libhb: fix or simplify several hacks involved with Libav support

For files that are demuxed by Libav, we must share the format context
with the decoder iso that it can obtain the codec context for each stream.
The code that did this was very convoluted and difficult to understand.
It is simplified by simply passing the context in hb_title_t.

Reader was closing stream files before the decoder was finished with the
context.  This created the need to delay the actual close and cache
the context.  Changed reader so it behaves more like the rest of handbrake's
work objects which lets us explicitly close after the decoders are finished.

Libav does some probing of the file when av_find_stream_info is called.
This probing leaves the format context in a bad state for some files and
causes subsequent reads or seeks to misbehave.  So open 2 contexts in
ffmpeg_open.  One is used only for probing, and the other only for reading.

decavcodec.c had 2 separate decoders for files demuxed by hb and files
demuxed by Libav.  They have been combined and simplified.

Previously, it was not possible to decode one source audio track multiple
times in order to fan it out to multiple output tracks if the file is
demuxed by Libav.  We were using the codec context from the format context.
Since there is only one of these for each stream, we could only do one
decode for each stream.  Use avcodec_copy_context to make copies of
the codec context and allow multiple decodes. This allows removal of
a lot of special case code for Libav streams that was necessary to
duplicate the output of the decoder.

Patch Libav's mkv demux to fix a seek problem.  This has been pushed 
upstreams, so the next time we update Libav, we must remove this patch.


git-svn-id: svn://svn.handbrake.fr/HandBrake/trunk@4141 b64f7644-9d1e-0410-96f1-a4d463321fa5

1 files changed, 71 insertions, 62 deletions

diff --git a/libhb/scan.c b/libhb/scan.c
index 34137c42b..8999dc15c 100644
--- a/libhb/scan.c
+++ b/libhb/scan.c
@@ -149,14 +149,21 @@ static void ScanFunc( void * _data )
             }
         }
     }
-    else if ( (data->stream = hb_stream_open( data->path, 0 ) ) != NULL )
-    {
-        hb_list_add( data->list_title, hb_stream_title_scan( data->stream ) );
-    }
     else
     {
-        hb_log( "scan: unrecognized file type" );
-        return;
+        hb_title_t * title = hb_title_init( data->path, 0 );
+        if ( (data->stream = hb_stream_open( data->path, title, 1 ) ) != NULL )
+        {
+            title = hb_stream_title_scan( data->stream, title );
+            if ( title )
+                hb_list_add( data->list_title, title );
+        }
+        else
+        {
+            hb_title_close( &title );
+            hb_log( "scan: unrecognized file type" );
+            return;
+        }
     }
 
     for( i = 0; i < hb_list_count( data->list_title ); )
@@ -338,17 +345,17 @@ static inline int clampBlack( int x )
     return x < 16 ? 16 : x;
 }
 
-static int row_all_dark( hb_title_t *title, uint8_t* luma, int row )
+static int row_all_dark( hb_work_info_t *info, uint8_t* luma, int row )
 {
-    luma += title->width * row;
+    luma += info->width * row;
 
     // compute the average luma value of the row
     int i, avg = 0;
-    for ( i = 0; i < title->width; ++i )
+    for ( i = 0; i < info->width; ++i )
     {
         avg += clampBlack( luma[i] );
     }
-    avg /= title->width;
+    avg /= info->width;
     if ( avg >= DARK )
         return 0;
 
@@ -356,7 +363,7 @@ static int row_all_dark( hb_title_t *title, uint8_t* luma, int row )
     // all pixels are within +-16 of the average (this range is fairly coarse
     // but there's a lot of quantization noise for luma values near black
     // so anything less will fail to crop because of the noise).
-    for ( i = 0; i < title->width; ++i )
+    for ( i = 0; i < info->width; ++i )
     {
         if ( absdiff( avg, clampBlack( luma[i] ) ) > 16 )
             return 0;
@@ -364,11 +371,11 @@ static int row_all_dark( hb_title_t *title, uint8_t* luma, int row )
     return 1;
 }
 
-static int column_all_dark( hb_title_t *title, uint8_t* luma, int top, int bottom,
+static int column_all_dark( hb_work_info_t *info, uint8_t* luma, int top, int bottom,
                             int col )
 {
-    int stride = title->width;
-    int height = title->height - top - bottom;
+    int stride = info->width;
+    int height = info->height - top - bottom;
     luma += stride * top + col;
 
