Some basics of AudioMapping.
[dcpomatic.git] / src / lib / ffmpeg_decoder.cc
index 3c607a7fb260f741392b42d221177e9857a72848..adf16c9408a36d39976d730e86e8c29a894ec505 100644 (file)
@@ -1,3 +1,5 @@
+/* -*- c-basic-offset: 8; default-tab-width: 8; -*- */
+
 /*
     Copyright (C) 2012 Carl Hetherington <cth@carlh.net>
 
@@ -41,7 +43,6 @@ extern "C" {
 #include "transcoder.h"
 #include "job.h"
 #include "filter.h"
-#include "options.h"
 #include "exceptions.h"
 #include "image.h"
 #include "util.h"
@@ -49,6 +50,9 @@ extern "C" {
 #include "ffmpeg_decoder.h"
 #include "filter_graph.h"
 #include "subtitle.h"
+#include "audio_buffers.h"
+
+#include "i18n.h"
 
 using std::cout;
 using std::string;
@@ -58,11 +62,15 @@ using std::list;
 using boost::shared_ptr;
 using boost::optional;
 using boost::dynamic_pointer_cast;
+using libdcp::Size;
+
+boost::mutex FFmpegDecoder::_mutex;
 
-FFmpegDecoder::FFmpegDecoder (shared_ptr<Film> f, shared_ptr<const Options> o, Job* j)
-       : Decoder (f, o, j)
-       , VideoDecoder (f, o, j)
-       , AudioDecoder (f, o, j)
+FFmpegDecoder::FFmpegDecoder (shared_ptr<const Film> f, shared_ptr<const FFmpegContent> c, bool video, bool audio, bool subtitles)
+       : Decoder (f)
+       , VideoDecoder (f)
+       , AudioDecoder (f, c)
+       , _ffmpeg_content (c)
        , _format_context (0)
        , _video_stream (-1)
        , _frame (0)
@@ -72,6 +80,9 @@ FFmpegDecoder::FFmpegDecoder (shared_ptr<Film> f, shared_ptr<const Options> o, J
        , _audio_codec (0)
        , _subtitle_codec_context (0)
        , _subtitle_codec (0)
+       , _decode_video (video)
+       , _decode_audio (audio)
+       , _decode_subtitles (subtitles)
 {
        setup_general ();
        setup_video ();
@@ -81,10 +92,12 @@ FFmpegDecoder::FFmpegDecoder (shared_ptr<Film> f, shared_ptr<const Options> o, J
 
 FFmpegDecoder::~FFmpegDecoder ()
 {
+       boost::mutex::scoped_lock lm (_mutex);
+       
        if (_audio_codec_context) {
                avcodec_close (_audio_codec_context);
        }
-       
+
        if (_video_codec_context) {
                avcodec_close (_video_codec_context);
        }
@@ -103,15 +116,15 @@ FFmpegDecoder::setup_general ()
 {
        av_register_all ();
 
-       if (avformat_open_input (&_format_context, _film->content_path().c_str(), 0, 0) < 0) {
-               throw OpenFileError (_film->content_path ());
+       if (avformat_open_input (&_format_context, _ffmpeg_content->file().string().c_str(), 0, 0) < 0) {
+               throw OpenFileError (_ffmpeg_content->file().string ());
        }
 
        if (avformat_find_stream_info (_format_context, 0) < 0) {
-               throw DecodeError ("could not find stream information");
+               throw DecodeError (_("could not find stream information"));
        }
 
-       /* Find video, audio and subtitle streams and choose the first of each */
+       /* Find video, audio and subtitle streams */
 
        for (uint32_t i = 0; i < _format_context->nb_streams; ++i) {
                AVStream* s = _format_context->streams[i];
@@ -122,96 +135,86 @@ FFmpegDecoder::setup_general ()
                        /* This is a hack; sometimes it seems that _audio_codec_context->channel_layout isn't set up,
                           so bodge it here.  No idea why we should have to do this.
                        */
-                       
+
                        if (s->codec->channel_layout == 0) {
                                s->codec->channel_layout = av_get_default_channel_layout (s->codec->channels);
                        }
                        
