@appium/coresim 1.4.0 → 1.5.0

package/src/native/sim_video_stream.mm

@@ -0,0 +1,427 @@
+#include "sim_video_stream.h"
+
+#import <CoreMedia/CoreMedia.h>
+#import <CoreVideo/CoreVideo.h>
+#import <IOSurface/IOSurface.h>
+#import <VideoToolbox/VideoToolbox.h>
+
+#include <algorithm>
+#include <atomic>
+#include <ctime>
+
+#include "nserror_bridge.h"
+#include "safe_dispatch.h"
+#include "sim_screenshot.h"
+
+namespace coresim {
+
+namespace {
+
+NSString* const kVideoStreamErrorDomain = @"com.appium.coresim.VideoStream";
+
+NSError* MakeError(NSInteger code, NSString* message) {
+  return [NSError errorWithDomain:kVideoStreamErrorDomain code:code userInfo:@{NSLocalizedDescriptionKey : message}];
+}
+
+NSError* MakeStatusError(NSInteger code, NSString* what, OSStatus status) {
+  return MakeError(code, [NSString stringWithFormat:@"%@ (OSStatus %d)", what, static_cast<int>(status)]);
+}
+
+// CFAbsoluteTimeGetCurrent() is wall time — subject to backward NTP/manual clock corrections,
+// which would violate VTCompressionSessionEncodeFrame's requirement of strictly increasing PTS.
+// CLOCK_MONOTONIC_RAW never goes backward and isn't adjusted by NTP.
+double MonotonicSeconds() {
+  struct timespec ts;
+  clock_gettime(CLOCK_MONOTONIC_RAW, &ts);
+  return static_cast<double>(ts.tv_sec) + static_cast<double>(ts.tv_nsec) / 1e9;
+}
+
+void AppendAnnexB(std::vector<uint8_t>& out, const uint8_t* nal, size_t length) {
+  static const uint8_t kStartCode[4] = {0, 0, 0, 1};
+  out.insert(out.end(), kStartCode, kStartCode + 4);
+  out.insert(out.end(), nal, nal + length);
+}
+
+// VideoToolbox's compressed output is AVCC-framed (a 4-byte big-endian length prefix per NAL,
+// no start codes) — rewrites it into Annex-B, matching VideoAccessUnit's documented wire format.
+void AppendSampleBufferNALs(std::vector<uint8_t>& out, CMSampleBufferRef sampleBuffer) {
+  CMBlockBufferRef block = CMSampleBufferGetDataBuffer(sampleBuffer);
+  if (block == nullptr) {
+    return;
+  }
+  size_t totalLength = CMBlockBufferGetDataLength(block);
+  if (totalLength == 0) {
+    return;
+  }
+  // CMBlockBufferGetDataPointer's pointer only covers the contiguous region starting at the given
+  // offset, which for a segmented buffer (multiple backing memory blocks — CoreMedia's documented
+  // contract, not just a VideoToolbox implementation detail) can be far shorter than totalLength;
+  // reading up to totalLength through it would run past that region. CopyDataBytes stitches
+  // segments together into a caller-owned, guaranteed-contiguous copy instead.
