mirror of
https://github.com/mpv-player/mpv
synced 2024-12-10 08:56:04 +00:00
4eb1b755bb
The problem: once upon a time, windows idiots decided to try to store vorbis-in-ogg-in-avi. Of course this failed miserably, but they used the audio format tag 0xfffe for "extended" to do this. Later someone working on MPlayer somehow decided 0xfffe was the format for vorbis, which is nonsense, and now that's conflicting with real wav files with extended audio format. This patch changes demux_ogg (and mkv) to use sane fourcc's for vorbis and theora and gets rid of the 0xfffe nonsense so hopefully wav files with extended audio will work now. If there are problems, we'll have to find workarounds...and drive an 18-wheeler full of cola thru the house of whoever wrote this 0xfffe nonsense in MPlayer to begin with... git-svn-id: svn://svn.mplayerhq.hu/mplayer/trunk@14844 b3059339-0415-0410-9bf9-f77b7e298cf2
1524 lines
47 KiB
C
1524 lines
47 KiB
C
|
|
#include "config.h"
|
|
|
|
#ifdef HAVE_OGGVORBIS
|
|
|
|
#include <stdlib.h>
|
|
#include <stdio.h>
|
|
#include <string.h>
|
|
#include <assert.h>
|
|
#include <math.h>
|
|
|
|
#include "../mp_msg.h"
|
|
#include "../help_mp.h"
|
|
#include "stream.h"
|
|
#include "demuxer.h"
|
|
#include "stheader.h"
|
|
|
|
#define FOURCC_VORBIS mmioFOURCC('v', 'r', 'b', 's')
|
|
#define FOURCC_THEORA mmioFOURCC('t', 'h', 'e', 'o')
|
|
|
|
#ifdef TREMOR
|
|
#include <tremor/ogg.h>
|
|
#include <tremor/ivorbiscodec.h>
|
|
#else
|
|
#include <ogg/ogg.h>
|
|
#include <vorbis/codec.h>
|
|
#endif
|
|
|
|
#ifdef HAVE_OGGTHEORA
|
|
#include <theora/theora.h>
|
|
#endif
|
|
|
|
#define BLOCK_SIZE 4096
|
|
|
|
/// Vorbis decoder context : we need the vorbis_info for vorbis timestamping
|
|
/// Shall we put this struct def in a common header ?
|
|
typedef struct ov_struct_st {
|
|
vorbis_info vi; /* struct that stores all the static vorbis bitstream
|
|
settings */
|
|
vorbis_comment vc; /* struct that stores all the bitstream user comments */
|
|
vorbis_dsp_state vd; /* central working state for the packet->PCM decoder */
|
|
vorbis_block vb; /* local working space for packet->PCM decode */
|
|
float rg_scale; /* replaygain scale */
|
|
#ifdef TREMOR
|
|
int rg_scale_int;
|
|
#endif
|
|
} ov_struct_t;
|
|
|
|
/* Theora decoder context : we won't be able to interpret granule positions
|
|
* without using theora_granule_time with the theora_state of the stream.
|
|
* This is duplicated in `vd_theora.c'; put this in a common header?