                        _audio_streams.push_back (
-                               shared_ptr<AudioStream> (
-                                       new FFmpegAudioStream (stream_name (s), i, s->codec->sample_rate, s->codec->channel_layout)
-                                       )
+                               FFmpegAudioStream (stream_name (s), i, s->codec->sample_rate, s->codec->channels)
                                );
-                       
+
                } else if (s->codec->codec_type == AVMEDIA_TYPE_SUBTITLE) {
-                       _subtitle_streams.push_back (
-                               shared_ptr<SubtitleStream> (
-                                       new SubtitleStream (stream_name (s), i)
-                                       )
-                               );
+                       _subtitle_streams.push_back (FFmpegSubtitleStream (stream_name (s), i));
                }
        }
 
        if (_video_stream < 0) {
-               throw DecodeError ("could not find video stream");
+               throw DecodeError (N_("could not find video stream"));
        }
 
        _frame = avcodec_alloc_frame ();
        if (_frame == 0) {
-               throw DecodeError ("could not allocate frame");
+               throw DecodeError (N_("could not allocate frame"));
        }
 }
 
 void
 FFmpegDecoder::setup_video ()
 {
+       boost::mutex::scoped_lock lm (_mutex);
+       
        _video_codec_context = _format_context->streams[_video_stream]->codec;
        _video_codec = avcodec_find_decoder (_video_codec_context->codec_id);
 
        if (_video_codec == 0) {
-               throw DecodeError ("could not find video decoder");
+               throw DecodeError (_("could not find video decoder"));
        }
 
-       /* I think this prevents problems with green hash on decodes and
-          "changing frame properties on the fly is not supported by all filters"
-          messages with some content.  Although I'm not sure; needs checking.
-       */
-       AVDictionary* opts = 0;
-       av_dict_set (&opts, "threads", "1", 0);
-       
-       if (avcodec_open2 (_video_codec_context, _video_codec, &opts) < 0) {
-               throw DecodeError ("could not open video decoder");
+       if (avcodec_open2 (_video_codec_context, _video_codec, 0) < 0) {
+               throw DecodeError (N_("could not open video decoder"));
        }
 }
 
 void
 FFmpegDecoder::setup_audio ()
 {
-       if (!_audio_stream) {
+       boost::mutex::scoped_lock lm (_mutex);
+       
+       if (!_ffmpeg_content->audio_stream ()) {
                return;
        }
 
-       shared_ptr<FFmpegAudioStream> ffa = dynamic_pointer_cast<FFmpegAudioStream> (_audio_stream);
-       assert (ffa);
-       
-       _audio_codec_context = _format_context->streams[ffa->id()]->codec;
+       _audio_codec_context = _format_context->streams[_ffmpeg_content->audio_stream()->id]->codec;
        _audio_codec = avcodec_find_decoder (_audio_codec_context->codec_id);
 
        if (_audio_codec == 0) {
-               throw DecodeError ("could not find audio decoder");
+               throw DecodeError (_("could not find audio decoder"));
        }
 
        if (avcodec_open2 (_audio_codec_context, _audio_codec, 0) < 0) {
-               throw DecodeError ("could not open audio decoder");
+               throw DecodeError (N_("could not open audio decoder"));
        }
 }
 
 void
 FFmpegDecoder::setup_subtitle ()
 {
-       if (!_subtitle_stream) {
+       boost::mutex::scoped_lock lm (_mutex);
+       
+       if (!_ffmpeg_content->subtitle_stream() || _ffmpeg_content->subtitle_stream()->id >= int (_format_context->nb_streams)) {
                return;
        }
 
