[melted] / src / modules / avformat / producer_avformat.c

/*
 * producer_avformat.c -- avformat producer
 * Copyright (C) 2003-2004 Ushodaya Enterprises Limited
 * Author: Charles Yates <charles.yates@pandora.be>
 *
 * 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
 * the Free Software Foundation; either version 2 of the License, or
 * (at your option) any later version.
 *
 * This program is distributed in the hope that it will be useful,
 * but WITHOUT ANY WARRANTY; without even the implied warranty of
 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
 * GNU General Public License for more details.
 *
 * You should have received a copy of the GNU General Public License
 * along with this program; if not, write to the Free Software Foundation,
 * Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA.
 */

// Local header files
#include "producer_avformat.h"

// MLT Header files
#include <framework/mlt_frame.h>

// ffmpeg Header files
#include <avformat.h>

// System header files
#include <stdlib.h>
#include <string.h>
#include <pthread.h>
#include <math.h>

void avformat_lock( );
void avformat_unlock( );

// Forward references.
static int producer_open( mlt_producer this, char *file );
static int producer_get_frame( mlt_producer this, mlt_frame_ptr frame, int index );

/** Constructor for libavformat.
*/

mlt_producer producer_avformat_init( char *file )
{
        mlt_producer this = NULL;

        // Check that we have a non-NULL argument
        if ( file != NULL )
        {
                // Construct the producer
                this = calloc( 1, sizeof( struct mlt_producer_s ) );

                // Initialise it
                if ( mlt_producer_init( this, NULL ) == 0 )
                {
                        // Get the properties
                        mlt_properties properties = MLT_PRODUCER_PROPERTIES( this );

                        // Set the resource property (required for all producers)
                        mlt_properties_set( properties, "resource", file );

                        // Register our get_frame implementation
                        this->get_frame = producer_get_frame;

                        // Open the file
                        if ( producer_open( this, file ) != 0 )
                        {
                                // Clean up
                                mlt_producer_close( this );
                                this = NULL;
                        }
                }
        }

        return this;
}

/** Find the default streams.
*/

static void find_default_streams( AVFormatContext *context, int *audio_index, int *video_index )
{
        int i;

        // Allow for multiple audio and video streams in the file and select first of each (if available)
        for( i = 0; i < context->nb_streams; i++ ) 
        {
                // Get the codec context
                AVCodecContext *codec_context = context->streams[ i ]->codec;

                // Determine the type and obtain the first index of each type
                switch( codec_context->codec_type ) 
                {
                        case CODEC_TYPE_VIDEO:
                                if ( *video_index < 0 )
                                        *video_index = i;
                                break;
                        case CODEC_TYPE_AUDIO:
                                if ( *audio_index < 0 )
                                        *audio_index = i;
                                break;
                        default:
                                break;
                }
        }
}

/** Producer file destructor.
*/

static void producer_file_close( void *context )
{
        if ( context != NULL )
        {
                // Lock the mutex now
                avformat_lock( );

                // Close the file
                av_close_input_file( context );

                // Unlock the mutex now
                avformat_unlock( );
        }
}

/** Producer file destructor.
*/

static void producer_codec_close( void *codec )
{
        if ( codec != NULL )
        {
                // Lock the mutex now
                avformat_lock( );

                // Close the file
                avcodec_close( codec );

                // Unlock the mutex now
                avformat_unlock( );
        }
}

/** Open the file.
*/

static int producer_open( mlt_producer this, char *file )
{
        // Return an error code (0 == no error)
        int error = 0;

        // Context for avformat
        AVFormatContext *context = NULL;

        // Get the properties
        mlt_properties properties = MLT_PRODUCER_PROPERTIES( this );

        // We will treat everything with the producer fps
        double fps = mlt_properties_get_double( properties, "fps" );

        // Lock the mutex now
        avformat_lock( );
        
        // If "MRL", then create AVInputFormat
        AVInputFormat *format = NULL;
        AVFormatParameters *params = NULL;
        char *standard = NULL;
        char *mrl = strchr( file, ':' );