|
|
*/
|
|
#ifdef HAVE_OGGTHEORA
|
|
typedef struct theora_struct_st {
|
|
theora_state st;
|
|
theora_comment cc;
|
|
theora_info inf;
|
|
} theora_struct_t;
|
|
#endif
|
|
|
|
//// OggDS headers
|
|
// Header for the new header format
|
|
typedef struct stream_header_video
|
|
{
|
|
ogg_int32_t width;
|
|
ogg_int32_t height;
|
|
} stream_header_video;
|
|
|
|
typedef struct stream_header_audio
|
|
{
|
|
ogg_int16_t channels;
|
|
ogg_int16_t blockalign;
|
|
ogg_int32_t avgbytespersec;
|
|
} stream_header_audio;
|
|
|
|
typedef struct __attribute__((__packed__)) stream_header
|
|
{
|
|
char streamtype[8];
|
|
char subtype[4];
|
|
|
|
ogg_int32_t size; // size of the structure
|
|
|
|
ogg_int64_t time_unit; // in reference time
|
|
ogg_int64_t samples_per_unit;
|
|
ogg_int32_t default_len; // in media time
|
|
|
|
ogg_int32_t buffersize;
|
|
ogg_int16_t bits_per_sample;
|
|
|
|
ogg_int16_t padding;
|
|
|
|
union
|
|
{
|
|
// Video specific
|
|
stream_header_video video;
|
|
// Audio specific
|
|
stream_header_audio audio;
|
|
} sh;
|
|
} stream_header;
|
|
|
|
/// Our private datas
|
|
|
|
typedef struct ogg_syncpoint {
|
|
int64_t granulepos;
|
|
off_t page_pos;
|
|
} ogg_syncpoint_t;
|
|
|
|
/// A logical stream
|
|
typedef struct ogg_stream {
|
|
/// Timestamping stuff
|
|
float samplerate; /// granulpos 2 time
|
|
int64_t lastpos;
|
|
int32_t lastsize;
|
|
|
|
// Logical stream state
|
|
ogg_stream_state stream;
|
|
int hdr_packets;
|
|
int vorbis;
|
|
int theora;
|
|
int flac;
|
|
int text;
|
|
int id;
|
|
} ogg_stream_t;
|
|
|
|
typedef struct ogg_demuxer {
|
|
/// Physical stream state
|
|
ogg_sync_state sync;
|
|
/// Current page
|
|
ogg_page page;
|
|
/// Logical streams
|
|
ogg_stream_t *subs;
|
|
int num_sub;
|
|
ogg_syncpoint_t* syncpoints;
|
|
int num_syncpoint;
|
|
off_t pos, last_size;
|
|
int64_t final_granulepos;
|
|
|
|
/* Used for subtitle switching. */
|
|
int n_text;
|
|
int *text_ids;
|
|
char **text_langs;
|
|
} ogg_demuxer_t;
|
|
|
|
#define NUM_VORBIS_HDR_PACKETS 3
|
|
|
|
/// Some defines from OggDS
|
|
#define PACKET_TYPE_HEADER 0x01
|
|
#define PACKET_TYPE_BITS 0x07
|
|
#define PACKET_LEN_BITS01 0xc0
|
|
#define PACKET_LEN_BITS2 0x02
|
|
#define PACKET_IS_SYNCPOINT 0x08
|
|
|
|
extern int index_mode;
|
|
|
|
extern char *dvdsub_lang, *audio_lang;
|
|
extern int dvdsub_id;
|
|
extern int demux_aid_vid_mismatch;
|
|
|
|
//-------- subtitle support - should be moved to decoder layer, and queue
|
|
// - subtitles up in demuxer buffer...
|
|
|
|
#include "../subreader.h"
|
|
#include "../libvo/sub.h"
|
|
#define OGG_SUB_MAX_LINE 128
|
|
|
|
static subtitle ogg_sub;
|
|
extern subtitle* vo_sub;
|
|
static float clear_sub;
|
|
//FILE* subout;
|
|
|
|
static
|
|
uint16_t get_uint16 (const void *buf)
|
|
{
|
|
uint16_t ret;
|
|
unsigned char *tmp;
|
|
|
|
tmp = (unsigned char *) buf;
|
|
|
|
ret = tmp[1] & 0xff;
|
|
ret = (ret << 8) + (tmp[0] & 0xff);
|
|
|
|
return (ret);
|
|
}
|
|
|
|
static
|
|
uint32_t get_uint32 (const void *buf)
|
|
{
|
|
uint32_t ret;
|
|
unsigned char *tmp;
|
|
|
|
tmp = (unsigned char *) buf;
|
|
|
|
ret = tmp[3] & 0xff;
|
|
ret = (ret << 8) + (tmp[2] & 0xff);
|
|
ret = (ret << 8) + (tmp[1] & 0xff);
|
