Emit the large bits of pointless silence which are the cause of #252 in half-second...
[dcpomatic.git] / src / lib / ffmpeg_decoder.cc
index fcb2e82baabd84de4b0f6d53169a92831b920809..a51b521d0fe22918010a6976c5ae754dd9d8e051 100644 (file)
@@ -1,7 +1,5 @@
-/* -*- c-basic-offset: 8; default-tab-width: 8; -*- */
-
 /*
-    Copyright (C) 2012 Carl Hetherington <cth@carlh.net>
+    Copyright (C) 2012-2014 Carl Hetherington <cth@carlh.net>
 
     This program is free software; you can redistribute it and/or modify
     it under the terms of the GNU General Public License as published by
 #include <iomanip>
 #include <iostream>
 #include <stdint.h>
-#include <boost/lexical_cast.hpp>
+#include <sndfile.h>
 extern "C" {
-#include <tiffio.h>
 #include <libavcodec/avcodec.h>
 #include <libavformat/avformat.h>
-#include <libswscale/swscale.h>
-#include <libpostproc/postprocess.h>
 }
-#include <sndfile.h>
 #include "film.h"
-#include "format.h"
-#include "transcoder.h"
-#include "job.h"
 #include "filter.h"
 #include "exceptions.h"
 #include "image.h"
@@ -49,179 +40,117 @@ extern "C" {
 #include "log.h"
 #include "ffmpeg_decoder.h"
 #include "filter_graph.h"
-#include "subtitle.h"
 #include "audio_buffers.h"
+#include "ffmpeg_content.h"
+#include "image_proxy.h"
 
 #include "i18n.h"
 
+#define LOG_GENERAL(...) film->log()->log (String::compose (__VA_ARGS__), Log::TYPE_GENERAL);
+#define LOG_ERROR(...) film->log()->log (String::compose (__VA_ARGS__), Log::TYPE_ERROR);
+#define LOG_WARNING(...) film->log()->log (__VA_ARGS__, Log::TYPE_WARNING);
+
 using std::cout;
 using std::string;
 using std::vector;
 using std::stringstream;
 using std::list;
+using std::min;
+using std::pair;
 using boost::shared_ptr;
 using boost::optional;
 using boost::dynamic_pointer_cast;
 using libdcp::Size;
 
-boost::mutex FFmpegDecoder::_mutex;
-
-FFmpegDecoder::FFmpegDecoder (shared_ptr<const Film> f, shared_ptr<const FFmpegContent> c, bool video, bool audio, bool subtitles)
+FFmpegDecoder::FFmpegDecoder (shared_ptr<const Film> f, shared_ptr<const FFmpegContent> c, bool video, bool audio)
        : Decoder (f)
        , VideoDecoder (f, c)
        , AudioDecoder (f, c)
-       , _ffmpeg_content (c)
-       , _format_context (0)
-       , _video_stream (-1)
-       , _frame (0)
-       , _video_codec_context (0)
-       , _video_codec (0)
-       , _audio_codec_context (0)
-       , _audio_codec (0)
+       , SubtitleDecoder (f)
+       , FFmpeg (c)
        , _subtitle_codec_context (0)
        , _subtitle_codec (0)
        , _decode_video (video)
        , _decode_audio (audio)
-       , _decode_subtitles (subtitles)
+       , _pts_offset (0)
+       , _just_sought (false)
 {
-       setup_general ();
-       setup_video ();
-       setup_audio ();
        setup_subtitle ();
-}
 
-FFmpegDecoder::~FFmpegDecoder ()
-{
-       boost::mutex::scoped_lock lm (_mutex);
-       
-       if (_audio_codec_context) {
-               avcodec_close (_audio_codec_context);
-       }
+       /* Audio and video frame PTS values may not start with 0.  We want
+          to fiddle them so that:
 
-       if (_video_codec_context) {
-               avcodec_close (_video_codec_context);
-       }
+          1.  One of them starts at time 0.
+          2.  The first video PTS value ends up on a frame boundary.
 
