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 setting is only ever written by
70# the user — a failing automatic route is reported, never switched away from.
71CONF_STREAMING_MODE: Final[str] = "streaming_mode"
72# Provider marker that the compatibility-mode pins were reset once. Earlier
73# releases switched a player here themselves when its native control channel
74# failed (usually a network dropout), pinning it to a lane many devices reject
75# outright, so those machine-written values are returned to Automatic a single
76# time; a deliberate choice can simply be made again.
77CONF_COMPAT_PINS_REVIEWED: Final[str] = "compat_pins_reviewed"
78STREAMING_MODE_AUTO: Final[str] = "auto"
79STREAMING_MODE_AP2_PTP: Final[str] = "ap2_ptp"
80STREAMING_MODE_AP2_NTP: Final[str] = "ap2_ntp"
81STREAMING_MODE_AP2_COMPAT: Final[str] = "ap2_compat"
82STREAMING_MODE_RAOP: Final[str] = "raop"
83CONF_STORED_VOLUME: Final[str] = "stored_volume"
84CONF_COMPANION_CREDENTIALS: Final[str] = "companion_credentials"
85CONF_MRP_CREDENTIALS: Final[str] = "mrp_credentials"
86CONF_NATIVE_MRP_CREDENTIALS: Final[str] = "native_mrp_credentials"
87
88# Bundle id of the Music Assistant tvOS dashboard app, launched over Companion on
89# eligible Apple TVs (see tvos/docs/launch-contract.md).
90TVOS_APP_BUNDLE_ID: Final[str] = "io.music-assistant.tvos"
91
92AIRPLAY_DISCOVERY_TYPE: Final[str] = "_airplay._tcp.local."
93COMPANION_DISCOVERY_TYPE: Final[str] = "_companion-link._tcp.local."
94MRP_DISCOVERY_TYPE: Final[str] = "_mediaremotetv._tcp.local."
95RAOP_DISCOVERY_TYPE: Final[str] = "_raop._tcp.local."
96DACP_DISCOVERY_TYPE: Final[str] = "_dacp._tcp.local."
97
98# Floor for a late joiner's anchor, and the one it rests on whenever the binary
99# reports no readiness projection ([STATUS] clock_ready). A joiner cannot
100# honour an instant in the past, and that second case has no device evidence at
101# all, so this carries it the whole way: the command's trip to the binary, the
102# binary's own 250 ms floor, and the receiver seating the anchor. The value is
103# field-proven, not derived from those.
104# It is NOT headroom for the binary's post-commit clock verification, which
105# arms only when the receiver has still not probed by the time it reads the
106# START, and only for an anchor that clears the receiver queue depth plus
107# 500 ms - past ~2 s of effective depth, an anchor this close leaves no room.
108AIRPLAY_LATE_JOIN_MIN_HEADROOM_MS: Final[int] = 2500
109# How long a group start, a join or the Sendspin bridge waits for the binary's
110# first receiver-clock projection ([STATUS] clock_ready with state=probing|
111# ready). The binary emits its first line right after [STATUS] connected and
112# refreshes it about every 250 ms, and the projection exists from the receiver's
113# first probe (~1078 ms after connect on a cold device), so this only has to
114# cover a slower device before giving up and anchoring on the lead alone.
115# Independent of the binary's own stall report (state=stalled), which is a
116# slower, higher-confidence diagnosis of a receiver that never answers at all: a
117# wait that times out here has simply run out of planning time and falls back,
118# while a stall says the speaker will not play. Keep them apart - tightening the
119# stall report to meet this deadline would trade the margin that keeps it free
120# of false alarms.
121AIRPLAY_CLOCK_READY_TIMEOUT_MS: Final[int] = 2500
122# Lead added on top of a reported readiness instant, wherever one is waited for.
123# The binary refuses to place an anchor inside its own 250 ms floor, measured
124# from when IT reads the START command rather than when the server sends it (the
125# same trap as AIRPLAY_START_LEAD_MS), so the lead carries that floor plus 250 ms
126# for the command reaching the binary and for the convergence error of a
127# projection made from the receiver's very first probe.