|
ret = (ret << 8) + (tmp[0] & 0xff);
|
|
|
|
return (ret);
|
|
}
|
|
|
|
static
|
|
uint64_t get_uint64 (const void *buf)
|
|
{
|
|
uint64_t ret;
|
|
unsigned char *tmp;
|
|
|
|
tmp = (unsigned char *) buf;
|
|
|
|
ret = tmp[7] & 0xff;
|
|
ret = (ret << 8) + (tmp[6] & 0xff);
|
|
ret = (ret << 8) + (tmp[5] & 0xff);
|
|
ret = (ret << 8) + (tmp[4] & 0xff);
|
|
ret = (ret << 8) + (tmp[3] & 0xff);
|
|
ret = (ret << 8) + (tmp[2] & 0xff);
|
|
ret = (ret << 8) + (tmp[1] & 0xff);
|
|
ret = (ret << 8) + (tmp[0] & 0xff);
|
|
|
|
return (ret);
|
|
}
|
|
|
|
void demux_ogg_init_sub () {
|
|
int lcv;
|
|
if(!ogg_sub.text[0]) // not yet allocated
|
|
for (lcv = 0; lcv < SUB_MAX_TEXT; lcv++) {
|
|
ogg_sub.text[lcv] = (char*)malloc(OGG_SUB_MAX_LINE);
|
|
}
|
|
}
|
|
|
|
void demux_ogg_add_sub (ogg_stream_t* os,ogg_packet* pack) {
|
|
int lcv;
|
|
int line_pos = 0;
|
|
int ignoring = 0;
|
|
char *packet = pack->packet;
|
|
|
|
mp_msg(MSGT_DEMUX,MSGL_DBG2,"\ndemux_ogg_add_sub %02X %02X %02X '%s'\n",
|
|
(unsigned char)packet[0],
|
|
(unsigned char)packet[1],
|
|
(unsigned char)packet[2],
|
|
&packet[3]);
|
|
|
|
ogg_sub.lines = 0;
|
|
if (((unsigned char)packet[0]) == 0x88) { // some subtitle text
|
|
// Find data start
|
|
int32_t duration = 0;
|
|
int16_t hdrlen = (*packet & PACKET_LEN_BITS01)>>6, i;
|
|
hdrlen |= (*packet & PACKET_LEN_BITS2) <<1;
|
|
lcv = 1 + hdrlen;
|
|
for (i = hdrlen; i > 0; i--) {
|
|
duration <<= 8;
|
|
duration |= (unsigned char)packet[i];
|
|
}
|
|
if ((hdrlen > 0) && (duration > 0)) {
|
|
float pts;
|
|
if(pack->granulepos == -1)
|
|
pack->granulepos = os->lastpos + os->lastsize;
|
|
pts = (float)pack->granulepos/(float)os->samplerate;
|
|
clear_sub = 1.0 + pts + (float)duration/1000.0;
|
|
}
|
|
while (1) {
|
|
int c = packet[lcv++];
|
|
if(c=='\n' || c==0 || line_pos >= OGG_SUB_MAX_LINE-1){
|
|
ogg_sub.text[ogg_sub.lines][line_pos] = 0; // close sub
|
|
if(line_pos) ogg_sub.lines++;
|
|
if(!c || ogg_sub.lines>=SUB_MAX_TEXT) break; // EOL or TooMany
|
|
line_pos = 0;
|
|
}
|
|
switch (c) {
|
|
case '\r':
|
|
case '\n': // just ignore linefeeds for now
|
|
// their placement seems rather haphazard
|
|
break;
|
|
case '<': // some html markup, ignore for now
|
|
ignoring = 1;
|
|
break;
|
|
case '>':
|
|
ignoring = 0;
|
|
break;
|
|
default:
|
|
if(!ignoring)
|
|
ogg_sub.text[ogg_sub.lines][line_pos++] = c;
|
|
break;
|
|
}
|
|
}
|
|
}
|
|
|
|
mp_msg(MSGT_DEMUX,MSGL_DBG2,"Ogg sub lines: %d first: '%s'\n",
|
|
ogg_sub.lines, ogg_sub.text[0]);
|
|
#ifdef USE_ICONV
|
|
subcp_recode1(&ogg_sub);
|
|
#endif
|
|
vo_sub = &ogg_sub;
|
|
vo_osd_changed(OSDTYPE_SUBTITLE);
|
|
}
|
|
|
|
|
|
// get the logical stream of the current page
|
|
// fill os if non NULL and return the stream id
|
|
static int demux_ogg_get_page_stream(ogg_demuxer_t* ogg_d,ogg_stream_state** os) {
|
|
int id,s_no;
|
|
ogg_page* page = &ogg_d->page;
|
|
|
|
s_no = ogg_page_serialno(page);
|
|
|
|
for(id= 0; id < ogg_d->num_sub ; id++) {
|
|
if(s_no == ogg_d->subs[id].stream.serialno)
|
|
break;
|
|
}
|
|
|
|
if(id == ogg_d->num_sub) {
|
|
// If we have only one vorbis stream allow the stream id to change
|
|
// it's normal on radio stream (each song have an different id).
