BOOST_FOREACH.
[dcpomatic.git] / src / lib / ffmpeg_decoder.cc
index 6a8f4d4136dab4322b4b7713536b4e6019b09852..b050fc594a76f97ab72833c38792f62daf4f100d 100644 (file)
@@ -27,8 +27,9 @@
 #include "image.h"
 #include "util.h"
 #include "log.h"
+#include "dcpomatic_log.h"
 #include "ffmpeg_decoder.h"
-#include "subtitle_decoder.h"
+#include "text_decoder.h"
 #include "ffmpeg_audio_stream.h"
 #include "ffmpeg_subtitle_stream.h"
 #include "video_filter_graph.h"
@@ -39,8 +40,9 @@
 #include "film.h"
 #include "audio_decoder.h"
 #include "compose.hpp"
-#include "subtitle_content.h"
+#include "text_content.h"
 #include "audio_content.h"
+#include "frame_interval_checker.h"
 #include <dcp/subtitle_string.h>
 #include <sub/ssa_reader.h>
 #include <sub/subtitle.h>
@@ -49,7 +51,6 @@ extern "C" {
 #include <libavcodec/avcodec.h>
 #include <libavformat/avformat.h>
 }
-#include <boost/foreach.hpp>
 #include <boost/algorithm/string.hpp>
 #include <vector>
 #include <iomanip>
@@ -58,11 +59,6 @@ extern "C" {
 
 #include "i18n.h"
 
-#define LOG_GENERAL(...) _log->log (String::compose (__VA_ARGS__), LogEntry::TYPE_GENERAL);
-#define LOG_ERROR(...) _log->log (String::compose (__VA_ARGS__), LogEntry::TYPE_ERROR);
-#define LOG_WARNING_NC(...) _log->log (__VA_ARGS__, LogEntry::TYPE_WARNING);
-#define LOG_WARNING(...) _log->log (String::compose (__VA_ARGS__), LogEntry::TYPE_WARNING);
-
 using std::cout;
 using std::string;
 using std::vector;
@@ -71,21 +67,22 @@ using std::min;
 using std::pair;
 using std::max;
 using std::map;
-using boost::shared_ptr;
+using std::shared_ptr;
 using boost::is_any_of;
 using boost::split;
 using boost::optional;
-using boost::dynamic_pointer_cast;
+using std::dynamic_pointer_cast;
 using dcp::Size;
+using namespace dcpomatic;
 
-FFmpegDecoder::FFmpegDecoder (shared_ptr<const FFmpegContent> c, shared_ptr<Log> log, bool fast)
+FFmpegDecoder::FFmpegDecoder (shared_ptr<const Film> film, shared_ptr<const FFmpegContent> c, bool fast)
        : FFmpeg (c)
-       , _log (log)
+       , Decoder (film)
        , _have_current_subtitle (false)
 {
-       if (c->video) {
-               video.reset (new VideoDecoder (this, c, log));
-               _pts_offset = pts_offset (c->ffmpeg_audio_streams(), c->first_video(), c->active_video_frame_rate());
+       if (c->video && c->video->use()) {
+               video.reset (new VideoDecoder (this, c));
+               _pts_offset = pts_offset (c->ffmpeg_audio_streams(), c->first_video(), c->active_video_frame_rate(film));
                /* It doesn't matter what size or pixel format this is, it just needs to be black */
                _black_image.reset (new Image (AV_PIX_FMT_RGB24, dcp::Size (128, 128), true));
                _black_image->make_black ();
@@ -94,12 +91,12 @@ FFmpegDecoder::FFmpegDecoder (shared_ptr<const FFmpegContent> c, shared_ptr<Log>
        }
 
        if (c->audio) {
-               audio.reset (new AudioDecoder (this, c->audio, log, fast));
+               audio.reset (new AudioDecoder (this, c->audio, fast));
        }
 
-       if (c->subtitle) {
+       if (c->only_text()) {
                /* XXX: this time here should be the time of the first subtitle, not 0 */
-               subtitle.reset (new SubtitleDecoder (this, c->subtitle, log, ContentTime()));
+               text.push_back (shared_ptr<TextDecoder> (new TextDecoder (this, c->only_text(), ContentTime())));
        }
 