-       _subtitle_codec_context = _format_context->streams[_subtitle_stream->id()]->codec;
+       _subtitle_codec_context = _format_context->streams[_ffmpeg_content->subtitle_stream()->id]->codec;
        _subtitle_codec = avcodec_find_decoder (_subtitle_codec_context->codec_id);
 
        if (_subtitle_codec == 0) {
-               throw DecodeError ("could not find subtitle decoder");
+               throw DecodeError (_("could not find subtitle decoder"));
        }
        
        if (avcodec_open2 (_subtitle_codec_context, _subtitle_codec, 0) < 0) {
-               throw DecodeError ("could not open subtitle decoder");
+               throw DecodeError (N_("could not open subtitle decoder"));
        }
 }
 
@@ -220,149 +223,40 @@ bool
 FFmpegDecoder::pass ()
 {
        int r = av_read_frame (_format_context, &_packet);
-       
+
        if (r < 0) {
                if (r != AVERROR_EOF) {
                        /* Maybe we should fail here, but for now we'll just finish off instead */
                        char buf[256];
                        av_strerror (r, buf, sizeof(buf));
-                       _film->log()->log (String::compose ("error on av_read_frame (%1) (%2)", buf, r));
+                       _film->log()->log (String::compose (N_("error on av_read_frame (%1) (%2)"), buf, r));
                }
-               
+
                /* Get any remaining frames */
                
                _packet.data = 0;
                _packet.size = 0;
-
+               
                /* XXX: should we reset _packet.data and size after each *_decode_* call? */
-
-               int frame_finished;
-
-               while (avcodec_decode_video2 (_video_codec_context, _frame, &frame_finished, &_packet) >= 0 && frame_finished) {
-                       filter_and_emit_video (_frame);
+               
+               if (_decode_video) {
+                       while (decode_video_packet ());
                }
 
-               if (_audio_stream && _opt->decode_audio) {
-                       while (avcodec_decode_audio4 (_audio_codec_context, _frame, &frame_finished, &_packet) >= 0 && frame_finished) {
-                               int const data_size = av_samples_get_buffer_size (
-                                       0, _audio_codec_context->channels, _frame->nb_samples, audio_sample_format (), 1
-                                       );
-
-                               assert (_audio_codec_context->channels == _film->audio_channels());
-                               Audio (deinterleave_audio (_frame->data[0], data_size));
-                       }
+               if (_ffmpeg_content->audio_stream() && _decode_audio) {
+                       decode_audio_packet ();
                }
-
+                       
                return true;
        }
 
        avcodec_get_frame_defaults (_frame);
 