        // AV option (0 = both, 1 = video, 2 = audio)
        int av = 0;
        
        // Setting lowest log level
        av_log_set_level( -1 );

        // Only if there is not a protocol specification that avformat can handle
        if ( mrl && !url_exist( file ) )
        {
                // 'file' becomes format abbreviation
                mrl[0] = 0;
        
                // Lookup the format
                format = av_find_input_format( file );
                
                // Eat the format designator
                file = ++mrl;
                
                if ( format )
                {
                        // Allocate params
                        params = calloc( sizeof( AVFormatParameters ), 1 );
                        
                        // These are required by video4linux (defaults)
                        params->width = 640;
                        params->height = 480;
                        params->time_base= (AVRational){1,25};
                        params->device = file;
                        params->channels = 2;
                        params->sample_rate = 48000;
                }
                
                // Parse out params
                mrl = strchr( file, '?' );
                while ( mrl )
                {
                        mrl[0] = 0;
                        char *name = strdup( ++mrl );
                        char *value = strchr( name, ':' );
                        if ( value )
                        {
                                value[0] = 0;
                                value++;
                                char *t = strchr( value, '&' );
                                if ( t )
                                        t[0] = 0;
                                if ( !strcmp( name, "frame_rate" ) )
                                        params->time_base.den = atoi( value );
                                else if ( !strcmp( name, "frame_rate_base" ) )
                                        params->time_base.num = atoi( value );
                                else if ( !strcmp( name, "sample_rate" ) )
                                        params->sample_rate = atoi( value );
                                else if ( !strcmp( name, "channels" ) )
                                        params->channels = atoi( value );
                                else if ( !strcmp( name, "width" ) )
                                        params->width = atoi( value );
                                else if ( !strcmp( name, "height" ) )
                                        params->height = atoi( value );
                                else if ( !strcmp( name, "standard" ) )
                                {
                                        standard = strdup( value );
                                        params->standard = standard;
                                }
                                else if ( !strcmp( name, "av" ) )
                                        av = atoi( value );
                        }
                        free( name );
                        mrl = strchr( mrl, '&' );
                }
        }

        // Now attempt to open the file
        error = av_open_input_file( &context, file, format, 0, params ) < 0;
        
        // Cleanup AVFormatParameters
        free( standard );
        free( params );

        // If successful, then try to get additional info
        if ( error == 0 )
        {
                // Get the stream info
                error = av_find_stream_info( context ) < 0;

                // Continue if no error
                if ( error == 0 )
                {
                        // We will default to the first audio and video streams found
                        int audio_index = -1;
                        int video_index = -1;
                        int av_bypass = 0;

                        // Now set properties where we can (use default unknowns if required)
                        if ( context->duration != AV_NOPTS_VALUE ) 
                        {
                                // This isn't going to be accurate for all formats
                                mlt_position frames = ( mlt_position )( ( ( double )context->duration / ( double )AV_TIME_BASE ) * fps );
                                mlt_properties_set_position( properties, "out", frames - 2 );
                                mlt_properties_set_position( properties, "length", frames - 1 );
                        }

                        // Find default audio and video streams
                        find_default_streams( context, &audio_index, &video_index );

            if ( context->start_time != AV_NOPTS_VALUE )
                mlt_properties_set_double( properties, "_start_time", context->start_time );
                        
                        // Check if we're seekable (something funny about mpeg here :-/)
                        if ( strcmp( file, "pipe:" ) && strncmp( file, "http://", 6 ) )
                                mlt_properties_set_int( properties, "seekable", av_seek_frame( context, -1, mlt_properties_get_double( properties, "_start_time" ), AVSEEK_FLAG_BACKWARD ) >= 0 );
                        else
                                av_bypass = 1;

                        // Store selected audio and video indexes on properties
                        mlt_properties_set_int( properties, "audio_index", audio_index );
                        mlt_properties_set_int( properties, "video_index", video_index );

