feat: step to adjacent decoded video frames

Checks/playback.mm

@@ -31,12 +31,109 @@ static PlayerState until(Player *player, const std::function<bool(PlayerState)>
     std::exit(1);
 }
 
+static PlayerState frameCompleted(Player *player, uint64_t revision, double pts) {
+    auto state = until(player, [revision](auto s) {
+        return !s.seek_pending && s.revision != revision && s.status == PLAYER_PAUSED;
+    });
+    if (std::abs(state.position - pts) >= .0001) {
+        std::fprintf(stderr, "Frame PTS: expected %.4f, got %.4f\n", pts, state.position);
+        require(false, "step selects exact adjacent presentation timestamp");
+    }
+    return state;
+}
+
+static PlayerState step(Player *player, int direction, double pts) {
+    auto revision = player_get_state(player).revision;
+    require(player_step_frame(player, direction), "frame step accepted");
+    return frameCompleted(player, revision, pts);
+}
+
+static void frameSteps(Player *player, MTKView *view, const std::string &directory) {
+    // Independent literals from the fixture's integer-millisecond setpts input.
+    constexpr double pts[] = {0, .1, .2, .3, .4, .65, .9, 1.15, 1.4, 1.5, 1.6, 1.7};
+    require(player_open(player, (directory + "/frame-step.mp4").c_str()), "VFR open accepted");
+    until(player, [](auto s) { return !s.seek_pending && s.status == PLAYER_PAUSED; });
+    require(!player_step_frame(player, 0) && !player_step_frame(player, 2) &&
+            !player_step_frame(player, -2), "only adjacent frame directions accepted");
+    step(player, -1, 0); // First boundary also ensures the initial preview is ready.
+    for (unsigned i = 1; i < sizeof(pts) / sizeof(pts[0]); ++i) {
+        step(player, 1, pts[i]);
+        [view draw];
+    }
+    step(player, 1, 1.7);
+    for (int i = 10; i >= 0; --i) step(player, -1, pts[i]);
+    step(player, -1, 0);
+
+    auto revision = player_get_state(player).revision;
+    require(player_step_frame(player, 1) && player_step_frame(player, 1) &&
+            player_step_frame(player, 1) && player_step_frame(player, 1),
+            "repeated queued steps accepted");
+    frameCompleted(player, revision, .4);
+    revision = player_get_state(player).revision;
+    require(player_step_frame(player, 1) && player_step_frame(player, 1) &&
+            player_step_frame(player, -1), "mixed queued steps accepted");
+    frameCompleted(player, revision, .65);
+
+    revision = player_get_state(player).revision;
+    require(player_seek(player, .72), "non-keyframe seek accepted");
+    until(player, [revision](auto s) {
+        return !s.seek_pending && s.revision != revision && s.status == PLAYER_PAUSED;
+    });
+    // Seek previews .9, not the .72 authoritative seek clock and not keyframe zero.
+    step(player, 1, 1.15);
+    step(player, -1, .9);
+    step(player, -1, .65);
+    revision = player_get_state(player).revision;
+    require(player_seek(player, .72) && player_step_frame(player, 1) &&
+            player_step_frame(player, -1), "queued seek and steps accepted");
+    frameCompleted(player, revision, .9);
+    step(player, -1, .65);
+
+    require(player_play(player), "play before frame step accepted");
+    until(player, [](auto s) { return s.status == PLAYER_PLAYING && s.position > .67 && s.position < .8; });
+    auto frozen = step(player, 1, .9);
+    [view draw];
+    for (int i = 0; i < 20; ++i)
+        CFRunLoopRunInMode(kCFRunLoopDefaultMode, .005, false);
+    auto paused = player_get_state(player);
+    require(paused.status == PLAYER_PAUSED && !paused.seek_pending &&
+            std::abs(paused.position - frozen.position) < .0001,
+            "stepping during playback pauses and freezes audio/video timeline");
+    require(player_play(player), "resume stepped timeline accepted");
+    until(player, [](auto s) { return s.status == PLAYER_PLAYING && s.position > 1.19 && s.position < 1.32; });
+    require(player_pause(player), "pause resumed timeline accepted");
+    until(player, [](auto s) { return s.status == PLAYER_PAUSED; });
+    // Resumed presentation has advanced from .9 to 1.15; no pre-step frame remains.