+  std::vector<uint8_t> data(totalLength);
+  if (CMBlockBufferCopyDataBytes(block, 0, totalLength, data.data()) != kCMBlockBufferNoErr) {
+    return;
+  }
+  const uint8_t* dataPointer = data.data();
+  size_t offset = 0;
+  while (offset + 4 <= totalLength) {
+    uint32_t nalLength = (static_cast<uint32_t>(dataPointer[offset]) << 24) |
+                         (static_cast<uint32_t>(dataPointer[offset + 1]) << 16) |
+                         (static_cast<uint32_t>(dataPointer[offset + 2]) << 8) | dataPointer[offset + 3];
+    offset += 4;
+    if (nalLength == 0 || offset + nalLength > totalLength) {
+      break;
+    }
+    AppendAnnexB(out, dataPointer + offset, nalLength);
+    offset += nalLength;
+  }
+}
+
+using ParameterSetAtIndexFn = OSStatus (*)(CMFormatDescriptionRef, size_t, const uint8_t**, size_t*, size_t*, int*);
+
+void AppendParameterSets(std::vector<uint8_t>& out, CMFormatDescriptionRef format, ParameterSetAtIndexFn getAtIndex) {
+  size_t count = 0;
+  if (getAtIndex(format, 0, nullptr, nullptr, &count, nullptr) != noErr) {
+    return;
+  }
+  for (size_t i = 0; i < count; i++) {
+    const uint8_t* bytes = nullptr;
+    size_t size = 0;
+    if (getAtIndex(format, i, &bytes, &size, nullptr, nullptr) == noErr) {
+      AppendAnnexB(out, bytes, size);
+    }
+  }
+}
+
+}  // namespace
+
+class VideoStreamSession::Impl {
+ public:
+  Impl(id device, VideoStreamOptions options, std::function<void(VideoAccessUnit)> onAccessUnit,
+       std::function<void(NSError*)> onError, std::function<void()> onEnd)
+      : device_(device),
+        options_(options),
+        onAccessUnit_(std::move(onAccessUnit)),
+        onError_(std::move(onError)),
+        onEnd_(std::move(onEnd)) {
+    queue_ = dispatch_queue_create("com.appium.coresim.videoStream", DISPATCH_QUEUE_SERIAL);
+  }
+
+  ~Impl() { Stop(); }
+
+  void Start() {
+    NSError* error = nil;
+    id descriptor = ResolveCaptureDisplay(device_, options_.displayId, &error);
+    if (descriptor == nil) {
+      throw NSErrorException(error);
+    }
+    id surfaceObj = CurrentDisplaySurface(descriptor);
+    if (surfaceObj == nil) {
+      throw NSErrorException(MakeError(4, @"The device's display surface is not available yet"));
+    }
+    IOSurfaceRef surface = (__bridge IOSurfaceRef)surfaceObj;
+    // Set up synchronously (not lazily on the first Tick()) so a setup failure rejects Start()
+    // directly rather than only reaching onError, which the caller may not be listening for yet.
+    NSError* setupError = nil;
+    if (!SetUpSession(surface, &setupError)) {
+      throw NSErrorException(setupError);
+    }
+    // Must be set before EncodeSurface below — both it and HandleEncodedSample measure elapsed
+    // time from this.
+    startTime_ = MonotonicSeconds();
+    // Encode immediately rather than waiting for a *changed* seed on the first tick, or the
+    // stream would stay silent until the display changes again. running_ is set true before this
+    // call (not after), since VTCompressionSessionEncodeFrame's output callback can in principle
+    // fire on another thread before this one returns — HandleEncodedSample discards samples while
+    // running_ is false, which would otherwise silently drop the stream's very first (keyframe)
+    // access unit. A failure resets it and tears session_ down itself here, rather than going
+    // through Stop()/onEnd_ (see coresim.mm — onEnd_ firing this early would double-release its
+    // ThreadSafeFunctions).
+    running_ = true;
+    bool encoded = false;
+    try {
+      encoded = EncodeSurface(surface);
+    } catch (...) {
+      running_ = false;
+      VTCompressionSessionInvalidate(session_);
+      CFRelease(session_);
+      session_ = nullptr;
+      throw;
+    }
+    // Only commit the seed once a frame was actually submitted — a transient pixel-buffer
+    // creation failure (EncodeSurface returning false) otherwise leaves lastSeed_ at its default
+    // 0, so the first Tick() sees the real seed as "changed" and retries automatically instead of
+    // the stream going silent forever on a display that never changes again.