-       if (_subtitle_codec_context) {
-               avcodec_close (_subtitle_codec_context);
-       }
+          Then we remove big initial gaps in PTS and we allow our
+          insertion of black frames to work.
 
-       av_free (_frame);
-       
-       avformat_close_input (&_format_context);
-}      
+          We will do:
+            audio_pts_to_use = audio_pts_from_ffmpeg + pts_offset;
+            video_pts_to_use = video_pts_from_ffmpeg + pts_offset;
+       */
 
-void
-FFmpegDecoder::setup_general ()
-{
-       av_register_all ();
+       bool const have_video = video && c->first_video();
+       bool const have_audio = audio && c->audio_stream() && c->audio_stream()->first_audio;
 
-       if (avformat_open_input (&_format_context, _ffmpeg_content->file().string().c_str(), 0, 0) < 0) {
-               throw OpenFileError (_ffmpeg_content->file().string ());
-       }
+       /* First, make one of them start at 0 */
 
-       if (avformat_find_stream_info (_format_context, 0) < 0) {
-               throw DecodeError (_("could not find stream information"));
+       if (have_audio && have_video) {
+               _pts_offset = - min (c->first_video().get(), c->audio_stream()->first_audio.get());
+       } else if (have_video) {
+               _pts_offset = - c->first_video().get();
+       } else if (have_audio) {
+               _pts_offset = - c->audio_stream()->first_audio.get();
        }
 
-       /* Find video, audio and subtitle streams */
-
-       for (uint32_t i = 0; i < _format_context->nb_streams; ++i) {
-               AVStream* s = _format_context->streams[i];
-               if (s->codec->codec_type == AVMEDIA_TYPE_VIDEO) {
-                       _video_stream = i;
-               } else if (s->codec->codec_type == AVMEDIA_TYPE_AUDIO) {
-
-                       /* 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<FFmpegAudioStream> (
-                                       new 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<FFmpegSubtitleStream> (new FFmpegSubtitleStream (stream_name (s), i)));
+       /* Now adjust both so that the video pts starts on a frame */
+       if (have_video && have_audio) {
+               double first_video = c->first_video().get() + _pts_offset;
+               double const old_first_video = first_video;
+               
+               /* Round the first video up to a frame boundary */
+               if (fabs (rint (first_video * c->video_frame_rate()) - first_video * c->video_frame_rate()) > 1e-6) {
+                       first_video = ceil (first_video * c->video_frame_rate()) / c->video_frame_rate ();
                }
-       }
-
-       if (_video_stream < 0) {
-               throw DecodeError (N_("could not find video stream"));
-       }
 
-       _frame = avcodec_alloc_frame ();
-       if (_frame == 0) {
-               throw DecodeError (N_("could not allocate frame"));
+               _pts_offset += first_video - old_first_video;
        }
 }
 
-void
-FFmpegDecoder::setup_video ()
+FFmpegDecoder::~FFmpegDecoder ()
 {
        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"));
-       }
-
-       if (avcodec_open2 (_video_codec_context, _video_codec, 0) < 0) {
-               throw DecodeError (N_("could not open video decoder"));
+       if (_subtitle_codec_context) {
+               avcodec_close (_subtitle_codec_context);
        }
 }
 
 void
-FFmpegDecoder::setup_audio ()
+FFmpegDecoder::flush ()
 {
-       boost::mutex::scoped_lock lm (_mutex);
+       /* Get any remaining frames */
        
-       if (!_ffmpeg_content->audio_stream ()) {
-               return;
-       }
-
-       _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"));
-       }
-
-       if (avcodec_open2 (_audio_codec_context, _audio_codec, 0) < 0) {
-               throw DecodeError (N_("could not open audio decoder"));
-       }
-}
-
-void
-FFmpegDecoder::setup_subtitle ()
-{
-       boost::mutex::scoped_lock lm (_mutex);
+       _packet.data = 0;
+       _packet.size = 0;
        
