Files
chanora/crates/chanora_audio/src/mobile_voice_backend.rs
T
Edison Jwa d59da05f93 refactor(audio): share AudioHandler between iOS and macOS, bump deps
crates/chanora_audio/src/engine.rs: drop the macOS-specific event-queue producer/consumer path; macOS now uses the iOS-style direct AudioHandler::fill_buffer in the VPIO render callback. The shared AudioHandler is an Arc<Mutex<...>>; the realtime callback uses try_lock so it never blocks on the tokio decode task (see ios_voice_unit.rs render callback).

crates/chanora_audio/src/mobile_voice_backend.rs: update VoiceAudioParams cfg gates — handler is now the iOS/macOS/desktop shape (Arc<Mutex<AudioHandler<SessionAudioId>>>), event_producer is Android-only.

crates/chanora_audio/src/lib.rs: widen the audio_event_queue module visibility to test so the macOS-specific path can be exercised by the unit test suite.

Cargo.toml: bump cpal 0.17.3 -> 0.18.0, jni 0.21 -> 0.22.4, windows 0.54 -> 0.62, criterion 0.5 -> 0.8. Cargo.lock follows.
2026-06-07 23:27:46 +09:00

814 lines
32 KiB
Rust

//! Cross-platform mobile voice audio backend trait (SDD-111).
//!
//! `MobileVoiceAudioBackend` abstracts the lifetime of a paired
//! low-latency voice input/output stream pair on mobile platforms.
//! It is implemented on Android by
//! [`crate::android_voice_unit::AndroidVoiceUnit`] (SDD-111). On
//! iOS the existing [`crate::ios_voice_unit::IosVoiceUnit`] will be
//! back-filled to this trait under SDD-117 — until that lands the
//! iOS implementation does **not** implement this trait, and iOS
//! audio continues to flow through its dedicated path.
//!
//! Desktop targets (Windows / macOS / Linux) do not use this trait;
//! `cpal` (and SDL on Linux) own desktop capture/playback per the
//! existing audio engine design.
// TODO(SDD-117): back-fill `IosVoiceUnit` to implement this trait
// so the engine can hold a single `Box<dyn MobileVoiceAudioBackend>`
// across iOS and Android.
use std::fmt;
use std::sync::atomic::{AtomicBool, AtomicU32};
use std::sync::{Arc, Mutex};
use tokio::sync::mpsc;
use tsclientlib::audio::AudioHandler;
use crate::engine::SessionAudioId;
use crate::AudioEffects;
use chanora_protocol::OutPacket;
/// Events the audio-callback thread or platform JNI listener can post
/// to the tokio side of the engine (SDD-115). Audio callbacks MUST
/// NOT mutate engine state directly; they marshal through this enum.
#[derive(Debug, Clone)]
pub enum BackendEvent {
/// Platform reported a stream-disconnect (route change, device
/// removal). Engine should schedule a reopen.
Disconnected,
/// `AudioManager.OnAudioFocusChangeListener` reported transient
/// loss with ducking allowed; engine may continue.
FocusTransientCanDuck,
/// Transient focus loss (e.g., incoming alarm). Engine should
/// pause capture; resume on `FocusGain`.
FocusTransient,
/// Focus regained.
FocusGain,
/// Permanent focus loss. Engine should leave the session.
FocusLost,
/// Bluetooth SCO state changed. P0 stance is "no crash"; the
/// engine logs and continues (SDD-115).
BluetoothScoStateChanged(i32),
}
/// Receiver end the engine retains. Constructed by the backend; see
/// [`MobileVoiceAudioBackend::take_event_rx`].
pub type BackendEventRx = mpsc::UnboundedReceiver<BackendEvent>;
/// Sender end shared with audio callbacks. Unbounded is deliberate:
/// dropping a route-change event because the engine is briefly behind
/// is preferable to blocking the audio thread.
pub type BackendEventTx = mpsc::UnboundedSender<BackendEvent>;
/// Audio session id reported by the platform after the input stream
/// has been opened. Required by SDD-113 to bind hardware effects
/// (`AcousticEchoCanceler`, `NoiseSuppressor`, `AutomaticGainControl`)
/// against the active capture session.
pub type AudioSessionId = i32;
/// Engine-owned state shared with mobile voice audio callbacks.