|
|
// But we (or the codec?) should check that the samplerate, etc
|
|
// doesn't change (for radio stream it's ok)
|
|
if(ogg_d->num_sub == 1 && ogg_d->subs[0].vorbis) {
|
|
ogg_stream_reset(&ogg_d->subs[0].stream);
|
|
ogg_stream_init(&ogg_d->subs[0].stream,s_no);
|
|
id = 0;
|
|
} else
|
|
return -1;
|
|
}
|
|
|
|
if(os)
|
|
*os = &ogg_d->subs[id].stream;
|
|
|
|
return id;
|
|
|
|
}
|
|
|
|
static unsigned char* demux_ogg_read_packet(ogg_stream_t* os,ogg_packet* pack,void *context,float* pts,int* flags, int samplesize) {
|
|
unsigned char* data;
|
|
|
|
*pts = 0;
|
|
*flags = 0;
|
|
|
|
if(os->vorbis) {
|
|
data = pack->packet;
|
|
if(*pack->packet & PACKET_TYPE_HEADER)
|
|
os->hdr_packets++;
|
|
else if (context )
|
|
{
|
|
vorbis_info *vi = &((ov_struct_t*)context)->vi;
|
|
|
|
// When we dump the audio, there is no vi, but we don't care of timestamp in this case
|
|
int32_t blocksize = vorbis_packet_blocksize(vi,pack) / samplesize;
|
|
// Calculate the timestamp if the packet don't have any
|
|
if(pack->granulepos == -1) {
|
|
pack->granulepos = os->lastpos;
|
|
if(os->lastsize > 0)
|
|
pack->granulepos += os->lastsize;
|
|
}
|
|
*pts = pack->granulepos / (float)vi->rate;
|
|
os->lastsize = blocksize;
|
|
os->lastpos = pack->granulepos;
|
|
}
|
|
# ifdef HAVE_OGGTHEORA
|
|
} else if (os->theora) {
|
|
/* we pass complete packets to theora, mustn't strip the header! */
|
|
data = pack->packet;
|
|
os->lastsize = 1;
|
|
|
|
/* header packets beginn on 1-bit: thus check (*data&0x80). We don't
|
|
have theora_state st, until all header packets were passed to the
|
|
decoder. */
|
|
if (context != NULL && !(*data&0x80))
|
|
{
|
|
theora_info *thi = ((theora_struct_t*)context)->st.i;
|
|
int keyframe_granule_shift=_ilog(thi->keyframe_frequency_force-1);
|
|
int64_t iframemask = (1 << keyframe_granule_shift) - 1;
|
|
|
|
if (pack->granulepos >= 0)
|
|
{
|
|
os->lastpos = pack->granulepos >> keyframe_granule_shift;
|
|
os->lastpos += pack->granulepos & iframemask;
|
|
*flags = ((pack->granulepos & iframemask) == 0);
|
|
}
|
|
else
|
|
{
|
|
os->lastpos++;
|
|
}
|
|
pack->granulepos = os->lastpos;
|
|
*pts = (double)os->lastpos / (double)os->samplerate;
|
|
}
|
|
#endif /* HAVE_OGGTHEORA */
|
|
# ifdef HAVE_FLAC
|
|
} else if (os->flac) {
|
|
/* we pass complete packets to flac, mustn't strip the header! */
|
|
data = pack->packet;
|
|
#endif /* HAVE_FLAC */
|
|
} else {
|
|
if(*pack->packet & PACKET_TYPE_HEADER)
|
|
os->hdr_packets++;
|
|
else {
|
|
// Find data start
|
|
int16_t hdrlen = (*pack->packet & PACKET_LEN_BITS01)>>6;
|
|
hdrlen |= (*pack->packet & PACKET_LEN_BITS2) <<1;
|
|
data = pack->packet + 1 + hdrlen;
|
|
// Calculate the timestamp
|
|
if(pack->granulepos == -1)
|
|
