Logging improvements to allow prettier displays in the server GUI.

[dcpomatic.git] / src / lib / ffmpeg.cc

diff --git a/src/lib/ffmpeg.cc b/src/lib/ffmpeg.cc
index 60eea6ec7756b626cf76fb996f552940196730b9..6103ccd50ed98863afeb31f6087cddb251226765 100644 (file)
--- a/src/lib/ffmpeg.cc
+++ b/src/lib/ffmpeg.cc
@@ -1,5 +1,5 @@
 /*
 /*

-    Copyright (C) 2013 Carl Hetherington <cth@carlh.net>
+    Copyright (C) 2013-2015 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
 
     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


@@ -17,25 +17,36 @@
 
 */
 
 
 */
 

+#include "ffmpeg.h"
+#include "ffmpeg_content.h"
+#include "film.h"
+#include "exceptions.h"
+#include "util.h"
+#include "raw_convert.h"
+#include "log.h"
+#include "ffmpeg_subtitle_stream.h"
+#include "ffmpeg_audio_stream.h"
+#include "compose.hpp"

 extern "C" {
 #include <libavcodec/avcodec.h>
 #include <libavformat/avformat.h>
 #include <libswscale/swscale.h>
 }
 extern "C" {
 #include <libavcodec/avcodec.h>
 #include <libavformat/avformat.h>
 #include <libswscale/swscale.h>
 }

-#include "ffmpeg.h"
-#include "ffmpeg_content.h"
-#include "exceptions.h"
-#include "util.h"
+#include <boost/algorithm/string.hpp>
+#include <boost/foreach.hpp>
+#include <iostream>

 
 #include "i18n.h"
 
 using std::string;
 using std::cout;
 
 #include "i18n.h"
 
 using std::string;
 using std::cout;

-using std::stringstream;
+using std::cerr;
+using std::vector;

 using boost::shared_ptr;
 using boost::shared_ptr;

-using boost::lexical_cast;
+using boost::optional;

 
 boost::mutex FFmpeg::_mutex;
 
 boost::mutex FFmpeg::_mutex;

+boost::weak_ptr<Log> FFmpeg::_ffmpeg_log;

 
 FFmpeg::FFmpeg (boost::shared_ptr<const FFmpegContent> c)
        : _ffmpeg_content (c)
 
 FFmpeg::FFmpeg (boost::shared_ptr<const FFmpegContent> c)
        : _ffmpeg_content (c)


@@ -47,8 +58,7 @@ FFmpeg::FFmpeg (boost::shared_ptr<const FFmpegContent> c)
        , _video_stream (-1)
 {
        setup_general ();
        , _video_stream (-1)
 {
        setup_general ();

-       setup_video ();
-       setup_audio ();
+       setup_decoders ();

 }
 
 FFmpeg::~FFmpeg ()
 }
 
 FFmpeg::~FFmpeg ()


@@ -56,14 +66,10 @@ FFmpeg::~FFmpeg ()
        boost::mutex::scoped_lock lm (_mutex);
 
        for (uint32_t i = 0; i < _format_context->nb_streams; ++i) {
        boost::mutex::scoped_lock lm (_mutex);
 
        for (uint32_t i = 0; i < _format_context->nb_streams; ++i) {

-               AVCodecContext* context = _format_context->streams[i]->codec;
-               if (context->codec_type == AVMEDIA_TYPE_VIDEO || context->codec_type == AVMEDIA_TYPE_AUDIO) {
-                       avcodec_close (context);
-               }
+               avcodec_close (_format_context->streams[i]->codec);

        }
 
        av_frame_free (&_frame);
        }
 
        av_frame_free (&_frame);