                        // Fetch the width, height and aspect ratio
                        if ( video_index != -1 )
                        {
                                AVCodecContext *codec_context = context->streams[ video_index ]->codec;
                                mlt_properties_set_int( properties, "width", codec_context->width );
                                mlt_properties_set_int( properties, "height", codec_context->height );
                                mlt_properties_set_double( properties, "aspect_ratio", av_q2d( codec_context->sample_aspect_ratio ) );
                        }
                        
                        // We're going to cheat here - for a/v files, we will have two contexts (reasoning will be clear later)
                        if ( av == 0 && !av_bypass && audio_index != -1 && video_index != -1 )
                        {
                                // We'll use the open one as our video_context
                                mlt_properties_set_data( properties, "video_context", context, 0, producer_file_close, NULL );

                                // And open again for our audio context
                                av_open_input_file( &context, file, NULL, 0, NULL );
                                av_find_stream_info( context );

                                // Audio context
                                mlt_properties_set_data( properties, "audio_context", context, 0, producer_file_close, NULL );
                        }
                        else if ( av != 2 && video_index != -1 )
                        {
                                // We only have a video context
                                mlt_properties_set_data( properties, "video_context", context, 0, producer_file_close, NULL );
                        }
                        else if ( audio_index != -1 )
                        {
                                // We only have an audio context
                                mlt_properties_set_data( properties, "audio_context", context, 0, producer_file_close, NULL );
                        }
                        else
                        {
                                // Something has gone wrong
                                error = -1;
                        }

                        mlt_properties_set_int( properties, "av_bypass", av_bypass );
                }
        }

        // Unlock the mutex now
        avformat_unlock( );

        return error;
}

/** Convert a frame position to a time code.
*/

static double producer_time_of_frame( mlt_producer this, mlt_position position )
{
        // Get the properties
        mlt_properties properties = MLT_PRODUCER_PROPERTIES( this );

        // Obtain the fps
        double fps = mlt_properties_get_double( properties, "fps" );

        // Do the calc
        return ( double )position / fps;
}

static inline void convert_image( AVFrame *frame, uint8_t *buffer, int pix_fmt, mlt_image_format format, int width, int height )
{
        if ( format == mlt_image_yuv420p )
        {
                AVPicture pict;
                pict.data[0] = buffer;
                pict.data[1] = buffer + width * height;
                pict.data[2] = buffer + ( 3 * width * height ) / 2;
                pict.linesize[0] = width;
                pict.linesize[1] = width >> 1;
                pict.linesize[2] = width >> 1;
                img_convert( &pict, PIX_FMT_YUV420P, (AVPicture *)frame, pix_fmt, width, height );
        }
        else if ( format == mlt_image_rgb24 )
        {
                AVPicture output;
                avpicture_fill( &output, buffer, PIX_FMT_RGB24, width, height );
                img_convert( &output, PIX_FMT_RGB24, (AVPicture *)frame, pix_fmt, width, height );
        }
        else
        {
                AVPicture output;
                avpicture_fill( &output, buffer, PIX_FMT_YUV422, width, height );
                img_convert( &output, PIX_FMT_YUV422, (AVPicture *)frame, pix_fmt, width, height );
        }
}

/** Get an image from a frame.
*/

static int producer_get_image( mlt_frame frame, uint8_t **buffer, mlt_image_format *format, int *width, int *height, int writable )
{
        // Get the properties from the frame
        mlt_properties frame_properties = MLT_FRAME_PROPERTIES( frame );

        // Obtain the frame number of this frame
        mlt_position position = mlt_properties_get_position( frame_properties, "avformat_position" );

        // Get the producer 
        mlt_producer this = mlt_properties_get_data( frame_properties, "avformat_producer", NULL );

        // Get the producer properties
        mlt_properties properties = MLT_PRODUCER_PROPERTIES( this );

        // Fetch the video_context
        AVFormatContext *context = mlt_properties_get_data( properties, "video_context", NULL );