#[cfg_attr(target_os = "ios", derive(Clone))]
pub(crate) struct VoiceAudioParams {
/// Opus-encoded voice packets sent on this channel toward the
/// protocol layer.
pub voice_out_tx: mpsc::Sender<OutPacket>,
/// PTT transmission gate — true when the user holds the PTT key.
pub transmit_active: Arc<AtomicBool>,
/// Counter incremented per encoded frame sent.
pub frames_sent: Arc<AtomicU32>,
/// Pre-encode amplitude scale (1.0 = unity).
pub mic_gain: f32,
/// AudioHandler owned by the Android output callback.
#[cfg(target_os = "android")]
pub handler: AudioHandler<SessionAudioId>,
/// Producer used by Android engine tasks to feed the output callback.
#[cfg(target_os = "android")]
pub event_producer: crate::audio_event_queue::AudioEventProducer,
/// AudioHandler that inbound decode+mix feeds into; the output
/// callback pulls mixed stereo f32 from it. iOS, macOS, and desktop
/// share this `Arc<Mutex<...>>` shape; the realtime callback uses
/// `try_lock` so it never blocks on the tokio decode task (see
/// `ios_voice_unit.rs` render callback).
#[cfg(not(target_os = "android"))]
pub handler: Arc<Mutex<AudioHandler<SessionAudioId>>>,
/// Master output gain (f32 bits stored in AtomicU32 for lock-free
/// cross-thread read from the realtime audio callback).
pub output_gain: Arc<AtomicU32>,
/// True = output silence regardless of incoming voice frames.
pub output_muted: Arc<AtomicBool>,
/// Optional TransmitModeSelector for VoiceActivity transmit mode.
/// The capture callback calls set_voice_activity_open on this when
/// VAD detects speech. None means VoiceActivity mode is disabled.
pub voice_activity_selector: Option<Arc<crate::TransmitModeSelector>>,
/// Shared audio-processing config (WebRTC APM flags, VAD backend).
pub audio_processing_config: Arc<Mutex<crate::AudioProcessingConfig>>,
/// Shared audio-processing statistics for diagnostics.
pub audio_processing_stats: Arc<crate::SharedAudioProcessingStats>,
}
/// Achieved performance-mode reported by the platform after stream
/// open (SDD-112 item 4).
#[derive(Debug, Clone, Copy, PartialEq, Eq)]
pub enum AchievedPerformanceMode {
/// Low-latency tier granted (SRS-210 target).
LowLatency,
/// Power-saving tier (latency tier not granted).
PowerSaving,
/// No specific tier — platform default.
None,
}
impl fmt::Display for AchievedPerformanceMode {
fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
match self {
Self::LowLatency => f.write_str("LowLatency"),
Self::PowerSaving => f.write_str("PowerSaving"),
Self::None => f.write_str("None"),
}
}
}
/// Achieved sharing-mode reported by the platform after stream open
/// (SDD-112 item 7 / SRS-214).
#[derive(Debug, Clone, Copy, PartialEq, Eq)]
pub enum AchievedSharingMode {
/// `Exclusive` granted — lowest latency path.
Exclusive,
/// Fallback `Shared` was used.
Shared,
}
impl fmt::Display for AchievedSharingMode {
fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
match self {
Self::Exclusive => f.write_str("Exclusive"),
Self::Shared => f.write_str("Shared"),
}
}
}
/// Error surface for the mobile voice backend (SDD-111). Kept
/// stringly-typed at the boundary to avoid leaking platform error
/// types out of the crate.
#[derive(Debug, Clone)]
pub enum BackendError {
/// Stream open failed (config rejected, device busy, permissions).
OpenFailed(String),
/// Stream start/stop/close failed.
LifecycleFailed(String),
/// Audio device disconnected — engine should reopen via the
/// state machine. Audio callback threads MUST NOT mutate engine
/// state directly; this variant is delivered through the
/// error-callback channel and consumed by a regular tokio task.
ErrorDisconnected,
/// JNI / platform service error.
Platform(String),
/// Configuration rejected at construction (SDD-112 item 9 — the
/// engine asked for a config that the backend can prove is
/// impossible without dispatching a platform open).