-       shared_ptr<FFmpegAudioStream> ffa = dynamic_pointer_cast<FFmpegAudioStream> (_audio_stream);
-       
-       if (_packet.stream_index == _video_stream) {
-
-               int frame_finished;
-               int const r = avcodec_decode_video2 (_video_codec_context, _frame, &frame_finished, &_packet);
-               if (r >= 0 && frame_finished) {
-
-                       if (r != _packet.size) {
-                               _film->log()->log (String::compose ("Used only %1 bytes of %2 in packet", r, _packet.size));
-                       }
-
-                       /* Where we are in the output, in seconds */
-                       double const out_pts_seconds = video_frame() / frames_per_second();
-
-                       /* Where we are in the source, in seconds */
-                       double const source_pts_seconds = av_q2d (_format_context->streams[_packet.stream_index]->time_base)
-                               * av_frame_get_best_effort_timestamp(_frame);
-
-                       _film->log()->log (
-                               String::compose ("Source video frame ready; source at %1, output at %2", source_pts_seconds, out_pts_seconds),
-                               Log::VERBOSE
-                               );
-
-                       if (!_first_video) {
-                               _first_video = source_pts_seconds;
-                       }
-
-                       /* Difference between where we are and where we should be */
-                       double const delta = source_pts_seconds - _first_video.get() - out_pts_seconds;
-                       double const one_frame = 1 / frames_per_second();
-
-                       /* Insert frames if required to get out_pts_seconds up to pts_seconds */
-                       if (delta > one_frame) {
-                               int const extra = rint (delta / one_frame);
-                               for (int i = 0; i < extra; ++i) {
-                                       repeat_last_video ();
-                                       _film->log()->log (
-                                               String::compose (
-                                                       "Extra video frame inserted at %1s; source frame %2, source PTS %3 (at %4 fps)",
-                                                       out_pts_seconds, video_frame(), source_pts_seconds, frames_per_second()
-                                                       )
-                                               );
-                               }
-                       }
-
-                       if (delta > -one_frame) {
-                               /* Process this frame */
-                               filter_and_emit_video (_frame);
-                       } else {
-                               /* Otherwise we are omitting a frame to keep things right */
-                               _film->log()->log (String::compose ("Frame removed at %1s", out_pts_seconds));
-                       }
-               }
-
-       } else if (ffa && _packet.stream_index == ffa->id() && _opt->decode_audio) {
-
-               int frame_finished;
-               if (avcodec_decode_audio4 (_audio_codec_context, _frame, &frame_finished, &_packet) >= 0 && frame_finished) {
-
-                       /* Where we are in the source, in seconds */
-                       double const source_pts_seconds = av_q2d (_format_context->streams[_packet.stream_index]->time_base)
-                               * av_frame_get_best_effort_timestamp(_frame);
-
-                       /* We only decode audio if we've had our first video packet through, and if it
-                          was before this packet.  Until then audio is thrown away.
-                       */
-                               
-                       if (_first_video && _first_video.get() <= source_pts_seconds) {
-
-                               if (!_first_audio) {
-                                       _first_audio = source_pts_seconds;
-                                       
-                                       /* This is our first audio frame, and if we've arrived here we must have had our
-                                          first video frame.  Push some silence to make up any gap between our first
-                                          video frame and our first audio.
-                                       */
-                       
-                                       /* frames of silence that we must push */
-                                       int const s = rint ((_first_audio.get() - _first_video.get()) * ffa->sample_rate ());
-                                       
-                                       _film->log()->log (
-                                               String::compose (
-                                                       "First video at %1, first audio at %2, pushing %3 audio frames of silence for %4 channels (%5 bytes per sample)",
-                                                       _first_video.get(), _first_audio.get(), s, ffa->channels(), bytes_per_audio_sample()
-                                                       )
-                                               );
-                                       
-                                       if (s) {
-                                               shared_ptr<AudioBuffers> audio (new AudioBuffers (ffa->channels(), s));
-                                               audio->make_silent ();
-                                               Audio (audio);
-                                       }
-                               }
-
-                               int const data_size = av_samples_get_buffer_size (
-                                       0, _audio_codec_context->channels, _frame->nb_samples, audio_sample_format (), 1
-                                       );
-                               
-                               assert (_audio_codec_context->channels == _film->audio_channels());
-                               Audio (deinterleave_audio (_frame->data[0], data_size));
-                       }
-               }
-                       
-       } else if (_subtitle_stream && _packet.stream_index == _subtitle_stream->id() && _opt->decode_subtitles && _first_video) {
+       if (_packet.stream_index == _video_stream && _decode_video) {
+               decode_video_packet ();
+       } else if (_ffmpeg_content->audio_stream() && _packet.stream_index == _ffmpeg_content->audio_stream()->id && _decode_audio) {
+               decode_audio_packet ();
+       } else if (_ffmpeg_content->subtitle_stream() && _packet.stream_index == _ffmpeg_content->subtitle_stream()->id && _decode_subtitles) {
 
                int got_subtitle;
                AVSubtitle sub;
@@ -373,7 +267,7 @@ FFmpegDecoder::pass ()
                        if (sub.num_rects > 0) {
                                shared_ptr<TimedSubtitle> ts;
                                try {
-                                       emit_subtitle (shared_ptr<TimedSubtitle> (new TimedSubtitle (sub, _first_video.get())));
+                                       emit_subtitle (shared_ptr<TimedSubtitle> (new TimedSubtitle (sub)));
                                } catch (...) {
                                        /* some problem with the subtitle; we probably didn't understand it */
                                }
@@ -388,34 +282,34 @@ FFmpegDecoder::pass ()
        return false;
 }
 
