music-assistant-server

1"""Constants for the AirPlay provider."""
2
3from __future__ import annotations
4
5from dataclasses import replace
6from enum import IntEnum, StrEnum
7from typing import Final
8
9from music_assistant_models.enums import ContentType, PlayerFeature
10from music_assistant_models.media_items import AudioFormat
11
12from music_assistant.constants import CONF_ENTRY_SYNC_ADJUST
13
14DOMAIN = "airplay"
15
16
17class StreamingProtocol(IntEnum):
18    """AirPlay streaming protocol versions."""
19
20    RAOP = 1  # AirPlay 1 (RAOP)
21    AIRPLAY2 = 2  # AirPlay 2
22
23
24class AirPlayRemoteCommand(StrEnum):
25    """Transport commands received from an AirPlay receiver."""
26
27    PLAY = "play"
28    PAUSE = "pause"
29    PLAY_PAUSE = "play_pause"
30    NEXT = "next"
31    PREVIOUS = "previous"
32
33
34class ClockReadiness(StrEnum):
35    """
36    How a receiver's clock readiness resolved for an anchor decision.
37
38    Only PROJECTED carries an instant; the rest all mean "anchor on the lead
39    alone", but for very different reasons - one is a device that will not play
40    at all, and treating them alike hides it.
41    """
42
43    # The binary projected when the receiver's clock becomes usable.
44    PROJECTED = "projected"
45    # NTP timing: there is no receiver clock to wait for.
46    NOT_APPLICABLE = "not_applicable"
47    # The receiver never answered our PTP clock and will render silence.
48    STALLED = "stalled"
49    # Nothing arrived within the wait: a slow device (retryable) or a receiver
50    # whose readiness went unreported.
51    UNREPORTED = "unreported"
52
53
54CONF_PASSWORD: Final[str] = "password"
55# Storage-only marker (no config entry) set when the device rejected the stored
56# password, so the player keeps asking for setup across restarts until a working
57# password is entered.
58CONF_PASSWORD_INVALID: Final[str] = "password_invalid"
59# Provider marker that the stored password verdicts were reviewed once. Releases
60# that could not tell a password challenge apart from a flat refusal wrote the
61# key above for both, so what they left behind is no evidence about a password
62# and is dropped a single time; a device that really challenges marks itself
63# again on its next connect.
64CONF_PASSWORD_MARKERS_REVIEWED: Final[str] = "password_markers_reviewed"
65CONF_IGNORE_VOLUME: Final[str] = "ignore_volume"
66CONF_ENCRYPTION: Final[str] = "encryption"
67# Advanced per-device streaming mode: pins the protocol/timing lane for
68# receivers whose automatic route misbehaves. Options are offered per device
69# capability; Automatic is the default and the only value MA itself writes away
70# from: a receiver that never answers PTP is switched to NTP, while one whose
71# native control channel conclusively fails is switched to compatibility mode.
72CONF_STREAMING_MODE: Final[str] = "streaming_mode"
73STREAMING_MODE_AUTO: Final[str] = "auto"
74STREAMING_MODE_AP2_PTP: Final[str] = "ap2_ptp"
75STREAMING_MODE_AP2_NTP: Final[str] = "ap2_ntp"
76STREAMING_MODE_AP2_COMPAT: Final[str] = "ap2_compat"
77STREAMING_MODE_RAOP: Final[str] = "raop"
78CONF_STORED_VOLUME: Final[str] = "stored_volume"
79CONF_COMPANION_CREDENTIALS: Final[str] = "companion_credentials"
80CONF_MRP_CREDENTIALS: Final[str] = "mrp_credentials"
81CONF_NATIVE_MRP_CREDENTIALS: Final[str] = "native_mrp_credentials"
82
83# Bundle id of the Music Assistant tvOS dashboard app, launched over Companion on
84# eligible Apple TVs (see tvos/docs/launch-contract.md).
85TVOS_APP_BUNDLE_ID: Final[str] = "io.music-assistant.tvos"
86
87AIRPLAY_DISCOVERY_TYPE: Final[str] = "_airplay._tcp.local."
88COMPANION_DISCOVERY_TYPE: Final[str] = "_companion-link._tcp.local."
89MRP_DISCOVERY_TYPE: Final[str] = "_mediaremotetv._tcp.local."
90RAOP_DISCOVERY_TYPE: Final[str] = "_raop._tcp.local."
91DACP_DISCOVERY_TYPE: Final[str] = "_dacp._tcp.local."
