final linux version

.gitignore

@@ -20,5 +20,8 @@ build
 lib/linux-x86_64
 build-ffmpeg
 
+.DS_Store
+*.user
+
 
 /ffmpeg-audio/.qtcreator

Makefile

@@ -10,8 +10,12 @@ all:
 
 install: all
 	mkdir -p lib/$(SUBDIR)
+	@echo "copying from src/$(SUBDIR) to lib/$(SUBDIR)"
+	exit
+	(cd src/$(SUBDIR);tar cf - . ) | (cd lib/$(SUBDIR); tar xvf - )
 	FILES=`ls build/*.so build-ffmpeg/*.so` 2>/dev/null; if [ "$$FILES" != "" ]; then cp $$FILES lib/$(SUBDIR); fi
 	FILES=`ls build/*.dll build-ffmpeg/*.dll` 2>/dev/null; if [ "$$FILES" != "" ]; then cp $$FILES lib/$(SUBDIR); fi
+	FILES=`ls build/*.dylib build-ffmpeg/*.dylib` 2>/dev/null; if [ "$$FILES" != "" ]; then cp $$FILES lib/$(SUBDIR); fi
 
 test: install
 	cp lib/linux-x86_64/*.so ~/.local/share/racket/racket-sound-lib/linux-x86_64

README.md

@@ -1,3 +1,14 @@
 # racket-sound-lib
 
 Binary packages for the racket-sound FFI binding
+
+## Using on Mac OS X 
+
+Make sure you have libao, libFLAC, mpg123 and ffmpeg-full installed using brew.
+
+% brew install libao
+% brew install flac
+% brew install mpg123
+% brew install ffmpeg-full
+
+

ao-play-async/CMakeLists.txt

@@ -5,6 +5,8 @@ project(ao-play-async LANGUAGES C)
 set(CMAKE_CXX_STANDARD 17)
 set(CMAKE_CXX_STANDARD_REQUIRED ON)
 
+include_directories(${CMAKE_CURRENT_SOURCE_DIR}/..)
+
 add_library(ao-play-async SHARED
     ao_playasync.c
     ao_playasync.h
@@ -16,6 +18,12 @@ if(WIN32)
     target_link_directories(ao-play-async PRIVATE ../lib/windows-x86_64)
 endif()
 
+if(APPLE)
+    set(AO_HOME /opt/homebrew)
+    include_directories(${AO_HOME}/include)
+    target_link_directories(ao-play-async PRIVATE ${AO_HOME}/lib)
+endif()
+
 if(WIN32)
     target_link_libraries(ao-play-async PRIVATE libao-1.2.2 winmm.lib ksuser.lib)
 else()

ao-play-async/ao_playasync.h

@@ -11,7 +11,7 @@
 #define AOPLAYASYNC_EXPORT extern
 #endif
 
-#define VERSION 2
+#include "../ffi_version.h"
 
 typedef enum {
     ao = 1,
@@ -46,4 +46,5 @@ AOPLAYASYNC_EXPORT double ao_volume_async(void *ao_handle);
 AOPLAYASYNC_EXPORT int ao_bufsize_async(void *handle);
 AOPLAYASYNC_EXPORT int ao_real_output_bits_async(void *handle);
 
+
 #endif // AO_PLAYASYNC_H

ffi_version.h

@@ -0,0 +1,10 @@
+#ifndef __FFI_VERSION__
+#define __FFI_VERSION__
+
+#define VERSION_MAJOR 1
+#define VERSION_MINOR 0
+#define VERSION_PATCH 1
+
+#define ffi_version() ((VERSION_MAJOR << 16) + (VERSION_MINOR << 8) + VERSION_PATCH)
+
+#endif

ffmpeg-audio/CMakeLists.txt

@@ -8,6 +8,8 @@ set(CMAKE_CXX_STANDARD_REQUIRED ON)
 add_library(ffmpeg_audio SHARED
     ffmpeg_audio.cpp
     ffmpeg_audio.h
+    ../ffi_version.h
+    ffmpeg_audio_refactored.cpp
 )
 
 add_executable(demo_ffmpeg_audio
@@ -21,6 +23,11 @@ if(WIN32)
     target_link_directories(ffmpeg_audio PRIVATE ${FFMPEG_LIB})
     set(FFMPEG_LIBRARIES avcodec.lib avformat.lib swresample.lib avutil.lib avdevice.lib avfilter.lib swscale.lib)
     target_link_directories(demo_ffmpeg_audio PRIVATE ${FFMPEG_LIB})
+elseif(APPLE)
+    target_link_directories(ffmpeg_audio PRIVATE /opt/homebrew/opt/ffmpeg-full/lib)
+    include_directories(/opt/homebrew/opt/ffmpeg-full/include)
+    set(FFMPEG_LIBRARIES avcodec avformat swresample avutil avdevice avfilter swscale)
+    target_link_directories(demo_ffmpeg_audio PRIVATE /opt/homebrew/opt/ffmpeg-full/lib)
 else()
     set(FFMPEG_LIBRARIES avcodec avformat swresample avutil avdevice avfilter swscale)
 endif()

ffmpeg-audio/ffmpeg_audio.cpp

@@ -13,6 +13,7 @@
  */
 
 #include "ffmpeg_audio.h"
+#include "../ffi_version.h"
 