+    if (encoded) {
+      lastSeed_ = IOSurfaceGetSeed(surface);
+    }
+
+    double interval = 1.0 / std::max(options_.fps, 1.0);
+    dispatch_source_t timer = dispatch_source_create(DISPATCH_SOURCE_TYPE_TIMER, 0, 0, queue_);
+    dispatch_source_set_timer(timer, dispatch_time(DISPATCH_TIME_NOW, 0),
+                              static_cast<uint64_t>(interval * NSEC_PER_SEC),
+                              static_cast<uint64_t>(interval * NSEC_PER_SEC / 10));
+    // `this` outlives the timer: Stop()/StopFromQueue() always drain or outrun it before `this`
+    // can be destroyed (see their comments below).
+    dispatch_source_set_event_handler(timer, ^{
+      Tick();
+    });
+    timer_ = timer;
+    dispatch_resume(timer_);
+  }
+
+  // Callable from any thread except `queue_` itself (would deadlock on the dispatch_sync below).
+  void Stop() {
+    if (!running_.exchange(false)) {
+      return;  // idempotent
+    }
+    if (timer_ != nullptr) {
+      dispatch_source_cancel(timer_);
+      // Blocks until any in-flight Tick() finishes — by then running_ is already false, so it
+      // won't touch session_ again.
+      dispatch_sync(queue_, ^{
+                    });
+      timer_ = nullptr;
+    }
+    TearDownSessionAndFireEnd();
+  }
+
+  void RequestKeyFrame() { forceKeyFrame_ = true; }
+
+ private:
+  // Same as Stop() minus the dispatch_sync barrier — only safe from within Tick() itself, already
+  // serialized on `queue_`; would race a concurrent Tick() from any other thread.
+  void StopFromQueue() {
+    if (!running_.exchange(false)) {
+      return;  // idempotent — e.g. an external Stop() already won this race
+    }
+    if (timer_ != nullptr) {
+      dispatch_source_cancel(timer_);
+      timer_ = nullptr;
+    }
+    TearDownSessionAndFireEnd();
+  }
+
+  void TearDownSessionAndFireEnd() {
+    if (session_ != nullptr) {
+      // Flushes and blocks until every already-submitted frame's callback has returned — without
+      // this, a frame submitted just before Stop() could fire after onEnd_ releases the
+      // ThreadSafeFunctions below (a use-after-release; see CLAUDE.md).
+      VTCompressionSessionCompleteFrames(session_, kCMTimeInvalid);
+      VTCompressionSessionInvalidate(session_);
+      CFRelease(session_);
+      session_ = nullptr;
+    }
+    if (onEnd_) {
+      onEnd_();
+    }
+  }
+
+  void Tick() {
+    if (!running_) {
+      return;
+    }
+    if (pendingErrorTeardown_) {
+      // HandleEncodedSample (below) can run on VideoToolbox's own callback thread, where it's
+      // unsafe to tear down directly — StopFromQueue()/CompleteFrames() are only safe already
+      // serialized on `queue_` (here), and CompleteFrames specifically would deadlock waiting on
+      // its own still-executing callback. It sets this flag instead; picked up on the very next
+      // tick (queue_-serialized, safe) rather than via a raw cross-thread dispatch, since nothing
+      // here can outlive `this` the way a block captured on another thread otherwise could.
+      StopFromQueue();
+      return;
+    }
+    @autoreleasepool {
+      try {
+        // Re-resolved every tick (like CaptureScreenshot does), not cached once in Start(), so a
+        // deleted device or disconnected display surfaces a real error instead of Tick() quietly
+        // doing nothing forever.