128AIRPLAY_CLOCK_READY_LEAD_MS: Final[int] = 500
129# Default receiver buffer depth per device family: (manufacturer wildcard,
130# model wildcard, firmware wildcard) -> depth in ms, matched case-insensitively
131# in order, first match wins; unmatched devices stay on Automatic (the binary's
132# stock depth). The table is EMPTY since the buffered (type 103) stream became
133# the auto-route for the receivers that used to need a deepened queue: the
134# LinkPlay pipelines that starved on the realtime stream (WiiM at 1750 ms,
135# Edifier MS50A silent below 2500 ms) manage their own buffer on the buffered
136# stream and play fine on Automatic. The per-player depth setting remains as
137# an advanced override; extend the table only for devices that starve on the
138# route they actually take.
139AIRPLAY_BUFFER_DEPTH_DEFAULTS: Final[tuple[tuple[str, str, str, int], ...]] = ()
140# Per-player override of the splice receiver-queue depth in ms (0 = automatic).
141CONF_BUFFER_DEPTH: Final[str] = "buffer_depth"
142# How long a plain (non-join) START waits for the binary's [STATUS] started ack.
143# A strict buffered receiver can reject its anchor until its clock is seated;
144# cliairplay then makes up to 12 attempts, 500 ms apart. Cover that 5.5-second
145# retry span plus control-response and status-delivery margin.
146AIRPLAY_START_ACK_TIMEOUT_MS: Final[int] = 7000
147# A join can additionally withhold its ack while receiver-clock verification
148# settles. Keep its independently named bound aligned with the buffered retry
149# span too; command failures still answer either wait immediately.
150AIRPLAY_JOIN_START_ACK_TIMEOUT_MS: Final[int] = 7000
151# How far the content a corrected anchor actually cut may fall short of the cut
152# it asked for before the reported media position is re-based and the shortfall
153# reported. The binary derives the cut it took from the bytes it discarded, so a
154# few ms of byte quantization is expected and correcting for it would only
155# jitter the base; anything larger means the cut really did end early (the input
156# ran out inside it, or a teardown settled it) and the position is over-advanced
157# by that much for the rest of the anchor.
158AIRPLAY_CONTENT_CUT_TOLERANCE_MS: Final[int] = 20
159# Anchor leads for a readiness-confirmed START (cold and warm alike), solo and
160# group: the session only anchors after the binary confirmed the connection
161# ([STATUS] connected) and the new audio flowing ([STATUS] audio), so a lead no
162# longer guesses at setup or transcoder spin-up time. Both cover the receiver
163# re-anchor (accepted down to ~150 ms in the flush-ladder measurements) plus
164# the command's trip down the pipe; the group one also covers fanning the
165# shared instant out to every member. The pipe margin is what keeps them
166# workable: the binary rejects any instant inside its own 250 ms floor,
167# measured from when IT reads the command, and corrects a miss to that floor
168# plus another full lead — so a lead sitting exactly on the floor misses by the
169# delivery time every time and turns a start into a group-wide re-anchor ladder.
170AIRPLAY_START_LEAD_MS: Final[int] = 400
171AIRPLAY_GROUP_START_LEAD_MS: Final[int] = 500
172# Cold GROUP starts anchor further out: a receiver on a brand-new session
173# still acquires its PTP slave lock (~1.7-2.3 s measured on Sonos) and cannot
174# render at an anchor inside that window - it starts late and the group opens
175# audibly out of sync. Warm re-anchors reuse a locked clock and keep the
176# short leads above; solo cold starts have no sync partner to miss.
177AIRPLAY_COLD_GROUP_START_LEAD_MS: Final[int] = 2500
178# Margin added on top of a member's reported warm lead (the splice-timeline
179# queue depth; that timeline is the default for every native AirPlay 2 session)
180# when anchoring a warm re-start: covers the command round-trips between the
181# flush acks and the shared START so every member's skip target lands beyond
182# its queued audio.