+    step(player, -1, .9);
+    require(player_play(player), "play stepped media through EOF");
+    until(player, [](auto s) { return s.status == PLAYER_ENDED; });
+    step(player, 1, 1.7);
+    step(player, -1, 1.6);
+    step(player, 1, 1.7);
+    step(player, 1, 1.7);
+    require(player_open(player, (directory + "/video.mp4").c_str()), "video-only stepping open");
+    until(player, [](auto s) { return s.status == PLAYER_PAUSED; });
+    revision = player_get_state(player).revision;
+    require(player_seek(player, 3.9), "seek near video-only last frame");
+    until(player, [revision](auto s) { return !s.seek_pending && s.revision != revision; });
+    step(player, 1, 95.0 / 24);
+    require(player_play(player), "resume video-only last frame");
+    auto ended = until(player, [](auto s) { return s.status == PLAYER_ENDED; });
+    require(std::abs(ended.position - 4) < .0001, "last stepped frame retains its presentation duration");
+    step(player, -1, 94.0 / 24);
+    require(player_close(player), "close stepped media accepted");
+    until(player, [](auto s) { return s.status == PLAYER_EMPTY; });
+    std::puts("PASS exact VFR/B-frame steps/queued steps/seek reference/pause/resume/boundaries");
+}
+
 int main(int argc, char **argv) {
     @autoreleasepool {
         require(argc == 2, "fixture directory argument required");
         [NSApplication sharedApplication];
         std::unique_ptr<Player, decltype(&player_destroy)> player(player_create(), player_destroy);
         require(bool(player), "player creation");
+        require(!player_step_frame(nullptr, 1) && !player_step_frame(player.get(), 1),
+                "null and empty players reject frame steps");
         void *native = player_macos_make_video_view(player.get());
         require(native != nullptr, "Metal view and shaders initialize");
         MTKView *view = (__bridge_transfer MTKView *)native;
@@ -56,6 +153,9 @@ int main(int argc, char **argv) {
                     "audio track selection");
             double aspect = !opened.has_video ? 0 : std::string(fixture) == "anamorphic.mp4" ? 8.0 / 3 : 16.0 / 9;
             require(std::abs(opened.video_aspect_ratio - aspect) < .0001, "display aspect includes non-square pixels");
+            if (!opened.has_video)
+                require(!player_step_frame(player.get(), 1) && !player_step_frame(player.get(), -1),
+                        "audio-only media rejects frame steps");
             require(!player_seek(player.get(), NAN), "non-finite seek rejected");
             require(player_play(player.get()), "play accepted");
             until(player.get(), [](auto s) { return s.status == PLAYER_PLAYING && s.position > .35; });
@@ -98,6 +198,7 @@ int main(int argc, char **argv) {
                     "close clears media state and pending seeks");
             std::printf("PASS %s: open/play/pause/seek/end/replay/close\n", fixture);
         }
+        frameSteps(player.get(), view, argv[1]);
         require(player_open(player.get(), (std::string(argv[1]) + "/hdr.mkv").c_str()), "HDR open accepted");
         auto hdr = until(player.get(), [](auto s) { return s.status == PLAYER_ERROR; }, true);
         require(std::string(hdr.error).find("HDR") != std::string::npos, "HDR reports unsupported output");

Makefile

@@ -41,6 +41,8 @@ check: .build/playback-check check-audio
 	ffmpeg -hide_banner -loglevel error -y -i .build/fixtures/av.mp4 -an -c:v copy .build/fixtures/video.mp4
 	ffmpeg -hide_banner -loglevel error -y -i .build/fixtures/av.mp4 -an -vf setsar=3/2 -c:v mpeg4 .build/fixtures/anamorphic.mp4
 	ffmpeg -hide_banner -loglevel error -y -i .build/fixtures/av.mp4 -vn -c:a pcm_s16le .build/fixtures/audio.wav
+	# Literal frame PTS: 0,.1,.2,.3,.4,.65,.9,1.15,1.4,1.5,1.6,1.7; reordered B-frames across short GOPs.
+	ffmpeg -hide_banner -loglevel error -y -f lavfi -i testsrc2=size=320x180:rate=10:duration=1.2 -f lavfi -i sine=frequency=440:sample_rate=44100:duration=2 -vf "settb=1/1000,setpts='if(lt(N,5),N*100,if(lt(N,9),400+(N-4)*250,1400+(N-8)*100))'" -fps_mode vfr -enc_time_base:v 1:1000 -c:v mpeg4 -g 4 -bf 2 -video_track_timescale 1000 -c:a aac .build/fixtures/frame-step.mp4
 	ffmpeg -hide_banner -loglevel error -y -f lavfi -i testsrc2=size=160x90:rate=12 -t 1 -vf 'format=yuv420p10le,setparams=color_primaries=bt2020:color_trc=smpte2084:colorspace=bt2020nc' -c:v ffv1 .build/fixtures/hdr.mkv
 	.build/playback-check .build/fixtures
 

README.md

@@ -27,12 +27,15 @@ Opening video fits the window to its display aspect ratio, including non-square
 | `K` / `Space` | Toggle playback |
 | `M` | Toggle mute, restoring the previous audible level |
 | `J` / `L` | Seek backward / forward ten seconds |
+| `,` / `.` | Step to the previous / next video frame; leave playback paused |
 | `F` | Toggle fullscreen |
 | `P` | Toggle keep-on-top (windowed mode) |
 | `O` | Open the file picker to switch files |
 | `Q` | Quit through normal application shutdown |
 
-Single-key shortcuts apply only to the playback window and do not intercept text editing, file-picker input, or Command/Control/Option combinations. Held toggle keys do not repeatedly toggle; held `J`/`L` can continue seeking. `,` / `.` frame stepping and `C` captions are unbound until supported.