+        NSError* resolveError = nil;
+        id descriptor = ResolveCaptureDisplay(device_, options_.displayId, &resolveError);
+        if (descriptor == nil) {
+          if (onError_) {
+            onError_(resolveError);
+          }
+          StopFromQueue();
+          return;
+        }
+        id surfaceObj = CurrentDisplaySurface(descriptor);
+        if (surfaceObj == nil) {
+          return;  // transient — the connection may not have a frame ready yet, try again next tick
+        }
+        IOSurfaceRef surface = (__bridge IOSurfaceRef)surfaceObj;
+        uint32_t seed = IOSurfaceGetSeed(surface);
+        if (seed == lastSeed_) {
+          return;  // unchanged since the last tick — mirrors CoreSimulator's own recorder, which
+                   // only encodes a frame when the display actually changes (see CLAUDE.md)
+        }
+        // Only commit the new seed once EncodeSurface actually submits it — a transient failure
+        // (pixel-buffer creation) must leave lastSeed_ stale so the next tick retries this same
+        // frame instead of silently going quiet until the display changes again.
+        if (EncodeSurface(surface)) {
+          lastSeed_ = seed;
+        }
+      } catch (const std::exception& e) {
+        // Without this, an exception here (e.g. a dropped display-proxy connection) would escape
+        // this bare GCD timer handler uncaught and crash the whole process (see CLAUDE.md).
+        if (onError_) {
+          onError_(MakeError(3, [NSString stringWithFormat:@"Video stream encoding failed: %s", e.what()]));
+        }
+        StopFromQueue();
+      }
+    }
+  }
+
+  bool SetUpSession(IOSurfaceRef surface, NSError** error) {
+    int32_t width = static_cast<int32_t>(IOSurfaceGetWidth(surface));
+    int32_t height = static_cast<int32_t>(IOSurfaceGetHeight(surface));
+    CMVideoCodecType codecType =
+        options_.codec == VideoStreamCodec::kHEVC ? kCMVideoCodecType_HEVC : kCMVideoCodecType_H264;
+    OSStatus status = VTCompressionSessionCreate(kCFAllocatorDefault, width, height, codecType, nullptr, nullptr,
+                                                 kCFAllocatorDefault, OutputCallback, this, &session_);
+    if (status != noErr) {
+      *error = MakeStatusError(1, @"Failed to create a VTCompressionSession", status);
+      return false;
+    }
+    // Clamped like Start()'s timer interval — an unvalidated 0 here would set MaxKeyFrameInterval
+    // to an out-of-spec value.
+    double fps = std::max(options_.fps, 1.0);
+    status = VTSessionSetProperty(session_, kVTCompressionPropertyKey_RealTime, kCFBooleanTrue);
+    if (status == noErr) {
+      status = VTSessionSetProperty(session_, kVTCompressionPropertyKey_AllowFrameReordering, kCFBooleanFalse);
+    }
+    if (status == noErr) {
+      status = VTSessionSetProperty(session_, kVTCompressionPropertyKey_AverageBitRate,
+                                    (__bridge CFNumberRef) @(options_.bitrate));
+    }
+    if (status == noErr) {
+      status =
+          VTSessionSetProperty(session_, kVTCompressionPropertyKey_ExpectedFrameRate, (__bridge CFNumberRef) @(fps));
+    }
+    if (status == noErr) {
+      status = VTSessionSetProperty(session_, kVTCompressionPropertyKey_MaxKeyFrameInterval,
+                                    (__bridge CFNumberRef) @(static_cast<int>(fps * 2)));
+    }
+    if (status != noErr) {
+      *error = MakeStatusError(2, @"Failed to configure the VTCompressionSession", status);
+      VTCompressionSessionInvalidate(session_);
+      CFRelease(session_);
+      session_ = nullptr;
+      return false;
+    }
+    VTCompressionSessionPrepareToEncodeFrames(session_);
+    return true;
+  }
+
+  // Returns whether a frame was actually submitted to the encoder — false for a transient
+  // pixel-buffer creation failure the caller should retry, as opposed to a real encode failure
+  // (thrown, not returned, since that tears down the whole stream).
