feat(ptt): code-side initial split — transmit_active / capability badge / sanitizer

Implements the gen2 v0.9.3 doc baseline's first slice of code work:

  * SRS-201: split the audio engine's `ptt` AtomicBool into the
    authoritative `transmit_active` flag. The legacy `set_ptt` /
    `ptt` accessors are retained as `#[doc(hidden)]` thin wrappers
    so the existing bridge command and the existing Flutter
    hold-to-talk UI keep compiling.
  * SAD-075 / SDD-089 acknowledged at the type level: only
    `AudioEngine::set_transmit_active` (or its legacy alias)
    mutates the flag; the encoder feed reads it once per outbound
    frame and never writes.
  * SDD-082: new `chanora_audio::ptt` module ships the
    `PttCapabilityLevel` enum (`L0Focused`, `L1GlobalShortcut`,
    `L2GlobalHoldToTalk`, `L3GlobalWithMouseButtons`,
    `L4DeviceAware` reserved) with a stable `as_str` mapping and
    an `is_global` classifier.
  * SDD-087: `PttBackendDescriptor::focused()` constant value for
    the universal Focused-PTT fallback. The struct shape carries
    only privacy-safe fields (`level`, `backend_id`,
    `bound_input_class`) — a key code cannot fit through this
    surface by construction (DEC-027).
  * SAD-077 / SDD-090: `RedactingLogLayer` now hosts the
    `PttBanCheckVisitor` and the `PTT_BANNED_FIELDS` constant
    (`key_code`, `scan_code`, `virtual_key`, `vk`, `keysym`,
    `keysym_string`, `key_sequence`, `key_press_history`,
    `key_timing`). Any record whose field set names a banned key
    is dropped before reaching the in-memory log sink or the
    user-initiated diagnostic export. The check is structural and
    runs ahead of formatting / redaction.
  * `SessionEvent::PttCapability` carries the diagnostics-safe
    descriptor through the broadcast event stream;
    `chanora_core::ChanoraSession::start_audio` publishes the
    Focused-PTT descriptor when the audio engine starts (SRS-196
    / SDD-091).
  * `BridgeEvent::PttCapability` mirrors the event across the
    FFI boundary. flutter_rust_bridge codegen regenerated.
  * Flutter `_AudioControls` renders a capability badge above the
    PTT button: a globe icon for Global levels, a focus-frame
    icon for `L0Focused`, plus a Tooltip exposing the bound input
    class. New ARB key `pttCapabilityBadge(level, backend)` in
    `app_en.arb` and `app_zh.arb`.

Per-platform global PTT backends (`WindowsRawInputBackend`,
`MacOSEventTapBackend`, `LinuxGnomeWaylandBackend`) and the
`MissedKeyUpWatchdog` task land in a separate follow-up commit;
this milestone ships only PTT-L0 universally so the application's
runtime capability reporting is honest from day one.

Tests
-----

* `chanora_audio` rises from 1 to 4 unit tests covering
  `PttCapabilityLevel::as_str`, `is_global`, and the
  `PttBackendDescriptor::focused()` shape contract.
* `chanora_diagnostics` rises from 9 to 11 unit tests covering
  the new `PttBanCheckVisitor` over every banned field name and
  the `PTT_BANNED_FIELDS` stability assertion.
* Workspace total: 53 unit + integration tests, all green with
  `CHANORA_DISABLE_KEYRING=1` (was 49 at v1.0.0-rc.2).
* `flutter analyze`: clean.
* `cargo deny check`: advisories ok, bans ok, licenses ok,
  sources ok.
* `cargo about generate`: zero warnings (license inventory
  regenerated).
* `tools/dump_flutter_licenses.sh`: 94 packages, zero without
  LICENSE.
* Linux x86_64 release bundle builds clean.

No Android live verification in this commit per the user's note
that the test device was removed. Android arm64-v8a continues to
build via the same `cargo ndk` path; runtime reporting on Android
is `L0Focused` for the foreseeable future.