        // Get the video_index
        int index = mlt_properties_get_int( properties, "video_index" );

        // Obtain the expected frame numer
        mlt_position expected = mlt_properties_get_position( properties, "_video_expected" );

        // Calculate the real time code
        double real_timecode = producer_time_of_frame( this, position );

        // Get the video stream
        AVStream *stream = context->streams[ index ];

        // Get codec context
        AVCodecContext *codec_context = stream->codec;

        // Packet
        AVPacket pkt;

        // Get the conversion frame
        AVFrame *av_frame = mlt_properties_get_data( properties, "av_frame", NULL );

        // Special case pause handling flag
        int paused = 0;

        // Special case ffwd handling
        int ignore = 0;

        // Current time calcs
        double current_time = mlt_properties_get_double( properties, "_current_time" );

        // We may want to use the source fps if available
        double source_fps = mlt_properties_get_double( properties, "source_fps" );

        // Get the seekable status
        int seekable = mlt_properties_get_int( properties, "seekable" );

        // Generate the size in bytes
        int size = 0; 

        // Hopefully provide better support for streams...
        int av_bypass = mlt_properties_get_int( properties, "av_bypass" );

        // Set the result arguments that we know here (only *buffer is now required)
        *width = codec_context->width;
        *height = codec_context->height;

        switch ( *format )
        {
                case mlt_image_yuv420p:
                        size = *width * 3 * ( *height + 1 ) / 2;
                        break;
                case mlt_image_rgb24:
                        size = *width * ( *height + 1 ) * 3;
                        break;
                default:
                        *format = mlt_image_yuv422;
                        size = *width * ( *height + 1 ) * 2;
                        break;
        }

        // Set this on the frame properties
        mlt_properties_set_int( frame_properties, "width", *width );
        mlt_properties_set_int( frame_properties, "height", *height );

        // Construct the output image
        *buffer = mlt_pool_alloc( size );

        // Seek if necessary
        if ( position != expected )
        {
                if ( position + 1 == expected )
                {
                        // We're paused - use last image
                        paused = 1;
                }
                else if ( !seekable && position > expected && ( position - expected ) < 250 )
                {
                        // Fast forward - seeking is inefficient for small distances - just ignore following frames
                        ignore = position - expected;
                }
                else if ( seekable && ( position < expected || position - expected >= 12 ) )
                {
                        // Set to the real timecode
                        av_seek_frame( context, -1, mlt_properties_get_double( properties, "_start_time" ) + real_timecode * 1000000.0, AVSEEK_FLAG_BACKWARD );
        
                        // Remove the cached info relating to the previous position
                        mlt_properties_set_double( properties, "_current_time", real_timecode );

                        mlt_properties_set_data( properties, "av_frame", NULL, 0, NULL, NULL );
                        av_frame = NULL;
                }
        }
        
        // Duplicate the last image if necessary
        if ( av_frame != NULL && ( paused || mlt_properties_get_double( properties, "_current_time" ) >= real_timecode ) &&
                 av_bypass == 0 )
        {
                // Duplicate it
                convert_image( av_frame, *buffer, codec_context->pix_fmt, *format, *width, *height );

                // Set this on the frame properties
                mlt_properties_set_data( frame_properties, "image", *buffer, size, ( mlt_destructor )mlt_pool_release, NULL );
        }
        else
        {
                int ret = 0;
                int got_picture = 0;
                int must_decode = 1;

                // Temporary hack to improve intra frame only
                if ( !strcmp( codec_context->codec->name, "mjpeg" ) )
                        must_decode = 0;

                av_init_packet( &pkt );

                // Construct an AVFrame for YUV422 conversion
                if ( av_frame == NULL )
                {
                        av_frame = avcodec_alloc_frame( );
                        mlt_properties_set_data( properties, "av_frame", av_frame, 0, av_free, NULL );
                }

                while( ret >= 0 && !got_picture )
                {
                        // Read a packet
                        ret = av_read_frame( context, &pkt );