-       if (!_ffmpeg_content->subtitle_stream() || _ffmpeg_content->subtitle_stream()->id >= int (_format_context->nb_streams)) {
-               return;
-       }
-
-       _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"));
+       /* XXX: should we reset _packet.data and size after each *_decode_* call? */
+       
+       if (_decode_video) {
+               while (decode_video_packet ()) {}
        }
        
-       if (avcodec_open2 (_subtitle_codec_context, _subtitle_codec, 0) < 0) {
-               throw DecodeError (N_("could not open subtitle decoder"));
+       if (_ffmpeg_content->audio_stream() && _decode_audio) {
+               decode_audio_packet ();
        }
-}
 
+       /* Stop us being asked for any more data */
+       _video_position = _ffmpeg_content->video_length_after_3d_combine ();
+       _audio_position = _ffmpeg_content->audio_length ();
+}
 
-bool
+void
 FFmpegDecoder::pass ()
 {
        int r = av_read_frame (_format_context, &_packet);
@@ -231,57 +160,29 @@ FFmpegDecoder::pass ()
                        /* 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 (N_("error on av_read_frame (%1) (%2)"), buf, r));
+                       shared_ptr<const Film> film = _film.lock ();
+                       assert (film);
+                       LOG_ERROR (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? */
-               
-               if (_decode_video) {
-                       while (decode_video_packet ());
-               }
-
-               if (_ffmpeg_content->audio_stream() && _decode_audio) {
-                       decode_audio_packet ();
-               }
-                       
-               return true;
+               flush ();
+               return;
        }
 
-       avcodec_get_frame_defaults (_frame);
+       shared_ptr<const Film> film = _film.lock ();
+       assert (film);
 
-       if (_packet.stream_index == _video_stream && _decode_video) {
+       int const si = _packet.stream_index;
+       
+       if (si == _video_stream && _decode_video) {
                decode_video_packet ();
-       } else if (_ffmpeg_content->audio_stream() && _packet.stream_index == _ffmpeg_content->audio_stream()->id && _decode_audio) {
+       } else if (_ffmpeg_content->audio_stream() && _ffmpeg_content->audio_stream()->uses_index (_format_context, si) && _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;
-               if (avcodec_decode_subtitle2 (_subtitle_codec_context, &sub, &got_subtitle, &_packet) && got_subtitle) {
-                       /* Sometimes we get an empty AVSubtitle, which is used by some codecs to
-                          indicate that the previous subtitle should stop.
-                       */
-                       if (sub.num_rects > 0) {
-                               shared_ptr<TimedSubtitle> ts;
-                               try {
-                                       emit_subtitle (shared_ptr<TimedSubtitle> (new TimedSubtitle (sub)));
-                               } catch (...) {
-                                       /* some problem with the subtitle; we probably didn't understand it */
-                               }
-                       } else {
-                               emit_subtitle (shared_ptr<TimedSubtitle> ());
-                       }
-                       avsubtitle_free (&sub);
-               }
+       } else if (_ffmpeg_content->subtitle_stream() && _ffmpeg_content->subtitle_stream()->uses_index (_format_context, si) && film->with_subtitles ()) {
+               decode_subtitle_packet ();
        }
-       
+
        av_free_packet (&_packet);
-       return false;
 }
 
 /** @param data pointer to array of pointers to buffers.
@@ -302,6 +203,23 @@ FFmpegDecoder::deinterleave_audio (uint8_t** data, int size)
        shared_ptr<AudioBuffers> audio (new AudioBuffers (_ffmpeg_content->audio_channels(), frames));
 
        switch (audio_sample_format()) {
+       case AV_SAMPLE_FMT_U8:
+       {
+               uint8_t* p = reinterpret_cast<uint8_t *> (data[0]);
+               int sample = 0;
+               int channel = 0;
+               for (int i = 0; i < total_samples; ++i) {
+                       audio->data(channel)[sample] = float(*p++) / (1 << 23);
+
+                       ++channel;
+                       if (channel == _ffmpeg_content->audio_channels()) {
+                               channel = 0;
+                               ++sample;
+                       }
+               }
+       }
+       break;
+       
        case AV_SAMPLE_FMT_S16:
        {
                int16_t* p = reinterpret_cast<int16_t *> (data[0]);
@@ -380,183 +298,99 @@ FFmpegDecoder::deinterleave_audio (uint8_t** data, int size)
        return audio;
 }
 