This commit is contained in:
EdisonJwa
2026-05-15 15:02:03 +08:00
parent 02ffadfa52
commit 7b21916049
16 changed files with 642 additions and 31 deletions
+49 -18
View File
@@ -77,7 +77,13 @@ impl Default for AudioEngineConfig {
/// Running audio engine. Drop = stop.
pub struct AudioEngine {
ptt: Arc<AtomicBool>,
/// `transmit_active` is the authoritative gate for outbound
/// voice — the Opus encoder feed consults this flag once per
/// 20 ms frame. PTT subsystems (focused widget, future
/// Windows / macOS / Linux global backends) drive this flag
/// through [`Self::set_transmit_active`]; nothing else is
/// permitted to flip it (SAD-075 / SDD-089).
transmit_active: Arc<AtomicBool>,
frames_sent: Arc<AtomicU32>,
frames_received: Arc<AtomicU32>,
/// Master output gain as f32 bits in an AtomicU32. Default 1.0.
@@ -181,7 +187,7 @@ impl AudioEngine {
}
}
let ptt = Arc::new(AtomicBool::new(cfg.ptt_initial));
let transmit_active = Arc::new(AtomicBool::new(cfg.ptt_initial));
let frames_sent = Arc::new(AtomicU32::new(0));
let frames_received = Arc::new(AtomicU32::new(0));
let output_gain = Arc::new(AtomicU32::new(1.0_f32.to_bits()));
@@ -196,7 +202,7 @@ impl AudioEngine {
let capture_result = try_open_capture(
&in_dev,
voice_out_tx,
ptt.clone(),
transmit_active.clone(),
frames_sent.clone(),
cfg.mic_gain,
);
@@ -293,7 +299,7 @@ impl AudioEngine {
});
Ok(Self {
ptt,
transmit_active,
frames_sent,
frames_received,
output_gain,
@@ -316,19 +322,41 @@ impl AudioEngine {
info!(target: "chanora_audio", "audio engine stopped");
}
/// Set the push-to-talk active state. When false, captured audio
/// is discarded before encoding. No-op if capture is inactive.
pub fn set_ptt(&self, active: bool) {
self.ptt.store(active, Ordering::Relaxed);
/// Set the **transmission gate** (SRS-201). When true the
/// encoder feed is allowed to emit Opus frames; when false the
/// captured audio is discarded before encoding. This is the
/// only writer permitted on `transmit_active` (SAD-075 /
/// SDD-089). Push-to-Talk subsystems — focused PTT today,
/// per-platform global backends in a follow-up — call this
/// method exclusively. No-op when capture is inactive.
pub fn set_transmit_active(&self, active: bool) {
self.transmit_active.store(active, Ordering::Relaxed);
}
/// Current PTT state.
/// Current transmit gate state.
pub fn transmit_active(&self) -> bool {
self.transmit_active.load(Ordering::Relaxed)
}
/// Legacy alias for [`Self::set_transmit_active`]. Retained so
/// the existing bridge `set_ptt` command and the existing
/// Flutter UI continue to compile during the v0.9.3 PTT
/// migration (SRS-201 splits the conceptual `ptt` flag into
/// `transmit_active` / `capture_active`).
#[doc(hidden)]
pub fn set_ptt(&self, active: bool) {
self.set_transmit_active(active);
}
/// Legacy alias for [`Self::transmit_active`].
#[doc(hidden)]
pub fn ptt(&self) -> bool {
self.ptt.load(Ordering::Relaxed)
self.transmit_active()
}
/// True if the capture stream opened. When false, the engine
/// runs in playback-only mode and PTT is a no-op.