92
93# Floor for a late joiner's anchor, and the one it rests on whenever the binary
94# reports no readiness projection ([STATUS] clock_ready). A joiner cannot
95# honour an instant in the past, and that second case has no device evidence at
96# all, so this carries it the whole way: the command's trip to the binary, the
97# binary's own 250 ms floor, and the receiver seating the anchor. The value is
98# field-proven, not derived from those.
99# It is NOT headroom for the binary's post-commit clock verification, which
100# arms only when the receiver has still not probed by the time it reads the
101# START, and only for an anchor that clears the receiver queue depth plus
102# 500 ms - past ~2 s of effective depth, an anchor this close leaves no room.
103AIRPLAY_LATE_JOIN_MIN_HEADROOM_MS: Final[int] = 2500
104# How long a group start, a join or the Sendspin bridge waits for the binary's
105# first receiver-clock projection ([STATUS] clock_ready with state=probing|
106# ready). The binary emits its first line right after [STATUS] connected and
107# refreshes it about every 250 ms, and the projection exists from the receiver's
108# first probe (~1078 ms after connect on a cold device), so this only has to
109# cover a slower device before giving up and anchoring on the lead alone.
110# Independent of the binary's own stall report (state=stalled), which is a
111# slower, higher-confidence diagnosis of a receiver that never answers at all: a
112# wait that times out here has simply run out of planning time and falls back,
113# while a stall says the speaker will not play. Keep them apart - tightening the
114# stall report to meet this deadline would trade the margin that keeps it free
115# of false alarms.
116AIRPLAY_CLOCK_READY_TIMEOUT_MS: Final[int] = 2500
117# Lead added on top of a reported readiness instant, wherever one is waited for.
118# The binary refuses to place an anchor inside its own 250 ms floor, measured
119# from when IT reads the START command rather than when the server sends it (the
120# same trap as AIRPLAY_START_LEAD_MS), so the lead carries that floor plus 250 ms
121# for the command reaching the binary and for the convergence error of a
122# projection made from the receiver's very first probe.
123AIRPLAY_CLOCK_READY_LEAD_MS: Final[int] = 500
124# Default receiver buffer depth per device family: (manufacturer wildcard,
125# model wildcard, firmware wildcard) -> depth in ms, matched case-insensitively
126# in order, first match wins; unmatched devices stay on Automatic (the binary's
127# stock depth). The table is EMPTY since the buffered (type 103) stream became
128# the auto-route for the receivers that used to need a deepened queue: the
129# LinkPlay pipelines that starved on the realtime stream (WiiM at 1750 ms,
130# Edifier MS50A silent below 2500 ms) manage their own buffer on the buffered
131# stream and play fine on Automatic. The per-player depth setting remains as
132# an advanced override; extend the table only for devices that starve on the
133# route they actually take.
134AIRPLAY_BUFFER_DEPTH_DEFAULTS: Final[tuple[tuple[str, str, str, int], ...]] = ()
135# Per-player override of the splice receiver-queue depth in ms (0 = automatic).
136CONF_BUFFER_DEPTH: Final[str] = "buffer_depth"
137# How long a plain (non-join) START waits for the binary's [STATUS] started ack.
138# A strict buffered receiver can reject its anchor until its clock is seated;
139# cliairplay then makes up to 12 attempts, 500 ms apart. Cover that 5.5-second
140# retry span plus control-response and status-delivery margin.
141AIRPLAY_START_ACK_TIMEOUT_MS: Final[int] = 7000
142# A join can additionally withhold its ack while receiver-clock verification
143# settles. Keep its independently named bound aligned with the buffered retry
144# span too; command failures still answer either wait immediately.
145AIRPLAY_JOIN_START_ACK_TIMEOUT_MS: Final[int] = 7000
146# How far the content a corrected anchor actually cut may fall short of the cut
147# it asked for before the reported media position is re-based and the shortfall
148# reported. The binary derives the cut it took from the bytes it discarded, so a
149# few ms of byte quantization is expected and correcting for it would only
150# jitter the base; anything larger means the cut really did end early (the input
151# ran out inside it, or a teardown settled it) and the position is over-advanced
152# by that much for the rest of the anchor.