-float
-FFmpegDecoder::video_frame_rate () const
-{
-       AVStream* s = _format_context->streams[_video_stream];
-
-       if (s->avg_frame_rate.num && s->avg_frame_rate.den) {
-               return av_q2d (s->avg_frame_rate);
-       }
-
-       return av_q2d (s->r_frame_rate);
-}
-
 AVSampleFormat
 FFmpegDecoder::audio_sample_format () const
 {
-       if (_audio_codec_context == 0) {
+       if (!_ffmpeg_content->audio_stream()) {
                return (AVSampleFormat) 0;
        }
        
-       return _audio_codec_context->sample_fmt;
-}
-
-libdcp::Size
-FFmpegDecoder::native_size () const
-{
-       return libdcp::Size (_video_codec_context->width, _video_codec_context->height);
-}
-
-PixelFormat
-FFmpegDecoder::pixel_format () const
-{
-       return _video_codec_context->pix_fmt;
+       return audio_codec_context()->sample_fmt;
 }
 
 int
-FFmpegDecoder::time_base_numerator () const
-{
-       return _video_codec_context->time_base.num;
-}
-
-int
-FFmpegDecoder::time_base_denominator () const
-{
-       return _video_codec_context->time_base.den;
-}
-
-int
-FFmpegDecoder::sample_aspect_ratio_numerator () const
+FFmpegDecoder::bytes_per_audio_sample () const
 {
-       return _video_codec_context->sample_aspect_ratio.num;
+       return av_get_bytes_per_sample (audio_sample_format ());
 }
 
-int
-FFmpegDecoder::sample_aspect_ratio_denominator () const
+void
+FFmpegDecoder::seek (VideoContent::Frame frame, bool accurate)
 {
-       return _video_codec_context->sample_aspect_ratio.den;
-}
+       double const time_base = av_q2d (_format_context->streams[_video_stream]->time_base);
 
-string
-FFmpegDecoder::stream_name (AVStream* s) const
-{
-       stringstream n;
+       /* If we are doing an accurate seek, our initial shot will be 5 frames (5 being
+          a number plucked from the air) earlier than we want to end up.  The loop below
+          will hopefully then step through to where we want to be.
+       */
+       int initial = frame;
 
-       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;
-               }
+       if (accurate) {
+               initial -= 5;
        }
 
-       if (n.str().empty()) {
-               n << N_("unknown");
+       if (initial < 0) {
+               initial = 0;
        }
 
-       return n.str ();
-}
-
-int
-FFmpegDecoder::bytes_per_audio_sample () const
-{
-       return av_get_bytes_per_sample (audio_sample_format ());
-}
+       /* Initial seek time in the stream's timebase */
+       int64_t const initial_vt = ((initial / _ffmpeg_content->video_frame_rate()) - _pts_offset) / time_base;
 
-bool
-FFmpegDecoder::seek (double p)
-{
-       return do_seek (p, false, false);
-}
+       av_seek_frame (_format_context, _video_stream, initial_vt, AVSEEK_FLAG_BACKWARD);
 
-bool
-FFmpegDecoder::seek_back ()
-{
-       if (last_content_time() < 2.5) {
-               return true;
+       avcodec_flush_buffers (video_codec_context());
+       if (_subtitle_codec_context) {
+               avcodec_flush_buffers (_subtitle_codec_context);
        }
-       
-       return do_seek (last_content_time() - 2.5 / video_frame_rate(), true, true);
-}
 