/// runs in playback-only mode and the transmit gate is a
/// no-op (no frames will ever be encoded).
pub fn capture_active(&self) -> bool {
self.capture_active
}
@@ -380,7 +408,7 @@ impl Drop for AudioEngine {
fn try_open_capture(
in_dev: &cpal::Device,
voice_out_tx: mpsc::Sender<OutPacket>,
ptt: Arc<AtomicBool>,
transmit_active: Arc<AtomicBool>,
frames_sent: Arc<AtomicU32>,
mic_gain: f32,
) -> Result<cpal::Stream, AudioError> {
@@ -405,7 +433,7 @@ fn try_open_capture(
in_channels,
mic_gain,
voice_out_tx,
ptt,
transmit_active,
frames_sent,
)));
@@ -433,7 +461,9 @@ struct CaptureState {
resample_pos: f64,
opus_out: [u8; MAX_OPUS_FRAME],
voice_out_tx: mpsc::Sender<OutPacket>,
ptt: Arc<AtomicBool>,
/// The PTT transmission gate. Read once per outbound frame; the
/// CaptureState never mutates this flag.
transmit_active: Arc<AtomicBool>,
frames_sent: Arc<AtomicU32>,
}
@@ -444,7 +474,7 @@ impl CaptureState {
in_channels: usize,
mic_gain: f32,
voice_out_tx: mpsc::Sender<OutPacket>,
ptt: Arc<AtomicBool>,
transmit_active: Arc<AtomicBool>,
frames_sent: Arc<AtomicU32>,
) -> Self {
Self {
@@ -456,15 +486,16 @@ impl CaptureState {
resample_pos: 0.0,
opus_out: [0u8; MAX_OPUS_FRAME],
voice_out_tx,
ptt,
transmit_active,
frames_sent,
}
}
/// Consume an arbitrary-rate, multichannel cpal buffer; produce
/// 48 kHz mono frames; encode and send on PTT.
/// 48 kHz mono frames; encode and send when `transmit_active`
/// is true (PTT engaged).
fn ingest<T: ToF32 + Copy>(&mut self, buf: &[T]) {
if !self.ptt.load(Ordering::Relaxed) {
if !self.transmit_active.load(Ordering::Relaxed) {
// Drain accumulator while muted so we don't pop on PTT release.
self.pcm_accum.clear();
return;
+2
View File
@@ -29,8 +29,10 @@
#![warn(missing_docs)]
mod engine;
pub mod ptt;
pub use engine::{AudioEngine, AudioEngineConfig};
pub use ptt::{PttCapabilityLevel, PttBackendDescriptor};
use thiserror::Error;
+161
View File
@@ -0,0 +1,161 @@
//! Desktop Push-to-Talk capability model (SRS-195 / SRS-196 / SAD-071 /
//! SDD-081 / SDD-082).
//!
//! This module defines the typed `PttCapabilityLevel` enum and a
//! lightweight `PttBackendDescriptor` value the audio engine
//! publishes to upstream consumers so the UI can render the live
//! capability badge (SDD-091) and the release verification record
//! can carry per-platform evidence (SysDes-148).
//!
//! Concrete platform backends (`WindowsRawInputBackend`,
//! `MacOSEventTapBackend`, `LinuxGnomeWaylandBackend`) are deferred
//! to a follow-up code milestone; this commit lands the trait shape
//! and the universal `FocusedPttBackend` constant value so the
//! current Flutter-side hold-to-talk widget reports its capability
//! honestly through the bridge.
use core::fmt;
/// Detected runtime capability of the active desktop PTT backend.
///
/// The reported value shall match runtime behaviour — a backend
/// that *could* deliver `L2GlobalHoldToTalk` but lacks the
/// user-granted permission or the required compositor support
/// reports `L0Focused` (SRS-196 / SysRS-298).
///
/// `L4DeviceAware` is reserved per the gen2 v0.9.3 baseline
/// (SDD-082) and is **not** produced by any MVP implementation.
#[derive(Debug, Clone, Copy, PartialEq, Eq, Hash)]
pub enum PttCapabilityLevel {
/// Focused PTT. Press / release works only while the
/// application window has input focus. Mandatory baseline on
/// every desktop platform per SysRS-296.
L0Focused,
/// Global shortcut activation. The OS recognises a global
/// accelerator and notifies the application, but hold-to-talk
/// semantics may be approximated rather than guaranteed.
L1GlobalShortcut,
/// Global hold-to-talk. Press and release events are delivered
/// while the application is not focused.
L2GlobalHoldToTalk,
/// Global hold-to-talk plus mouse side buttons (typically
/// Mouse4 / Mouse5, sometimes labelled "back" / "forward").