 #include <algorithm>
 #include <cstdint>
@@ -1021,3 +1022,8 @@ const char *fmpg_int_version2string(int ver)
 
     return version;
 }
+
+int fmpg_version()
+{
+    return ffi_version();
+}

ffmpeg-audio/ffmpeg_audio.h

@@ -150,6 +150,7 @@ FFMPEG_EXTERN int64_t fmpg_sample_position(fmpg_instance *instance);
 /* Approximate start time of the current decoded block in seconds. */
 FFMPEG_EXTERN double fmpg_timecode(fmpg_instance *instance);
 
+FFMPEG_EXTERN int fmpg_version();
 FFMPEG_EXTERN const char *fmpg_ffmpeg_version();
 FFMPEG_EXTERN const char *fmpg_int_version2string(int ver);
 FFMPEG_EXTERN int fmpg_compatible_ffmpeg();

ffmpeg-audio/ffmpeg_audio_refactored.cpp

@@ -0,0 +1,900 @@
+/*
+ * Audio-only FFmpeg wrapper with a plain C ABI.
+ *
+ * This implementation intentionally hides FFmpeg concepts from the public API:
+ *
+ *   - no stream_index in the API;
+ *   - no AVPacket/package object in the API;
+ *   - no explicit decoder object in the API;
+ *   - no metadata/tag API exposed to C callers.
+ *
+ * Internally the instance owns everything needed to decode one selected audio
+ * stream. The caller simply opens a file and repeatedly calls fmpg_decode_next().
+ */
+
+#include "ffmpeg_audio.h"
+#include "../ffi_version.h"
+
+#include <algorithm>
+#include <cstdint>
+#include <cstdio>
+#include <cstdlib>
+#include <cstring>
+#include <limits>
+#include <vector>
+
+extern "C" {
+#include <libavcodec/avcodec.h>
+#include <libavformat/avformat.h>
+#include <libavutil/avutil.h>
+#include <libavutil/channel_layout.h>
+#include <libavutil/samplefmt.h>
+#include <libswresample/swresample.h>
+}
+
+#define MSG0(type, msg) fprintf(stderr, type ":");fprintf(stderr, "%s", __FUNCTION__);fprintf(stderr, ": %s\n", msg)
+#define MSG1(type, msg, a) fprintf(stderr, type ":");fprintf(stderr, "%s", __FUNCTION__);fprintf(stderr, ": " msg "\n", a)
+#define MSG2(type, msg, a, b) fprintf(stderr, type ":");fprintf(stderr, "%s", __FUNCTION__);fprintf(stderr, ": " msg "\n", a, b)
+#define MSG3(type, msg, a, b, c) fprintf(stderr, type ":");fprintf(stderr, "%s", __FUNCTION__);fprintf(stderr, ": " msg "\n", a, b, c)
+
+#define INFO0(msg) MSG0("info", msg)
+#define INFO1(msg, a) MSG1("info", msg, a)
+#define INFO2(msg, a, b) MSG2("info", msg, a, b)
+#define INFO3(msg, a, b, c) MSG3("info", msg, a, b, c)
+
+#define ERROR0(msg) MSG0("error", msg)
+#define ERROR1(msg, a) MSG1("error", msg, a)
+#define ERROR2(msg, a, b) MSG2("error", msg, a, b)
+#define ERROR3(msg, a, b, c) MSG3("error", msg, a, b, c)
+
+static constexpr int FMPG_OUTPUT_BITS = 32;
+static constexpr int FMPG_OUTPUT_BYTES = 4;
+static constexpr AVSampleFormat FMPG_OUTPUT_FMT = AV_SAMPLE_FMT_S32;
+
+struct audio_info_storage {
+    int audio_stream_count = 0;
+    int selected_stream_index = -1; /* Internal FFmpeg stream index. */
+    int sample_rate = 0;
+    int channels = 0;
+    int64_t duration_ms = -1;
+    int64_t duration_samples = -1;  /* Output sample frames. */
+
+    void clear()
+    {
+        audio_stream_count = 0;
+        selected_stream_index = -1;
+        sample_rate = 0;
+        channels = 0;
+        duration_ms = -1;
+        duration_samples = -1;
+    }
+};
+
+struct decoder_storage {
+    const AVCodec *codec = nullptr;
+    AVCodecContext *codec_ctx = nullptr;
+    AVFrame *frame = nullptr;
+    SwrContext *swr_ctx = nullptr;
+
+    std::vector<uint8_t> pcm;
+
+    bool eof_seen = false;
+    bool decoder_drained = false;
+
+    double timecode = 0.0;
+
+    int64_t last_samples = 0;
+    int64_t buffer_start_sample = 0;
+    int64_t next_sample_position = 0;
+
+    /* >= 0 while a seek has requested us to discard decoded pre-roll samples. */
+    int64_t discard_until_sample = -1;
+
+    void clear_output()
+    {
+        pcm.clear();
+        last_samples = 0;
+        buffer_start_sample = next_sample_position;
+    }
+
+    void free_ffmpeg()
+    {
+        avcodec_free_context(&codec_ctx);
+        av_frame_free(&frame);
+        swr_free(&swr_ctx);
+        codec = nullptr;
+        pcm.clear();
+        eof_seen = false;
+        decoder_drained = false;
+        timecode = 0.0;
+        last_samples = 0;
+        buffer_start_sample = 0;
+        next_sample_position = 0;
+        discard_until_sample = -1;
+    }
+
+    ~decoder_storage()
+    {
+        free_ffmpeg();
+    }
+};
+
+struct __fmpg_instance__ {
+    bool opened = false;
+    AVFormatContext *format_ctx = nullptr;
+    audio_info_storage audio_info;
+    decoder_storage decoder;
+
+    ~__fmpg_instance__()
+    {
+        if (format_ctx) {
+            avformat_close_input(&format_ctx);