-bool
-FFmpegDecoder::seek_forward ()
-{
-       if (last_content_time() >= (video_length() - video_frame_rate())) {
-               return true;
+       /* This !accurate is piling hack upon hack; setting _just_sought to true
+          even with accurate == true defeats our attempt to align the start
+          of the video and audio.  Here we disable that defeat when accurate == true
+          i.e. when we are making a DCP rather than just previewing one.
+          Ewww.  This should be gone in 2.0.
+       */
+       if (!accurate) {
+               _just_sought = true;
        }
        
-       return do_seek (last_content_time() - 0.5 / video_frame_rate(), true, true);
-}
-
-bool
-FFmpegDecoder::do_seek (double p, bool backwards, bool accurate)
-{
-       int64_t const vt = p / av_q2d (_format_context->streams[_video_stream]->time_base);
-
-       int const r = av_seek_frame (_format_context, _video_stream, vt, backwards ? AVSEEK_FLAG_BACKWARD : 0);
-
-       avcodec_flush_buffers (_video_codec_context);
-       if (_subtitle_codec_context) {
-               avcodec_flush_buffers (_subtitle_codec_context);
+       _video_position = frame;
+       
+       if (frame == 0 || !accurate) {
+               /* We're already there, or we're as close as we need to be */
+               return;
        }
 
-       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;
-                                       }
-                               }
-                       }
-                       
+       while (1) {
+               int r = av_read_frame (_format_context, &_packet);
+               if (r < 0) {
+                       return;
+               }
+
+               if (_packet.stream_index != _video_stream) {
                        av_free_packet (&_packet);
+                       continue;
                }
-       }
                
-       return r < 0;
-}
-
-void
-FFmpegDecoder::film_changed (Film::Property p)
-{
-       switch (p) {
-       case Film::FILTERS:
-       {
-               boost::mutex::scoped_lock lm (_filter_graphs_mutex);
-               _filter_graphs.clear ();
-       }
-       break;
+               int finished = 0;
+               r = avcodec_decode_video2 (video_codec_context(), _frame, &finished, &_packet);
+               if (r >= 0 && finished) {
+                       _video_position = rint (
+                               (av_frame_get_best_effort_timestamp (_frame) * time_base + _pts_offset) * _ffmpeg_content->video_frame_rate()
+                               );
 
-       default:
-               break;
+                       if (_video_position >= (frame - 1)) {
+                               av_free_packet (&_packet);
+                               break;
+                       }
+               }
+               
+               av_free_packet (&_packet);
        }
-}
 
-/** @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();
+       /* _video_position should be the next thing to be emitted, which will the one after the thing
+          we just saw.
+       */
+       _video_position++;
 }
 
 void
@@ -567,29 +401,51 @@ FFmpegDecoder::decode_audio_packet ()
        */
        
        AVPacket copy_packet = _packet;
-
+       
        while (copy_packet.size > 0) {
 
                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) {
+               int const decode_result = avcodec_decode_audio4 (audio_codec_context(), _frame, &frame_finished, &copy_packet);
+               if (decode_result < 0) {
+                       shared_ptr<const Film> film = _film.lock ();
+                       assert (film);
+                       LOG_ERROR ("avcodec_decode_audio4 failed (%1)", decode_result);
+                       return;
+               }
+
+               if (frame_finished) {
                        
+                       if (_audio_position == 0) {
                                /* 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);
+                               double const pts = av_q2d (_format_context->streams[copy_packet.stream_index]->time_base)
+                                       * av_frame_get_best_effort_timestamp(_frame) + _pts_offset;
+
+                               if (pts > 0) {
+                                       /* Emit some silence */
+                                       int64_t frames = pts * _ffmpeg_content->content_audio_frame_rate ();
+                                       while (frames > 0) {
+                                               int64_t const this_time = min (frames, (int64_t) _ffmpeg_content->content_audio_frame_rate() / 2);
+                                               
+                                               shared_ptr<AudioBuffers> silence (
+                                                       new AudioBuffers (_ffmpeg_content->audio_channels(), this_time)
+                                                       );
+                                       
+                                               silence->make_silent ();
+                                               audio (silence, _audio_position);
+                                               frames -= this_time;
+                                       }
+                               }
                        }
                        
