[GLinnik21]

PMS names DTS dca (every DTS entry in the profiles shipped with PMS, the PMS API spec, the #584 server log). Our route check and the Load payload only knew dts, so a DTS track from a real server never direct-played, in Auto or in Force Direct Play — even on a set whose own codec table lists DTS. That is the Force Direct Play refusal reported in discussion #695. Nobody saw it because the synthetic fixture declares dts and tests/mock_pms.py echoed it back.

Stacked on #898; this diff is against that branch.

Change

• One canonicaliser, devcaps::audio_codec_id (lowercase, dca→dts, ec-3→eac3), used wherever codec names are compared: device-table membership and channel bounds (Caps::audio_has/audio_supports), the Load payload (engine::audio_payload_codec), and FFmpeg's stream matcher (ff.rs audio_name_matches, so a multi-track file can't pick TrueHD while the Load says DTS). Inbound data stays raw: track memory compares raw codec strings.
• The profile we send PMS spells DTS dca everywhere PMS reads a codec name (direct-play list, transcode target, add-limitation scopeName) through one profile_codec_name.
• Force Direct Play DTS rule (Caps::forced_admits_dts): DTS is offered only when the TV's table has a DTS row with a channel count — the same evidence Auto uses — but without the channel ceiling. Planner and forced profile use the same predicate. Before this, Forced offered DTS to every set; with the spelling fixed, that would send DTS to sets with no DTS decoder (dts_restore_plus measured a C5 whose table lists DTS without one). Consequence: a set patched with base dts_restore (which never edits the table) still gets a refusal in Forced — now one that names DTS (Force Direct Play: the audio refusal names the codec #898).
• DTS-HD MA 7.1: Auto still refuses it against a 6-channel ceiling (today's documented rule). We feed only the DTS core.
• The mock server speaks PMS's dca; tests/run.py maps the Load name back for its comparison.
• Test seam: devcaps::caps_on_this_thread_for_test (thread-local, test-support only, same pattern as i18n::language_on_this_thread_for_test).
• Docs: agent-reference, playback-settings, pms-api updated.

Verification

• Red first, then green: devcaps dca membership/channel bounds and the Forced row rule (compile-red); engine audio_payload_codec("dca") (left: None, right: Some(("DTS", 4))); auto_and_forced_agree_on_the_dca_spelling; ff.rs the_payload_codec_matches_ffmpegs_name_across_spellings; end-to-end a_dca_track_direct_plays_on_a_set_with_a_dts_row (Auto, DTS-row table, MDE directplay → /library/parts/ URL and a /decision request carrying dca, red with the dca arm disabled); transcoder profile tests red with profile_codec_name made the identity.
• Never seen red: forced_direct_play_refuses_dca_without_a_dts_row (Forced refused dca before too; guards the no-Load rule) and a_selected_dca_track_is_picked_when_the_table_has_a_dts_row (written after the fix).
• On the committed merged tree: make check ok (plx_media 1699 passed; 0 failed; 64 ignored), make test-hostsim C=plx_media 1853 passed; 0 failed; 64 ignored, make check-nodef clean. Firmware-compat review: no FFI, link or DT_NEEDED change (source-level; ARM build left to CI).

TV check (dev set, webOS 4.5, debug build of this branch, guest on tests/mock_pms.py, panel off, muted)

The mock served the synthetic H.264 + DTS 5.1 file with codec="dca" (as PMS does). The event log:

• devcaps: … audio=aac,ac3,eac3,dts and pms: server 0 version=…-synthetic
• decision: part=directplay video=copy general=Some(1000) mde=Some(1000) -> DIRECT PLAY
• load: v=H264 a="DTS" …, pipeline str=audio/x-dts
• clock advancing at 1x through pos=20s (play≈1000pm, timeline playing t=18s/60s), no playing error.

Not verified:

• Audible output (muted run).
• That a DTS Load a set cannot decode ends in a terminal error rather than PlxNative (LG webOS): playback never starts on files with DTS-HD MA audio — server chooses Direct Play, player hangs #584's spinner — the dev set decodes DTS, so this needs a debug trigger that forces such a Load; follow-up.
• Whether PMS also accepts dts in a profile is untested and now moot: we send dca.

🤖 Generated with Claude Code