+Single-key shortcuts apply only to the playback window and do not intercept text editing, file-picker input, or Command/Control/Option combinations. Held toggle keys do not repeatedly toggle; held `J`/`L` continue seeking and held `,`/`.` continue stepping. `C` captions remains unbound.
+
+Frame stepping uses decoded presentation timestamps, not an average-FPS time offset, including variable-frame-rate video and reordered B-frames. Steps start from the displayed frame, stop at the first/last frame, and keep the timeline paused. Audio-only files disable stepping; normal playback resumes with audio synchronized to the selected frame.
 
 - Existing shortcuts remain: **⌘O** opens files, **←/→** seeks five seconds, **↑/↓** adjusts volume by five percent, **⌃⌘F** toggles fullscreen, and **Esc** leaves fullscreen. Double-clicking video also toggles fullscreen.
 - The timeline slider seeks on release and holds the requested position until queued seeking finishes. Both slider tracks span the native thumb-center travel range; the orange fill ends at the current volume thumb center. The volume thumb is opaque light gray.
@@ -43,8 +46,8 @@ Only implemented actions have controls; there are no placeholder playlist, subti
 
 ## Where to change things
 
-- `Sources/Core/player.h`: the C interface and authoritative state snapshot. Playback commands enqueue work; poll state for completion. `seek_pending` is set synchronously when a seek is accepted and clears when all queued seeks on that timeline finish. Volume set/get is synchronous and uses a lock-free target gain. Callers own the opaque handle and must stop using it before destruction.
-- `Sources/Core/player.cpp`: single demux/decode worker, bounded buffers, timestamps, audio-master synchronization, paused previews, decoder/resampler drain, seeking, errors, and shutdown. A seek stops output and flushes all old timeline state before resuming.
+- `Sources/Core/player.h`: the C interface and authoritative state snapshot. Playback commands enqueue work; poll state for completion. `seek_pending` is set synchronously when a seek or frame step is accepted and clears when all queued seeks/steps on that timeline finish. `player_step_frame` accepts only `-1` or `+1`, requires ready video, and publishes the selected frame's timestamp as `position`. Volume set/get is synchronous and uses a lock-free target gain. Callers own the opaque handle and must stop using it before destruction.
+- `Sources/Core/player.cpp`: single demux/decode worker, bounded buffers, timestamps, audio-master synchronization, paused previews, exact frame stepping, decoder/resampler drain, seeking, errors, and shutdown. A seek stops output and flushes all old timeline state before resuming.
 - `Sources/Core/output.hpp` and `Sources/macOS/Output.mm`: native output contract, allocation-free/nonblocking Core Audio callback and bounded audio ring, latency-aware presentation clock, retained frame mailbox, and Metal YUV conversion. The view can outlive the player safely.
 - `Sources/App/PlayerApp.swift`: MainActor state polling, file panel/Finder/drop handling, video host, AppKit glass and native slider cells, floating controls, visibility, video-fitted sizing, and native window integration.
 - `Makefile`: compiler/linker commands and app bundle construction. No generated Xcode project or additional build dependency.
@@ -55,7 +58,7 @@ Only implemented actions have controls; there are no placeholder playlist, subti
 make check
 ```
 
-Generates short fixtures and exercises the real playback interface and native outputs: audio/video, video-only, audio-only, non-square-pixel display aspect, pause, paused/playing and queued seek acknowledgement, completion, replay, close, HDR rejection, failed-open recovery, and destruction while playing. It initializes the actual Metal shaders and uses an offscreen native window. It requires a working audio device and graphical macOS session, and plays short sine-wave audio. It is a runtime smoke check, not a mocked clock/device test.