InvalidConfig {
/// Human-readable reason; not user-facing.
reason: String,
},
}
impl fmt::Display for BackendError {
fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
match self {
Self::OpenFailed(m) => write!(f, "open failed: {m}"),
Self::LifecycleFailed(m) => write!(f, "lifecycle failed: {m}"),
Self::ErrorDisconnected => f.write_str("audio device disconnected"),
Self::Platform(m) => write!(f, "platform: {m}"),
Self::InvalidConfig { reason } => write!(f, "invalid config: {reason}"),
}
}
}
impl std::error::Error for BackendError {}
/// Voice-stream configuration passed into the backend at `open()`
/// (SDD-112). Immutable after construction — the engine treats this
/// as a request; the platform may grant something weaker and the
/// backend reports the achieved values via `achieved_*` accessors.
#[derive(Debug, Clone)]
pub struct AndroidVoiceStreamConfig {
/// Requested sample rate. Target is 48 kHz (SRS-210); the
/// engine resamples if the device grants its native rate.
pub sample_rate: u32,
/// Requested channel count. Mono in / mono out (SRS-210).
pub channel_count: u32,
/// Request `LowLatency` performance mode (SRS-210).
pub request_low_latency: bool,
/// Request `Exclusive` sharing mode with `Shared` fallback
/// (SRS-214 / SDD-112 item 7).
pub request_exclusive: bool,
/// Voice-effect toggles the engine plans to engage. The Android
/// backend tries hardware AEC/NS/AGC (SDD-113); failures fall
/// back to the engine's existing software path.
pub effects: AudioEffects,
}
impl Default for AndroidVoiceStreamConfig {
fn default() -> Self {
Self {
sample_rate: 48_000,
channel_count: 1,
request_low_latency: false,
request_exclusive: false,
effects: AudioEffects::default(),
}
}
}
impl AndroidVoiceStreamConfig {
/// Validating constructor (SDD-112 item 9). Rejects sample rates
/// outside `(0, 192_000]` and channel counts outside `[1, 2]`
/// before the platform-open round-trip can fail.
pub fn new(
sample_rate: u32,
channel_count: u32,
request_low_latency: bool,
request_exclusive: bool,
effects: AudioEffects,
) -> Result<Self, BackendError> {
if sample_rate == 0 || sample_rate > 192_000 {
return Err(BackendError::InvalidConfig {
reason: format!("sample_rate {sample_rate} out of range (0, 192000]"),
});
}
if !(1..=2).contains(&channel_count) {
return Err(BackendError::InvalidConfig {
reason: format!("channel_count {channel_count} out of range [1, 2]"),
});
}
Ok(Self {
sample_rate,
channel_count,
request_low_latency,
request_exclusive,
effects,
})
}
}
/// SDD-111 trait. The engine drives a backend through this surface.
///
/// Object-safety: kept object-safe so the engine can hold a
/// `Box<dyn MobileVoiceAudioBackend>` once SDD-117 back-fills iOS.
/// All methods take `&mut self` because lifecycle transitions
/// require exclusive access; the audio callback thread does **not**
/// reach engine state through this trait — it marshals events
/// through the error channel exposed by [`Self::take_error_rx`].
pub trait MobileVoiceAudioBackend: Send {
/// Start audio flowing. Must be called after `open()` (which
/// the constructor performs).
fn start(&mut self) -> Result<(), BackendError>;
/// Stop audio flow. Streams remain open and `start` may be
/// called again.
fn stop(&mut self) -> Result<(), BackendError>;
/// Close streams and release platform resources. After this
/// the backend is unusable.
///
/// NOTE: SDD-111 item 1 specifies a consuming `self` signature.
/// We retain `&mut self` to preserve trait object-safety
/// (`Box<dyn MobileVoiceAudioBackend>` is the target shape per
/// SDD-117); the engine treats post-`close()` instances as
/// drop-on-next-statement.
fn close(&mut self) -> Result<(), BackendError>;
/// Take the backend event receiver (SDD-111 item 1 / SDD-115).