L3GlobalWithMouseButtons,
/// Device-aware PTT. Reserved; no MVP implementation produces
/// this value (SDD-082).
L4DeviceAware,
}
impl PttCapabilityLevel {
/// Short identifier used by the diagnostics sanitizer and by
/// the release verification record. Stable across releases —
/// release notes and platform-test traces compare against these
/// strings.
pub fn as_str(self) -> &'static str {
match self {
Self::L0Focused => "L0Focused",
Self::L1GlobalShortcut => "L1GlobalShortcut",
Self::L2GlobalHoldToTalk => "L2GlobalHoldToTalk",
Self::L3GlobalWithMouseButtons => "L3GlobalWithMouseButtons",
Self::L4DeviceAware => "L4DeviceAware",
}
}
/// True when the level represents a global (non-focused)
/// behaviour. UI consumers use this to gate the "global PTT
/// available" affordance.
pub fn is_global(self) -> bool {
matches!(
self,
Self::L1GlobalShortcut
| Self::L2GlobalHoldToTalk
| Self::L3GlobalWithMouseButtons
| Self::L4DeviceAware
)
}
}
impl fmt::Display for PttCapabilityLevel {
fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
f.write_str(self.as_str())
}
}
/// Diagnostics-safe descriptor of the active PTT backend. Carries
/// only the fields the privacy policy permits in the user-initiated
/// diagnostic export (SRS-202 / DEC-027): the detected capability
/// level, a fixed `backend_id` string per implementation, and an
/// optional bound-input class (`"keyboard"`, `"mouse-side-button"`,
/// …). The raw key code, scan code, virtual-key value, or keysym
/// of any user binding is **never** part of this structure.
#[derive(Debug, Clone, PartialEq, Eq)]
pub struct PttBackendDescriptor {
/// Detected runtime capability.
pub level: PttCapabilityLevel,
/// Stable identifier per implementation. Examples:
/// `"focused"`, `"raw-input"`, `"low-level-hook"`,
/// `"event-tap"`, `"gnome-wayland-portal"`.
pub backend_id: &'static str,
/// Coarse description of the bound input. `None` when no
/// binding is active. The value is a stable category string,
/// never a key code.
pub bound_input_class: Option<&'static str>,
}
impl PttBackendDescriptor {
/// The universal Focused-PTT fallback (SDD-087). Every desktop
/// platform reports this value until a platform-specific
/// global backend lands in a follow-up code milestone.
pub const fn focused() -> Self {
Self {
level: PttCapabilityLevel::L0Focused,
backend_id: "focused",
bound_input_class: Some("keyboard"),
}
}
}
impl Default for PttBackendDescriptor {
fn default() -> Self {
Self::focused()
}
}
#[cfg(test)]
mod tests {
use super::*;
#[test]
fn level_as_str_is_stable() {
assert_eq!(PttCapabilityLevel::L0Focused.as_str(), "L0Focused");
assert_eq!(
PttCapabilityLevel::L2GlobalHoldToTalk.as_str(),
"L2GlobalHoldToTalk"
);
assert_eq!(
PttCapabilityLevel::L3GlobalWithMouseButtons.as_str(),
"L3GlobalWithMouseButtons"
);
}
#[test]
fn level_is_global_classification() {
assert!(!PttCapabilityLevel::L0Focused.is_global());
assert!(PttCapabilityLevel::L1GlobalShortcut.is_global());
assert!(PttCapabilityLevel::L2GlobalHoldToTalk.is_global());
assert!(PttCapabilityLevel::L3GlobalWithMouseButtons.is_global());
assert!(PttCapabilityLevel::L4DeviceAware.is_global());
}
#[test]
fn focused_descriptor_carries_only_safe_fields() {
let d = PttBackendDescriptor::focused();
assert_eq!(d.level, PttCapabilityLevel::L0Focused);
assert_eq!(d.backend_id, "focused");
assert_eq!(d.bound_input_class, Some("keyboard"));
// Compile-time check: the struct shape itself excludes
// anything that could carry a key code (DEC-027).
let _: &str = d.backend_id;
let _: Option<&str> = d.bound_input_class;
}
}
+25
View File
@@ -510,6 +510,22 @@ pub enum BridgeEvent {
/// Latest client count.
clients: u32,
},
/// Detected desktop Push-to-Talk capability (gen2 v0.9.3,
/// DEC-023..028). The fields carry only privacy-safe values per
/// DEC-027: the capability level, a stable backend identifier,
/// and the bound input class. No key codes, scan codes, or
/// virtual-key values cross this boundary.