+/** @param data pointer to array of pointers to buffers.
+ *  Only the first buffer will be used for non-planar data, otherwise there will be one per channel.
+ */
 shared_ptr<AudioBuffers>
-FFmpegDecoder::deinterleave_audio (uint8_t* data, int size)
+FFmpegDecoder::deinterleave_audio (uint8_t** data, int size)
 {
-       assert (_film->audio_channels());
+       assert (_ffmpeg_content->audio_channels());
        assert (bytes_per_audio_sample());
 
-       shared_ptr<FFmpegAudioStream> ffa = dynamic_pointer_cast<FFmpegAudioStream> (_audio_stream);
-       assert (ffa);
-       
        /* Deinterleave and convert to float */
 
-       assert ((size % (bytes_per_audio_sample() * ffa->channels())) == 0);
+       assert ((size % (bytes_per_audio_sample() * _ffmpeg_content->audio_channels())) == 0);
 
        int const total_samples = size / bytes_per_audio_sample();
-       int const frames = total_samples / _film->audio_channels();
-       shared_ptr<AudioBuffers> audio (new AudioBuffers (ffa->channels(), frames));
+       int const frames = total_samples / _ffmpeg_content->audio_channels();
+       shared_ptr<AudioBuffers> audio (new AudioBuffers (_ffmpeg_content->audio_channels(), frames));
 
        switch (audio_sample_format()) {
        case AV_SAMPLE_FMT_S16:
        {
-               int16_t* p = (int16_t *) data;
+               int16_t* p = reinterpret_cast<int16_t *> (data[0]);
                int sample = 0;
                int channel = 0;
                for (int i = 0; i < total_samples; ++i) {
                        audio->data(channel)[sample] = float(*p++) / (1 << 15);
 
                        ++channel;
-                       if (channel == _film->audio_channels()) {
+                       if (channel == _ffmpeg_content->audio_channels()) {
                                channel = 0;
                                ++sample;
                        }
@@ -423,16 +317,27 @@ FFmpegDecoder::deinterleave_audio (uint8_t* data, int size)
        }
        break;
 
+       case AV_SAMPLE_FMT_S16P:
+       {
+               int16_t** p = reinterpret_cast<int16_t **> (data);
+               for (int i = 0; i < _ffmpeg_content->audio_channels(); ++i) {
+                       for (int j = 0; j < frames; ++j) {
+                               audio->data(i)[j] = static_cast<float>(p[i][j]) / (1 << 15);
+                       }
+               }
+       }
+       break;
+       
        case AV_SAMPLE_FMT_S32:
        {
-               int32_t* p = (int32_t *) data;
+               int32_t* p = reinterpret_cast<int32_t *> (data[0]);
                int sample = 0;
                int channel = 0;
                for (int i = 0; i < total_samples; ++i) {
-                       audio->data(channel)[sample] = float(*p++) / (1 << 31);
+                       audio->data(channel)[sample] = static_cast<float>(*p++) / (1 << 31);
 
                        ++channel;
-                       if (channel == _film->audio_channels()) {
+                       if (channel == _ffmpeg_content->audio_channels()) {
                                channel = 0;
                                ++sample;
                        }
@@ -442,14 +347,14 @@ FFmpegDecoder::deinterleave_audio (uint8_t* data, int size)
 
        case AV_SAMPLE_FMT_FLT:
        {
-               float* p = reinterpret_cast<float*> (data);
+               float* p = reinterpret_cast<float*> (data[0]);
                int sample = 0;
                int channel = 0;
                for (int i = 0; i < total_samples; ++i) {
                        audio->data(channel)[sample] = *p++;
 