/// Returns `None` after the first call — the engine is the sole
/// owner of the receiver. Audio callbacks retain only the sender.
fn take_event_rx(&mut self) -> Option<BackendEventRx>;
/// Sample rate the platform actually granted (Hz). SDD-111 item 1.
fn achieved_sample_rate(&self) -> u32;
/// Achieved input preset (SDD-112 item 6, SDD-111 item 1).
fn achieved_input_preset(&self) -> AchievedInputPreset;
/// Frames-per-burst the platform reported (SDD-112 item 4).
fn achieved_frames_per_burst(&self) -> u32;
/// Platform audio session id for the input stream, if known.
/// Required for SDD-113 hardware-effect binding.
fn session_id(&self) -> Option<AudioSessionId>;
/// Achieved performance-mode of the input stream (SDD-112).
fn achieved_input_performance_mode(&self) -> AchievedPerformanceMode;
/// Achieved sharing-mode of the input stream (SDD-112).
fn achieved_input_sharing_mode(&self) -> AchievedSharingMode;
/// Achieved performance-mode of the output stream (SDD-112).
fn achieved_output_performance_mode(&self) -> AchievedPerformanceMode;
/// Achieved sharing-mode of the output stream (SDD-112).
fn achieved_output_sharing_mode(&self) -> AchievedSharingMode;
}
// --- Pure helpers (test-enabling, SWE4-UV-048/049/051) -----------
//
// These helpers describe the unit-level decisions encoded in the
// Android backend's fallback ladders and the SRS-210 latency tier
// classifier. They are intentionally pure (no platform handles,
// no IO) so they can be unit-tested on the host target without
// AAudio / JNI. They mirror the production ladder in
// `android_voice_unit::AndroidVoiceUnit::open_input_fallback` and
// the SDD-116 matrix basis described in SDD-111/112.
//
// Production code (the Android backend) currently inlines the
// ladder for clarity; these helpers are kept here as the canonical
// reference and as the verification surface for SWE.4. If the
// ladder ever changes, both this helper and the inline code MUST
// move together — see SWE4-UV-048/049.
/// Input-preset choice for the Android backend (SDD-112 item 6).
/// Mirrors `oboe::InputPreset` without depending on it so the
/// helper is host-testable.
#[derive(Debug, Clone, Copy, PartialEq, Eq)]
pub enum InputPresetChoice {
/// Voice-communication preset (preferred, primary).
VoiceCommunication,
/// Voice-performance preset (first fallback rung).
VoicePerformance,
/// Generic / unprocessed (last rung).
Generic,
}
/// Returns the next input-preset to try given the presets already
/// attempted, or `None` when the ladder is exhausted (SDD-112
/// item 6, SWE4-UV-048).
///
/// Ladder: VoiceCommunication -> VoicePerformance -> Generic.
pub fn next_input_preset_after(attempted: &[InputPresetChoice]) -> Option<InputPresetChoice> {
const LADDER: [InputPresetChoice; 3] = [
InputPresetChoice::VoiceCommunication,
InputPresetChoice::VoicePerformance,
InputPresetChoice::Generic,
];
LADDER.iter().copied().find(|p| !attempted.contains(p))
}
/// Sharing-mode choice for the Android backend (SDD-112 item 7).
#[derive(Debug, Clone, Copy, PartialEq, Eq)]
pub enum SharingModeChoice {
/// `Exclusive` — lowest-latency path; first try.
Exclusive,
/// `Shared` — fallback when `Exclusive` is denied.
Shared,
}
/// Returns the next sharing-mode to try (SDD-112 item 7,
/// SWE4-UV-049). Ladder: Exclusive -> Shared -> exhausted.
pub fn next_sharing_mode_after(attempted: &[SharingModeChoice]) -> Option<SharingModeChoice> {
const LADDER: [SharingModeChoice; 2] =
[SharingModeChoice::Exclusive, SharingModeChoice::Shared];
LADDER.iter().copied().find(|m| !attempted.contains(m))
}
/// Latency tier per SRS-210 (SWE4-UV-051). Pure mapping from the
/// platform-granted `AchievedPerformanceMode` to the SRS-210
/// latency-tier decision the engine uses for diagnostics and
/// adaptive jitter-buffer sizing.
#[derive(Debug, Clone, Copy, PartialEq, Eq)]
pub enum LatencyTier {
/// 150 ms target tier (low-latency granted).
LowLatency,
/// 250 ms fallback tier (no low-latency).