+        }
+    }
+};
+
+static int count_audio_streams(const AVFormatContext *ctx)
+{
+    if (!ctx) {
+        return 0;
+    }
+
+    int count = 0;
+    for (unsigned i = 0; i < ctx->nb_streams; ++i) {
+        const AVCodecParameters *par = ctx->streams[i]->codecpar;
+        if (par && par->codec_type == AVMEDIA_TYPE_AUDIO) {
+            ++count;
+        }
+    }
+    return count;
+}
+
+static int64_t milliseconds_from_seconds(double seconds)
+{
+    if (seconds < 0.0) {
+        return -1;
+    }
+    return static_cast<int64_t>(seconds * 1000.0 + 0.5);
+}
+
+static int64_t samples_from_seconds(double seconds, int sample_rate)
+{
+    if (seconds < 0.0 || sample_rate <= 0) {
+        return -1;
+    }
+    return static_cast<int64_t>(seconds * static_cast<double>(sample_rate) +
+                                0.5);
+}
+
+static double stream_duration_seconds(const AVStream *stream)
+{
+    if (!stream || stream->duration == AV_NOPTS_VALUE) {
+        return -1.0;
+    }
+    return static_cast<double>(stream->duration) * av_q2d(stream->time_base);
+}
+
+static double format_duration_seconds(const AVFormatContext *ctx)
+{
+    if (!ctx || ctx->duration == AV_NOPTS_VALUE) {
+        return -1.0;
+    }
+    return static_cast<double>(ctx->duration) /
+           static_cast<double>(AV_TIME_BASE);
+}
+
+static int64_t timestamp_to_samples(int64_t timestamp,
+                                    const AVStream *stream,
+                                    int sample_rate)
+{
+    if (!stream || timestamp == AV_NOPTS_VALUE || sample_rate <= 0) {
+        return -1;
+    }
+
+    const double seconds = static_cast<double>(timestamp) *
+                           av_q2d(stream->time_base);
+    return samples_from_seconds(seconds, sample_rate);
+}
+
+static bool fill_audio_info(fmpg_instance *self)
+{
+    AVFormatContext *ctx = self->format_ctx;
+
+    self->audio_info.clear();
+    self->audio_info.audio_stream_count = count_audio_streams(ctx);
+
+    const int best = av_find_best_stream(ctx,
+                                         AVMEDIA_TYPE_AUDIO,
+                                         -1,
+                                         -1,
+                                         nullptr,
+                                         0);
+    if (best < 0) {
+        return false;
+    }
+
+    AVStream *stream = ctx->streams[best];
+    const AVCodecParameters *par = stream->codecpar;
+
+    if (!par || par->codec_type != AVMEDIA_TYPE_AUDIO ||
+        par->sample_rate <= 0 || par->ch_layout.nb_channels <= 0) {
+        return false;
+    }
+
+    self->audio_info.selected_stream_index = best;
+    self->audio_info.sample_rate = par->sample_rate;
+    self->audio_info.channels = par->ch_layout.nb_channels;
+
+    double seconds = stream_duration_seconds(stream);
+    if (seconds < 0.0) {
+        seconds = format_duration_seconds(ctx);
+    }
+
+    self->audio_info.duration_ms = milliseconds_from_seconds(seconds);
+    self->audio_info.duration_samples =
+        samples_from_seconds(seconds, self->audio_info.sample_rate);
+
+    return true;
+}
+
+static bool instance_ready(const fmpg_instance *instance)
+{
+    return instance && instance->opened && instance->format_ctx &&
+           instance->audio_info.selected_stream_index >= 0 &&
+           instance->decoder.codec_ctx && instance->decoder.swr_ctx;
+}
+
+static bool init_codec_context(fmpg_instance *self)
+{
+    decoder_storage &dec = self->decoder;
+    const int stream_index = self->audio_info.selected_stream_index;
+    const AVCodecParameters *par =
+        self->format_ctx->streams[stream_index]->codecpar;
+
+    dec.codec = avcodec_find_decoder(par->codec_id);
+    if (!dec.codec) {
+        return false;
+    }
+
+    dec.codec_ctx = avcodec_alloc_context3(dec.codec);
+    if (!dec.codec_ctx) {
+        return false;
+    }
+
+    if (avcodec_parameters_to_context(dec.codec_ctx, par) < 0) {
+        return false;
+    }
+
+    if (avcodec_open2(dec.codec_ctx, dec.codec, nullptr) < 0) {
+        return false;
+    }
+
+    dec.frame = av_frame_alloc();
+    return dec.frame != nullptr;
+}
+
+static bool init_resampler(fmpg_instance *self)
+{
+    decoder_storage &dec = self->decoder;
+    const AVChannelLayout *layout = &dec.codec_ctx->ch_layout;
+
+    if (layout->nb_channels <= 0 || dec.codec_ctx->sample_rate <= 0) {
+        return false;
+    }
+
+    if (swr_alloc_set_opts2(&dec.swr_ctx,
+                            layout,
+                            FMPG_OUTPUT_FMT,
+                            dec.codec_ctx->sample_rate,