153AIRPLAY_CONTENT_CUT_TOLERANCE_MS: Final[int] = 20
154# Anchor leads for a readiness-confirmed START (cold and warm alike), solo and
155# group: the session only anchors after the binary confirmed the connection
156# ([STATUS] connected) and the new audio flowing ([STATUS] audio), so a lead no
157# longer guesses at setup or transcoder spin-up time. Both cover the receiver
158# re-anchor (accepted down to ~150 ms in the flush-ladder measurements) plus
159# the command's trip down the pipe; the group one also covers fanning the
160# shared instant out to every member. The pipe margin is what keeps them
161# workable: the binary rejects any instant inside its own 250 ms floor,
162# measured from when IT reads the command, and corrects a miss to that floor
163# plus another full lead — so a lead sitting exactly on the floor misses by the
164# delivery time every time and turns a start into a group-wide re-anchor ladder.
165AIRPLAY_START_LEAD_MS: Final[int] = 400
166AIRPLAY_GROUP_START_LEAD_MS: Final[int] = 500
167# Cold GROUP starts anchor further out: a receiver on a brand-new session
168# still acquires its PTP slave lock (~1.7-2.3 s measured on Sonos) and cannot
169# render at an anchor inside that window - it starts late and the group opens
170# audibly out of sync. Warm re-anchors reuse a locked clock and keep the
171# short leads above; solo cold starts have no sync partner to miss.
172AIRPLAY_COLD_GROUP_START_LEAD_MS: Final[int] = 2500
173# Margin added on top of a member's reported warm lead (the splice-timeline
174# queue depth; that timeline is the default for every native AirPlay 2 session)
175# when anchoring a warm re-start: covers the command round-trips between the
176# flush acks and the shared START so every member's skip target lands beyond
177# its queued audio.
178AIRPLAY_SPLICE_LEAD_MARGIN_MS: Final[int] = 150
179
180# Floor for the late-join PCM ring, which has to hold every sample between the
181# audible position and the write head: a joiner's anchor maps onto content the
182# group was already fed but has not played yet. That distance - the write-head
183# lead - is the sum of every buffer between the session's byte counter and the
184# speaker, and it is far larger than the binary's own ring alone:
185#   ~5.7 s  the per-member ffmpeg and the pipes around it (measured 5.2 s
186#           steady / 5.7 s peak at 44.1 kHz/16-bit, which is the worst case in
187#           SECONDS - the same buffers hold ~4.3 s at the 32-bit hi-res carrier
188#           rate, and low-delay ffmpeg flags do not shrink them)
189#   ~4.0 s  the cliairplay ring: the binary's own lead (2000 ms default,
190#           clamped to the device-reported window) plus its 2 s of slack
191#   ~2.0 s  the receiver's own buffer
192# ~11.7 s in total, against 8.9-11.0 s measured across native AirPlay 2
193# sessions (Apple TV and Sonos, joining in both directions). This floor is that
194# sum rounded up, and it is only a floor: the lead is measured per session and
195# the ring grows to match, because a receiver that reports a wider lead window
196# or buffers more deeply moves the sum with nothing to announce it.
197AIRPLAY_LATE_JOIN_RING_MIN_SECONDS: Final[float] = 12.0
198# Grown on top of the largest lead a session has actually shown, so a lead that
199# drifts up between two joins cannot clip the oldest sample a joiner needs.
200AIRPLAY_LATE_JOIN_RING_MARGIN_SECONDS: Final[float] = 2.0
201# Hard bound on the ring, which is per session (one ring feeds every member) and
202# costs byte_rate x seconds. Bounded in BYTES rather than seconds on purpose:
203# the seconds the ring must hold are largest at the LOWEST byte rate (see the
204# measurements above), so a single byte bound buys ~35 s at 44.1 kHz/16-bit -
205# three times the measured lead, where the seconds are actually needed - and
206# still ~16 s at the 32-bit hi-res carrier, where the pipeline holds fewer
207# seconds anyway. 6 MiB per playing group is a few percent of the server's idle
208# footprint.
209AIRPLAY_LATE_JOIN_RING_MAX_BYTES: Final[int] = 6 * 1024 * 1024
210
211# Delay (seconds) before automatically re-joining a group member whose
212# cliairplay process died unexpectedly mid-session (e.g. the device rode out a
213# network blackout longer than the binary's own keepalive tolerance). A single
214# attempt keeps the behaviour predictable: it waits long enough for a short
215# blackout to clear, and if the device is still gone the player is left idle.
216# Staged retries can be reintroduced by adding entries to the tuple.
217AIRPLAY_REJOIN_ATTEMPT_DELAYS: Final[tuple[int, ...]] = (5,)
218
219# Shared audible instant for a native announcement over a live stream: now +
220# the largest member span + this margin. A member can only mix the clip into
221# audio it has not delivered yet, and its span (warm_lead_ms on the splice
222# timeline, the reported lead_ms otherwise) is how far its delivery head runs
223# ahead of the audible position - so the earliest instant EVERY member can
224# honor lies one max-span out. The margin covers fanning the command out over
225# the pipes and the per-member arm processing, so one shared instant stays
226# feasible for all members.