                        // We only deal with video from the selected video_index
                        if ( ret >= 0 && pkt.stream_index == index && pkt.size > 0 )
                        {
                                // Determine time code of the packet
                                if ( pkt.dts != AV_NOPTS_VALUE )
                                        current_time = av_q2d( stream->time_base ) * pkt.dts;
                                else
                                        current_time = real_timecode;

                                // Decode the image
                                if ( must_decode || current_time >= real_timecode )
                                        ret = avcodec_decode_video( codec_context, av_frame, &got_picture, pkt.data, pkt.size );

                                if ( got_picture )
                                {
                                        // Handle ignore
                                        if ( ( int )( current_time * 100 ) < ( int )( real_timecode * 100 ) - 7 )
                                        {
                                                ignore = 0;
                                                got_picture = 0;
                                        }
                                        else if ( current_time >= real_timecode )
                                        {
                                                ignore = 0;
                                        }
                                        else if ( ignore -- )
                                        {
                                                got_picture = 0;
                                        }
                                }
                        }

                        // Now handle the picture if we have one
                        if ( got_picture )
                        {
                                mlt_properties_set_int( frame_properties, "progressive", !av_frame->interlaced_frame );
                                mlt_properties_set_int( frame_properties, "top_field_first", av_frame->top_field_first );

                                convert_image( av_frame, *buffer, codec_context->pix_fmt, *format, *width, *height );

                                mlt_properties_set_data( frame_properties, "image", *buffer, size, (mlt_destructor)mlt_pool_release, NULL );

                                if ( current_time == 0 && source_fps != 0 )
                                {
                                        double fps = mlt_properties_get_double( properties, "fps" );
                                        current_time = ceil( source_fps * ( double )position / fps ) * ( 1 / source_fps );
                                        mlt_properties_set_double( properties, "_current_time", current_time );
                                }
                                else
                                {
                                        mlt_properties_set_double( properties, "_current_time", current_time );
                                }
                        }

                        // We're finished with this packet regardless
                        av_free_packet( &pkt );
                }
        }
        
        // Set the field order property for this frame
        mlt_properties_set_int( frame_properties, "top_field_first", 
                mlt_properties_get_int( properties, "top_field_first" ) );

        // Regardless of speed, we expect to get the next frame (cos we ain't too bright)
        mlt_properties_set_position( properties, "_video_expected", position + 1 );

        return 0;
}

/** Set up video handling.
*/

static void producer_set_up_video( mlt_producer this, mlt_frame frame )
{
        // Get the properties
        mlt_properties properties = MLT_PRODUCER_PROPERTIES( this );

        // Fetch the video_context
        AVFormatContext *context = mlt_properties_get_data( properties, "video_context", NULL );

        // Get the video_index
        int index = mlt_properties_get_int( properties, "video_index" );

        // Get the frame properties
        mlt_properties frame_properties = MLT_FRAME_PROPERTIES( frame );

        if ( context != NULL && index != -1 )
        {
                // Get the video stream
                AVStream *stream = context->streams[ index ];

                // Get codec context
                AVCodecContext *codec_context = stream->codec;

                // Get the codec
                AVCodec *codec = mlt_properties_get_data( properties, "video_codec", NULL );

                // Initialise the codec if necessary
                if ( codec == NULL )
                {
                        // Find the codec
                        codec = avcodec_find_decoder( codec_context->codec_id );

                        // If we don't have a codec and we can't initialise it, we can't do much more...
                        if ( codec != NULL && avcodec_open( codec_context, codec ) >= 0 )
                        {
                                // Now store the codec with its destructor
                                mlt_properties_set_data( properties, "video_codec", codec_context, 0, producer_codec_close, NULL );
                        }
                        else
                        {
                                // Remember that we can't use this later
                                mlt_properties_set_int( properties, "video_index", -1 );
                        }
                }