pack->granulepos = os->lastpos + (os->lastsize ? os->lastsize : 1);
|
|
// If we alredy have a timestamp it can be a syncpoint
|
|
if(*pack->packet & PACKET_IS_SYNCPOINT)
|
|
*flags = 1;
|
|
*pts = pack->granulepos/os->samplerate;
|
|
// Save the packet length and timestamp
|
|
os->lastsize = 0;
|
|
while(hdrlen) {
|
|
os->lastsize <<= 8;
|
|
os->lastsize |= pack->packet[hdrlen];
|
|
hdrlen--;
|
|
}
|
|
os->lastpos = pack->granulepos;
|
|
}
|
|
}
|
|
return data;
|
|
}
|
|
|
|
// check if clang has substring from comma separated langlist
|
|
static int demux_ogg_check_lang(char *clang, char *langlist)
|
|
{
|
|
char *c;
|
|
|
|
if (!langlist || !*langlist)
|
|
return 0;
|
|
while ((c = strchr(langlist, ',')))
|
|
{
|
|
if (!strncasecmp(clang, langlist, c - langlist))
|
|
return 1;
|
|
langlist = &c[1];
|
|
}
|
|
if (!strncasecmp(clang, langlist, strlen(langlist)))
|
|
return 1;
|
|
return 0;
|
|
}
|
|
|
|
static int demux_ogg_sub_reverse_id(demuxer_t *demuxer, int id);
|
|
|
|
/// Try to print out comments and also check for LANGUAGE= tag
|
|
static void demux_ogg_check_comments(demuxer_t *d, ogg_stream_t *os, int id, vorbis_comment *vc)
|
|
{
|
|
char *hdr, *val;
|
|
char **cmt = vc->user_comments;
|
|
int index;
|
|
ogg_demuxer_t *ogg_d = (ogg_demuxer_t *)d->priv;
|
|
|
|
while(*cmt)
|
|
{
|
|
hdr = NULL;
|
|
if (!strncasecmp(*cmt, "ENCODED_USING=", 14))
|
|
{
|
|
hdr = "Software";
|
|
val = *cmt + 14;
|
|
}
|
|
else if (!strncasecmp(*cmt, "LANGUAGE=", 9))
|
|
{
|
|
val = *cmt + 9;
|
|
if (identify)
|
|
{
|
|
if (ogg_d->subs[id].text)
|
|
mp_msg(MSGT_GLOBAL, MSGL_INFO, "ID_SID_%d_LANG=%s\n", ogg_d->subs[id].id, val);
|
|
else if (id != d->video->id)
|
|
mp_msg(MSGT_GLOBAL, MSGL_INFO, "ID_AID_%d_LANG=%s\n", ogg_d->subs[id].id, val);
|
|
}
|
|
if (ogg_d->subs[id].text)
|
|
mp_msg(MSGT_DEMUX, MSGL_INFO, "[Ogg] Language for -sid %d is '-slang \"%s\"'\n", ogg_d->subs[id].id, val);
|
|
// copy this language name into the array
|
|
index = demux_ogg_sub_reverse_id(d, id);
|
|
if (index >= 0) {
|
|
// in case of malicious files with more than one lang per track:
|
|
if (ogg_d->text_langs[index]) free(ogg_d->text_langs[index]);
|
|
ogg_d->text_langs[index] = strdup(val);
|
|
}
|
|
// check for -slang if subs are uninitialized yet
|
|
if (os->text && d->sub->id == -1 && demux_ogg_check_lang(val, dvdsub_lang))
|
|
{
|
|
d->sub->id = id;
|
|
dvdsub_id = index;
|
|
mp_msg(MSGT_DEMUX, MSGL_V, "Ogg demuxer: Displaying subtitle stream id %d which matched -slang %s\n", id, val);
|
|
}
|
|
else
|
|
hdr = "Language";
|
|
}
|
|
else if (!strncasecmp(*cmt, "ENCODER_URL=", 12))
|
|
{
|
|
hdr = "Encoder URL";
|
|
val = *cmt + 12;
|
|
}
|
|
else if (!strncasecmp(*cmt, "TITLE=", 6))
|
|
{
|
|
hdr = "Name";
|
|
val = *cmt + 6;