+Generates short fixtures and exercises the real playback interface and native outputs: audio/video, video-only, audio-only, non-square-pixel display aspect, pause, paused/playing and queued seek acknowledgement, exact bidirectional VFR/B-frame stepping across keyframes, queued/mixed steps, seek-to-step transitions, first/last-frame boundaries, stepping pause/resume, completion, replay, close, HDR rejection, failed-open recovery, and destruction while playing. It initializes the actual Metal shaders and uses an offscreen native window. It requires a working audio device and graphical macOS session, and plays short sine-wave audio. It is a runtime smoke check, not a mocked clock/device test.
 
 `make check-audio` runs just the audio-quality check. A check-only hook captures actual Core Audio callback PCM into preallocated storage and compares every sample against FFmpeg's reference output for PCM/AAC, 44.1/48 kHz, and coarse container timestamps. It checks fractional gain, silent muted startup, smooth live volume changes, and unchanged playback completion. A separate gapped fixture verifies that real timestamp gaps remain silent. Release builds contain no capture hook.
 
@@ -69,4 +72,6 @@ Visual correctness, perceived A/V synchronization, and control usability still n
 
 Local regular files, automatically selected best audio/video tracks, software decoding, and SDR output. PQ/HLG HDR is rejected rather than rendered incorrectly. No subtitles, selectable tracks, VideoToolbox decoding, rotation/deinterlacing, or network playback yet. Mid-track audio format changes report an error.
 
+Frame stepping re-decodes from an earlier keyframe and rebuilds audio at the selected timestamp. Long-GOP or unindexed files can therefore step slowly; frames are not cached across the entire file. Stepping requires usable decoded presentation timestamps and reports an error rather than inventing frame times when they are absent.
+
 Audio buffering is bounded to one second plus one decoded block (up to four seconds); video is bounded to 64 frames/128 MiB. Extremely interleaved files can drop far-future video frames to avoid starving audio; the ceiling and upgrade path are marked at `queueVideo()`. Device/stream latency is accounted for, but device changes during playback are not handled yet.

Sources/App/PlayerApp.swift

@@ -52,6 +52,7 @@ final class PlayerModel: ObservableObject {
             (state.status == PLAYER_PAUSED || state.status == PLAYER_PLAYING || state.status == PLAYER_ENDED)
     }
     var canSeek: Bool { canTogglePlayback && state.duration.isFinite && state.duration > 0 }
+    var canStepFrame: Bool { canTogglePlayback && state.has_video && !isSeeking }
     var canOpen: Bool { handle != nil && !isBusy }
     var canClose: Bool { handle != nil && !isBusy && state.status != PLAYER_EMPTY }
     var canSetVolume: Bool { handle != nil && !isBusy && state.has_audio && state.status != PLAYER_ERROR }
@@ -155,6 +156,14 @@ final class PlayerModel: ObservableObject {
         revealControls()
     }
 
+    func stepFrame(_ direction: Int32) {
+        guard canStepFrame, let handle else { return }
+        let accepted = player_step_frame(handle, direction)
+        if accepted { seekPending = true }
+        commandError = accepted ? nil : "Could not queue the frame step. Please try again."
+        revealControls()
+    }
+
     func setVolume(_ value: Double) {
         guard let handle else { return }
         if player_set_volume(handle, Float(value)) {
@@ -697,6 +706,8 @@ private struct WindowChrome: NSViewRepresentable {
             switch key {
             case "j": model.seekRelative(-10)
             case "l": model.seekRelative(10)
+            case ",": model.stepFrame(-1)
+            case ".": model.stepFrame(1)
             case "k", " ":
                 if !event.isARepeat { model.togglePlayback() }
             case "m":
@@ -825,6 +836,9 @@ struct PlayerApp: App {
                 Button("Seek Back 10 Seconds") { model.seekRelative(-10) }.disabled(!model.canSeek)
                 Button("Seek Forward 10 Seconds") { model.seekRelative(10) }.disabled(!model.canSeek)
                 Divider()
+                Button("Previous Frame") { model.stepFrame(-1) }.disabled(!model.canStepFrame)
+                Button("Next Frame") { model.stepFrame(1) }.disabled(!model.canStepFrame)
+                Divider()
                 Button("Volume Up") { model.setVolume(min(1, model.volume + 0.05)) }
                     .keyboardShortcut(.upArrow, modifiers: []).disabled(!model.canSetVolume)
                 Button("Volume Down") { model.setVolume(max(0, model.volume - 0.05)) }

Sources/Core/player.cpp

@@ -46,7 +46,9 @@ using Frame = std::unique_ptr<AVFrame, FrameDeleter>;
 struct VideoFrame {
     Frame frame;
     double pts;
+    double end;
     std::size_t bytes;
+    uint64_t timestampOccurrence = 0;
 };
 struct Track {
     AVCodecContext *codec = nullptr;
@@ -54,18 +56,26 @@ struct Track {
     bool drainSent = false;
     bool drained = false;
     double nextPts = 0;
+    int64_t countedTimestamp = AV_NOPTS_VALUE;
+    uint64_t timestampOccurrence = 0;
 };
-enum class Action { Open, Play, Pause, Seek, Close };
+enum class Action { Open, Play, Pause, Seek, Step, Close };
 struct Command {
     Action action;
     std::string path;
     double seconds = 0;
+    int direction = 0;
     uint64_t epoch = 0;
 };
 
 void copyError(char *destination, const char *message) noexcept {
     std::snprintf(destination, 512, "%s", message ? message : "Unknown playback error");
 }
+
+int64_t presentationTimestamp(const AVFrame *frame) {
+    return frame->best_effort_timestamp != AV_NOPTS_VALUE ?