+                            layout,
+                            dec.codec_ctx->sample_fmt,
+                            dec.codec_ctx->sample_rate,
+                            0,
+                            nullptr) < 0) {
+        return false;
+    }
+
+    return swr_init(dec.swr_ctx) >= 0;
+}
+
+static bool init_decoder(fmpg_instance *self)
+{
+    self->decoder.free_ffmpeg();
+    return init_codec_context(self) && init_resampler(self);
+}
+
+int fmpg_compatible_ffmpeg()
+{
+    int compiled_avformat_major = LIBAVFORMAT_VERSION_MAJOR;
+    int compiled_avcodec_major = LIBAVCODEC_VERSION_MAJOR;
+    int compiled_swresample_major = LIBSWRESAMPLE_VERSION_MAJOR;
+    int compiled_avutil_major = LIBAVUTIL_VERSION_MAJOR;
+
+    int current_avformat_major = AV_VERSION_MAJOR(avformat_version());
+    int current_avcodec_major = AV_VERSION_MAJOR(avcodec_version());
+    int current_swresample_major = AV_VERSION_MAJOR(swresample_version());
+    int current_avutil_major = AV_VERSION_MAJOR(avutil_version());
+
+    auto comp = [](const char *lib, int cv, int rv) {
+        if (cv != rv) {
+            ERROR3("FFMPEG %s Major versions not equal. Compile time: %d, runtime: %d", lib, cv, rv);
+        }
+    };
+
+    comp("AVFormat", compiled_avformat_major, current_avformat_major);
+    comp("AVCodec", compiled_avcodec_major, current_avcodec_major);
+    comp("SWResample", compiled_swresample_major, current_swresample_major);
+    comp("AVUtil", compiled_avutil_major, current_avutil_major);
+
+    int compatible = (compiled_avformat_major == current_avformat_major) &&
+                     (compiled_avcodec_major == current_avcodec_major) &&
+                     (compiled_swresample_major == current_swresample_major) &&
+                     (compiled_avutil_major == current_avutil_major);
+
+    return compatible;
+}
+
+fmpg_instance *fmpg_init(void)
+{
+    if (!fmpg_compatible_ffmpeg()) {
+        ERROR0("Compiled major ffmpeg version ≃ runtime major version, not compatible.");
+        return nullptr;
+    }
+
+    try {
+        return new fmpg_instance();
+    } catch (...) {
+        return nullptr;
+    }
+}
+
+void fmpg_free(fmpg_instance *instance)
+{
+    delete instance;
+}
+
+int fmpg_open_file(fmpg_instance *instance, const char *filename)
+{
+    if (!instance || instance->opened || !filename) {
+        return 0;
+    }
+
+    if (instance->format_ctx != nullptr) {
+        return 0;
+    }
+
+    int r = avformat_open_input(&instance->format_ctx,
+                                filename,
+                                nullptr,
+                                nullptr);
+    if (r < 0) {
+        fmpg_close(instance);
+        return 0;
+    }
+
+
+    if (avformat_find_stream_info(instance->format_ctx, nullptr) < 0) {
+        fmpg_close(instance);
+        return 0;
+    }
+
+
+    if (!fill_audio_info(instance) || !init_decoder(instance)) {
+        fmpg_close(instance);
+        return 0;
+    }
+
+    instance->opened = true;
+    return 1;
+}
+
+void fmpg_close(fmpg_instance *instance)
+{
+    if (!instance) {
+        return;
+    }
+
+    instance->decoder.free_ffmpeg();
+
+    if (instance->format_ctx) {
+        avformat_close_input(&instance->format_ctx);
+    }
+
+    instance->opened = false;
+    instance->audio_info.clear();
+}
+
+int fmpg_is_open(fmpg_instance *instance)
+{
+    return instance_ready(instance) ? 1 : 0;
+}
+
+int fmpg_audio_stream_count(fmpg_instance *instance)
+{
+    return instance && instance->opened ? instance->audio_info.audio_stream_count
+                                        : 0;
+}
+
+int fmpg_audio_sample_rate(fmpg_instance *instance)
+{
+    return instance_ready(instance) ? instance->audio_info.sample_rate : 0;
+}
+
+int fmpg_audio_channels(fmpg_instance *instance)
+{
+    return instance_ready(instance) ? instance->audio_info.channels : 0;
+}
+
+int fmpg_audio_bits_per_sample(fmpg_instance *)
+{
+    return FMPG_OUTPUT_BITS;
+}
+
+int fmpg_audio_bytes_per_sample(fmpg_instance *)
+{
+    return FMPG_OUTPUT_BYTES;
+}
+
+int64_t fmpg_duration_ms(fmpg_instance *instance)
+{
+    return instance_ready(instance) ? instance->audio_info.duration_ms : -1;
+}
+
+int64_t fmpg_duration_samples(fmpg_instance *instance)
+{
+    return instance_ready(instance) ? instance->audio_info.duration_samples : -1;
+}
+
+static bool append_bytes(decoder_storage &dec,
+                         const uint8_t *src,
+                         size_t bytes)
+{
+    if (!bytes) {
+        return true;
+    }
+
+    if (bytes > static_cast<size_t>(std::numeric_limits<int>::max()) -