                        ++channel;
-                       if (channel == _film->audio_channels()) {
+                       if (channel == _ffmpeg_content->audio_channels()) {
                                channel = 0;
                                ++sample;
                        }
@@ -459,23 +364,22 @@ FFmpegDecoder::deinterleave_audio (uint8_t* data, int size)
                
        case AV_SAMPLE_FMT_FLTP:
        {
-               float* p = reinterpret_cast<float*> (data);
-               for (int i = 0; i < _film->audio_channels(); ++i) {
-                       memcpy (audio->data(i), p, frames * sizeof(float));
-                       p += frames;
+               float** p = reinterpret_cast<float**> (data);
+               for (int i = 0; i < _ffmpeg_content->audio_channels(); ++i) {
+                       memcpy (audio->data(i), p[i], frames * sizeof(float));
                }
        }
        break;
 
        default:
-               throw DecodeError (String::compose ("Unrecognised audio sample format (%1)", static_cast<int> (audio_sample_format())));
+               throw DecodeError (String::compose (_("Unrecognised audio sample format (%1)"), static_cast<int> (audio_sample_format())));
        }
 
        return audio;
 }
 
 float
-FFmpegDecoder::frames_per_second () const
+FFmpegDecoder::video_frame_rate () const
 {
        AVStream* s = _format_context->streams[_video_stream];
 
@@ -496,10 +400,10 @@ FFmpegDecoder::audio_sample_format () const
        return _audio_codec_context->sample_fmt;
 }
 
-Size
+libdcp::Size
 FFmpegDecoder::native_size () const
 {
-       return Size (_video_codec_context->width, _video_codec_context->height);
+       return libdcp::Size (_video_codec_context->width, _video_codec_context->height);
 }
 
 PixelFormat
@@ -536,22 +440,24 @@ string
 FFmpegDecoder::stream_name (AVStream* s) const
 {
        stringstream n;
-       
-       AVDictionaryEntry const * lang = av_dict_get (s->metadata, "language", 0, 0);
-       if (lang) {
-               n << lang->value;
-       }
-       
-       AVDictionaryEntry const * title = av_dict_get (s->metadata, "title", 0, 0);
-       if (title) {
-               if (!n.str().empty()) {
-                       n << " ";
+
+       if (s->metadata) {
+               AVDictionaryEntry const * lang = av_dict_get (s->metadata, N_("language"), 0, 0);
+               if (lang) {
+                       n << lang->value;
+               }
+               
+               AVDictionaryEntry const * title = av_dict_get (s->metadata, N_("title"), 0, 0);
+               if (title) {
+                       if (!n.str().empty()) {
+                               n << N_(" ");
+                       }
+                       n << title->value;
                }
-               n << title->value;
        }
 
        if (n.str().empty()) {
-               n << "unknown";
+               n << N_("unknown");
        }
 
        return n.str ();
@@ -563,94 +469,167 @@ FFmpegDecoder::bytes_per_audio_sample () const
        return av_get_bytes_per_sample (audio_sample_format ());
 }
 
-void
-FFmpegDecoder::set_audio_stream (shared_ptr<AudioStream> s)
+bool
+FFmpegDecoder::seek (double p)
 {
-       AudioDecoder::set_audio_stream (s);
-       setup_audio ();
+       return do_seek (p, false, false);
 }
 
-void
-FFmpegDecoder::set_subtitle_stream (shared_ptr<SubtitleStream> s)
+bool
+FFmpegDecoder::seek_back ()
 {
-       VideoDecoder::set_subtitle_stream (s);
-       setup_subtitle ();
+       if (last_content_time() < 2.5) {
+               return true;
+       }
+       
+       return do_seek (last_content_time() - 2.5 / video_frame_rate(), true, true);
 }
 