+        frame->best_effort_timestamp : frame->pts;
+}
 }
 
 struct Player {
@@ -74,6 +84,7 @@ struct Player {
     std::condition_variable wake;
     std::deque<Command> commands;
     PlayerState state{};
+    uint64_t publishedEpoch = 0;
     std::array<char, 512> requestedError{};
     bool errorRequested = false;
     uint64_t errorEpoch = 0;
@@ -96,6 +107,9 @@ struct Player {
     int receiving = -1;
     std::deque<VideoFrame> video;
     std::size_t videoBytes = 0;
+    VideoFrame presented{};
+    int64_t skipPresentedTimestamp = AV_NOPTS_VALUE;
+    uint64_t skipPresentedOccurrence = 0;
     std::vector<float> pcm;
     std::size_t pcmFrames = 0;
     std::size_t pcmOffset = 0;
@@ -127,6 +141,10 @@ struct Player {
         std::lock_guard<std::mutex> lock(mutex);
         if (stopping.load(std::memory_order_relaxed) || commands.size() >= commandLimit)
             return false;
+        if (command.action == Action::Step &&
+            (!state.has_video || publishedEpoch != requestedEpoch.load(std::memory_order_relaxed) ||
+             (state.status != PLAYER_PAUSED && state.status != PLAYER_PLAYING && state.status != PLAYER_ENDED)))
+            return false;
         const bool newTimeline = command.action == Action::Open || command.action == Action::Close;
         command.epoch = requestedEpoch.load(std::memory_order_relaxed) + (newTimeline ? 1 : 0);
         commands.push_back(std::move(command));
@@ -137,7 +155,8 @@ struct Player {
             state.status = PLAYER_OPENING;
             state.revision = revision;
         }
-        if (commands.back().action == Action::Seek) state.seek_pending = true;
+        if (commands.back().action == Action::Seek || commands.back().action == Action::Step)
+            state.seek_pending = true;
         wake.notify_one();
         return true;
     }
@@ -175,14 +194,16 @@ struct Player {
     void publish() {
         std::lock_guard<std::mutex> lock(mutex);
         if (activeEpoch != requestedEpoch.load(std::memory_order_relaxed)) return;
+        publishedEpoch = activeEpoch;
         state.status = status;
         state.position = position;
         state.duration = duration;
         state.has_audio = tracks[0].codec != nullptr;
         state.has_video = tracks[1].codec != nullptr;
-        // The worker publishes only outside seek execution; retain pending for later queued seeks.
+        // Commands remain pending while executing; publication acknowledges only completed work.