+                    dec.pcm.size()) {
+        return false;
+    }
+
+    try {
+        const size_t old_size = dec.pcm.size();
+        dec.pcm.resize(old_size + bytes);
+        std::memcpy(dec.pcm.data() + old_size, src, bytes);
+        return true;
+    } catch (...) {
+        return false;
+    }
+}
+
+static bool append_converted_frame(fmpg_instance *self,
+                                   const AVFrame *frame)
+{
+    decoder_storage &dec = self->decoder;
+    const int channels = self->audio_info.channels;
+    const int sample_rate = self->audio_info.sample_rate;
+
+    if (channels <= 0 || frame->nb_samples <= 0) {
+        return true;
+    }
+
+    const int max_out_samples = swr_get_out_samples(dec.swr_ctx,
+                                                    frame->nb_samples);
+    if (max_out_samples <= 0) {
+        return false;
+    }
+
+    const int max_bytes = av_samples_get_buffer_size(nullptr,
+                                                     channels,
+                                                     max_out_samples,
+                                                     FMPG_OUTPUT_FMT,
+                                                     1);
+    if (max_bytes <= 0) {
+        return false;
+    }
+
+    std::vector<uint8_t> tmp(static_cast<size_t>(max_bytes));
+    uint8_t *out_planes[1] = { tmp.data() };
+
+    const int out_samples = swr_convert(dec.swr_ctx,
+                                        out_planes,
+                                        max_out_samples,
+                                        const_cast<const uint8_t **>(frame->data),
+                                        frame->nb_samples);
+    if (out_samples < 0) {
+        return false;
+    }
+
+    const int used_bytes = av_samples_get_buffer_size(nullptr,
+                                                      channels,
+                                                      out_samples,
+                                                      FMPG_OUTPUT_FMT,
+                                                      1);
+    if (used_bytes < 0) {
+        return false;
+    }
+
+    const int stream_index = self->audio_info.selected_stream_index;
+    const AVStream *stream = self->format_ctx->streams[stream_index];
+
+    int64_t frame_start = timestamp_to_samples(frame->best_effort_timestamp,
+                                               stream,
+                                               sample_rate);
+    if (frame_start < 0) {
+        frame_start = dec.next_sample_position;
+    }
+
+    int64_t keep_start = frame_start;
+    int keep_samples = out_samples;
+    size_t byte_offset = 0;
+
+    /*
+     * After seeking, FFmpeg may first return decoded samples from before the
+     * requested position. Discard them so public sample positions refer to the
+     * actual music position requested by the caller.
+     */
+    if (dec.discard_until_sample >= 0) {
+        const int64_t target = dec.discard_until_sample;
+        const int64_t frame_end = frame_start + out_samples;
+
+        if (frame_end <= target) {
+            dec.next_sample_position = frame_end;
+            return true;
+        }
+
+        if (frame_start < target) {
+            const int64_t drop = target - frame_start;
+            if (drop > 0 && drop < out_samples) {
+                byte_offset = static_cast<size_t>(drop) *
+                              static_cast<size_t>(channels) *
+                              FMPG_OUTPUT_BYTES;
+                keep_samples = static_cast<int>(out_samples - drop);
+                keep_start = target;
+            }
+        }
+
+        dec.discard_until_sample = -1;
+    }
+
+    if (keep_samples <= 0) {
+        dec.next_sample_position = frame_start + out_samples;
+        return true;
+    }
+
+    if (dec.pcm.empty()) {
+        dec.buffer_start_sample = keep_start;
+        dec.timecode = static_cast<double>(keep_start) /
+                       static_cast<double>(sample_rate);
+    }
+
+    const size_t keep_bytes = static_cast<size_t>(keep_samples) *
+                              static_cast<size_t>(channels) *
+                              FMPG_OUTPUT_BYTES;
+
+    if (!append_bytes(dec, tmp.data() + byte_offset, keep_bytes)) {
+        return false;
+    }
+
+    dec.last_samples += keep_samples;
+    dec.next_sample_position = keep_start + keep_samples;
+    return true;
+}
+
+static int receive_available_frames(fmpg_instance *self)
+{
+    decoder_storage &dec = self->decoder;
+    int produced = 0;
+
+    for (;;) {