+  bool EncodeSurface(IOSurfaceRef surface) {
+    CVPixelBufferRef pixelBuffer = nullptr;
+    CVReturn cvStatus = CVPixelBufferCreateWithIOSurface(kCFAllocatorDefault, surface, nullptr, &pixelBuffer);
+    if (cvStatus != kCVReturnSuccess || pixelBuffer == nullptr) {
+      return false;  // transient — try again next tick rather than tearing down the whole stream
+    }
+    CMTime pts = CMTimeMake(static_cast<int64_t>((MonotonicSeconds() - startTime_) * 1000000), 1000000);
+    NSDictionary* frameProperties = nil;
+    if (forceKeyFrame_.exchange(false)) {
+      frameProperties = @{(__bridge NSString*)kVTEncodeFrameOptionKey_ForceKeyFrame : @YES};
+    }
+    OSStatus status = VTCompressionSessionEncodeFrame(session_, pixelBuffer, pts, kCMTimeInvalid,
+                                                      (__bridge CFDictionaryRef)frameProperties, nullptr, nullptr);
+    CVPixelBufferRelease(pixelBuffer);
+    if (status != noErr) {
+      throw std::runtime_error([[NSString stringWithFormat:@"VTCompressionSessionEncodeFrame failed (OSStatus %d)",
+                                                           static_cast<int>(status)] UTF8String]);
+    }
+    return true;
+  }
+
+  static void OutputCallback(void* outputCallbackRefCon, void* /*sourceFrameRefCon*/, OSStatus status,
+                             VTEncodeInfoFlags /*infoFlags*/, CMSampleBufferRef sampleBuffer) {
+    static_cast<Impl*>(outputCallbackRefCon)->HandleEncodedSample(status, sampleBuffer);
+  }
+
+  void HandleEncodedSample(OSStatus status, CMSampleBufferRef sampleBuffer) {
+    if (!running_) {
+      return;
+    }
+    if (status != noErr) {
+      // Only the first failure is reported — pendingErrorTeardown_ doubles as the report-once
+      // gate, since Tick() (the only place that consumes it) only ever needs to see it once too.
+      if (!pendingErrorTeardown_.exchange(true)) {
+        if (onError_) {
+          onError_(MakeStatusError(3, @"VideoToolbox reported an encoding failure", status));
+        }
+      }
+      return;
+    }
+    if (sampleBuffer == nullptr) {
+      return;
+    }
+    bool isKeyFrame = true;
+    CFArrayRef attachments = CMSampleBufferGetSampleAttachmentsArray(sampleBuffer, false);
+    if (attachments != nullptr && CFArrayGetCount(attachments) > 0) {
+      CFDictionaryRef attachment = static_cast<CFDictionaryRef>(CFArrayGetValueAtIndex(attachments, 0));
+      isKeyFrame = !CFDictionaryContainsKey(attachment, kCMSampleAttachmentKey_NotSync);
+    }
+
+    // From the sample's own presentation timestamp (as submitted in EncodeSurface), not a fresh
+    // wall-clock read here — this callback can fire well after encoding actually happened, and
+    // resampling "now" would report a later, jittery timestamp than when the frame was captured.
+    CMTime pts = CMSampleBufferGetPresentationTimeStamp(sampleBuffer);
+    VideoAccessUnit unit;
+    unit.isKeyFrame = isKeyFrame;
+    unit.sequence = sequence_++;
+    unit.timestampMicros = pts.timescale != 0 ? (pts.value * 1000000 / pts.timescale) : 0;
+    if (isKeyFrame) {
+      CMFormatDescriptionRef format = CMSampleBufferGetFormatDescription(sampleBuffer);
+      if (format != nullptr) {
+        if (options_.codec == VideoStreamCodec::kHEVC) {
+          AppendParameterSets(unit.data, format, CMVideoFormatDescriptionGetHEVCParameterSetAtIndex);
+        } else {
+          AppendParameterSets(unit.data, format, CMVideoFormatDescriptionGetH264ParameterSetAtIndex);
+        }
+      }
+    }
+    AppendSampleBufferNALs(unit.data, sampleBuffer);
+    if (onAccessUnit_) {
+      onAccessUnit_(std::move(unit));
+    }
+  }
+
+  id device_;
+  VideoStreamOptions options_;
+  std::function<void(VideoAccessUnit)> onAccessUnit_;
+  std::function<void(NSError*)> onError_;
+  std::function<void()> onEnd_;
+
+  dispatch_queue_t queue_ = nullptr;
+  dispatch_source_t timer_ = nullptr;
+  VTCompressionSessionRef session_ = nullptr;
+  uint32_t lastSeed_ = 0;
+  // VideoToolbox's output callback isn't documented as single-threaded, so this is read-modify-
+  // written atomically rather than assuming HandleEncodedSample never runs concurrently.