        _next_time.resize (_format_context->nb_streams);
@@ -115,7 +112,7 @@ FFmpegDecoder::flush ()
 
        /* XXX: should we reset _packet.data and size after each *_decode_* call? */
 
-       while (video && decode_video_packet ()) {}
+       while (video && decode_video_packet()) {}
 
        if (audio) {
                decode_audio_packet ();
@@ -123,21 +120,21 @@ FFmpegDecoder::flush ()
 
        /* Make sure all streams are the same length and round up to the next video frame */
 
-       FrameRateChange const frc = _ffmpeg_content->film()->active_frame_rate_change(_ffmpeg_content->position());
-       ContentTime full_length (_ffmpeg_content->full_length(), frc);
+       FrameRateChange const frc = film()->active_frame_rate_change(_ffmpeg_content->position());
+       ContentTime full_length (_ffmpeg_content->full_length(film()), frc);
        full_length = full_length.ceil (frc.source);
        if (video) {
                double const vfr = _ffmpeg_content->video_frame_rate().get();
                Frame const f = full_length.frames_round (vfr);
-               Frame v = video->position().frames_round (vfr) + 1;
+               Frame v = video->position(film()).get_value_or(ContentTime()).frames_round(vfr) + 1;
                while (v < f) {
-                       video->emit (shared_ptr<const ImageProxy> (new RawImageProxy (_black_image)), v);
+                       video->emit (film(), shared_ptr<const ImageProxy> (new RawImageProxy (_black_image)), v);
                        ++v;
                }
        }
 
-       BOOST_FOREACH (shared_ptr<FFmpegAudioStream> i, _ffmpeg_content->ffmpeg_audio_streams ()) {
-               ContentTime a = audio->stream_position(i);
+       for (auto i: _ffmpeg_content->ffmpeg_audio_streams ()) {
+               ContentTime a = audio->stream_position(film(), i);
                /* Unfortunately if a is 0 that really means that we don't know the stream position since
                   there has been no data on it since the last seek.  In this case we'll just do nothing
                   here.  I'm not sure if that's the right idea.
@@ -147,7 +144,7 @@ FFmpegDecoder::flush ()
                                ContentTime to_do = min (full_length - a, ContentTime::from_seconds (0.1));
                                shared_ptr<AudioBuffers> silence (new AudioBuffers (i->channels(), to_do.frames_ceil (i->frame_rate())));
                                silence->make_silent ();
-                               audio->emit (i, silence, a);
+                               audio->emit (film(), i, silence, a, true);
                                a += to_do;
                        }
                }
@@ -182,9 +179,9 @@ FFmpegDecoder::pass ()
        int const si = _packet.stream_index;
        shared_ptr<const FFmpegContent> fc = _ffmpeg_content;
 
-       if (_video_stream && si == _video_stream.get() && !video->ignore()) {
+       if (_video_stream && si == _video_stream.get() && video && !video->ignore()) {
                decode_video_packet ();
-       } else if (fc->subtitle_stream() && fc->subtitle_stream()->uses_index(_format_context, si) && !subtitle->ignore()) {
+       } else if (fc->subtitle_stream() && fc->subtitle_stream()->uses_index(_format_context, si) && !only_text()->ignore()) {
                decode_subtitle_packet ();
        } else {
                decode_audio_packet ();
@@ -202,9 +199,14 @@ FFmpegDecoder::deinterleave_audio (shared_ptr<FFmpegAudioStream> stream) const
 {
        DCPOMATIC_ASSERT (bytes_per_audio_sample (stream));
 
+DCPOMATIC_DISABLE_WARNINGS
        int const size = av_samples_get_buffer_size (
                0, stream->stream(_format_context)->codec->channels, _frame->nb_samples, audio_sample_format (stream), 1
                );
+DCPOMATIC_ENABLE_WARNINGS
+
+       /* XXX: can't we just use _frame->nb_samples directly here? */
+       /* XXX: can't we use swr_convert() to do the format conversion? */
 