         state.seek_pending = std::any_of(commands.begin(), commands.end(), [this](const Command &command) {
-            return command.action == Action::Seek && command.epoch == activeEpoch;
+            return (command.action == Action::Seek || command.action == Action::Step) &&
+                command.epoch == activeEpoch;
         });
         state.video_aspect_ratio = 0;
         if (const auto *video = tracks[1].codec; video && video->width > 0 && video->height > 0) {
@@ -199,6 +220,8 @@ struct Player {
         output.reset();
         video.clear();
         videoBytes = 0;
+        presented = {};
+        skipPresentedTimestamp = AV_NOPTS_VALUE;
         swr_free(&resampler);
         av_channel_layout_uninit(&inputLayout);
         for (auto &track : tracks) {
@@ -253,6 +276,8 @@ struct Player {
             if (!track.codec) continue;
             avcodec_flush_buffers(track.codec);
             track.drainSent = track.drained = false;
+            track.countedTimestamp = AV_NOPTS_VALUE;
+            track.timestampOccurrence = 0;
             const AVStream *stream = format->streams[track.index];
             track.nextPts = stream->start_time == AV_NOPTS_VALUE ? target :
                 std::max(target, stream->start_time * av_q2d(stream->time_base) - origin);
@@ -265,6 +290,8 @@ struct Player {
         av_frame_unref(decoded);
         video.clear();
         videoBytes = 0;
+        presented = {};
+        skipPresentedTimestamp = AV_NOPTS_VALUE;
         pcmFrames = pcmOffset = 0;
         packetPending = demuxEnded = resamplerDrained = false;
         receiving = -1;
@@ -394,6 +421,108 @@ struct Player {
         if (resume) start();
     }
 
+    void seekVideo(int64_t timestamp, double target) {
+        output.pause();
+        check(av_seek_frame(format, tracks[1].index, timestamp, AVSEEK_FLAG_BACKWARD),
+              "Seeking video keyframe");
+        resetTimeline(target);
+        status = PLAYER_PAUSED;
+    }
+
+    void presentVideo(VideoFrame frame) {
+        output.present(frame.frame.get());
+        presented = std::move(frame);
+        previewShown = true;
+    }
+
+    VideoFrame nextSteppingFrame() {
+        while (video.empty() && !tracks[1].drained) {
+            if (interrupted()) throw Cancelled{};
+            if (!decodeOne(true)) {
+                std::unique_lock<std::mutex> lock(mutex);
+                wake.wait_for(lock, std::chrono::milliseconds(5), [this] { return interrupted(); });
+            }
+        }
+        if (video.empty()) return {};
+        VideoFrame frame = std::move(video.front());
+        video.pop_front();
+        videoBytes -= frame.bytes;
+        return frame;
+    }
+
+    void step(int direction) {
+        if (!format || !tracks[1].codec) return;
+        pause();
+        status = PLAYER_PAUSED;
+        // A seek followed immediately by a queued step must first establish its preview.
+        while (!presented.frame) {
+            auto frame = nextSteppingFrame();
+            if (!frame.frame) return;
+            if (frame.pts + sampleTime * 0.5 < seekTarget) continue;
+            presentVideo(std::move(frame));
+        }
+        VideoFrame selected = std::move(presented);
+        const int64_t reference = presentationTimestamp(selected.frame.get());
+        const uint64_t referenceOccurrence = selected.timestampOccurrence;
+        if (reference == AV_NOPTS_VALUE)
+            throw std::runtime_error("Frame stepping requires video presentation timestamps");
+        const AVStream *stream = format->streams[tracks[1].index];
+        const long double ticksPerSecond = 1.0L / av_q2d(stream->time_base);
+        const long double originTimestamp = std::clamp(std::floor(origin * ticksPerSecond),
+            static_cast<long double>(std::numeric_limits<int64_t>::min() + 1),
+            static_cast<long double>(std::numeric_limits<int64_t>::max()));
+        const int64_t beginning = std::min(reference, std::max(static_cast<int64_t>(originTimestamp),
+            stream->start_time == AV_NOPTS_VALUE ? static_cast<int64_t>(originTimestamp) : stream->start_time));
+        int64_t target = reference > beginning ? reference - 1 : reference;
+        long double backoff = std::max(1.0L, ticksPerSecond);
+        for (;;) {
+            // FFmpeg's decoder emits presentation order, including delayed B-frames at drain.
+            // ponytail: a one-GOP/unindexed file can require a prefix scan; retained frames stay bounded.
+            seekVideo(target, 0);
+            VideoFrame previous{};
+            bool sawReference = false;
+            for (;;) {
+                auto frame = nextSteppingFrame();
+                if (!frame.frame) break;
+                const int64_t timestamp = presentationTimestamp(frame.frame.get());
+                const bool atReference = timestamp == reference &&
+                    frame.timestampOccurrence == referenceOccurrence;
+                const bool beforeReference = timestamp < reference ||
+                    (timestamp == reference && frame.timestampOccurrence < referenceOccurrence);
+                if (atReference) sawReference = true;
+                if (direction > 0) {
+                    if (!beforeReference && !atReference) {
+                        if (sawReference) selected = std::move(frame);
+                        break;
+                    }
+                } else {
+                    if (!beforeReference) break;
+                    if (timestamp >= beginning) previous = std::move(frame);
+                }
+            }
+            if ((direction > 0 && sawReference) || previous.frame || target == beginning) {
+                if (previous.frame) selected = std::move(previous);
+                break;
+            }
+            // Some demuxers land later than requested. Expand the keyframe search, not an FPS guess.
+            target = static_cast<int64_t>(std::max(static_cast<long double>(beginning),
+                static_cast<long double>(reference) - backoff));
+            backoff *= 2;
+        }
+        const int64_t selectedTimestamp = presentationTimestamp(selected.frame.get());
+        const uint64_t selectedOccurrence = selected.timestampOccurrence;
+        const double selectedPts = selected.pts;
+        // Reuse the verified preroll anchor: timestamp seeks can land after a B-frame's PTS.