-void
-FFmpegDecoder::filter_and_emit_video (AVFrame* frame)
+bool
+FFmpegDecoder::seek_forward ()
 {
-       shared_ptr<FilterGraph> graph;
-
-       list<shared_ptr<FilterGraph> >::iterator i = _filter_graphs.begin();
-       while (i != _filter_graphs.end() && !(*i)->can_process (Size (frame->width, frame->height), (AVPixelFormat) frame->format)) {
-               ++i;
+       if (last_content_time() >= (video_length() - video_frame_rate())) {
+               return true;
        }
+       
+       return do_seek (last_content_time() - 0.5 / video_frame_rate(), true, true);
+}
 
-       if (i == _filter_graphs.end ()) {
-               graph.reset (new FilterGraph (_film, this, _opt->apply_crop, Size (frame->width, frame->height), (AVPixelFormat) frame->format));
-               _filter_graphs.push_back (graph);
-               _film->log()->log (String::compose ("New graph for %1x%2, pixel format %3", frame->width, frame->height, frame->format));
-       } else {
-               graph = *i;
-       }
+bool
+FFmpegDecoder::do_seek (double p, bool backwards, bool accurate)
+{
+       int64_t const vt = p / av_q2d (_format_context->streams[_video_stream]->time_base);
 
-       list<shared_ptr<Image> > images = graph->process (frame);
+       int const r = av_seek_frame (_format_context, _video_stream, vt, backwards ? AVSEEK_FLAG_BACKWARD : 0);
 
-       for (list<shared_ptr<Image> >::iterator i = images.begin(); i != images.end(); ++i) {
-               emit_video (*i);
+       avcodec_flush_buffers (_video_codec_context);
+       if (_subtitle_codec_context) {
+               avcodec_flush_buffers (_subtitle_codec_context);
        }
+
+       if (accurate) {
+               while (1) {
+                       int r = av_read_frame (_format_context, &_packet);
+                       if (r < 0) {
+                               return true;
+                       }
+                       
+                       avcodec_get_frame_defaults (_frame);
+                       
+                       if (_packet.stream_index == _video_stream) {
+                               int finished = 0;
+                               int const r = avcodec_decode_video2 (_video_codec_context, _frame, &finished, &_packet);
+                               if (r >= 0 && finished) {
+                                       int64_t const bet = av_frame_get_best_effort_timestamp (_frame);
+                                       if (bet > vt) {
+                                               break;
+                                       }
+                               }
+                       }
+                       
+                       av_free_packet (&_packet);
+               }
+       }
+               
+       return r < 0;
 }
 
-shared_ptr<FFmpegAudioStream>
-FFmpegAudioStream::create (string t, optional<int> v)
+void
+FFmpegDecoder::film_changed (Film::Property p)
 {
-       if (!v) {
-               /* version < 1; no type in the string, and there's only FFmpeg streams anyway */
-               return shared_ptr<FFmpegAudioStream> (new FFmpegAudioStream (t, v));
+       switch (p) {
+       case Film::CROP:
+       case Film::FILTERS:
+       {
+               boost::mutex::scoped_lock lm (_filter_graphs_mutex);
+               _filter_graphs.clear ();
        }
+       break;
 
-       stringstream s (t);
-       string type;
-       s >> type;
-       if (type != "ffmpeg") {
-               return shared_ptr<FFmpegAudioStream> ();
+       default:
+               break;
        }
+}
 
-       return shared_ptr<FFmpegAudioStream> (new FFmpegAudioStream (t, v));
+/** @return Length (in video frames) according to our content's header */
+ContentVideoFrame
+FFmpegDecoder::video_length () const
+{
+       return (double(_format_context->duration) / AV_TIME_BASE) * video_frame_rate();
 }
 