        /* Deinterleave and convert to float */
 
@@ -311,6 +313,7 @@ FFmpegDecoder::deinterleave_audio (shared_ptr<FFmpegAudioStream> stream) const
        case AV_SAMPLE_FMT_FLTP:
        {
                float** p = reinterpret_cast<float**> (_frame->data);
+               DCPOMATIC_ASSERT (_frame->channels <= channels);
                /* Sometimes there aren't as many channels in the _frame as in the stream */
                for (int i = 0; i < _frame->channels; ++i) {
                        memcpy (data[i], p[i], frames * sizeof(float));
@@ -331,7 +334,9 @@ FFmpegDecoder::deinterleave_audio (shared_ptr<FFmpegAudioStream> stream) const
 AVSampleFormat
 FFmpegDecoder::audio_sample_format (shared_ptr<FFmpegAudioStream> stream) const
 {
+DCPOMATIC_DISABLE_WARNINGS
        return stream->stream (_format_context)->codec->sample_fmt;
+DCPOMATIC_ENABLE_WARNINGS
 }
 
 int
@@ -361,6 +366,7 @@ FFmpegDecoder::seek (ContentTime time, bool accurate)
        if (_video_stream) {
                stream = _video_stream;
        } else {
+               DCPOMATIC_ASSERT (_ffmpeg_content->audio);
                shared_ptr<FFmpegAudioStream> s = dynamic_pointer_cast<FFmpegAudioStream> (_ffmpeg_content->audio->stream ());
                if (s) {
                        stream = s->index (_format_context);
@@ -380,19 +386,33 @@ FFmpegDecoder::seek (ContentTime time, bool accurate)
                AVSEEK_FLAG_BACKWARD
                );
 
+       {
+               /* Force re-creation of filter graphs to reset them and hence to make sure
+                  they don't have any pre-seek frames knocking about.
+               */
+               boost::mutex::scoped_lock lm (_filter_graphs_mutex);
+               _filter_graphs.clear ();
+       }
+
        if (video_codec_context ()) {
                avcodec_flush_buffers (video_codec_context());
        }
 
-       BOOST_FOREACH (shared_ptr<FFmpegAudioStream> i, ffmpeg_content()->ffmpeg_audio_streams()) {
+DCPOMATIC_DISABLE_WARNINGS
+       for (auto i: ffmpeg_content()->ffmpeg_audio_streams()) {
                avcodec_flush_buffers (i->stream(_format_context)->codec);
        }
+DCPOMATIC_ENABLE_WARNINGS
 
        if (subtitle_codec_context ()) {
                avcodec_flush_buffers (subtitle_codec_context ());
        }
 
        _have_current_subtitle = false;
+
+       for (auto& i: _next_time) {
+               i = optional<ContentTime>();
+       }
 }
 
 void
@@ -417,6 +437,7 @@ FFmpegDecoder::decode_audio_packet ()
                return;
        }
 
+DCPOMATIC_DISABLE_WARNINGS
        while (copy_packet.size > 0) {
 
                int frame_finished;
@@ -441,12 +462,14 @@ FFmpegDecoder::decode_audio_packet ()
                        shared_ptr<AudioBuffers> data = deinterleave_audio (*stream);
 
                        ContentTime ct;
-                       if (_frame->pts == AV_NOPTS_VALUE && _next_time[stream_index]) {
+                       if (_frame->pts == AV_NOPTS_VALUE) {
                                /* In some streams we see not every frame coming through with a timestamp; for those
                                   that have AV_NOPTS_VALUE we need to work out the timestamp ourselves.  This is
                                   particularly noticeable with TrueHD streams (see #1111).
                                */
-                               ct = *_next_time[stream_index];
+                               if (_next_time[stream_index]) {
+                                       ct = *_next_time[stream_index];
+                               }
                        } else {
                                ct = ContentTime::from_seconds (
                                        av_frame_get_best_effort_timestamp (_frame) *
@@ -476,10 +499,11 @@ FFmpegDecoder::decode_audio_packet ()
                                        to_string(_pts_offset)
                                        );
                        }
+DCPOMATIC_ENABLE_WARNINGS
 