+        const int ret = avcodec_receive_frame(dec.codec_ctx, dec.frame);
+
+        if (ret == AVERROR(EAGAIN)) {
+            return produced;
+        }
+
+        if (ret == AVERROR_EOF) {
+            dec.decoder_drained = true;
+            return produced;
+        }
+
+        if (ret < 0) {
+            return -1;
+        }
+
+        if (!append_converted_frame(self, dec.frame)) {
+            av_frame_unref(dec.frame);
+            return -1;
+        }
+
+        produced = dec.last_samples > 0 ? 1 : produced;
+        av_frame_unref(dec.frame);
+    }
+}
+
+static bool read_selected_audio_packet(fmpg_instance *self, AVPacket *pkt)
+{
+    const int wanted_stream = self->audio_info.selected_stream_index;
+
+    for (;;) {
+        const int ret = av_read_frame(self->format_ctx, pkt);
+        if (ret < 0) {
+            return false;
+        }
+
+        if (pkt->stream_index == wanted_stream) {
+            return true;
+        }
+
+        av_packet_unref(pkt);
+    }
+}
+
+static int drain_resampler(fmpg_instance *self)
+{
+    decoder_storage &dec = self->decoder;
+    const int channels = self->audio_info.channels;
+    const int sample_rate = self->audio_info.sample_rate;
+    int produced = 0;
+
+    for (;;) {
+        const int delay = static_cast<int>(swr_get_delay(dec.swr_ctx,
+                                                        sample_rate));
+        if (delay <= 0) {
+            break;
+        }
+
+        const int max_bytes = av_samples_get_buffer_size(nullptr,
+                                                         channels,
+                                                         delay,
+                                                         FMPG_OUTPUT_FMT,
+                                                         1);
+        if (max_bytes <= 0) {
+            break;
+        }
+
+        std::vector<uint8_t> tmp(static_cast<size_t>(max_bytes));
+        uint8_t *out_planes[1] = { tmp.data() };
+
+        const int out_samples = swr_convert(dec.swr_ctx,
+                                            out_planes,
+                                            delay,
+                                            nullptr,
+                                            0);
+        if (out_samples <= 0) {
+            break;
+        }
+
+        const int used_bytes = av_samples_get_buffer_size(nullptr,
+                                                          channels,
+                                                          out_samples,
+                                                          FMPG_OUTPUT_FMT,
+                                                          1);
+        if (used_bytes < 0) {
+            break;
+        }
+
+        if (dec.pcm.empty()) {
+            dec.buffer_start_sample = dec.next_sample_position;
+            dec.timecode = static_cast<double>(dec.buffer_start_sample) /
+                           static_cast<double>(sample_rate);
+        }
+
+        if (!append_bytes(dec, tmp.data(), static_cast<size_t>(used_bytes))) {
+            return -1;
+        }
+
+        dec.last_samples += out_samples;
+        dec.next_sample_position += out_samples;
+        produced = 1;
+    }
+
+    return produced;
+}
+
+int fmpg_decode_next(fmpg_instance *instance)
+{
+    if (!instance_ready(instance)) {
+        return 0;
+    }
+
+    decoder_storage &dec = instance->decoder;
+    dec.clear_output();
+
+    /* First return any frames that are already pending in the decoder. */
+    int produced = receive_available_frames(instance);
+    if (produced < 0) {
+        return 0;
+    }
+    if (produced > 0 && !dec.pcm.empty()) {
+        return 1;
+    }
+
+    AVPacket *pkt = av_packet_alloc();
+    if (!pkt) {
+        return 0;
+    }
+
+    while (!dec.eof_seen) {
+        if (!read_selected_audio_packet(instance, pkt)) {
+            dec.eof_seen = true;
+            av_packet_unref(pkt);
+            break;
+        }
+
+        int ret = avcodec_send_packet(dec.codec_ctx, pkt);
+        av_packet_unref(pkt);
+
+        if (ret == AVERROR(EAGAIN)) {
+            produced = receive_available_frames(instance);
+            if (produced < 0) {
+                av_packet_free(&pkt);
+                return 0;
+            }
+            if (produced > 0 && !dec.pcm.empty()) {
+                av_packet_free(&pkt);
+                return 1;
+            }
+            continue;
+        }
+
+        if (ret < 0) {
+            av_packet_free(&pkt);
+            return 0;
+        }
+
+        produced = receive_available_frames(instance);
+        if (produced < 0) {