Fallback,
}
/// SDD-116 / SRS-210 classifier: maps the achieved performance mode
/// to a latency tier. Pure function — no platform dependencies.
pub fn latency_tier_for(achieved: AchievedPerformanceMode) -> LatencyTier {
match achieved {
AchievedPerformanceMode::LowLatency => LatencyTier::LowLatency,
AchievedPerformanceMode::PowerSaving | AchievedPerformanceMode::None => {
LatencyTier::Fallback
}
}
}
// --- SDD-112 / SDD-113 / SDD-116 diagnostics snapshot -------------
//
// `AndroidAudioDiagnostics` is the cross-platform-shaped snapshot
// exported by the diagnostics bundle (`chanora_diagnostics`) and the
// FRB bridge so that android-p0-acceptance TC-14 / TC-15 / TC-16 can
// collect per-device evidence for the SDD-116 verification matrix.
// All fields are device-side technical scalars; per SDD-090 no PII,
// permission state, or server identity is admitted into this struct.
/// Per-effect engine that satisfied a hardware-effect request
/// (SDD-113 item 5). `Hardware` means the JNI `create()` returned
/// a live `GlobalRef` and `setEnabled(true)` succeeded; `Software`
/// means the engine's cross-platform AEC/NS/AGC path is carrying
/// the effect; `None` means the effect was not requested at all.
#[derive(Debug, Clone, Copy, PartialEq, Eq)]
pub enum EffectEngine {
/// Platform hardware effect engaged on the active session id.
Hardware,
/// Software-engine fallback path engaged for the effect.
Software,
/// Neither path engaged (effect not requested).
None,
}
impl fmt::Display for EffectEngine {
fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
match self {
Self::Hardware => f.write_str("hardware"),
Self::Software => f.write_str("software"),
Self::None => f.write_str("none"),
}
}
}
/// Per-effect engagement record (SDD-113 item 7). `engaged` is the
/// observable "this effect is doing something" bit; `engine` records
/// which path satisfied the request so SRS-212 verification can
/// distinguish hardware vs software fulfilment from a single export.
#[derive(Debug, Clone, Copy, PartialEq, Eq)]
pub struct EffectEngagement {
/// Whether the effect is currently active (hardware or software).
pub engaged: bool,
/// Which engine satisfied the request.
pub engine: EffectEngine,
}
impl Default for EffectEngagement {
fn default() -> Self {
Self {
engaged: false,
engine: EffectEngine::None,
}
}
}
/// Achieved input-preset reported after stream-open (SDD-112 item 6).
/// `Unknown` is used when the platform / wrapper does not expose a
/// readback of the achieved preset; SDD-116 item 3 requires the
/// attempt sequence to be observable, and the achieved value is
/// preserved here whenever the wrapper can surface it.
#[derive(Debug, Clone, Copy, PartialEq, Eq)]
pub enum AchievedInputPreset {
/// `VoiceCommunication` preset granted (primary).
VoiceCommunication,
/// `VoicePerformance` preset granted (first fallback rung).
VoicePerformance,
/// `Generic` preset granted (last rung).
Generic,
/// Platform did not expose a preset readback.
Unknown,
}
impl fmt::Display for AchievedInputPreset {
fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
match self {
Self::VoiceCommunication => f.write_str("VoiceCommunication"),
Self::VoicePerformance => f.write_str("VoicePerformance"),
Self::Generic => f.write_str("Generic"),
Self::Unknown => f.write_str("Unknown"),
}
}
}
/// Diagnostics snapshot of the Android voice-audio session
/// (SDD-112 item 4 / SDD-113 item 7 / SDD-116 item 3).
///
/// Populated once at `AndroidVoiceUnit::open()` and cleared on
/// `close()`. The struct is cross-platform-compilable (the surrounding
/// `chanora_diagnostics::AndroidAudioReport` re-exports it) so the
/// diagnostics bundle has a stable schema even on non-Android hosts —
/// on non-Android the diagnostics report carries `None` for this
/// section.
///
/// Per SDD-090 every field here is a device-side technical scalar.
/// No server hostname, account id, permission state, or PII admitted.