                        /* Give this data provided there is some, and its time is sane */
                        if (ct >= ContentTime() && data->frames() > 0) {
-                               audio->emit (*stream, data, ct);
+                               audio->emit (film(), *stream, data, ct);
                        }
                }
 
@@ -494,9 +518,11 @@ FFmpegDecoder::decode_video_packet ()
        DCPOMATIC_ASSERT (_video_stream);
 
        int frame_finished;
+DCPOMATIC_DISABLE_WARNINGS
        if (avcodec_decode_video2 (video_codec_context(), _frame, &frame_finished, &_packet) < 0 || !frame_finished) {
                return false;
        }
+DCPOMATIC_ENABLE_WARNINGS
 
        boost::mutex::scoped_lock lm (_filter_graphs_mutex);
 
@@ -525,9 +551,11 @@ FFmpegDecoder::decode_video_packet ()
 
                if (i->second != AV_NOPTS_VALUE) {
                        double const pts = i->second * av_q2d (_format_context->streams[_video_stream.get()]->time_base) + _pts_offset.seconds ();
+
                        video->emit (
+                               film(),
                                shared_ptr<ImageProxy> (new RawImageProxy (image)),
-                               llrint (pts * _ffmpeg_content->active_video_frame_rate ())
+                               llrint(pts * _ffmpeg_content->active_video_frame_rate(film()))
                                );
                } else {
                        LOG_WARNING_NC ("Dropping frame without PTS");
@@ -549,9 +577,9 @@ FFmpegDecoder::decode_subtitle_packet ()
        /* Stop any current subtitle, either at the time it was supposed to stop, or now if now is sooner */
        if (_have_current_subtitle) {
                if (_current_subtitle_to) {
-                       subtitle->emit_stop (min(*_current_subtitle_to, subtitle_period(sub).from + _pts_offset));
+                       only_text()->emit_stop (min(*_current_subtitle_to, subtitle_period(sub).from + _pts_offset));
                } else {
-                       subtitle->emit_stop (subtitle_period(sub).from + _pts_offset);
+                       only_text()->emit_stop (subtitle_period(sub).from + _pts_offset);
                }
                _have_current_subtitle = false;
        }
@@ -593,7 +621,7 @@ FFmpegDecoder::decode_subtitle_packet ()
        }
 
        if (_current_subtitle_to) {
-               subtitle->emit_stop (*_current_subtitle_to);
+               only_text()->emit_stop (*_current_subtitle_to);
        }
 
        avsubtitle_free (&sub);
@@ -610,17 +638,17 @@ FFmpegDecoder::decode_bitmap_subtitle (AVSubtitleRect const * rect, ContentTime
 #ifdef DCPOMATIC_HAVE_AVSUBTITLERECT_PICT
        /* 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
+       /* sub_p looks up into a BGRA palette which is at rect->pict.data[1];
           (i.e. first byte B, second G, third R, fourth A)
        */
-       uint32_t const * palette = (uint32_t *) rect->pict.data[1];
+       uint8_t const * palette = rect->pict.data[1];
 #else
        /* Start of the first line in the subtitle */
        uint8_t* sub_p = rect->data[0];
-       /* sub_p looks up into a BGRA palette which is here
-          (i.e. first byte B, second G, third R, fourth A)
+       /* sub_p looks up into a BGRA palette which is at rect->data[1].
+          (first byte B, second G, third R, fourth A)
        */
-       uint32_t const * palette = (uint32_t *) rect->data[1];
+       uint8_t const * palette = rect->data[1];
 #endif
        /* And the stream has a map of those palette colours to colours
           chosen by the user; created a `mapped' palette from those settings.
@@ -628,7 +656,7 @@ FFmpegDecoder::decode_bitmap_subtitle (AVSubtitleRect const * rect, ContentTime
        map<RGBA, RGBA> colour_map = ffmpeg_content()->subtitle_stream()->colours ();
        vector<RGBA> mapped_palette (rect->nb_colors);
        for (int i = 0; i < rect->nb_colors; ++i) {
-               RGBA c ((palette[i] & 0xff0000) >> 16, (palette[i] & 0xff00) >> 8, palette[i] & 0xff, (palette[i] & 0xff000000) >> 24);
+               RGBA c (palette[2], palette[1], palette[0], palette[3]);
                map<RGBA, RGBA>::const_iterator j = colour_map.find (c);
                if (j != colour_map.end ()) {
                        mapped_palette[i] = j->second;
@@ -639,29 +667,43 @@ FFmpegDecoder::decode_bitmap_subtitle (AVSubtitleRect const * rect, ContentTime
                        */
                        mapped_palette[i] = c;
                }
+               palette += 4;
        }
 