PttCapability {
/// Stable capability identifier (`"L0Focused"`,
/// `"L1GlobalShortcut"`, `"L2GlobalHoldToTalk"`,
/// `"L3GlobalWithMouseButtons"`, or `"L4DeviceAware"`).
level: String,
/// Stable backend identifier (e.g. `"focused"`).
backend_id: String,
/// Coarse bound input class (e.g. `"keyboard"`,
/// `"mouse-side-button"`); empty when no binding is active.
bound_input_class: String,
},
}
impl From<chanora_core::SessionEvent> for BridgeEvent {
@@ -534,6 +550,15 @@ impl From<chanora_core::SessionEvent> for BridgeEvent {
chanora_core::SessionEvent::SnapshotChanged { channels, clients } => {
BridgeEvent::SnapshotChanged { channels, clients }
}
chanora_core::SessionEvent::PttCapability {
level,
backend_id,
bound_input_class,
} => BridgeEvent::PttCapability {
level,
backend_id,
bound_input_class,
},
}
}
}
@@ -929,6 +929,16 @@ impl SseDecode for crate::api::BridgeEvent {
clients: var_clients,
};
}
7 => {
let mut var_level = <String>::sse_decode(deserializer);
let mut var_backendId = <String>::sse_decode(deserializer);
let mut var_boundInputClass = <String>::sse_decode(deserializer);
return crate::api::BridgeEvent::PttCapability {
level: var_level,
backend_id: var_backendId,
bound_input_class: var_boundInputClass,
};
}
_ => {
unimplemented!("");
}
@@ -1247,6 +1257,17 @@ impl flutter_rust_bridge::IntoDart for crate::api::BridgeEvent {
clients.into_into_dart().into_dart(),
]
.into_dart(),
crate::api::BridgeEvent::PttCapability {
level,
backend_id,
bound_input_class,
} => [
7.into_dart(),
level.into_into_dart().into_dart(),
backend_id.into_into_dart().into_dart(),
bound_input_class.into_into_dart().into_dart(),
]
.into_dart(),
_ => {
unimplemented!("");
}
@@ -1440,6 +1461,16 @@ impl SseEncode for crate::api::BridgeEvent {
<u32>::sse_encode(channels, serializer);
<u32>::sse_encode(clients, serializer);
}
crate::api::BridgeEvent::PttCapability {
level,
backend_id,
bound_input_class,
} => {
<i32>::sse_encode(7, serializer);
<String>::sse_encode(level, serializer);
<String>::sse_encode(backend_id, serializer);
<String>::sse_encode(bound_input_class, serializer);
}
_ => {
unimplemented!("");
}
+121
View File
@@ -384,8 +384,35 @@ impl InMemoryLogSink {
}
}
/// Field names that name a raw key value or key-press timing
/// sequence. Records carrying any of these names are dropped before
/// they reach the log sink (REDACT-PTT-001..006 in
/// `docs/security/diagnostic-redaction-audit-report.md`; SDD-090).
///
/// We compare by exact field name rather than a content scan —
/// partial-redaction false negatives are riskier than a missing
/// log line, and the audio-engine and PTT-backend code paths emit
/// records with stable field names that we control.
const PTT_BANNED_FIELDS: &[&str] = &[
"key_code",
"scan_code",
"virtual_key",
"vk",
"keysym",
"keysym_string",
"key_sequence",
"key_press_history",
"key_timing",
];
/// A `tracing` Layer that funnels records into an
/// [`InMemoryLogSink`]. Install during process init.