183AIRPLAY_SPLICE_LEAD_MARGIN_MS: Final[int] = 150
184
185# Floor for the late-join PCM ring, which has to hold every sample between the
186# audible position and the write head: a joiner's anchor maps onto content the
187# group was already fed but has not played yet. That distance - the write-head
188# lead - is the sum of every buffer between the session's byte counter and the
189# speaker, and it is far larger than the binary's own ring alone:
190#   ~5.7 s  the per-member ffmpeg and the pipes around it (measured 5.2 s
191#           steady / 5.7 s peak at 44.1 kHz/16-bit, which is the worst case in
192#           SECONDS - the same buffers hold ~4.3 s at the 32-bit hi-res carrier
193#           rate, and low-delay ffmpeg flags do not shrink them)
194#   ~4.0 s  the cliairplay ring: the binary's own lead (2000 ms default,
195#           clamped to the device-reported window) plus its 2 s of slack
196#   ~2.0 s  the receiver's own buffer
197# ~11.7 s in total, against 8.9-11.0 s measured across native AirPlay 2
198# sessions (Apple TV and Sonos, joining in both directions). This floor is that
199# sum rounded up, and it is only a floor: the lead is measured per session and
200# the ring grows to match, because a receiver that reports a wider lead window
201# or buffers more deeply moves the sum with nothing to announce it.
202AIRPLAY_LATE_JOIN_RING_MIN_SECONDS: Final[float] = 12.0
203# Grown on top of the largest lead a session has actually shown, so a lead that
204# drifts up between two joins cannot clip the oldest sample a joiner needs.
205AIRPLAY_LATE_JOIN_RING_MARGIN_SECONDS: Final[float] = 2.0
206# Hard bound on the ring, which is per session (one ring feeds every member) and
207# costs byte_rate x seconds. Bounded in BYTES rather than seconds on purpose:
208# the seconds the ring must hold are largest at the LOWEST byte rate (see the
209# measurements above), so a single byte bound buys ~35 s at 44.1 kHz/16-bit -
210# three times the measured lead, where the seconds are actually needed - and
211# still ~16 s at the 32-bit hi-res carrier, where the pipeline holds fewer
212# seconds anyway. 6 MiB per playing group is a few percent of the server's idle
213# footprint.
214AIRPLAY_LATE_JOIN_RING_MAX_BYTES: Final[int] = 6 * 1024 * 1024
215
216# Delay (seconds) before automatically re-joining a group member whose
217# cliairplay process died unexpectedly mid-session (e.g. the device rode out a
218# network blackout longer than the binary's own keepalive tolerance). A single
219# attempt keeps the behaviour predictable: it waits long enough for a short
220# blackout to clear, and if the device is still gone the player is left idle.
221# Staged retries can be reintroduced by adding entries to the tuple.
222AIRPLAY_REJOIN_ATTEMPT_DELAYS: Final[tuple[int, ...]] = (5,)
223
224# Shared audible instant for a native announcement over a live stream: now +
225# the largest member span + this margin. A member can only mix the clip into
226# audio it has not delivered yet, and its span (warm_lead_ms on the splice
227# timeline, the reported lead_ms otherwise) is how far its delivery head runs
228# ahead of the audible position - so the earliest instant EVERY member can
229# honor lies one max-span out. The margin covers fanning the command out over
230# the pipes and the per-member arm processing, so one shared instant stays
231# feasible for all members.
232AIRPLAY_ANNOUNCE_AT_MARGIN_MS: Final[int] = 300
233# Span assumed for a member whose binary reported neither a warm lead nor a
234# device lead (both read 0 = unreported): the binary's own default playback
235# lead, which bounds how far its delivery head can run ahead.
236AIRPLAY_ANNOUNCE_FALLBACK_SPAN_MS: Final[int] = 2000
237# Music gain (dB) under the clip while it plays; the binary ramps the duck in
238# and out itself. <= -60 mutes the music entirely. -18 dB puts the music
239# clearly in the background under speech (-12 was field-judged too shallow).