        /* Start of the output data */
-       uint32_t* out_p = (uint32_t *) image->data()[0];
+       uint8_t* out_p = 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;
+               uint8_t* out_line_p = out_p;
                for (int x = 0; x < rect->w; ++x) {
                        RGBA const p = mapped_palette[*sub_line_p++];
-                       /* XXX: this seems to be wrong to me (isn't the output image BGRA?) but it looks right on screen */
-                       *out_line_p++ = (p.a << 24) | (p.b << 16) | (p.g << 8) | p.r;
+                       *out_line_p++ = p.b;
+                       *out_line_p++ = p.g;
+                       *out_line_p++ = p.r;
+                       *out_line_p++ = p.a;
                }
 #ifdef DCPOMATIC_HAVE_AVSUBTITLERECT_PICT
                sub_p += rect->pict.linesize[0];
 #else
                sub_p += rect->linesize[0];
 #endif
-               out_p += image->stride()[0] / sizeof (uint32_t);
+               out_p += image->stride()[0];
        }
 
-       int const target_width = subtitle_codec_context()->width;
-       int const target_height = subtitle_codec_context()->height;
+       int target_width = subtitle_codec_context()->width;
+       if (target_width == 0 && video_codec_context()) {
+               /* subtitle_codec_context()->width == 0 has been seen in the wild but I don't
+                  know if it's supposed to mean something from FFmpeg's point of view.
+               */
+               target_width = video_codec_context()->width;
+       }
+       int target_height = subtitle_codec_context()->height;
+       if (target_height == 0 && video_codec_context()) {
+               target_height = video_codec_context()->height;
+       }
+       DCPOMATIC_ASSERT (target_width);
+       DCPOMATIC_ASSERT (target_height);
        dcpomatic::Rect<double> const scaled_rect (
                static_cast<double> (rect->x) / target_width,
                static_cast<double> (rect->y) / target_height,
@@ -669,7 +711,7 @@ FFmpegDecoder::decode_bitmap_subtitle (AVSubtitleRect const * rect, ContentTime
                static_cast<double> (rect->h) / target_height
                );
 
-       subtitle->emit_image_start (from, image, scaled_rect);
+       only_text()->emit_bitmap_start (from, image, scaled_rect);
 }
 
 void
@@ -679,21 +721,29 @@ FFmpegDecoder::decode_ass_subtitle (string ass, ContentTime from)
           produces a single format of Dialogue: lines...
        */
 
-       vector<string> bits;
-       split (bits, ass, is_any_of (","));
-       if (bits.size() < 10) {
+       int commas = 0;
+       string text;
+       for (size_t i = 0; i < ass.length(); ++i) {
+               if (commas < 9 && ass[i] == ',') {
+                       ++commas;
+               } else if (commas == 9) {
+                       text += ass[i];
+               }
+       }
+
+       if (text.empty ()) {
                return;
        }
 
        sub::RawSubtitle base;
        list<sub::RawSubtitle> raw = sub::SSAReader::parse_line (
                base,
-               bits[9],
+               text,
                _ffmpeg_content->video->size().width,
                _ffmpeg_content->video->size().height
                );
 
-       BOOST_FOREACH (sub::Subtitle const & i, sub::collect<list<sub::Subtitle> > (raw)) {
-               subtitle->emit_text_start (from, i);
+       for (auto const& i: sub::collect<list<sub::Subtitle> > (raw)) {
+               only_text()->emit_plain_start (from, i);
        }
 }