     // compute the average value of the column
@@ -495,7 +502,7 @@ static int DecodePreviews( hb_scan_t * data, hb_title_t * title )
     }
     else if (data->batch)
     {
-        data->stream = hb_stream_open( title->path, title );
+        data->stream = hb_stream_open( title->path, title, 1 );
     }
 
     for( i = 0; i < data->preview_count; i++ )
@@ -670,22 +677,13 @@ static int DecodePreviews( hb_scan_t * data, hb_title_t * title )
             }
             vid_decoder->close( vid_decoder );
             free( vid_decoder );
+            hb_log( "scan: could not get a video information" );
             continue;
         }
 
         remember_info( info_list, &vid_info );
 
-        if ( title->video_codec_name == NULL )
-        {
-            title->video_codec_name = strdup( vid_info.name );
-        }
-        title->width = vid_info.width;
-        title->height = vid_info.height;
-        title->rate = vid_info.rate;
-        title->rate_base = vid_info.rate_base;
-        title->video_bitrate = vid_info.bitrate;
-
-        if( title->rate_base == 1126125 )
+        if( vid_info.rate_base == 1126125 )
         {
             /* Frame FPS is 23.976 (meaning it's progressive), so
                start keeping track of how many are reporting at
@@ -700,7 +698,7 @@ static int DecodePreviews( hb_scan_t * data, hb_title_t * title )
                    which means we should be conservative and use
                    29.97 as the title's FPS for now.
                 */
-                title->rate_base = 900900;
+                vid_info.rate_base = 900900;
             }
             else
             {
@@ -711,16 +709,16 @@ static int DecodePreviews( hb_scan_t * data, hb_title_t * title )
                 {
                     hb_deep_log( 2, "Title's mostly NTSC Film, setting fps to 23.976");
                 }
-                title->rate_base = 1126125;
+                vid_info.rate_base = 1126125;
             }
         }
-        else if( title->rate_base == 900900 && progressive_count >= 6 )
+        else if( vid_info.rate_base == 900900 && progressive_count >= 6 )
         {
             /*
              * We've already deduced that the frame rate is 23.976, so set it
              * back again.
              */
-            title->rate_base = 1126125;
+            vid_info.rate_base = 1126125;
         }
 
         while( ( buf_es = hb_list_item( list_es, 0 ) ) )
@@ -730,7 +728,7 @@ static int DecodePreviews( hb_scan_t * data, hb_title_t * title )
         }
 
         /* Check preview for interlacing artifacts */
-        if( hb_detect_comb( vid_buf, title->width, title->height, 10, 30, 9, 10, 30, 9 ) )
+        if( hb_detect_comb( vid_buf, vid_info.width, vid_info.height, 10, 30, 9, 10, 30, 9 ) )
         {
             hb_deep_log( 2, "Interlacing detected in preview frame %i", i+1);
             interlaced_preview_count++;
@@ -744,7 +742,7 @@ static int DecodePreviews( hb_scan_t * data, hb_title_t * title )
             file_preview = fopen( filename, "wb" );
             if( file_preview )
             {
-                fwrite( vid_buf->data, title->width * title->height * 3 / 2,
+                fwrite( vid_buf->data, vid_info.width * vid_info.height * 3 / 2,
                         1, file_preview );
                 fclose( file_preview );
             }
@@ -758,7 +756,7 @@ static int DecodePreviews( hb_scan_t * data, hb_title_t * title )
 
 #define Y    vid_buf->data
         int top, bottom, left, right;
-        int h4 = title->height / 4, w4 = title->width / 4;
+        int h4 = vid_info.height / 4, w4 = vid_info.width / 4;
 
         // When widescreen content is matted to 16:9 or 4:3 there's sometimes
         // a thin border on the outer edge of the matte. On TV content it can be
@@ -768,11 +766,11 @@ static int DecodePreviews( hb_scan_t * data, hb_title_t * title )
         // we can crop the matte. The border width depends on the resolution
         // (12 pixels on 1080i looks visually the same as 4 pixels on 480i)
         // so we allow the border to be up to 1% of the frame height.
-        const int border = title->height / 100;
+        const int border = vid_info.height / 100;
 