+  std::atomic<uint64_t> sequence_{0};
+  double startTime_ = 0;
+  std::atomic<bool> running_{false};
+  // Set by HandleEncodedSample (possibly off queue_) on an encoder failure, consumed by the next
+  // Tick() (on queue_) — see both for why teardown can't just happen inline there.
+  std::atomic<bool> pendingErrorTeardown_{false};
+  std::atomic<bool> forceKeyFrame_{false};
+};
+
+VideoStreamSession::VideoStreamSession(id device, VideoStreamOptions options,
+                                       std::function<void(VideoAccessUnit)> onAccessUnit,
+                                       std::function<void(NSError*)> onError, std::function<void()> onEnd)
+    : impl_(std::make_unique<Impl>(device, options, std::move(onAccessUnit), std::move(onError), std::move(onEnd))) {}
+
+VideoStreamSession::~VideoStreamSession() = default;
+
+void VideoStreamSession::Start() { impl_->Start(); }
+
+void VideoStreamSession::Stop() { impl_->Stop(); }
+
+void VideoStreamSession::RequestKeyFrame() { impl_->RequestKeyFrame(); }
+
+}  // namespace coresim

package/src/native-simctl.ts

@@ -49,6 +49,8 @@ import {
   setContentSize,
   setIncreaseContrast,
 } from './commands/ui.js';
+import {isVideoRecording, startVideoRecording, stopVideoRecording} from './commands/video-recording.js';
+import {startVideoStream} from './commands/video-stream.js';
 import {getWebInspectorSocket} from './commands/webinspector.js';
 // Bare re-imports so `declare module './native-simctl.js'` augmentations in the command modules
 // (which add their methods to NativeSimctl's type) reach downstream consumers' emitted .d.ts
@@ -68,6 +70,8 @@ import './commands/process.js';
 import './commands/screenshot.js';
 import './commands/spawn.js';
 import './commands/ui.js';
+import './commands/video-recording.js';
+import './commands/video-stream.js';
 import './commands/webinspector.js';
 import {NativeSimUnavailableError} from './errors.js';
 import type {
@@ -247,6 +251,12 @@ Object.assign(NativeSimctl.prototype, {
   getScreenshot,
   getDisplays,
 
+  // video recording
+  startVideoRecording,
+  stopVideoRecording,
+  isVideoRecording,
+  startVideoStream,
+
   // webinspector
   getWebInspectorSocket,
 

package/src/types.ts

@@ -128,6 +128,54 @@ export interface ScreenshotOptions {
   quality?: number;
 }
 
+/** Options for `NativeSimctl.startVideoRecording`. */
+export interface VideoRecordingOptions {
+  /**
+   * Which display to record, by `id` from `getDisplays()`. Defaults to the primary display
+   * (falling back to the first renderable display if none is primary, e.g. tvOS).
+   */
+  displayId?: string;
+  /** Video codec — `'h264'` (default) or `'hevc'`. */
+  codec?: 'h264' | 'hevc';
+  /**
+   * For a non-rectangular display (e.g. a Dynamic Island cutout): `'ignored'` (default) saves the
+   * unmasked framebuffer, `'black'` renders the mask black, `'alpha'` is not supported and
+   * behaves like `'black'`.