-                       copy_packet.data += decode_result;
-                       copy_packet.size -= decode_result;
+                       int const data_size = av_samples_get_buffer_size (
+                               0, audio_codec_context()->channels, _frame->nb_samples, audio_sample_format (), 1
+                               );
+                       
+                       audio (deinterleave_audio (_frame->data, data_size), _audio_position);
                }
+                       
+               copy_packet.data += decode_result;
+               copy_packet.size -= decode_result;
        }
 }
 
@@ -597,40 +453,198 @@ bool
 FFmpegDecoder::decode_video_packet ()
 {
        int frame_finished;
-       if (avcodec_decode_video2 (_video_codec_context, _frame, &frame_finished, &_packet) < 0 || !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));
+               shared_ptr<const Film> film = _film.lock ();
+               assert (film);
+
+               graph.reset (new FilterGraph (_ffmpeg_content, 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));
+
+               LOG_GENERAL (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);
+
+       list<pair<shared_ptr<Image>, int64_t> > images = graph->process (_frame);
+
+       shared_ptr<const Film> film = _film.lock ();
+       assert (film);
+
+       for (list<pair<shared_ptr<Image>, int64_t> >::iterator i = images.begin(); i != images.end(); ++i) {
+
+               shared_ptr<Image> image = i->first;
+               
+               if (i->second != AV_NOPTS_VALUE) {
+
+                       double const pts = i->second * av_q2d (_format_context->streams[_video_stream]->time_base) + _pts_offset;
+
+                       if (_just_sought) {
+                               /* We just did a seek, so disable any attempts to correct for where we
+                                  are / should be.
+                               */
+                               _video_position = rint (pts * _ffmpeg_content->video_frame_rate ());
+                               _just_sought = false;
+                       }
+
+                       double const next = _video_position / _ffmpeg_content->video_frame_rate();
+                       double const one_frame = 1 / _ffmpeg_content->video_frame_rate ();
+                       double delta = pts - next;
+
+                       while (delta > one_frame) {
+                               /* This PTS is more than one frame forward in time of where we think we should be; emit
+                                  a black frame.
+                               */
+
+                               /* XXX: I think this should be a copy of the last frame... */
+                               boost::shared_ptr<Image> black (
+                                       new Image (
+                                               static_cast<AVPixelFormat> (_frame->format),
+                                               libdcp::Size (video_codec_context()->width, video_codec_context()->height),
+                                               true
+                                               )
+                                       );
+                               
+                               shared_ptr<const Film> film = _film.lock ();
+                               assert (film);
+
+                               black->make_black ();
+                               video (shared_ptr<ImageProxy> (new RawImageProxy (image, film->log())), false, _video_position);
+                               delta -= one_frame;
+                       }
+
+                       if (delta > -one_frame) {
+                               /* This PTS is within a frame of being right; emit this (otherwise it will be dropped) */
+                               video (shared_ptr<ImageProxy> (new RawImageProxy (image, film->log())), false, _video_position);
+                       }
+                               
                } else {
-                       _film->log()->log ("Dropping frame without PTS");
+                       LOG_WARNING ("Dropping frame without PTS");
                }
        }
 