         for ( top = border; top < h4; ++top )
         {
-            if ( ! row_all_dark( title, Y, top ) )
+            if ( ! row_all_dark( &vid_info, Y, top ) )
                 break;
         }
         if ( top <= border )
@@ -781,7 +779,7 @@ static int DecodePreviews( hb_scan_t * data, hb_title_t * title )
             // didn't check are dark or if we shouldn't crop at all.
             for ( top = 0; top < border; ++top )
             {
-                if ( ! row_all_dark( title, Y, top ) )
+                if ( ! row_all_dark( &vid_info, Y, top ) )
                     break;
             }
             if ( top >= border )
@@ -791,14 +789,14 @@ static int DecodePreviews( hb_scan_t * data, hb_title_t * title )
         }
         for ( bottom = border; bottom < h4; ++bottom )
         {
-            if ( ! row_all_dark( title, Y, title->height - 1 - bottom ) )
+            if ( ! row_all_dark( &vid_info, Y, vid_info.height - 1 - bottom ) )
                 break;
         }
         if ( bottom <= border )
         {
             for ( bottom = 0; bottom < border; ++bottom )
             {
-                if ( ! row_all_dark( title, Y, title->height - 1 - bottom ) )
+                if ( ! row_all_dark( &vid_info, Y, vid_info.height - 1 - bottom ) )
                     break;
             }
             if ( bottom >= border )
@@ -808,12 +806,12 @@ static int DecodePreviews( hb_scan_t * data, hb_title_t * title )
         }
         for ( left = 0; left < w4; ++left )
         {
-            if ( ! column_all_dark( title, Y, top, bottom, left ) )
+            if ( ! column_all_dark( &vid_info, Y, top, bottom, left ) )
                 break;
         }
         for ( right = 0; right < w4; ++right )
         {
-            if ( ! column_all_dark( title, Y, top, bottom, title->width - 1 - right ) )
+            if ( ! column_all_dark( &vid_info, Y, top, bottom, vid_info.width - 1 - right ) )
                 break;
         }
 
@@ -856,35 +854,46 @@ skip_preview:
         hb_work_info_t vid_info;
         most_common_info( info_list, &vid_info );
 
+        if ( title->video_codec_name == NULL )
+        {
+            title->video_codec_name = strdup( vid_info.name );
+        }
         title->width = vid_info.width;
         title->height = vid_info.height;
-        title->pixel_aspect_width = vid_info.pixel_aspect_width;
-        title->pixel_aspect_height = vid_info.pixel_aspect_height;
+        if ( vid_info.rate && vid_info.rate_base )
+        {
+            title->rate = vid_info.rate;
+            title->rate_base = vid_info.rate_base;
+        }
+        title->video_bitrate = vid_info.bitrate;
+
+        if( vid_info.pixel_aspect_width && vid_info.pixel_aspect_height )
+        {
+            title->pixel_aspect_width = vid_info.pixel_aspect_width;
+            title->pixel_aspect_height = vid_info.pixel_aspect_height;
+        }
 
         // compute the aspect ratio based on the storage dimensions and the
         // pixel aspect ratio (if supplied) or just storage dimensions if no PAR.
         title->aspect = (double)title->width / (double)title->height;
-        if( title->pixel_aspect_width && title->pixel_aspect_height )
+        title->aspect *= (double)title->pixel_aspect_width /
+                         (double)title->pixel_aspect_height;
+
+        // For unknown reasons some French PAL DVDs put the original
+        // content's aspect ratio into the mpeg PAR even though it's
+        // the wrong PAR for the DVD. Apparently they rely on the fact
+        // that DVD players ignore the content PAR and just use the
+        // aspect ratio from the DVD metadata. So, if the aspect computed
+        // from the PAR is different from the container's aspect we use
+        // the container's aspect & recompute the PAR from it.
+        if( title->container_aspect && (int)(title->aspect * 9) != (int)(title->container_aspect * 9) )
         {
-            title->aspect *= (double)title->pixel_aspect_width /
-                             (double)title->pixel_aspect_height;
-
-            // For unknown reasons some French PAL DVDs put the original
-            // content's aspect ratio into the mpeg PAR even though it's
-            // the wrong PAR for the DVD. Apparently they rely on the fact
-            // that DVD players ignore the content PAR and just use the
-            // aspect ratio from the DVD metadata. So, if the aspect computed
-            // from the PAR is different from the container's aspect we use
-            // the container's aspect & recompute the PAR from it.
-            if( title->container_aspect && (int)(title->aspect * 9) != (int)(title->container_aspect * 9) )
-            {
-                hb_log("scan: content PAR gives wrong aspect %.2f; "
-                       "using container aspect %.2f", title->aspect,
-                       title->container_aspect );
-                title->aspect = title->container_aspect;
-                hb_reduce( &title->pixel_aspect_width, &title->pixel_aspect_height,
-                           (int)(title->aspect * title->height + 0.5), title->width );
-            }
+            hb_log("scan: content PAR gives wrong aspect %.2f; "
+                   "using container aspect %.2f", title->aspect,
+                   title->container_aspect );
+            title->aspect = title->container_aspect;
+            hb_reduce( &title->pixel_aspect_width, &title->pixel_aspect_height,
+                       (int)(title->aspect * title->height + 0.5), title->width );
         }
 
         // don't try to crop unless we got at least 3 previews