///
/// Per DEC-027 the layer also drops any record that carries one of
/// [`PTT_BANNED_FIELDS`] in its field set; the structural check
/// runs before format / redaction so a banned record never reaches
/// the in-memory sink and therefore never reaches the user-initiated
/// diagnostic export.
#[derive(Debug, Clone)]
pub struct RedactingLogLayer {
sink: InMemoryLogSink,
@@ -403,6 +430,17 @@ where
S: Subscriber + for<'a> LookupSpan<'a>,
{
fn on_event(&self, event: &Event<'_>, _ctx: Context<'_, S>) {
// PttSanitizer (SAD-077 / SDD-090): drop any record whose
// field set names a raw key code, scan code, virtual-key
// value, keysym, or timing sequence. The structural check
// is fast — at most one allocation-free pass over the
// field list.
let mut ban_check = PttBanCheckVisitor::default();
event.record(&mut ban_check);
if ban_check.banned {
return;
}
let mut visitor = FormatVisitor::default();
event.record(&mut visitor);
let line = format!(
@@ -418,6 +456,40 @@ where
fn on_new_span(&self, _: &Attributes<'_>, _: &Id, _: Context<'_, S>) {}
}
/// Lightweight `tracing::field::Visit` implementation that only
/// notes whether any visited field name matches the PTT banned
/// list. Allocation-free.
#[derive(Default)]
struct PttBanCheckVisitor {
banned: bool,
}
impl PttBanCheckVisitor {
fn check(&mut self, name: &str) {
if !self.banned && PTT_BANNED_FIELDS.iter().any(|b| *b == name) {
self.banned = true;
}
}
}
impl Visit for PttBanCheckVisitor {
fn record_debug(&mut self, field: &Field, _value: &dyn std::fmt::Debug) {
self.check(field.name());
}
fn record_str(&mut self, field: &Field, _value: &str) {
self.check(field.name());
}
fn record_i64(&mut self, field: &Field, _value: i64) {
self.check(field.name());
}
fn record_u64(&mut self, field: &Field, _value: u64) {
self.check(field.name());
}
fn record_bool(&mut self, field: &Field, _value: bool) {
self.check(field.name());
}
}
#[derive(Default)]
struct FormatVisitor {
message: String,
@@ -577,4 +649,53 @@ mod tests {
assert!(!txt.contains("10.0.0.1"));
assert!(txt.contains("[ip]") || txt.contains(REDACTION_MARKER));
}
#[test]
fn ptt_ban_check_visitor_flags_banned_fields() {
// Direct test of the visitor (we don't spin up a full
// tracing subscriber for this).
let mut v = PttBanCheckVisitor::default();
v.check("backend_id"); // safe
assert!(!v.banned);
v.check("key_code"); // banned
assert!(v.banned);
let mut v2 = PttBanCheckVisitor::default();
for name in [
"scan_code",
"virtual_key",
"vk",
"keysym",
"keysym_string",
"key_sequence",
"key_press_history",
"key_timing",
] {
v2 = PttBanCheckVisitor::default();
v2.check(name);
assert!(v2.banned, "expected {name} to be banned");
}
// Allowed PTT fields stay safe.
let mut v3 = PttBanCheckVisitor::default();
for name in ["capability_level", "backend_id", "bound_input_class"] {
v3.check(name);
}
assert!(!v3.banned);
}
#[test]
fn ptt_banned_list_is_non_empty_and_stable() {
// Lightweight regression catch: the audit document
// REDACT-PTT-001..006 enumerates these exact names.
assert!(PTT_BANNED_FIELDS.contains(&"key_code"));
assert!(PTT_BANNED_FIELDS.contains(&"scan_code"));
assert!(PTT_BANNED_FIELDS.contains(&"virtual_key"));
assert!(PTT_BANNED_FIELDS.contains(&"keysym"));
assert!(PTT_BANNED_FIELDS.contains(&"key_sequence"));
// No accidental additions of safe field names.
for safe in ["capability_level", "backend_id", "bound_input_class"] {
assert!(!PTT_BANNED_FIELDS.contains(&safe));
}
}
}