Log on av_read_frame error (rather than aborting) and log on unused data in a packet.
[dcpomatic.git] / src / lib / ffmpeg_decoder.cc
index 4b0594adddfb0d7893f1ec7b87cb0d19da886c7d..09f699543c790c6755b59cb6d54530833c793039 100644 (file)
@@ -41,7 +41,6 @@ extern "C" {
 #include "transcoder.h"
 #include "job.h"
 #include "filter.h"
-#include "film_state.h"
 #include "options.h"
 #include "exceptions.h"
 #include "image.h"
@@ -50,11 +49,14 @@ extern "C" {
 #include "ffmpeg_decoder.h"
 #include "subtitle.h"
 
-using namespace std;
-using namespace boost;
+using std::cout;
+using std::string;
+using std::vector;
+using std::stringstream;
+using boost::shared_ptr;
 
-FFmpegDecoder::FFmpegDecoder (boost::shared_ptr<const FilmState> s, boost::shared_ptr<const Options> o, Job* j, Log* l, bool minimal, bool ignore_length)
-       : Decoder (s, o, j, l, minimal, ignore_length)
+FFmpegDecoder::FFmpegDecoder (shared_ptr<Film> f, shared_ptr<const Options> o, Job* j, bool minimal, bool ignore_length)
+       : Decoder (f, o, j, minimal, ignore_length)
        , _format_context (0)
        , _video_stream (-1)
        , _audio_stream (-1)
@@ -86,8 +88,9 @@ FFmpegDecoder::~FFmpegDecoder ()
        if (_subtitle_codec_context) {
                avcodec_close (_subtitle_codec_context);
        }
-       
+
        av_free (_frame);
+       
        avformat_close_input (&_format_context);
 }      
 
@@ -98,8 +101,8 @@ FFmpegDecoder::setup_general ()
        
        av_register_all ();
 
-       if ((r = avformat_open_input (&_format_context, _fs->content_path().c_str(), 0, 0)) != 0) {
-               throw OpenFileError (_fs->content_path ());
+       if ((r = avformat_open_input (&_format_context, _film->content_path().c_str(), 0, 0)) != 0) {
+               throw OpenFileError (_film->content_path ());
        }
 
        if (avformat_find_stream_info (_format_context, 0) < 0) {
@@ -109,43 +112,30 @@ FFmpegDecoder::setup_general ()
        /* Find video, audio and subtitle streams and choose the first of each */
 
        for (uint32_t i = 0; i < _format_context->nb_streams; ++i) {
-               if (_format_context->streams[i]->codec->codec_type == AVMEDIA_TYPE_VIDEO) {
+               AVStream* s = _format_context->streams[i];
+               if (s->codec->codec_type == AVMEDIA_TYPE_VIDEO) {
                        _video_stream = i;
-               } else if (_format_context->streams[i]->codec->codec_type == AVMEDIA_TYPE_AUDIO) {
+               } else if (s->codec->codec_type == AVMEDIA_TYPE_AUDIO) {
                        if (_audio_stream == -1) {
                                _audio_stream = i;
                        }
-                       _audio_streams.push_back (Stream (stream_name (_format_context->streams[i]), i));
-               } else if (_format_context->streams[i]->codec->codec_type == AVMEDIA_TYPE_SUBTITLE) {
+                       _audio_streams.push_back (AudioStream (stream_name (s), i, s->codec->channels));
+               } else if (s->codec->codec_type == AVMEDIA_TYPE_SUBTITLE) {
                        if (_subtitle_stream == -1) {
                                _subtitle_stream = i;
                        }
-                       _subtitle_streams.push_back (Stream (stream_name (_format_context->streams[i]), i));
+                       _subtitle_streams.push_back (SubtitleStream (stream_name (s), i));
                }
        }
 
        /* Now override audio and subtitle streams with those from the Film, if it has any */
 
-       if (_fs->audio_stream() != -1) {
-               vector<Stream>::iterator i = _audio_streams.begin ();
-               while (i != _audio_streams.end() && i->id != _fs->audio_stream()) {
-                       ++i;
-               }
-
-               if (i != _audio_streams.end()) {
-                       _audio_stream = _fs->audio_stream ();
-               }
+       if (_film->audio_stream_index() != -1) {
+               _audio_stream = _film->audio_stream().id();
        }
 
-       if (_fs->subtitle_stream() != -1) {
-               vector<Stream>::iterator i = _subtitle_streams.begin ();
-               while (i != _subtitle_streams.end() && i->id != _fs->subtitle_stream()) {
-                       ++i;
-               }
-
-               if (i != _subtitle_streams.end()) {
-                       _subtitle_stream = _fs->subtitle_stream ();
-               }
+       if (_film->subtitle_stream_index() != -1) {
+               _subtitle_stream = _film->subtitle_stream().id ();
        }
 
        if (_video_stream < 0) {
@@ -167,8 +157,15 @@ FFmpegDecoder::setup_video ()
        if (_video_codec == 0) {
                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, 0) < 0) {
+       if (avcodec_open2 (_video_codec_context, _video_codec, &opts) < 0) {
                throw DecodeError ("could not open video decoder");
        }
 }
@@ -224,9 +221,13 @@ bool
 FFmpegDecoder::do_pass ()
 {
        int r = av_read_frame (_format_context, &_packet);
+       
        if (r < 0) {
                if (r != AVERROR_EOF) {
-                       throw DecodeError ("error on av_read_frame");
+                       /* 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));
                }
                