|
|
}
|
|
if (hdr)
|
|
mp_msg(MSGT_DEMUX, MSGL_V, " %s: %s\n", hdr, val);
|
|
cmt++;
|
|
}
|
|
}
|
|
|
|
/// Calculate the timestamp and add the packet to the demux stream
|
|
// return 1 if the packet was added, 0 otherwise
|
|
static int demux_ogg_add_packet(demux_stream_t* ds,ogg_stream_t* os,int id,ogg_packet* pack) {
|
|
demuxer_t* d = ds->demuxer;
|
|
demux_packet_t* dp;
|
|
unsigned char* data;
|
|
float pts = 0;
|
|
int flags = 0;
|
|
void *context = NULL;
|
|
int samplesize = 1;
|
|
|
|
// If packet is an comment header then we try to get comments at first
|
|
if (pack->bytes >= 7 && !memcmp(pack->packet, "\003vorbis", 7))
|
|
{
|
|
vorbis_info vi;
|
|
vorbis_comment vc;
|
|
|
|
vorbis_info_init(&vi);
|
|
vorbis_comment_init(&vc);
|
|
vi.rate = 1L; // it's checked by vorbis_synthesis_headerin()
|
|
if(vorbis_synthesis_headerin(&vi, &vc, pack) == 0) // if no errors
|
|
demux_ogg_check_comments(d, os, id, &vc);
|
|
vorbis_comment_clear(&vc);
|
|
vorbis_info_clear(&vi);
|
|
}
|
|
if (os->text) {
|
|
if (id == d->sub->id) // don't want to add subtitles to the demuxer for now
|
|
demux_ogg_add_sub(os,pack);
|
|
return 0;
|
|
}
|
|
// If packet is an header we jump it except for vorbis and theora
|
|
// (PACKET_TYPE_HEADER bit doesn't even exist for theora ?!)
|
|
// We jump nothing for FLAC. Ain't this great? Packet contents have to be
|
|
// handled differently for each and every stream type. The joy! The joy!
|
|
if(!os->flac && ((*pack->packet & PACKET_TYPE_HEADER) &&
|
|
(ds != d->audio || ( ((sh_audio_t*)ds->sh)->format != FOURCC_VORBIS || os->hdr_packets >= NUM_VORBIS_HDR_PACKETS ) ) &&
|
|
(ds != d->video || (((sh_video_t*)ds->sh)->format != FOURCC_THEORA))))
|
|
return 0;
|
|
|
|
// For vorbis packet the packet is the data, for other codec we must jump
|
|
// the header
|
|
if(ds == d->audio && ((sh_audio_t*)ds->sh)->format == FOURCC_VORBIS) {
|
|
context = ((sh_audio_t *)ds->sh)->context;
|
|
samplesize = ((sh_audio_t *)ds->sh)->samplesize;
|
|
}
|
|
if (ds == d->video && ((sh_audio_t*)ds->sh)->format == FOURCC_THEORA)
|
|
context = ((sh_video_t *)ds->sh)->context;
|
|
data = demux_ogg_read_packet(os,pack,context,&pts,&flags,samplesize);
|
|
if(d->video->id < 0)
|
|
((sh_audio_t*)ds->sh)->delay = pts;
|
|
|
|
/// Clear subtitles if necessary (for broken files)
|
|
if ((clear_sub > 0) && (pts >= clear_sub)) {
|
|
ogg_sub.lines = 0;
|
|
vo_sub = &ogg_sub;
|
|
vo_osd_changed(OSDTYPE_SUBTITLE);
|
|
clear_sub = -1;
|
|
}
|
|
/// Send the packet
|
|
dp = new_demux_packet(pack->bytes-(data-pack->packet));
|
|
memcpy(dp->buffer,data,pack->bytes-(data-pack->packet));
|
|
dp->pts = pts;
|
|
dp->flags = flags;
|
|
ds_add_packet(ds,dp);
|
|
mp_msg(MSGT_DEMUX,MSGL_DBG2,"New dp: %p ds=%p pts=%5.3f len=%d flag=%d \n",