+        // Audio is rebuilt and trimmed to selectedPts, while video skips the displayed frame.
+        seekVideo(target, selectedPts);
+        if (interrupted()) throw Cancelled{};
+        videoEnd = selected.end;
+        presentVideo(std::move(selected));
+        skipPresentedTimestamp = selectedTimestamp;
+        skipPresentedOccurrence = selectedOccurrence;
+        position = selectedPts;
+    }
+
     void execute(const Command &command) {
         if (command.epoch != requestedEpoch.load(std::memory_order_relaxed)) return;
         activeEpoch = command.epoch;
@@ -402,14 +531,14 @@ struct Player {
             case Action::Play: start(); break;
             case Action::Pause: pause(); break;
             case Action::Seek: seek(command.seconds, status == PLAYER_PLAYING); break;
+            case Action::Step: step(command.direction); break;
             case Action::Close: release(); break;
         }
         publish();
     }
 
     double framePts(Track &track, const AVFrame *frame) {
-        const int64_t timestamp = frame->best_effort_timestamp != AV_NOPTS_VALUE ?
-            frame->best_effort_timestamp : frame->pts;
+        const int64_t timestamp = presentationTimestamp(frame);
         if (timestamp == AV_NOPTS_VALUE) return track.nextPts;
         return timestamp * av_q2d(format->streams[track.index]->time_base) - origin;
     }
@@ -522,17 +651,34 @@ struct Player {
             (!video.empty() && videoBytes > videoByteLimit - video.back().bytes);
     }
 
-    void queueVideo() {
+    void queueVideo(bool stepping = false) {
         auto &track = tracks[1];
         const auto transfer = decoded->color_trc == AVCOL_TRC_UNSPECIFIED ?
             track.codec->color_trc : decoded->color_trc;
         if (transfer == AVCOL_TRC_SMPTE2084 || transfer == AVCOL_TRC_ARIB_STD_B67)
             throw std::runtime_error("HDR PQ/HLG video is unsupported: the native view currently supports SDR only");
+        const int64_t timestamp = presentationTimestamp(decoded);
+        if (stepping && timestamp == AV_NOPTS_VALUE)
+            throw std::runtime_error("Frame stepping requires video presentation timestamps");
         const double pts = framePts(track, decoded);
         const double frameDuration = decoded->duration > 0 ?
             decoded->duration * av_q2d(format->streams[track.index]->time_base) : videoFrameTime;
         track.nextPts = pts + frameDuration;
-        if (pts + sampleTime * 0.5 < seekTarget || (decoded->flags & AV_FRAME_FLAG_DISCARD)) return;
+        if (decoded->flags & AV_FRAME_FLAG_DISCARD) return;
+        if (timestamp != track.countedTimestamp) {
+            track.countedTimestamp = timestamp;
+            track.timestampOccurrence = 0;
+        } else {
+            ++track.timestampOccurrence;
+        }
+        if (!stepping && pts + sampleTime * 0.5 < seekTarget) return;
+        if (!stepping && skipPresentedTimestamp != AV_NOPTS_VALUE) {
+            if (timestamp != AV_NOPTS_VALUE &&
+                (timestamp < skipPresentedTimestamp ||
+                 (timestamp == skipPresentedTimestamp && track.timestampOccurrence <= skipPresentedOccurrence)))
+                return;
+            skipPresentedTimestamp = AV_NOPTS_VALUE;
+        }
         const int size = av_image_get_buffer_size(static_cast<AVPixelFormat>(decoded->format),
                                                    decoded->width, decoded->height, 1);
         check(size, "Sizing decoded video");
@@ -548,16 +694,20 @@ struct Player {
         }
         Frame frame(av_frame_clone(decoded));
         if (!frame) throw std::bad_alloc();
-        VideoFrame item{std::move(frame), pts, bytes};
+        VideoFrame item{std::move(frame), pts, pts + frameDuration, bytes, track.timestampOccurrence};
         auto insertion = std::upper_bound(video.begin(), video.end(), pts,
             [](double value, const VideoFrame &entry) { return value < entry.pts; });
         video.insert(insertion, std::move(item));
         videoBytes += bytes;
     }
 
-    bool decodeOne() {
-        if (pcmOffset < pcmFrames) return false;
+    bool decodeOne(bool stepping = false) {
+        if (!stepping && pcmOffset < pcmFrames) return false;
         if (receiving >= 0) {
+            if (stepping && receiving == 0) {
+                receiving = -1;
+                return true;
+            }
             auto &track = tracks[receiving];
             if (receiving == 1 && videoFull() &&
                 (!tracks[0].codec || output.queued() >= 1024 || status != PLAYER_PLAYING))
@@ -573,11 +723,16 @@ struct Player {
                 return true;
             }
             check(result, "Decoding media");
-            if (receiving == 0) convertAudio(); else queueVideo();