                /* Get any remaining frames */
@@ -234,6 +235,8 @@ FFmpegDecoder::do_pass ()
                _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) {
@@ -245,8 +248,8 @@ FFmpegDecoder::do_pass ()
                                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 == _fs->audio_channels());
+
+                               assert (_audio_codec_context->channels == _film->audio_channels());
                                process_audio (_frame->data[0], data_size);
                        }
                }
@@ -254,37 +257,123 @@ FFmpegDecoder::do_pass ()
                return true;
        }
 
+       avcodec_get_frame_defaults (_frame);
+       
        if (_packet.stream_index == _video_stream) {
 
                int frame_finished;
-               if (avcodec_decode_video2 (_video_codec_context, _frame, &frame_finished, &_packet) >= 0 && frame_finished) {
-                       process_video (_frame);
+               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_index() / 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);
+
+                       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 frame inserted at %1s; DCP frame %2, source PTS %3",
+                                                       out_pts_seconds, video_frame_index(), source_pts_seconds
+                                                       )
+                                               );
+                               }
+                       }
+
+                       if (delta > -one_frame) {
+                               /* Process this frame */
+                               process_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 (_audio_stream >= 0 && _packet.stream_index == _audio_stream && _opt->decode_audio) {
-               
-               avcodec_get_frame_defaults (_frame);
-               
+
                int frame_finished;
                if (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 == _fs->audio_channels());
-                       process_audio (_frame->data[0], data_size);
-               }
+                       /* 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);
 
-       } else if (_subtitle_stream >= 0 && _packet.stream_index == _subtitle_stream && _opt->decode_subtitles) {
+                       /* 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()) * audio_sample_rate ());
+                                       
+                                       _film->log()->log (
+                                               String::compose (
+                                                       "First video at %1, first audio at %2, pushing %3 frames of silence for %4 channels (%5 bytes per sample)",
+                                                       _first_video.get(), _first_audio.get(), s, audio_channels(), bytes_per_audio_sample()
+                                                       )
+                                               );
+                                       
+                                       if (s) {
+                                               /* hence bytes */
+                                               int const b = s * audio_channels() * bytes_per_audio_sample();
+                                               
+                                               /* XXX: this assumes that it won't be too much, and there are shaky assumptions
+                                                  that all sound representations are silent with memset()ed zero data.
+                                               */
+                                               uint8_t silence[b];
+                                               memset (silence, 0, b);
+                                               process_audio (silence, b);
+                                       }
+                               }
+
+                               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());
+                               process_audio (_frame->data[0], data_size);
+                       }
+               }
+                       
+       } else if (_subtitle_stream >= 0 && _packet.stream_index == _subtitle_stream && _opt->decode_subtitles && _first_video) {
 
                int got_subtitle;
                AVSubtitle sub;
                if (avcodec_decode_subtitle2 (_subtitle_codec_context, &sub, &got_subtitle, &_packet) && got_subtitle) {
-                       /* I'm not entirely sure why, but sometimes we get an AVSubtitle with
-                          no AVSubtitleRects.
+                       /* 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) {
-                               process_subtitle (shared_ptr<TimedSubtitle> (new TimedSubtitle (sub)));
+                               process_subtitle (shared_ptr<TimedSubtitle> (new TimedSubtitle (sub, _first_video.get())));
+                       } else {
+                               process_subtitle (shared_ptr<TimedSubtitle> ());
                        }
                        avsubtitle_free (&sub);
                }
@@ -294,12 +383,6 @@ FFmpegDecoder::do_pass ()
        return false;
 }
 
-int
-FFmpegDecoder::length_in_frames () const
-{
-       return (_format_context->duration / AV_TIME_BASE) * frames_per_second ();
-}
-
 float
 FFmpegDecoder::frames_per_second () const
 {
@@ -394,32 +477,18 @@ FFmpegDecoder::has_subtitles () const
        return (_subtitle_stream != -1);
 }
 
-vector<Stream>
+vector<AudioStream>
 FFmpegDecoder::audio_streams () const
 {
        return _audio_streams;
 }
 
-vector<Stream>
+vector<SubtitleStream>
 FFmpegDecoder::subtitle_streams () const
 {
        return _subtitle_streams;
 }
 
-void
-FFmpegDecoder::set_audio_stream (int s)
-{
-       _audio_stream = s;
-       setup_audio ();
-}
-
-void
-FFmpegDecoder::set_subtitle_stream (int s)
-{
-       _subtitle_stream = s;
-       setup_subtitle ();
-}
-
 string
 FFmpegDecoder::stream_name (AVStream* s) const
 {
@@ -444,3 +513,4 @@ FFmpegDecoder::stream_name (AVStream* s) const
 
        return n.str ();
 }
+