Don't crash if the first packet in a stream has AV_NOPTS_VALUE;
[dcpomatic.git] / src / lib / ffmpeg_decoder.cc
index 909a9d443d1383ebca4179554fa9745c12e5acfb..69f366aade4e86f0e704a8041d4f781686823ed6 100644 (file)
@@ -27,6 +27,7 @@
 #include "image.h"
 #include "util.h"
 #include "log.h"
+#include "dcpomatic_log.h"
 #include "ffmpeg_decoder.h"
 #include "text_decoder.h"
 #include "ffmpeg_audio_stream.h"
@@ -41,6 +42,7 @@
 #include "compose.hpp"
 #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>
@@ -58,10 +60,6 @@ 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;
@@ -78,14 +76,14 @@ using boost::optional;
 using boost::dynamic_pointer_cast;
 using dcp::Size;
 
-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());
+               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 +92,12 @@ 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 TextDecoder (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 +113,7 @@ 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 +121,21 @@ 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);
+               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 +145,7 @@ 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);
                                a += to_do;
                        }
                }
@@ -184,7 +182,7 @@ FFmpegDecoder::pass ()
 
        if (_video_stream && si == _video_stream.get() && !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 ();
@@ -380,6 +378,14 @@ 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());
        }
@@ -393,6 +399,10 @@ FFmpegDecoder::seek (ContentTime time, bool accurate)
        }
 
        _have_current_subtitle = false;
+
+       BOOST_FOREACH (optional<ContentTime>& i, _next_time) {
+               i = optional<ContentTime>();
+       }
 }
 
 void
@@ -441,12 +451,14 @@ 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) *
@@ -479,7 +491,7 @@ FFmpegDecoder::decode_audio_packet ()
 
                        /* 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);
                        }
                }
 
@@ -525,9 +537,11 @@ 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 +563,9 @@ 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 +607,7 @@ FFmpegDecoder::decode_subtitle_packet ()
        }
 
        if (_current_subtitle_to) {
-               subtitle->emit_stop (*_current_subtitle_to);
+               only_text()->emit_stop (*_current_subtitle_to);
        }
 
        avsubtitle_free (&sub);
@@ -617,18 +631,18 @@ FFmpegDecoder::decode_bitmap_subtitle (AVSubtitleRect const * rect, ContentTime
 #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];
 #endif
        /* And the stream has a map of those palette colours to colours
           chosen by the user; created a `mapped' palette from those settings.
        */
        map<RGBA, RGBA> colour_map = ffmpeg_content()->subtitle_stream()->colours ();
        vector<RGBA> mapped_palette (rect->nb_colors);
+       uint8_t const * palette = rect->data[1];
        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 +653,43 @@ 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 +697,7 @@ FFmpegDecoder::decode_bitmap_subtitle (AVSubtitleRect const * rect, ContentTime
                static_cast<double> (rect->h) / target_height
                );
 
-       subtitle->emit_bitmap_start (from, image, scaled_rect);
+       only_text()->emit_bitmap_start (from, image, scaled_rect);
 }
 
 void
@@ -702,6 +730,6 @@ FFmpegDecoder::decode_ass_subtitle (string ass, ContentTime from)
                );
 
        BOOST_FOREACH (sub::Subtitle const & i, sub::collect<list<sub::Subtitle> > (raw)) {
-               subtitle->emit_plain_start (from, i);
+               only_text()->emit_plain_start (from, i);
        }
 }