240AIRPLAY_ANNOUNCE_DUCK_DB: Final[int] = -18
241# Silence prepended to every announcement clip. The binary holds the music duck
242# for the whole clip file, so this is a window in which the music is already
243# ducked and the announcement has not started yet: the announcement volume is
244# raised inside it, where it cannot be heard as the music getting louder. It
245# has to cover the duck's ramp plus the round trip of the volume command.
246AIRPLAY_ANNOUNCE_DUCK_LEAD_S: Final[float] = 0.5
247# Silence appended to every announcement clip, so the volume restore has a
248# cushion to land in. The binary holds the duck for the whole clip, so the
249# restore lands while the music is still ducked instead of racing the duck's
250# 200 ms tail ramp (a restore that lands after the ramp plays a moment of
251# full-level music at the still-bumped device volume).
252AIRPLAY_ANNOUNCE_DUCK_TAIL_S: Final[float] = 1.0
253# On top of the lead to the commanded instant: how long to wait for a member's
254# announce_started before treating that member as not announcing. An outdated
255# binary silently ignores the unknown command, so this bounded wait is also what
256# detects that, and the announcement then fails instead of playing nowhere.
257AIRPLAY_ANNOUNCE_STARTED_TIMEOUT_MS: Final[int] = 3000
258# On top of the clip's audible end: how long to wait for announce_done. The
259# wait stays bounded because a queue that ends mid-clip emits its eof, which
260# ends the status stream while the clip still plays out over the drain.
261AIRPLAY_ANNOUNCE_DONE_TIMEOUT_MS: Final[int] = 5000
262# Pad after the clip's audible end before the pre-announcement volume is
263# restored: covers the jitter between the acked instant and true audibility.
264AIRPLAY_ANNOUNCE_VOLUME_RESTORE_PAD_MS: Final[int] = 500
265# Where inside the ducked lead-in the announcement volume is raised: past the
266# duck's own ramp, with the rest of the lead left for the command to reach the
267# receiver before the announcement itself becomes audible.
268AIRPLAY_ANNOUNCE_VOLUME_BUMP_DELAY_MS: Final[int] = 200
269# The AirPlay volume parameter is linear dB: 0..100 maps onto -30..0 dB on
270# every flow (libraop raopcl_float_volume, reused verbatim by the native AP2
271# SET_PARAMETER path), so one volume point is exactly 0.3 dB of output. This
272# makes the announcement volume bump's effect on the music bed computable, and
273# the duck is deepened by the same amount to keep the music's perceived level
274# at the configured duck depth.
275AIRPLAY_VOLUME_DB_PER_POINT: Final[float] = 0.3
276
277# Cover art is rendered to a local JPEG for the binary to embed (the binary
278# does not fetch URLs). 512px keeps the SET_PARAMETER payload small while still
279# looking sharp on speaker apps and the Apple TV now-playing screen.
280AIRPLAY_ARTWORK_SIZE: Final[int] = 512
281# How long a track-change metadata push waits for that render, so the artwork
282# can ride the SENDMETA bundle and the receiver rewrites its now-playing state
283# once instead of twice (bare replace, then artwork). A render that misses the
284# budget is not abandoned: it keeps running and is delivered with the
285# stand-alone ARTWORK command once it completes.
286AIRPLAY_ARTWORK_RENDER_TIMEOUT: Final[float] = 1.5
287EXTERNAL_ARTWORK_PATH_PREFIX: Final[str] = "external_artwork"
288
289# Per-protocol credential storage keys
290CONF_RAOP_CREDENTIALS: Final[str] = "raop_credentials"
291CONF_AIRPLAY_CREDENTIALS: Final[str] = "airplay_credentials"
292
293# AirPlay serves the shared sync-adjust control as a non-advanced (always visible)
294# setting: the binary handles the playback lead automatically and does not apply
295# device-reported render latency, so sync_adjust is the primary way to compensate
296# a device wired to a TV / AV receiver / amplifier that adds its own audio delay.
297# The AirPlay-scoped strings spell out the sign; the shared entry stays advanced
298# for other providers.