-       

        avformat_close_input (&_format_context);
 }
 
        avformat_close_input (&_format_context);
 }
 


@@ -79,24 +85,50 @@ avio_seek_wrapper (void* data, int64_t offset, int whence)
        return reinterpret_cast<FFmpeg*>(data)->avio_seek (offset, whence);
 }
 
        return reinterpret_cast<FFmpeg*>(data)->avio_seek (offset, whence);
 }
 

+void
+FFmpeg::ffmpeg_log_callback (void* ptr, int level, const char* fmt, va_list vl)
+{
+       if (level > AV_LOG_WARNING) {
+               return;
+       }
+
+       char line[1024];
+       static int prefix = 0;
+       av_log_format_line (ptr, level, fmt, vl, line, sizeof (line), &prefix);
+       shared_ptr<Log> log = _ffmpeg_log.lock ();
+       if (log) {
+               string str (line);
+               boost::algorithm::trim (str);
+               log->log (String::compose ("FFmpeg: %1", str), LogEntry::TYPE_GENERAL);
+       } else {
+               cerr << line;
+       }
+}
+

 void
 FFmpeg::setup_general ()
 {
        av_register_all ();
 
 void
 FFmpeg::setup_general ()
 {
        av_register_all ();
 

+       /* This might not work too well in some cases of multiple FFmpeg decoders,
+          but it's probably good enough.
+       */
+       _ffmpeg_log = _ffmpeg_content->film()->log ();
+       av_log_set_callback (FFmpeg::ffmpeg_log_callback);
+

        _file_group.set_paths (_ffmpeg_content->paths ());
        _avio_buffer = static_cast<uint8_t*> (wrapped_av_malloc (_avio_buffer_size));
        _avio_context = avio_alloc_context (_avio_buffer, _avio_buffer_size, 0, this, avio_read_wrapper, 0, avio_seek_wrapper);
        _format_context = avformat_alloc_context ();
        _format_context->pb = _avio_context;
        _file_group.set_paths (_ffmpeg_content->paths ());
        _avio_buffer = static_cast<uint8_t*> (wrapped_av_malloc (_avio_buffer_size));
        _avio_context = avio_alloc_context (_avio_buffer, _avio_buffer_size, 0, this, avio_read_wrapper, 0, avio_seek_wrapper);
        _format_context = avformat_alloc_context ();
        _format_context->pb = _avio_context;

-       
+

        AVDictionary* options = 0;
        /* These durations are in microseconds, and represent how far into the content file
           we will look for streams.
        */
        AVDictionary* options = 0;
        /* These durations are in microseconds, and represent how far into the content file
           we will look for streams.
        */

-       av_dict_set (&options, "analyzeduration", lexical_cast<string> (5 * 60 * 1e6).c_str(), 0);
-       av_dict_set (&options, "probesize", lexical_cast<string> (5 * 60 * 1e6).c_str(), 0);
-       
+       av_dict_set (&options, "analyzeduration", raw_convert<string> (5 * 60 * 1000000).c_str(), 0);
+       av_dict_set (&options, "probesize", raw_convert<string> (5 * 60 * 1000000).c_str(), 0);
+

        if (avformat_open_input (&_format_context, 0, 0, &options) < 0) {
                throw OpenFileError (_ffmpeg_content->path(0).string ());
        }
        if (avformat_open_input (&_format_context, 0, 0, &options) < 0) {
                throw OpenFileError (_ffmpeg_content->path(0).string ());
        }


@@ -107,29 +139,48 @@ FFmpeg::setup_general ()
 
        /* Find video stream */
 
 
        /* Find video stream */
 

+       int video_stream_undefined_frame_rate = -1;
+

        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) {
        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;
+                       if (s->avg_frame_rate.num > 0 && s->avg_frame_rate.den > 0) {
+                               /* This is definitely our video stream */
+                               _video_stream = i;
+                       } else {
+                               /* This is our video stream if we don't get a better offer */
+                               video_stream_undefined_frame_rate = i;
+                       }