+            av_packet_free(&pkt);
+            return 0;
+        }
+        if (produced > 0 && !dec.pcm.empty()) {
+            av_packet_free(&pkt);
+            return 1;
+        }
+    }
+
+    av_packet_free(&pkt);
+
+    if (!dec.decoder_drained) {
+        const int ret = avcodec_send_packet(dec.codec_ctx, nullptr);
+        if (ret < 0 && ret != AVERROR_EOF) {
+            return 0;
+        }
+
+        produced = receive_available_frames(instance);
+        if (produced < 0) {
+            return 0;
+        }
+        if (produced > 0 && !dec.pcm.empty()) {
+            return 1;
+        }
+    }
+
+    produced = drain_resampler(instance);
+    return produced > 0 && !dec.pcm.empty() ? 1 : 0;
+}
+
+int fmpg_seek_ms(fmpg_instance *instance, int64_t target_pos_ms)
+{
+    if (!instance_ready(instance) || target_pos_ms < 0) {
+        return 0;
+    }
+
+    const int stream_index = instance->audio_info.selected_stream_index;
+    AVStream *stream = instance->format_ctx->streams[stream_index];
+
+    const int64_t pos_us = av_rescale(target_pos_ms, AV_TIME_BASE, 1000);
+    const int64_t stream_ts = av_rescale_q(pos_us,
+                                           AV_TIME_BASE_Q,
+                                           stream->time_base);
+
+    if (av_seek_frame(instance->format_ctx,
+                      stream_index,
+                      stream_ts,
+                      AVSEEK_FLAG_BACKWARD) < 0) {
+        return 0;
+    }
+
+    decoder_storage &dec = instance->decoder;
+    const int64_t target_samples = samples_from_seconds(target_pos_ms / 1000.0,
+                                                        instance->audio_info.sample_rate);
+
+    avcodec_flush_buffers(dec.codec_ctx);
+    swr_close(dec.swr_ctx);
+    if (swr_init(dec.swr_ctx) < 0) {
+        return 0;
+    }
+
+    dec.pcm.clear();
+    dec.last_samples = 0;
+    dec.buffer_start_sample = target_samples >= 0 ? target_samples : 0;
+    dec.next_sample_position = target_samples >= 0 ? target_samples : 0;
+    dec.discard_until_sample = target_samples;
+    dec.timecode = target_pos_ms / 1000.0;
+    dec.eof_seen = false;
+    dec.decoder_drained = false;
+
+    return 1;
+}
+
+const uint8_t *fmpg_buffer(fmpg_instance *instance)
+{
+    return instance && !instance->decoder.pcm.empty()
+               ? instance->decoder.pcm.data()
+               : nullptr;
+}
+
+int fmpg_buffer_size(fmpg_instance *instance)
+{
+    if (!instance || instance->decoder.pcm.size() >
+                         static_cast<size_t>(std::numeric_limits<int>::max())) {
+        return 0;
+    }
+    return static_cast<int>(instance->decoder.pcm.size());
+}
+
+int64_t fmpg_buffer_samples(fmpg_instance *instance)
+{
+    return instance ? instance->decoder.last_samples : 0;
+}
+
+int64_t fmpg_buffer_start_sample(fmpg_instance *instance)
+{
+    return instance ? instance->decoder.buffer_start_sample : 0;
+}
+
+int64_t fmpg_buffer_end_sample(fmpg_instance *instance)
+{
+    if (!instance) {
+        return 0;
+    }
+    return instance->decoder.buffer_start_sample + instance->decoder.last_samples;
+}
+
+int64_t fmpg_sample_position(fmpg_instance *instance)
+{
+    return instance ? instance->decoder.next_sample_position : 0;
+}
+
+double fmpg_timecode(fmpg_instance *instance)
+{
+    return instance ? instance->decoder.timecode : 0.0;
+}
+
+const char *fmpg_ffmpeg_version()
+{
+    static char *version = nullptr;
+
+    if (version == nullptr) {
+        version = static_cast<char *>(malloc(1024));
+    }
+    sprintf(version, "avformat: %u.%u.%u (%d), avcodec: %u.%u.%u (%d), swresample: %u.%u.%u (%d), avutil: %u.%u.%u (%d)",
+            LIBAVFORMAT_VERSION_MAJOR, LIBAVFORMAT_VERSION_MINOR, LIBAVFORMAT_VERSION_MICRO, LIBAVFORMAT_VERSION_INT,
+            LIBAVCODEC_VERSION_MAJOR, LIBAVCODEC_VERSION_MINOR, LIBAVCODEC_VERSION_MICRO, LIBAVCODEC_VERSION_INT,
+            LIBSWRESAMPLE_VERSION_MAJOR, LIBSWRESAMPLE_VERSION_MINOR, LIBSWRESAMPLE_VERSION_MICRO, LIBSWRESAMPLE_VERSION_INT,
+            LIBAVUTIL_VERSION_MAJOR, LIBAVUTIL_VERSION_MINOR, LIBAVUTIL_VERSION_MICRO, LIBAVUTIL_VERSION_INT
+            );
+    return version;
+}
+
+const char *fmpg_int_version2string(int ver)
+{
+    static char *version = nullptr;
+
+    if (version == nullptr) {
+        version = static_cast<char *>(malloc(1024));
+    }
+
+    int major = AV_VERSION_MAJOR(ver);
+    int minor = AV_VERSION_MINOR(ver);
+    int micro = AV_VERSION_MICRO(ver);
+    sprintf(version, "%u.%u.%u", major, minor, micro);
+
+    return version;
+}
+
+int fmpg_version()
+{
+    return ffi_version();
+}