-FFmpegAudioStream::FFmpegAudioStream (string t, optional<int> version)
+void
+FFmpegDecoder::decode_audio_packet ()
 {
-       stringstream n (t);
+       /* Audio packets can contain multiple frames, so we may have to call avcodec_decode_audio4
+          several times.
+       */
        
-       int name_index = 4;
-       if (!version) {
-               name_index = 2;
-               int channels;
-               n >> _id >> channels;
-               _channel_layout = av_get_default_channel_layout (channels);
-               _sample_rate = 0;
-       } else {
-               string type;
-               /* Current (marked version 1) */
-               n >> type >> _id >> _sample_rate >> _channel_layout;
-               assert (type == "ffmpeg");
-       }
+       AVPacket copy_packet = _packet;
+
+       while (copy_packet.size > 0) {
 
-       for (int i = 0; i < name_index; ++i) {
-               size_t const s = t.find (' ');
-               if (s != string::npos) {
-                       t = t.substr (s + 1);
+               int frame_finished;
+               int const decode_result = avcodec_decode_audio4 (_audio_codec_context, _frame, &frame_finished, &copy_packet);
+               if (decode_result >= 0) {
+                       if (frame_finished) {
+                       
+                               /* Where we are in the source, in seconds */
+                               double const source_pts_seconds = av_q2d (_format_context->streams[copy_packet.stream_index]->time_base)
+                                       * av_frame_get_best_effort_timestamp(_frame);
+                               
+                               int const data_size = av_samples_get_buffer_size (
+                                       0, _audio_codec_context->channels, _frame->nb_samples, audio_sample_format (), 1
+                                       );
+                               
+                               assert (_audio_codec_context->channels == _ffmpeg_content->audio_channels());
+                               Audio (deinterleave_audio (_frame->data, data_size), source_pts_seconds);
+                       }
+                       
+                       copy_packet.data += decode_result;
+                       copy_packet.size -= decode_result;
                }
        }
-
-       _name = t;
 }
 
-string
-FFmpegAudioStream::to_string () const
+bool
+FFmpegDecoder::decode_video_packet ()
 {
-       return String::compose ("ffmpeg %1 %2 %3 %4", _id, _sample_rate, _channel_layout, _name);
-}
+       int frame_finished;
+       if (avcodec_decode_video2 (_video_codec_context, _frame, &frame_finished, &_packet) < 0 || !frame_finished) {
+               return false;
+       }
+               
+       boost::mutex::scoped_lock lm (_filter_graphs_mutex);
+       
+       shared_ptr<FilterGraph> graph;
+       
+       list<shared_ptr<FilterGraph> >::iterator i = _filter_graphs.begin();
+       while (i != _filter_graphs.end() && !(*i)->can_process (libdcp::Size (_frame->width, _frame->height), (AVPixelFormat) _frame->format)) {
+               ++i;
+       }
+       
+       if (i == _filter_graphs.end ()) {
+               graph.reset (new FilterGraph (_film, this, libdcp::Size (_frame->width, _frame->height), (AVPixelFormat) _frame->format));
+               _filter_graphs.push_back (graph);
+               _film->log()->log (String::compose (N_("New graph for %1x%2, pixel format %3"), _frame->width, _frame->height, _frame->format));
+       } else {
+               graph = *i;
+       }
+       
+       list<shared_ptr<Image> > images = graph->process (_frame);
+       
+       for (list<shared_ptr<Image> >::iterator i = images.begin(); i != images.end(); ++i) {
+               int64_t const bet = av_frame_get_best_effort_timestamp (_frame);
+               if (bet != AV_NOPTS_VALUE) {
+                       /* XXX: may need to insert extra frames / remove frames here ...
+                          (as per old Matcher)
+                       */
+                       emit_video (*i, false, bet * av_q2d (_format_context->streams[_video_stream]->time_base) * TIME_HZ);
+               } else {
+                       _film->log()->log ("Dropping frame without PTS");
+               }
+       }
 
+       return true;
+}