                }
        }
 
                }
        }
 

+       /* Files from iTunes sometimes have two video streams, one with the avg_frame_rate.num and .den set
+          to zero.  Only use such a stream if there is no alternative.
+       */
+       if (_video_stream == -1 && video_stream_undefined_frame_rate != -1) {
+               _video_stream = video_stream_undefined_frame_rate;
+       }
+

        if (_video_stream < 0) {
                throw DecodeError (N_("could not find video stream"));
        }
 
        if (_video_stream < 0) {
                throw DecodeError (N_("could not find video stream"));
        }
 

-       /* Hack: if the AVStreams have zero IDs, put some in.  We
-          use the IDs so that we can cope with VOBs, in which streams
+       /* Hack: if the AVStreams have duplicate IDs, replace them with our
+          own.  We use the IDs so that we can cope with VOBs, in which streams

           move about in index but remain with the same ID in different
           move about in index but remain with the same ID in different

-          VOBs.  However, some files have all-zero IDs, hence this hack.
+          VOBs.  However, some files have duplicate IDs, hence this hack.

        */
        */

-          
-       uint32_t i = 0;
-       while (i < _format_context->nb_streams && _format_context->streams[i]->id == 0) {
-               ++i;
+
+       bool duplicates = false;
+       for (uint32_t i = 0; i < _format_context->nb_streams; ++i) {
+               for (uint32_t j = i + 1; j < _format_context->nb_streams; ++j) {
+                       if (_format_context->streams[i]->id == _format_context->streams[j]->id) {
+                               duplicates = true;
+                       }
+               }

        }
 
        }
 

-       if (i == _format_context->nb_streams) {
+       if (duplicates) {

                /* Put in our own IDs */
                for (uint32_t i = 0; i < _format_context->nb_streams; ++i) {
                        _format_context->streams[i]->id = i;
                /* Put in our own IDs */
                for (uint32_t i = 0; i < _format_context->nb_streams; ++i) {
                        _format_context->streams[i]->id = i;


@@ -143,46 +194,32 @@ FFmpeg::setup_general ()
 }
 
 void
 }
 
 void

-FFmpeg::setup_video ()
-{
-       boost::mutex::scoped_lock lm (_mutex);
-
-       assert (_video_stream >= 0);
-       AVCodecContext* context = _format_context->streams[_video_stream]->codec;
-       AVCodec* codec = avcodec_find_decoder (context->codec_id);
-
-       if (codec == 0) {
-               throw DecodeError (_("could not find video decoder"));
-       }
-
-       if (avcodec_open2 (context, codec, 0) < 0) {
-               throw DecodeError (N_("could not open video decoder"));
-       }
-}
-
-void
-FFmpeg::setup_audio ()
+FFmpeg::setup_decoders ()

 {
        boost::mutex::scoped_lock lm (_mutex);
 
        for (uint32_t i = 0; i < _format_context->nb_streams; ++i) {
                AVCodecContext* context = _format_context->streams[i]->codec;
 {
        boost::mutex::scoped_lock lm (_mutex);
 
        for (uint32_t i = 0; i < _format_context->nb_streams; ++i) {
                AVCodecContext* context = _format_context->streams[i]->codec;

-               if (context->codec_type != AVMEDIA_TYPE_AUDIO) {
-                       continue;
-               }
-               
+

                AVCodec* codec = avcodec_find_decoder (context->codec_id);
                AVCodec* codec = avcodec_find_decoder (context->codec_id);

-               if (codec == 0) {
-                       throw DecodeError (_("could not find audio decoder"));
-               }
-               
-               if (avcodec_open2 (context, codec, 0) < 0) {
-                       throw DecodeError (N_("could not open audio decoder"));
+               if (codec) {
+
+                       /* This option disables decoding of DCA frame footers in our patched version
+                          of FFmpeg.  I believe these footers are of no use to us, and they can cause
+                          problems when FFmpeg fails to decode them (mantis #352).
+                       */
+                       AVDictionary* options = 0;
+                       av_dict_set (&options, "disable_footer", "1", 0);
+
+                       if (avcodec_open2 (context, codec, &options) < 0) {
+                               throw DecodeError (N_("could not open decoder"));
+                       }