lib/Building Win32.txt

@@ -0,0 +1,17 @@
+Building for Win32.
+
+Extract the taglib sources.
+
+Add this in CMakeLists.txt:
+
+set(ZLIB_LIBRARY "c:/devel/libraries/nwin64/lib/zlib.lib")
+set(ZLIB_INCLUDE_DIR "c:/devel/libraries/nwin64/include")
+set(BUILD_TESTING OFF)
+set(BUILD_SHARED_LIBS ON)
+
+Or any place you have zlib.lib stored.
+
+Copy the shared libraries to ../lib/windows_<arch>/
+
+
+

lib/Makefile

@@ -0,0 +1,15 @@
+
+TAGLIB=taglib-2.2.1
+
+all:
+	tar xf ${TAGLIB}.tar.gz
+	SUBDIR=`racket -e "(display (format \"~a-~a\" (system-type 'os*) (system-type 'arch)))"`; \
+	(cd ${TAGLIB};  cmake -DBUILD_SHARED_LIBS=ON -DCMAKE_INSTALL_PREFIX=../lib/$$SUBDIR -DCMAKE_BUILD_TYPE=Release .)
+	(cd ${TAGLIB}; make)
+	(cd ${TAGLIB}; make install)
+	SUBDIR=`racket -e "(display (format \"~a-~a\" (system-type 'os*) (system-type 'arch)))"`; \
+	(cd lib/$$SUBDIR/lib; tar cf - *so *.so.*) | (cd ../lib/$$SUBDIR; tar xvf - )
+
+clean:
+	rm -rf ${TAGLIB}
+	rm -rf lib

src/linux-x86_64/libtag.so

@@ -0,0 +1 @@
+libtag.so.2

src/linux-x86_64/libtag.so.2

@@ -0,0 +1 @@
+libtag.so.2.2.1

src/linux-x86_64/libtag_c.so

@@ -0,0 +1 @@
+libtag_c.so.2

src/linux-x86_64/libtag_c.so.2

@@ -0,0 +1 @@
+libtag_c.so.2.2.1