|
|
dp, ds, pts, dp->len, flags);
|
|
return 1;
|
|
}
|
|
|
|
/// if -forceidx build a table of all syncpoints to make seeking easier
|
|
/// otherwise try to get at least the final_granulepos
|
|
void demux_ogg_scan_stream(demuxer_t* demuxer) {
|
|
ogg_demuxer_t* ogg_d = demuxer->priv;
|
|
stream_t *s = demuxer->stream;
|
|
ogg_sync_state* sync = &ogg_d->sync;
|
|
ogg_page* page= &ogg_d->page;
|
|
ogg_stream_state* oss;
|
|
ogg_stream_t* os;
|
|
ogg_packet op;
|
|
int np,sid,p,samplesize=1;
|
|
void *context = NULL;
|
|
off_t pos, last_pos;
|
|
pos = last_pos = demuxer->movi_start;
|
|
|
|
// Reset the stream
|
|
if(index_mode == 2) {
|
|
stream_seek(s,demuxer->movi_start);
|
|
} else {
|
|
//the 270000 are just a wild guess
|
|
stream_seek(s,max(ogg_d->pos,demuxer->movi_end-270000));
|
|
}
|
|
ogg_sync_reset(sync);
|
|
|
|
// Get the serial number of the stream we use
|
|
if(demuxer->video->id >= 0) {
|
|
sid = demuxer->video->id;
|
|
/* demux_ogg_read_packet needs decoder context for Theora streams */
|
|
if (((sh_video_t*)demuxer->video->sh)->format == FOURCC_THEORA)
|
|
context = ((sh_video_t*)demuxer->video->sh)->context;
|
|
}
|
|
else {
|
|
sid = demuxer->audio->id;
|
|
/* demux_ogg_read_packet needs decoder context for Vorbis streams */
|
|
if(((sh_audio_t*)demuxer->audio->sh)->format == FOURCC_VORBIS) {
|
|
context = ((sh_audio_t*)demuxer->audio->sh)->context;
|
|
samplesize = ((sh_audio_t*)demuxer->audio->sh)->samplesize;
|
|
}
|
|
}
|
|
os = &ogg_d->subs[sid];
|
|
oss = &os->stream;
|
|
|
|
while(1) {
|
|
np = ogg_sync_pageseek(sync,page);
|
|
if(np < 0) { // We had to skip some bytes
|
|
if(index_mode == 2) mp_msg(MSGT_DEMUX,MSGL_ERR,"Bad page sync while building syncpoints table (%d)\n",-np);
|
|
pos += -np;
|
|
continue;
|
|
}
|
|
if(np <= 0) { // We need more data
|
|
char* buf = ogg_sync_buffer(sync,BLOCK_SIZE);
|
|
int len = stream_read(s,buf,BLOCK_SIZE);
|
|
if(len == 0 && s->eof)
|
|
break;
|
|
ogg_sync_wrote(sync,len);
|
|
continue;
|
|
}
|
|
// The page is ready
|
|
//ogg_sync_pageout(sync,page);
|
|
if(ogg_page_serialno(page) != os->stream.serialno) { // It isn't a page from the stream we want
|
|
pos += np;
|
|
continue;
|
|
}
|
|
if(ogg_stream_pagein(oss,page) != 0) {
|
|
mp_msg(MSGT_DEMUX,MSGL_ERR,"Pagein error ????\n");
|
|
pos += np;
|
|
continue;
|
|
}
|
|
p = 0;
|
|
while(ogg_stream_packetout(oss,&op) == 1) {
|
|
float pts;
|
|
int flags;
|
|
demux_ogg_read_packet(os,&op,context,&pts,&flags,samplesize);
|
|
if(op.granulepos >= 0) ogg_d->final_granulepos = op.granulepos;
|
|
if(index_mode == 2 && (flags || (os->vorbis && op.granulepos >= 0))) {
|
|
ogg_d->syncpoints = (ogg_syncpoint_t*)realloc(ogg_d->syncpoints,(ogg_d->num_syncpoint+1)*sizeof(ogg_syncpoint_t));