///
/// Verification: SWE4-UV-052 follow-through. With these fields now
/// exposed, an instrumented test on a real device can assert
/// `achieved.usage == VoiceCommunication` and
/// `achieved.content_type == Speech` against the recorded export;
/// today the values are pinned by SDD-112 item 5 on the requested
/// side and the achieved side is `Unknown` until an AAudio readback
/// path is wired.
#[derive(Debug, Clone)]
pub struct AndroidAudioDiagnostics {
// Requested side (SDD-112 items 4..7) — fixed by SDD-112.
/// Requested performance mode for the current Android request profile.
pub requested_performance_mode: &'static str,
/// Requested output usage (always "VoiceCommunication" in P0).
pub requested_usage: &'static str,
/// Requested output content type (always "Speech" in P0).
pub requested_content_type: &'static str,
/// Requested input preset (always "VoiceCommunication" first).
pub requested_input_preset: &'static str,
/// Requested sharing mode for the current Android request profile.
pub requested_sharing_mode: &'static str,
/// Requested sample rate (Hz).
pub requested_sample_rate_hz: u32,
// Achieved side (SDD-112 item 4 / 7 + SDD-116 item 3).
/// Achieved performance mode after stream open.
pub achieved_performance_mode: AchievedPerformanceMode,
/// Achieved sharing mode after stream open.
pub achieved_sharing_mode: AchievedSharingMode,
/// Achieved input preset (`Unknown` if the wrapper does not
/// expose a readback today).
pub achieved_input_preset: AchievedInputPreset,
/// Sample rate AAudio actually delivered (Hz).
pub achieved_sample_rate_hz: u32,
/// Frames-per-burst the platform reported.
pub achieved_frames_per_burst: u32,
// Per-effect engagement (SDD-113 item 7 / SRS-212).
/// Acoustic echo cancellation engagement.
pub aec: EffectEngagement,
/// Noise suppression engagement.
pub ns: EffectEngagement,
/// Automatic gain control engagement.
pub agc: EffectEngagement,
/// Latency tier derived from `achieved_performance_mode` per
/// SDD-116 / SRS-210 via [`latency_tier_for`].
pub latency_tier: LatencyTier,
}
impl AndroidAudioDiagnostics {
/// Render the snapshot as the SDD-116 verification-matrix YAML
/// fragment that the diagnostics bundle embeds. Lines are
/// indented two spaces to nest cleanly under an `[android audio]`
/// section header.
///
/// Per SDD-090 / SDD-077: no PII, no permission state, no
/// server identity is admitted by this serialiser. Only the
/// device-side technical scalars on this struct are emitted.
pub fn to_yaml_fragment(&self) -> String {
let latency_tier = match self.latency_tier {
LatencyTier::LowLatency => "LowLatency",
LatencyTier::Fallback => "Fallback",
};
format!(
" requested:\n\
\x20\x20\x20\x20performance_mode: \"{rpm}\"\n\
\x20\x20\x20\x20usage: \"{ru}\"\n\
\x20\x20\x20\x20content_type: \"{rct}\"\n\
\x20\x20\x20\x20input_preset: \"{rip}\"\n\
\x20\x20\x20\x20sharing_mode: \"{rsm}\"\n\
\x20\x20\x20\x20sample_rate_hz: {rsr}\n\
\x20\x20achieved:\n\
\x20\x20\x20\x20performance_mode: \"{apm}\"\n\
\x20\x20\x20\x20sharing_mode: \"{ashm}\"\n\
\x20\x20\x20\x20input_preset: \"{aip}\"\n\
\x20\x20\x20\x20sample_rate_hz: {asr}\n\
\x20\x20\x20\x20frames_per_burst: {afpb}\n\
\x20\x20effects:\n\
\x20\x20\x20\x20aec:\n\
\x20\x20\x20\x20\x20\x20engaged: {aece}\n\
\x20\x20\x20\x20\x20\x20engine: \"{aecg}\"\n\
\x20\x20\x20\x20ns:\n\
\x20\x20\x20\x20\x20\x20engaged: {nse}\n\
\x20\x20\x20\x20\x20\x20engine: \"{nsg}\"\n\
\x20\x20\x20\x20agc:\n\
\x20\x20\x20\x20\x20\x20engaged: {agce}\n\
\x20\x20\x20\x20\x20\x20engine: \"{agcg}\"\n\
\x20\x20latency_tier: \"{lt}\"\n",
rpm = self.requested_performance_mode,
ru = self.requested_usage,
rct = self.requested_content_type,
rip = self.requested_input_preset,
rsm = self.requested_sharing_mode,
rsr = self.requested_sample_rate_hz,
apm = self.achieved_performance_mode,
ashm = self.achieved_sharing_mode,
aip = self.achieved_input_preset,
asr = self.achieved_sample_rate_hz,
afpb = self.achieved_frames_per_burst,
aece = self.aec.engaged,
aecg = self.aec.engine,
nse = self.ns.engaged,
nsg = self.ns.engine,
agce = self.agc.engaged,
agcg = self.agc.engine,
lt = latency_tier,
)
}
}
// --- Process-global diagnostics slot (SDD-116) -------------------
//
// `AndroidVoiceUnit::open()` publishes its diagnostics here on a
// successful open, and `close()`/`Drop` clears the slot. The bridge
// reads it synchronously from `export_diagnostics()` so the user-
// initiated export bundle (DEC-016) carries the SDD-116 evidence
// without an async hop.
