, hygiene and PMSKit; native/simulator proof levels and the native Original hold-only correction remain documented. The investigation stays active in
docs/research/native-hdr-validation.md.Issue #316 remains open for the separate DV implementation and acceptance work. Physical HDR presentation, mixed-display behavior and HDR10+ dynamic-metadata processing are not established by this merge; equivalent Jellyfin/Emby live acceptance is also untested here. No DV candidate, P5 guard relaxation or implicit encoding change was included.
Investigate DV delivery and validate cross-backend playback (1.7.1) (#…
jlipworth commentedon Sep 15, 2026
PR #318 is merged and reachable on
origin/mainat33318438(coordinated 1.7.1 build 1).This ships the reviewed partial-support scope: narrowly verified Jellyfin copy-HLS waiting behavior and diagnostic/probe improvements. The P7 HDR10-base experiment remains default-off and DEBUG macOS only; Release does not enable it. Generic Plex profile, P5 guards and explicit video-encoding consent remain unchanged.
Exact reviewed-tree validation: full macOS 637 tests twice, iOS 621, tvOS 339 with normal runner exit; hermetic 1,707 Swift Testing plus 119 XCTest; fresh iPhone/TV semantic fixtures with inspected attachments; all five premerge CI contexts passed. Historical intermittent runner/deadline failures remain documented rather than retroactively waived. Research stays active in
docs/research/dv-p7-decoder-boundary.md; full DV reconstruction, physical HDR presentation and broader profile acceptance are not established by this merge. Physical-device installation is being verified separately and is not playback acceptance.Preserving the issue's existing closed state; this comment does not claim universal DV support or an App Store/TestFlight release.
Harden experimental P7 fallback initialization identity (#320)
jlipworth commentedon Sep 15, 2026
PR #320 is merged and reachable on
origin/mainat4ed709a4. The published tree matches the reviewed PR head exactly.Landed scope: per-open initialization-representation pinning/rejection, deterministic replacement-wait coverage, and the isolated deletion-persistence test prerequisite. The P7 HDR10-base candidate remains default-off, DEBUG macOS only; Release, Plex Generic, P5 guards, and explicit video-encoding consent are unchanged.
All five PR/push CI contexts passed before merge. Exact-revision native/package/simulator results and their limitations remain recorded in the PR and active
docs/research/dv-p7-decoder-boundary.md; this closeout does not turn historical failures or blocked runs into passes. Fresh closeout documentation/hygiene checks passed.Remaining research/acceptance: exact-source Plex rebind and fresh reopen (saved binding returned 404); Jellyfin sampling-gap investigation (no post-hold pass); missing canonical visionOS golden-simulator pointer; broader source/later-segment equivalence and physical HDR/DV/audio validation. No production enablement or App Store upload occurred.
Preserving the issue's existing closed state for the reviewed partial-support scope, not claiming universal Dolby Vision support. Research remains active rather than archived.