                // No codec, no show...
                if ( codec != NULL )
                {
                        double source_fps = 0;
                        int norm_aspect_ratio = mlt_properties_get_int( properties, "norm_aspect_ratio" );
                        double force_aspect_ratio = mlt_properties_get_double( properties, "force_aspect_ratio" );
                        double aspect_ratio;

                        // XXX: We won't know the real aspect ratio until an image is decoded
                        // but we do need it now (to satisfy filter_resize) - take a guess based
                        // on pal/ntsc
                        if ( force_aspect_ratio > 0.0 )
                        {
                                aspect_ratio = force_aspect_ratio;
                        }
                        else if ( !norm_aspect_ratio && codec_context->sample_aspect_ratio.num > 0 )
                        {
                                aspect_ratio = av_q2d( codec_context->sample_aspect_ratio );
                        }
                        else
                        {
                                int is_pal = mlt_properties_get_double( properties, "fps" ) == 25.0;
                                aspect_ratio = is_pal ? 59.0/54.0 : 10.0/11.0;
                        }

                        // Determine the fps
                        source_fps = ( double )codec_context->time_base.den / ( codec_context->time_base.num == 0 ? 1 : codec_context->time_base.num );

                        // We'll use fps if it's available
                        if ( source_fps > 0 && source_fps < 30 )
                                mlt_properties_set_double( properties, "source_fps", source_fps );
                        mlt_properties_set_double( properties, "aspect_ratio", aspect_ratio );
                        
                        // Set the width and height
                        mlt_properties_set_int( frame_properties, "width", codec_context->width );
                        mlt_properties_set_int( frame_properties, "height", codec_context->height );
                        mlt_properties_set_double( frame_properties, "aspect_ratio", aspect_ratio );

                        mlt_frame_push_get_image( frame, producer_get_image );
                        mlt_properties_set_data( frame_properties, "avformat_producer", this, 0, NULL, NULL );
                }
                else
                {
                        mlt_properties_set_int( frame_properties, "test_image", 1 );
                }
        }
        else
        {
                mlt_properties_set_int( frame_properties, "test_image", 1 );
        }
}

/** Get the audio from a frame.
*/

static int producer_get_audio( mlt_frame frame, int16_t **buffer, mlt_audio_format *format, int *frequency, int *channels, int *samples )
{
        // Get the properties from the frame
        mlt_properties frame_properties = MLT_FRAME_PROPERTIES( frame );

        // Obtain the frame number of this frame
        mlt_position position = mlt_properties_get_position( frame_properties, "avformat_position" );

        // Get the producer 
        mlt_producer this = mlt_properties_get_data( frame_properties, "avformat_producer", NULL );

        // Get the producer properties
        mlt_properties properties = MLT_PRODUCER_PROPERTIES( this );

        // Fetch the audio_context
        AVFormatContext *context = mlt_properties_get_data( properties, "audio_context", NULL );

        // Get the audio_index
        int index = mlt_properties_get_int( properties, "audio_index" );

        // Get the seekable status
        int seekable = mlt_properties_get_int( properties, "seekable" );

        // Obtain the expected frame numer
        mlt_position expected = mlt_properties_get_position( properties, "_audio_expected" );

        // Obtain the resample context if it exists (not always needed)
        ReSampleContext *resample = mlt_properties_get_data( properties, "audio_resample", NULL );

        // Obtain the audio buffer
        int16_t *audio_buffer = mlt_properties_get_data( properties, "audio_buffer", NULL );

        // Get amount of audio used
        int audio_used =  mlt_properties_get_int( properties, "_audio_used" );

        // Calculate the real time code
        double real_timecode = producer_time_of_frame( this, position );

        // Get the audio stream
        AVStream *stream = context->streams[ index ];

        // Get codec context
        AVCodecContext *codec_context = stream->codec;

        // Packet
        AVPacket pkt;

        // Number of frames to ignore (for ffwd)
        int ignore = 0;

        // Flag for paused (silence) 
        int paused = 0;