227AIRPLAY_ANNOUNCE_AT_MARGIN_MS: Final[int] = 300
228# Span assumed for a member whose binary reported neither a warm lead nor a
229# device lead (both read 0 = unreported): the binary's own default playback
230# lead, which bounds how far its delivery head can run ahead.
231AIRPLAY_ANNOUNCE_FALLBACK_SPAN_MS: Final[int] = 2000
232# Music gain (dB) under the clip while it plays; the binary ramps the duck in
233# and out itself. <= -60 mutes the music entirely. -18 dB puts the music
234# clearly in the background under speech (-12 was field-judged too shallow).
235AIRPLAY_ANNOUNCE_DUCK_DB: Final[int] = -18
236# Silence prepended to every announcement clip. The binary holds the music duck
237# for the whole clip file, so this is a window in which the music is already
238# ducked and the announcement has not started yet: the announcement volume is
239# raised inside it, where it cannot be heard as the music getting louder. It
240# has to cover the duck's ramp plus the round trip of the volume command.
241AIRPLAY_ANNOUNCE_DUCK_LEAD_S: Final[float] = 0.5
242# Silence appended to every announcement clip, so the volume restore has a
243# cushion to land in. The binary holds the duck for the whole clip, so the
244# restore lands while the music is still ducked instead of racing the duck's
245# 200 ms tail ramp (a restore that lands after the ramp plays a moment of
246# full-level music at the still-bumped device volume).
247AIRPLAY_ANNOUNCE_DUCK_TAIL_S: Final[float] = 1.0
248# On top of the lead to the commanded instant: how long to wait for a member's
249# announce_started before treating that member as not announcing. An outdated
250# binary silently ignores the unknown command, so this bounded wait is also what
251# detects that, and the announcement then fails instead of playing nowhere.
252AIRPLAY_ANNOUNCE_STARTED_TIMEOUT_MS: Final[int] = 3000
253# On top of the clip's audible end: how long to wait for announce_done. The
254# wait stays bounded because a queue that ends mid-clip emits its eof, which
255# ends the status stream while the clip still plays out over the drain.
256AIRPLAY_ANNOUNCE_DONE_TIMEOUT_MS: Final[int] = 5000
257# Pad after the clip's audible end before the pre-announcement volume is
258# restored: covers the jitter between the acked instant and true audibility.
259AIRPLAY_ANNOUNCE_VOLUME_RESTORE_PAD_MS: Final[int] = 500
260# Where inside the ducked lead-in the announcement volume is raised: past the
261# duck's own ramp, with the rest of the lead left for the command to reach the
262# receiver before the announcement itself becomes audible.
263AIRPLAY_ANNOUNCE_VOLUME_BUMP_DELAY_MS: Final[int] = 200
264# The AirPlay volume parameter is linear dB: 0..100 maps onto -30..0 dB on
265# every flow (libraop raopcl_float_volume, reused verbatim by the native AP2
266# SET_PARAMETER path), so one volume point is exactly 0.3 dB of output. This
267# makes the announcement volume bump's effect on the music bed computable, and
268# the duck is deepened by the same amount to keep the music's perceived level
269# at the configured duck depth.
270AIRPLAY_VOLUME_DB_PER_POINT: Final[float] = 0.3
271
272# Cover art is rendered to a local JPEG for the binary to embed (the binary
273# does not fetch URLs). 512px keeps the SET_PARAMETER payload small while still
274# looking sharp on speaker apps and the Apple TV now-playing screen.
275AIRPLAY_ARTWORK_SIZE: Final[int] = 512
276# How long a track-change metadata push waits for that render, so the artwork
277# can ride the SENDMETA bundle and the receiver rewrites its now-playing state
278# once instead of twice (bare replace, then artwork). A render that misses the
279# budget is not abandoned: it keeps running and is delivered with the
280# stand-alone ARTWORK command once it completes.
281AIRPLAY_ARTWORK_RENDER_TIMEOUT: Final[float] = 1.5
282EXTERNAL_ARTWORK_PATH_PREFIX: Final[str] = "external_artwork"
283
284# Per-protocol credential storage keys
285CONF_RAOP_CREDENTIALS: Final[str] = "raop_credentials"
286CONF_AIRPLAY_CREDENTIALS: Final[str] = "airplay_credentials"
287
288# AirPlay serves the shared sync-adjust control as a non-advanced (always visible)
289# setting: the binary handles the playback lead automatically and does not apply
290# device-reported render latency, so sync_adjust is the primary way to compensate
291# a device wired to a TV / AV receiver / amplifier that adds its own audio delay.