299CONF_ENTRY_SYNC_ADJUST_AIRPLAY = replace(CONF_ENTRY_SYNC_ADJUST, advanced=False)
300
301# Interactive setup-flow input keys (transient PIN/password form fields and the
302# optional "set up now?" choice for the control pairing steps).
303CONF_PAIRING_PIN: Final[str] = "pairing_pin"
304# every AirPlay pairing PIN (streaming, Companion, MRP) is 4 digits
305PAIRING_PIN_FORMAT: Final[str] = "####"
306CONF_PAIRING_PASSWORD: Final[str] = "pairing_password"
307CONF_COMPANION_PAIRING_PIN: Final[str] = "companion_pairing_pin"
308CONF_MRP_PAIRING_PIN: Final[str] = "mrp_pairing_pin"
309CONF_PAIR_NOW: Final[str] = "pair_now"
310
311FALLBACK_VOLUME: Final[int] = 20
312AIRPLAY_VOLUME_MUTE: Final[float] = -144.0
313# How long a volume we sent ourselves keeps the device's own volume reports from
314# being acted on. A receiver echoes every level it is given back over DACP, and
315# an echo that arrives after the next level was already sent would otherwise be
316# read as the user turning the knob and written straight back to the device.
317AIRPLAY_VOLUME_ECHO_GRACE_S: Final[float] = 2.0
318
319AIRPLAY_PCM_FORMAT = AudioFormat(
320    content_type=ContentType.from_bit_depth(16), sample_rate=44100, bit_depth=16
321)
322# Sample rates advertised for a receiver that supports 24-bit (AirPlay 2 flow
323# only). At 24-bit the cliairplay binary expects raw s32le on stdin and truncates
324# to 24-bit ALAC internally.
325AIRPLAY_HIRES_SAMPLE_RATES: Final[list[tuple[int, int]]] = [(44100, 24), (48000, 24)]
326
327# Bits in the audioFormat bit space (shared with the receiver's /info format
328# tables) that mark 24-bit ALAC: 44.1 kHz and 48 kHz respectively. Receivers
329# understate these - the Apple TV lists them for its buffered stream only, yet
330# renders them fine on the realtime stream - so a device advertising either bit
331# on either stream is treated as 24-bit capable.
332AIRPLAY_HIRES_AUDIO_FORMATS: Final[int] = (1 << 19) | (1 << 21)
333
334BASE_PLAYER_FEATURES: Final[set[PlayerFeature]] = {
335    PlayerFeature.PLAY_MEDIA,
336    PlayerFeature.PLAY_ANNOUNCEMENT,
337    PlayerFeature.SET_MEMBERS,
338    PlayerFeature.MULTI_DEVICE_DSP,
339    PlayerFeature.VOLUME_SET,
340    PlayerFeature.VOLUME_MUTE,
341}
342
343
344PIN_REQUIRED = 0x8
345PASSWORD_BIT = 0x80
346LEGACY_PAIRING_BIT = 0x200
347
348# Provider setting: opt-in for the shared PTP daemon's per-packet timing trace
349# (Announce/Sync/Follow_Up) when verbose logging is active. Off by default —
350# the trace floods the log and only matters for clock-sync debugging.
351CONF_VERBOSE_PTP_LOGGING: Final[str] = "verbose_ptp_logging"
352
353# The cliairplay binary tags no log levels on its output, so a genuine problem is
354# recognised by keyword and promoted to a warning that stays visible at normal levels.
355CLI_PROBLEM_MARKERS: Final[tuple[str, ...]] = ("error", "cannot", "failed", "unable")
356# Bound on how many of those promoted lines the shared PTP daemon may produce per
357# window before the rest are counted instead of logged. The markers above are
358# deliberately broad, and "error" is ordinary vocabulary in clock telemetry
359# (offset error, path delay error), so with the daemon's per-packet trace running
360# at ~10 lines/s a single matching line would otherwise fill the log at WARNING.
361# A burst still gets through, which is what a real one-shot daemon failure is.
362PTP_DAEMON_WARN_BURST: Final[int] = 5
363PTP_DAEMON_WARN_WINDOW: Final[float] = 60.0
364