        // Check for resample and create if necessary
        if ( resample == NULL && codec_context->channels <= 2 )
        {
                // Create the resampler
                resample = audio_resample_init( *channels, codec_context->channels, *frequency, codec_context->sample_rate );

                // And store it on properties
                mlt_properties_set_data( properties, "audio_resample", resample, 0, ( mlt_destructor )audio_resample_close, NULL );
        }
        else if ( resample == NULL )
        {
                *channels = codec_context->channels;
                *frequency = codec_context->sample_rate;
        }

        // Check for audio buffer and create if necessary
        if ( audio_buffer == NULL )
        {
                // Allocate the audio buffer
                audio_buffer = mlt_pool_alloc( AVCODEC_MAX_AUDIO_FRAME_SIZE * sizeof( int16_t ) );

                // And store it on properties for reuse
                mlt_properties_set_data( properties, "audio_buffer", audio_buffer, 0, ( mlt_destructor )mlt_pool_release, NULL );
        }

        // Seek if necessary
        if ( position != expected )
        {
                if ( position + 1 == expected )
                {
                        // We're paused - silence required
                        paused = 1;
                }
                else if ( !seekable && position > expected && ( position - expected ) < 250 )
                {
                        // Fast forward - seeking is inefficient for small distances - just ignore following frames
                        ignore = position - expected;
                }
                else if ( position < expected || position - expected >= 12 )
                {
                        // Set to the real timecode
                        if ( av_seek_frame( context, -1, mlt_properties_get_double( properties, "_start_time" ) + real_timecode * 1000000.0, AVSEEK_FLAG_BACKWARD ) != 0 )
                                paused = 1;

                        // Clear the usage in the audio buffer
                        audio_used = 0;
                }
        }

        // Get the audio if required
        if ( !paused )
        {
                int ret = 0;
                int got_audio = 0;
                int16_t *temp = mlt_pool_alloc( sizeof( int16_t ) * AVCODEC_MAX_AUDIO_FRAME_SIZE );

                av_init_packet( &pkt );

                while( ret >= 0 && !got_audio )
                {
                        // Check if the buffer already contains the samples required
                        if ( audio_used >= *samples && ignore == 0 )
                        {
                                got_audio = 1;
                                break;
                        }

                        // Read a packet
                        ret = av_read_frame( context, &pkt );

                int len = pkt.size;
                uint8_t *ptr = pkt.data;
                        int data_size;

                        // We only deal with audio from the selected audio_index
                        while ( ptr != NULL && ret >= 0 && pkt.stream_index == index && len > 0 )
                        {
                                // Decode the audio
                                ret = avcodec_decode_audio( codec_context, temp, &data_size, ptr, len );

                                if ( ret < 0 )
                                {
                                        ret = 0;
                                        break;
                                }

                                len -= ret;
                                ptr += ret;

                                if ( data_size > 0 )
                                {
                                        if ( resample != NULL )
                                        {
                                                audio_used += audio_resample( resample, &audio_buffer[ audio_used * *channels ], temp, data_size / ( codec_context->channels * sizeof( int16_t ) ) );
                                        }
                                        else
                                        {
                                                memcpy( &audio_buffer[ audio_used * *channels ], temp, data_size );
                                                audio_used += data_size / ( codec_context->channels * sizeof( int16_t ) );
                                        }

                                        // Handle ignore
                                        while ( ignore && audio_used > *samples )
                                        {
                                                ignore --;
                                                audio_used -= *samples;
                                                memmove( audio_buffer, &audio_buffer[ *samples * *channels ], audio_used * sizeof( int16_t ) );
                                        }
                                }

                                // If we're behind, ignore this packet
                                float current_pts = av_q2d( stream->time_base ) * pkt.pts;
                                if ( seekable && ( !ignore && current_pts <= ( real_timecode - 0.02 ) ) )
                                        ignore = 1;
                        }

                        // We're finished with this packet regardless
                        av_free_packet( &pkt );
                }

                *buffer = mlt_pool_alloc( *samples * *channels * sizeof( int16_t ) );
                mlt_properties_set_data( frame_properties, "audio", *buffer, 0, ( mlt_destructor )mlt_pool_release, NULL );