//
// Cross-platform-safe: the slot exists on every target; it stays
// `None` outside Android, and the diagnostics report renders an
// absent `android_audio` section accordingly.
static ANDROID_AUDIO_DIAGNOSTICS: std::sync::OnceLock<
std::sync::Mutex<Option<AndroidAudioDiagnostics>>,
> = std::sync::OnceLock::new();
fn android_audio_diagnostics_slot() -> &'static std::sync::Mutex<Option<AndroidAudioDiagnostics>> {
ANDROID_AUDIO_DIAGNOSTICS.get_or_init(|| std::sync::Mutex::new(None))
}
/// Publish a fresh diagnostics snapshot. Called by
/// `AndroidVoiceUnit::open()` after both streams open and hardware-
/// effect attach completes (SDD-112 item 10 / SDD-113 item 7).
pub fn publish_android_audio_diagnostics(d: AndroidAudioDiagnostics) {
if let Ok(mut g) = android_audio_diagnostics_slot().lock() {
*g = Some(d);
}
}
/// Clear the diagnostics slot. Called by `AndroidVoiceUnit::close()`
/// and `Drop` so a stale snapshot does not survive past the voice
/// session.
pub fn clear_android_audio_diagnostics() {
if let Ok(mut g) = android_audio_diagnostics_slot().lock() {
*g = None;
}
}
/// Read the current diagnostics snapshot. Returns `None` when no
/// voice session is active (and always returns `None` on non-Android
/// targets). Cross-platform-callable from the FRB bridge.
pub fn current_android_audio_diagnostics() -> Option<AndroidAudioDiagnostics> {
android_audio_diagnostics_slot()
.lock()
.ok()
.and_then(|g| g.clone())
}
#[cfg(test)]
mod tests {
use super::*;
/// SWE4-UV-047: default config records the conservative shared
/// voice-input request profile, mono 48 kHz, and default effects.
#[test]
fn swe4_uv_047_default_config_records_requested_values() {
let cfg = AndroidVoiceStreamConfig::default();
assert_eq!(cfg.sample_rate, 48_000);
assert_eq!(cfg.channel_count, 1);
assert!(!cfg.request_low_latency);
assert!(!cfg.request_exclusive);
// AudioEffects defaults are all-on per DEC-007..010.
assert!(cfg.effects.aec);
assert!(cfg.effects.noise_suppression);
assert!(cfg.effects.agc);
}
/// SWE4-UV-047: requested fields are independently settable so
/// a non-default request (e.g., power-saving, shared) is faithfully
/// recorded by the builder/struct.
#[test]
fn swe4_uv_047_config_allows_non_default_requests() {
let cfg = AndroidVoiceStreamConfig {
sample_rate: 16_000,
channel_count: 1,
request_low_latency: false,
request_exclusive: false,
..AndroidVoiceStreamConfig::default()
};
assert_eq!(cfg.sample_rate, 16_000);
assert!(!cfg.request_low_latency);
assert!(!cfg.request_exclusive);
}
/// SWE4-UV-048: ladder starts at VoiceCommunication, falls to
/// VoicePerformance, then Generic, then exhausts.