                }
                }

+
+               /* We are silently ignoring any failures to find suitable decoders here */

        }
 }
 
        }
 }
 

-

 AVCodecContext *
 FFmpeg::video_codec_context () const
 {
 AVCodecContext *
 FFmpeg::video_codec_context () const
 {


@@ -190,9 +227,13 @@ FFmpeg::video_codec_context () const
 }
 
 AVCodecContext *
 }
 
 AVCodecContext *

-FFmpeg::audio_codec_context () const
+FFmpeg::subtitle_codec_context () const

 {
 {

-       return _ffmpeg_content->audio_stream()->stream(_format_context)->codec;
+       if (!_ffmpeg_content->subtitle_stream ()) {
+               return 0;
+       }
+
+       return _ffmpeg_content->subtitle_stream()->stream(_format_context)->codec;

 }
 
 int
 }
 
 int


@@ -207,6 +248,75 @@ FFmpeg::avio_seek (int64_t const pos, int whence)
        if (whence == AVSEEK_SIZE) {
                return _file_group.length ();
        }
        if (whence == AVSEEK_SIZE) {
                return _file_group.length ();
        }

-       
+

        return _file_group.seek (pos, whence);
 }
        return _file_group.seek (pos, whence);
 }

+
+FFmpegSubtitlePeriod
+FFmpeg::subtitle_period (AVSubtitle const & sub)
+{
+       ContentTime const packet_time = ContentTime::from_seconds (static_cast<double> (sub.pts) / AV_TIME_BASE);
+
+       if (sub.end_display_time == static_cast<uint32_t> (-1)) {
+               /* End time is not known */
+               return FFmpegSubtitlePeriod (packet_time + ContentTime::from_seconds (sub.start_display_time / 1e3));
+       }
+
+       return FFmpegSubtitlePeriod (
+               packet_time + ContentTime::from_seconds (sub.start_display_time / 1e3),
+               packet_time + ContentTime::from_seconds (sub.end_display_time / 1e3)
+               );
+}
+
+/** Compute the pts offset to use given a set of audio streams and some video details.
+ *  Sometimes these parameters will have just been determined by an Examiner, sometimes
+ *  they will have been retrieved from a piece of Content, hence the need for this method
+ *  in FFmpeg.
+ */
+ContentTime
+FFmpeg::pts_offset (vector<shared_ptr<FFmpegAudioStream> > audio_streams, optional<ContentTime> first_video, double video_frame_rate) const
+{
+       /* Audio and video frame PTS values may not start with 0.  We want
+          to fiddle them so that:
+
+          1.  One of them starts at time 0.
+          2.  The first video PTS value ends up on a frame boundary.
+
+          Then we remove big initial gaps in PTS and we allow our
+          insertion of black frames to work.
+
+          We will do:
+            audio_pts_to_use = audio_pts_from_ffmpeg + pts_offset;
+            video_pts_to_use = video_pts_from_ffmpeg + pts_offset;
+       */
+
+       /* First, make one of them start at 0 */
+
+       ContentTime po = ContentTime::min ();
+
+       if (first_video) {
+               po = - first_video.get ();
+       }
+
+       BOOST_FOREACH (shared_ptr<FFmpegAudioStream> i, audio_streams) {
+               if (i->first_audio) {
+                       po = max (po, - i->first_audio.get ());
+               }
+       }
+
+       /* If the offset is positive we would be pushing things from a -ve PTS to be played.
+          I don't think we ever want to do that, as it seems things at -ve PTS are not meant
+          to be seen (use for alignment bars etc.); see mantis #418.
+       */
+       if (po > ContentTime ()) {
+               po = ContentTime ();
+       }
+
+       /* Now adjust so that the video pts starts on a frame */
+       if (first_video) {
+               ContentTime const fvc = first_video.get() + po;
+               po += fvc.round_up (video_frame_rate) - fvc;
+       }
+
+       return po;
+}