                // Now handle the audio if we have enough
                if ( audio_used >= *samples )
                {
                        memcpy( *buffer, audio_buffer, *samples * *channels * sizeof( int16_t ) );
                        audio_used -= *samples;
                        memmove( audio_buffer, &audio_buffer[ *samples * *channels ], audio_used * *channels * sizeof( int16_t ) );
                }
                else
                {
                        memset( *buffer, 0, *samples * *channels * sizeof( int16_t ) );
                }
                
                // Store the number of audio samples still available
                mlt_properties_set_int( properties, "_audio_used", audio_used );

                // Release the temporary audio
                mlt_pool_release( temp );
        }
        else
        {
                // Get silence and don't touch the context
                mlt_frame_get_audio( frame, buffer, format, frequency, channels, samples );
        }

        // Regardless of speed (other than paused), we expect to get the next frame
        if ( !paused )
                mlt_properties_set_position( properties, "_audio_expected", position + 1 );

        return 0;
}

/** Set up audio handling.
*/

static void producer_set_up_audio( mlt_producer this, mlt_frame frame )
{
        // Get the properties
        mlt_properties properties = MLT_PRODUCER_PROPERTIES( this );

        // Fetch the audio_context
        AVFormatContext *context = mlt_properties_get_data( properties, "audio_context", NULL );

        // Get the audio_index
        int index = mlt_properties_get_int( properties, "audio_index" );

        // Deal with audio context
        if ( context != NULL && index != -1 )
        {
                // Get the frame properties
                mlt_properties frame_properties = MLT_FRAME_PROPERTIES( frame );

                // Get the audio stream
                AVStream *stream = context->streams[ index ];

                // Get codec context
                AVCodecContext *codec_context = stream->codec;

                // Get the codec
                AVCodec *codec = mlt_properties_get_data( properties, "audio_codec", NULL );

                // Initialise the codec if necessary
                if ( codec == NULL )
                {
                        // Find the codec
                        codec = avcodec_find_decoder( codec_context->codec_id );

                        // If we don't have a codec and we can't initialise it, we can't do much more...
                        if ( codec != NULL && avcodec_open( codec_context, codec ) >= 0 )
                        {
                                // Now store the codec with its destructor
                                mlt_properties_set_data( properties, "audio_codec", codec_context, 0, producer_codec_close, NULL );

                        }
                        else
                        {
                                // Remember that we can't use this later
                                mlt_properties_set_int( properties, "audio_index", -1 );
                        }
                }

                // No codec, no show...
                if ( codec != NULL )
                {
                        mlt_frame_push_audio( frame, producer_get_audio );
                        mlt_properties_set_data( frame_properties, "avformat_producer", this, 0, NULL, NULL );
                        mlt_properties_set_int( frame_properties, "frequency", codec_context->sample_rate );
                        mlt_properties_set_int( frame_properties, "channels", codec_context->channels );
                }
        }
}

/** Our get frame implementation.
*/

static int producer_get_frame( mlt_producer this, mlt_frame_ptr frame, int index )
{
        // Create an empty frame
        *frame = mlt_frame_init( );

        // Update timecode on the frame we're creating
        mlt_frame_set_position( *frame, mlt_producer_position( this ) );

        // Set the position of this producer
        mlt_properties_set_position( MLT_FRAME_PROPERTIES( *frame ), "avformat_position", mlt_producer_frame( this ) );

        // Set up the video
        producer_set_up_video( this, *frame );

        // Set up the audio
        producer_set_up_audio( this, *frame );

        // Set the aspect_ratio
        mlt_properties_set_double( MLT_FRAME_PROPERTIES( *frame ), "aspect_ratio", mlt_properties_get_double( MLT_PRODUCER_PROPERTIES( this ), "aspect_ratio" ) );

        // Calculate the next timecode
        mlt_producer_prepare_next( this );

        return 0;
}