#[test]
fn swe4_uv_048_input_preset_fallback_ladder_order() {
assert_eq!(
next_input_preset_after(&[]),
Some(InputPresetChoice::VoiceCommunication)
);
assert_eq!(
next_input_preset_after(&[InputPresetChoice::VoiceCommunication]),
Some(InputPresetChoice::VoicePerformance)
);
assert_eq!(
next_input_preset_after(&[
InputPresetChoice::VoiceCommunication,
InputPresetChoice::VoicePerformance
]),
Some(InputPresetChoice::Generic)
);
assert_eq!(
next_input_preset_after(&[
InputPresetChoice::VoiceCommunication,
InputPresetChoice::VoicePerformance,
InputPresetChoice::Generic
]),
None,
"ladder exhausts after all three presets"
);
}
/// SWE4-UV-049: sharing-mode ladder remains Exclusive -> Shared -> exhausted.
/// The default request profile may start at Shared, but the ladder still
/// exists for explicitly opt-in low-latency / exclusive experiments.
#[test]
fn swe4_uv_049_sharing_mode_fallback_ladder_order() {
assert_eq!(
next_sharing_mode_after(&[]),
Some(SharingModeChoice::Exclusive)
);
assert_eq!(
next_sharing_mode_after(&[SharingModeChoice::Exclusive]),
Some(SharingModeChoice::Shared)
);
assert_eq!(
next_sharing_mode_after(&[SharingModeChoice::Exclusive, SharingModeChoice::Shared]),
None
);
}
/// SWE4-UV-051: LowLatency achieved maps to the 150 ms tier.
#[test]
fn swe4_uv_051_lowlatency_maps_to_lowlatency_tier() {
assert_eq!(
latency_tier_for(AchievedPerformanceMode::LowLatency),
LatencyTier::LowLatency
);
}
/// SWE4-UV-051: PowerSaving / None achieved map to the fallback tier.
#[test]
fn swe4_uv_051_non_lowlatency_maps_to_fallback_tier() {
assert_eq!(
latency_tier_for(AchievedPerformanceMode::PowerSaving),
LatencyTier::Fallback
);
assert_eq!(
latency_tier_for(AchievedPerformanceMode::None),
LatencyTier::Fallback
);
}
/// SWE4-UV-052: output stream is conceptually pinned to
/// VoiceCommunication usage + Speech content; the struct does
/// not expose any "music"/"media" alternative — only
/// request_low_latency + request_exclusive are variable. This
/// test guards against accidental introduction of a usage
/// override field on the config struct.
#[test]
fn swe4_uv_052_config_does_not_expose_alt_usage_or_content_type() {
// Compile-time field check: AndroidVoiceStreamConfig has no
// `usage` / `content_type` fields. If a future change adds
// one, this destructuring breaks and the author must update
// SWE4-UV-052 deliberately.
let AndroidVoiceStreamConfig {
sample_rate: _,
channel_count: _,
request_low_latency: _,
request_exclusive: _,
effects: _,
} = AndroidVoiceStreamConfig::default();
}
/// SWE4-UV-047: BackendError variants render without panicking
/// and carry their context (verifies the error surface used by
/// open()/start()/stop()/close()).
#[test]
fn swe4_uv_047_backend_error_display() {
assert!(BackendError::OpenFailed("x".into())
.to_string()
.contains("open failed"));
assert!(BackendError::LifecycleFailed("y".into())
.to_string()
.contains("lifecycle failed"));
assert!(BackendError::ErrorDisconnected
.to_string()
.contains("disconnected"));
assert!(BackendError::Platform("z".into())
.to_string()
.contains("platform"));
}
/// SWE4-UV-047: achieved enums Display stably (used in
/// structured "stream_opened" diagnostics; SRS-210 traces grep
/// on these tokens).
#[test]
fn swe4_uv_047_achieved_enum_display_is_stable() {
assert_eq!(
AchievedPerformanceMode::LowLatency.to_string(),
"LowLatency"
);
assert_eq!(
AchievedPerformanceMode::PowerSaving.to_string(),
"PowerSaving"
);
assert_eq!(AchievedPerformanceMode::None.to_string(), "None");
assert_eq!(AchievedSharingMode::Exclusive.to_string(), "Exclusive");
assert_eq!(AchievedSharingMode::Shared.to_string(), "Shared");
}
}