        return true;
 }
+
+       
+void
+FFmpegDecoder::setup_subtitle ()
+{
+       boost::mutex::scoped_lock lm (_mutex);
+       
+       if (!_ffmpeg_content->subtitle_stream()) {
+               return;
+       }
+
+       _subtitle_codec_context = _ffmpeg_content->subtitle_stream()->stream(_format_context)->codec;
+       if (_subtitle_codec_context == 0) {
+               throw DecodeError (N_("could not find subtitle stream"));
+       }
+
+       _subtitle_codec = avcodec_find_decoder (_subtitle_codec_context->codec_id);
+
+       if (_subtitle_codec == 0) {
+               throw DecodeError (N_("could not find subtitle decoder"));
+       }
+       
+       if (avcodec_open2 (_subtitle_codec_context, _subtitle_codec, 0) < 0) {
+               throw DecodeError (N_("could not open subtitle decoder"));
+       }
+}
+
+bool
+FFmpegDecoder::done () const
+{
+       bool const vd = !_decode_video || (_video_position >= _ffmpeg_content->video_length());
+       bool const ad = !_decode_audio || !_ffmpeg_content->audio_stream() || (_audio_position >= _ffmpeg_content->audio_length());
+       return vd && ad;
+}
+       
+void
+FFmpegDecoder::decode_subtitle_packet ()
+{
+       int got_subtitle;
+       AVSubtitle sub;
+       if (avcodec_decode_subtitle2 (_subtitle_codec_context, &sub, &got_subtitle, &_packet) < 0 || !got_subtitle) {
+               return;
+       }
+
+       /* Sometimes we get an empty AVSubtitle, which is used by some codecs to
+          indicate that the previous subtitle should stop.
+       */
+       if (sub.num_rects <= 0) {
+               subtitle (shared_ptr<Image> (), dcpomatic::Rect<double> (), 0, 0);
+               return;
+       } else if (sub.num_rects > 1) {
+               throw DecodeError (_("multi-part subtitles not yet supported"));
+       }
+               
+       /* Subtitle PTS in seconds (within the source, not taking into account any of the
+          source that we may have chopped off for the DCP)
+       */
+       double const packet_time = (static_cast<double> (sub.pts ) / AV_TIME_BASE) + _pts_offset;
+
+       /* hence start time for this sub */
+       Time const from = (packet_time + (double (sub.start_display_time) / 1e3)) * TIME_HZ;
+       Time const to = (packet_time + (double (sub.end_display_time) / 1e3)) * TIME_HZ;
+
+       AVSubtitleRect const * rect = sub.rects[0];
+
+       if (rect->type != SUBTITLE_BITMAP) {
+               throw DecodeError (_("non-bitmap subtitles not yet supported"));
+       }
+
+       /* Note RGBA is expressed little-endian, so the first byte in the word is R, second
+          G, third B, fourth A.
+       */
+       shared_ptr<Image> image (new Image (PIX_FMT_RGBA, libdcp::Size (rect->w, rect->h), true));
+
+       /* Start of the first line in the subtitle */
+       uint8_t* sub_p = rect->pict.data[0];
+       /* sub_p looks up into a BGRA palette which is here
+          (i.e. first byte B, second G, third R, fourth A)
+       */
+       uint32_t const * palette = (uint32_t *) rect->pict.data[1];
+       /* Start of the output data */
+       uint32_t* out_p = (uint32_t *) image->data()[0];
+
+       for (int y = 0; y < rect->h; ++y) {
+               uint8_t* sub_line_p = sub_p;
+               uint32_t* out_line_p = out_p;
+               for (int x = 0; x < rect->w; ++x) {
+                       uint32_t const p = palette[*sub_line_p++];
+                       *out_line_p++ = ((p & 0xff) << 16) | (p & 0xff00) | ((p & 0xff0000) >> 16) | (p & 0xff000000);
+               }
+               sub_p += rect->pict.linesize[0];
+               out_p += image->stride()[0] / sizeof (uint32_t);
+       }
+
+       libdcp::Size const vs = _ffmpeg_content->video_size ();
+
+       subtitle (
+               image,
+               dcpomatic::Rect<double> (
+                       static_cast<double> (rect->x) / vs.width,
+                       static_cast<double> (rect->y) / vs.height,
+                       static_cast<double> (rect->w) / vs.width,
+                       static_cast<double> (rect->h) / vs.height
+                       ),
+               from,
+               to
+               );
+                         
+       
+       avsubtitle_free (&sub);
+}