Merge #319's latest into the landing branch

ROADMAP.md

@@ -128,6 +128,7 @@ mutated and truncated movies. Field reports from the first public week are track
 - **2026-10-12:** `soundcraft-cli convert` rejects an unparsable or zero `--rate` and an unknown `--bit-depth` (exit 1, nothing written) instead of converting with a default (#256)
 - **2026-10-12:** Timecode labels with huge fields no longer overflow: frame counts are computed in 128 bits and clamp, so `setup.current_timecode` and `session.format_time` saturate instead of failing (#248)
 - **2026-10-12:** `session.format_time` with the smallest bar number no longer fails with an internal error: the bar offset saturates (#249)
+- **2026-10-12 (audio device settings):** Setup › Playback Engine chooses the audio system, output and input device, sample rate and buffer size, kept per machine with the preferences; 16/32-bit integer devices are converted; ASIO is available behind the opt-in `asio` build feature (verified with an SSL 2+ at 128 samples). Channel routing, one duplex clock and latency compensation remain (G-01). No readiness percentage change.
 - **2026-10-10 (UI docking):** Tracks and Groups, Clips, MIDI Editor, Memory Locations, and Undo History use movable panels. Edit and Mix retain separate arrangements; Clips and MIDI Editor belong to Edit. Transport, plug-in, and specialist utility windows retain their existing hosts. Panel moves and reset are available as UI commands, and saved UI preferences retain arrangements. Common toolbar and track record-enable buttons share interaction, accessible labels and keyboard focus through `craft-ui`, retaining blue tool selection and red record-arm states. No readiness percentage change.
 - **2026-10-10** — Stage re-normalized to **pre-alpha** by the core-workflow gate (interface recording, MIDI recording and session interchange fail). Full re-measure against Pro Tools Ultimate 2026.4.1 and the standard
   progress-docs set (this page, target-app parity, gaps, roadmap, UI/format/hardware/plugin/

crates/playback/src/record.rs

@@ -188,6 +188,52 @@ where
     )
 }
 
+/// The input stream: device samples are converted to f32 in a scratch buffer allocated here (so
+/// the callback never allocates), then feed the monitor ring and, while armed, the capture.
+#[cfg(not(target_os = "freebsd"))]
+fn build_input<T>(
+    device: &cpal::Device,
+    config: &cpal::StreamConfig,
+    shared: &Arc<Shared>,
+    monitor: &Arc<InputRing>,
+) -> Result<cpal::Stream, cpal::BuildStreamError>
+where
+    T: cpal::SizedSample,
+    f32: cpal::FromSample<T>,
+{
+    use cpal::Sample;
+    use cpal::traits::DeviceTrait;
+    let sh = Arc::clone(shared);
+    let mon = Arc::clone(monitor);
+    let mut scratch = vec![0.0f32; 8192 * usize::from(config.channels).max(1)];
+    let chunk = scratch.len();
+    let mut owed = 0;
+    device.build_input_stream(
+        config,
+        move |data: &[T], _: &cpal::InputCallbackInfo| {
+            crate::mark_audio_thread();
+            for part in data.chunks(chunk) {
+                let Some(s) = scratch.get_mut(..part.len()) else { return };
+                for (o, x) in s.iter_mut().zip(part) {
+                    *o = f32::from_sample(*x);
+                }
+                mon.push(s);
+                if sh.armed.load(Ordering::Acquire) {
+                    sh.capture(s, &mut owed);
+                } else {
+                    owed = 0;
+                }
+            }
+        },
+        |e| {
+            // Possibly the audio thread (see `crate::mark_audio_thread`).
+            crate::mark_audio_thread();
+            log::warn!("input stream error: {e}");
+        },
+        None,
+    )
+}
+
 impl Recorder {
     /// Open the default input device. Errors when there is none (the caller shows a message).
     pub fn open() -> Result<Recorder, String> {