+            if (receiving == 0) convertAudio(); else queueVideo(stepping);
             av_frame_unref(decoded);
             return true;
         }
         if (packetPending) {
+            if (stepping && packet->stream_index == tracks[0].index) {
+                av_packet_unref(packet);
+                packetPending = false;
+                return true;
+            }
             const int slot = packet->stream_index == tracks[0].index ? 0 : 1;
             const int result = avcodec_send_packet(tracks[slot].codec, packet);
             if (result == AVERROR(EAGAIN)) {
@@ -593,7 +748,7 @@ struct Player {
         if (demuxEnded) {
             for (int slot = 0; slot < 2; ++slot) {
                 auto &track = tracks[slot];
-                if (!track.codec || track.drainSent) continue;
+                if (!track.codec || track.drainSent || (stepping && slot == 0)) continue;
                 const int result = avcodec_send_packet(track.codec, nullptr);
                 if (result == AVERROR(EAGAIN)) {
                     receiving = slot;
@@ -604,7 +759,7 @@ struct Player {
                 receiving = slot;
                 return true;
             }
-            if ((!tracks[0].codec || tracks[0].drained) && !resamplerDrained) {
+            if (!stepping && (!tracks[0].codec || tracks[0].drained) && !resamplerDrained) {
                 drainResampler();
                 return true;
             }
@@ -617,7 +772,8 @@ struct Player {
         }
         if (result == AVERROR(EAGAIN)) return false;
         check(result, "Reading media packet");
-        if (packet->stream_index != tracks[0].index && packet->stream_index != tracks[1].index) {
+        if ((packet->stream_index != tracks[0].index && packet->stream_index != tracks[1].index) ||
+            (stepping && packet->stream_index == tracks[0].index)) {
             av_packet_unref(packet);
         } else {
             packetPending = true;
@@ -626,16 +782,13 @@ struct Player {
     }
 
     void renderVideo(double now) {
-        Frame latest;
+        VideoFrame latest{};
         while (!video.empty() && video.front().pts <= now + 0.002) {
             videoBytes -= video.front().bytes;
-            latest = std::move(video.front().frame);
+            latest = std::move(video.front());
             video.pop_front();
         }
-        if (latest) {
-            output.present(latest.get());
-            previewShown = true;
-        }
+        if (latest.frame) presentVideo(std::move(latest));
     }
 
     bool tick() {
@@ -643,10 +796,10 @@ struct Player {
         bool progressed = pumpAudio();
         if (status == PLAYER_PAUSED) {
             if (!previewShown && !video.empty()) {
-                output.present(video.front().frame.get());
-                videoBytes -= video.front().bytes;
+                auto frame = std::move(video.front());
+                videoBytes -= frame.bytes;
                 video.pop_front();
-                previewShown = true;
+                presentVideo(std::move(frame));
                 progressed = true;
             }
             const bool needPreview = tracks[1].codec && !previewShown;
@@ -756,6 +909,10 @@ bool player_seek(Player *player, double seconds) {
     if (!std::isfinite(seconds)) return false;
     return submit(player, [=] { return player->enqueue({Action::Seek, {}, seconds}); });
 }
+bool player_step_frame(Player *player, int direction) {
+    if (direction != -1 && direction != 1) return false;
+    return submit(player, [=] { return player->enqueue({Action::Step, {}, 0, direction}); });
+}
 bool player_close(Player *player) {
     return submit(player, [=] { return player->enqueue({Action::Close, {}}); });
 }

Sources/Core/player.h

@@ -17,7 +17,7 @@ typedef struct PlayerState {
     double duration;
     bool has_audio;
     bool has_video;
-    // True from accepted seek until all queued seeks on this timeline finish.
+    // True until all accepted seeks/frame steps on this timeline finish.
     bool seek_pending;
     // Display width / height, including sample aspect ratio; zero without video.
     double video_aspect_ratio;
@@ -32,6 +32,9 @@ bool player_open(Player *player, const char *path);
 bool player_play(Player *player);
 bool player_pause(Player *player);
 bool player_seek(Player *player, double seconds);
+// Direction must be -1 (previous) or +1 (next). Requires ready video media.
+// Pauses playback and selects the adjacent decoded presentation frame.
+bool player_step_frame(Player *player, int direction);
 bool player_close(Player *player);
 // Synchronous, thread-safe gain; survives open, seek, and close. Zero mutes.
 // Rejects null handles and non-finite/out-of-range values without changing gain.