+   */
+  mask?: 'ignored' | 'alpha' | 'black';
+}
+
+/** Options for `NativeSimctl.startVideoStream`. */
+export interface VideoStreamOptions {
+  /**
+   * Which display to stream, by `id` from `getDisplays()`. Defaults to the primary display
+   * (falling back to the first renderable display if none is primary, e.g. tvOS).
+   */
+  displayId?: string;
+  /** Video codec — `'h264'` (default) or `'hevc'`. */
+  codec?: 'h264' | 'hevc';
+  /**
+   * Max frames/sec to poll the framebuffer at — an unchanged frame is never re-encoded, so this
+   * is an upper bound, not a guarantee. Must be >= 1. Defaults to 15.
+   */
+  fps?: number;
+  /** Target average bitrate, in bits/sec. Defaults to 2,000,000 (2 Mbps). */
+  bitrate?: number;
+}
+
+/**
+ * One encoded frame from `VideoStream.accessUnits()` — Annex-B NAL units. A keyframe's `data` has
+ * parameter sets (SPS/PPS, or VPS/SPS/PPS for HEVC) prepended, so it's self-decodable alone.
+ */
+export interface VideoAccessUnit {
+  data: Buffer;
+  isKeyFrame: boolean;
+  /** Monotonically increasing per stream, starting at 0. */
+  sequence: number;
+  /** Microseconds since the stream started. */
+  timestampMicros: number;
+}
+
 /**
  * Options for `NativeSimctl.spawnProcess`, passed through to CoreSimulator's
  * `spawnWithPath:options:terminationQueue:terminationHandler:error:`. Only keys confirmed
@@ -230,6 +278,24 @@ export interface NativeSpawnResult {
  */
 export type NativeSpawnExitCallback = (code: number | null, signal: number | null) => void;
 
+/** Raw shape of an access unit as the native addon delivers it — see {@link VideoAccessUnit}. */
+export interface NativeVideoAccessUnit {
+  data: Buffer;
+  isKeyFrame: boolean;
+  sequence: number;
+  timestampMicros: number;
+}
+
+export type NativeVideoAccessUnitCallback = (unit: NativeVideoAccessUnit) => void;
+export type NativeVideoErrorCallback = (err: Error) => void;
+
+/** A live encoder session, wrapped by `coresim.mm`'s `NativeVideoStream` — what `NativeDeviceHandle.startVideoStream()` resolves to. */
+export interface NativeVideoStreamHandle {
+  stop(): Promise<void>;
+  /** Forces the next encoded frame to be a keyframe — trivial in-memory flag, so synchronous. */
+  requestKeyFrame(): void;
+}
+
 /** A `SimDevice`, wrapped by `coresim.mm`'s `NativeDevice` — what `NativeSimctl`'s `_findDevice()` resolves to. */
 export interface NativeDeviceHandle {
   // Trivial in-memory accessors — kept synchronous on the native side (see coresim.mm), never a
@@ -278,6 +344,16 @@ export interface NativeDeviceHandle {
   getWebInspectorSocket(): Promise<string>;
   screenshot(options?: {format?: 'png' | 'jpeg'; displayId?: string; quality?: number}): Promise<Buffer>;
   getDisplays(): Promise<SimDisplayInfo[]>;
+  startVideoRecording(
+    outputFile: string,
+    options?: {displayId?: string; codec?: 'h264' | 'hevc'; mask?: 'ignored' | 'alpha' | 'black'},
+  ): Promise<void>;
+  stopVideoRecording(): Promise<void>;
+  startVideoStream(
+    options: {displayId?: string; codec?: 'h264' | 'hevc'; fps?: number; bitrate?: number} | undefined,
+    onAccessUnit: NativeVideoAccessUnitCallback,
+    onError: NativeVideoErrorCallback,
+  ): Promise<NativeVideoStreamHandle>;
   spawn(path: string, options: SpawnOptions | undefined, onExit: NativeSpawnExitCallback): Promise<NativeSpawnResult>;
 }