292# The AirPlay-scoped strings spell out the sign; the shared entry stays advanced
293# for other providers.
294CONF_ENTRY_SYNC_ADJUST_AIRPLAY = replace(CONF_ENTRY_SYNC_ADJUST, advanced=False)
295
296# Interactive setup-flow input keys (transient PIN/password form fields and the
297# optional "set up now?" choice for the control pairing steps).
298CONF_PAIRING_PIN: Final[str] = "pairing_pin"
299# every AirPlay pairing PIN (streaming, Companion, MRP) is 4 digits
300PAIRING_PIN_FORMAT: Final[str] = "####"
301CONF_PAIRING_PASSWORD: Final[str] = "pairing_password"
302CONF_COMPANION_PAIRING_PIN: Final[str] = "companion_pairing_pin"
303CONF_MRP_PAIRING_PIN: Final[str] = "mrp_pairing_pin"
304CONF_PAIR_NOW: Final[str] = "pair_now"
305
306FALLBACK_VOLUME: Final[int] = 20
307AIRPLAY_VOLUME_MUTE: Final[float] = -144.0
308# How long a volume we sent ourselves keeps the device's own volume reports from
309# being acted on. A receiver echoes every level it is given back over DACP, and
310# an echo that arrives after the next level was already sent would otherwise be
311# read as the user turning the knob and written straight back to the device.
312AIRPLAY_VOLUME_ECHO_GRACE_S: Final[float] = 2.0
313
314AIRPLAY_PCM_FORMAT = AudioFormat(
315    content_type=ContentType.from_bit_depth(16), sample_rate=44100, bit_depth=16
316)
317# Sample rates advertised for a receiver that supports 24-bit (AirPlay 2 flow
318# only). At 24-bit the cliairplay binary expects raw s32le on stdin and truncates
319# to 24-bit ALAC internally.
320AIRPLAY_HIRES_SAMPLE_RATES: Final[list[tuple[int, int]]] = [(44100, 24), (48000, 24)]
321
322# Bits in the audioFormat bit space (shared with the receiver's /info format
323# tables) that mark 24-bit ALAC: 44.1 kHz and 48 kHz respectively. Receivers
324# understate these - the Apple TV lists them for its buffered stream only, yet
325# renders them fine on the realtime stream - so a device advertising either bit
326# on either stream is treated as 24-bit capable.
327AIRPLAY_HIRES_AUDIO_FORMATS: Final[int] = (1 << 19) | (1 << 21)
328
329BASE_PLAYER_FEATURES: Final[set[PlayerFeature]] = {
330    PlayerFeature.PLAY_MEDIA,
331    PlayerFeature.PLAY_ANNOUNCEMENT,
332    PlayerFeature.SET_MEMBERS,
333    PlayerFeature.MULTI_DEVICE_DSP,
334    PlayerFeature.VOLUME_SET,
335    PlayerFeature.VOLUME_MUTE,
336}
337
338
339PIN_REQUIRED = 0x8
340PASSWORD_BIT = 0x80
341LEGACY_PAIRING_BIT = 0x200
342
343# Provider setting: opt-in for the shared PTP daemon's per-packet timing trace
344# (Announce/Sync/Follow_Up) when verbose logging is active. Off by default —
345# the trace floods the log and only matters for clock-sync debugging.
346CONF_VERBOSE_PTP_LOGGING: Final[str] = "verbose_ptp_logging"
347
348# The cliairplay binary tags no log levels on its output, so a genuine problem is
349# recognised by keyword and promoted to a warning that stays visible at normal levels.
350CLI_PROBLEM_MARKERS: Final[tuple[str, ...]] = ("error", "cannot", "failed", "unable")
351# Bound on how many of those promoted lines the shared PTP daemon may produce per
352# window before the rest are counted instead of logged. The markers above are
353# deliberately broad, and "error" is ordinary vocabulary in clock telemetry
354# (offset error, path delay error), so with the daemon's per-packet trace running
355# at ~10 lines/s a single matching line would otherwise fill the log at WARNING.
356# A burst still gets through, which is what a real one-shot daemon failure is.
357PTP_DAEMON_WARN_BURST: Final[int] = 5
358PTP_DAEMON_WARN_WINDOW: Final[float] = 60.0
359