|
|
ogg_d->syncpoints[ogg_d->num_syncpoint].granulepos = op.granulepos;
|
|
ogg_d->syncpoints[ogg_d->num_syncpoint].page_pos = (ogg_page_continued(page) && p == 0) ? last_pos : pos;
|
|
ogg_d->num_syncpoint++;
|
|
}
|
|
p++;
|
|
}
|
|
if(p > 1 || (p == 1 && ! ogg_page_continued(page)))
|
|
last_pos = pos;
|
|
pos += np;
|
|
if(index_mode == 2) mp_msg(MSGT_DEMUX,MSGL_INFO,"Building syncpoint table %d%%\r",(int)(pos*100/s->end_pos));
|
|
}
|
|
if(index_mode == 2) mp_msg(MSGT_DEMUX,MSGL_INFO,"\n");
|
|
|
|
if(index_mode == 2) mp_msg(MSGT_DEMUX,MSGL_V,"Ogg syncpoints table builed: %d syncpoints\n",ogg_d->num_syncpoint);
|
|
mp_msg(MSGT_DEMUX,MSGL_V,"Ogg stream length (granulepos): %lld\n",ogg_d->final_granulepos);
|
|
|
|
stream_reset(s);
|
|
stream_seek(s,demuxer->movi_start);
|
|
ogg_sync_reset(sync);
|
|
for(np = 0 ; np < ogg_d->num_sub ; np++) {
|
|
ogg_stream_reset(&ogg_d->subs[np].stream);
|
|
ogg_d->subs[np].lastpos = ogg_d->subs[np].lastsize = ogg_d->subs[np].hdr_packets = 0;
|
|
}
|
|
|
|
|
|
// Get the first page
|
|
while(1) {
|
|
np = ogg_sync_pageout(sync,page);
|
|
if(np <= 0) { // We need more data
|
|
char* buf = ogg_sync_buffer(sync,BLOCK_SIZE);
|
|
int len = stream_read(s,buf,BLOCK_SIZE);
|
|
if(len == 0 && s->eof) {
|
|
mp_msg(MSGT_DEMUX,MSGL_ERR,"EOF while trying to get the first page !!!!\n");
|
|
break;
|
|
}
|
|
|
|
ogg_sync_wrote(sync,len);
|
|
continue;
|
|
}
|
|
demux_ogg_get_page_stream(ogg_d,&oss);
|
|
ogg_stream_pagein(oss,page);
|
|
break;
|
|
}
|
|
|
|
}
|
|
|
|
extern void print_wave_header(WAVEFORMATEX *h);
|
|
extern void print_video_header(BITMAPINFOHEADER *h);
|
|
|
|
/** \brief Return the number of subtitle tracks in the file.
|
|
|
|
\param demuxer The demuxer for which the number of subtitle tracks
|
|
should be returned.
|
|
*/
|
|
int demux_ogg_num_subs(demuxer_t *demuxer) {
|
|
ogg_demuxer_t *ogg_d = (ogg_demuxer_t *)demuxer->priv;
|
|
return ogg_d->n_text;
|
|
}
|
|
|
|
/** \brief Change the current subtitle stream and return its ID.
|
|
|
|
\param demuxer The demuxer whose subtitle stream will be changed.
|
|
\param new_num The number of the new subtitle track. The number must be
|
|
between 0 and ogg_d->n_text - 1.
|
|
|
|
\returns The Ogg stream number ( = page serial number) of the newly selected
|
|
track.
|
|
*/
|
|
int demux_ogg_sub_id(demuxer_t *demuxer, int index) {
|
|
ogg_demuxer_t *ogg_d = (ogg_demuxer_t *)demuxer->priv;
|
|
return (index < 0) ? index : (index >= ogg_d->n_text) ? -1 : ogg_d->text_ids[index];
|
|
}
|
|
|
|
/** \brief Translate the ogg track number into the subtitle number.
|
|
* \param demuxer The demuxer about whose subtitles we are inquiring.
|
|
* \param id The ogg track number of the subtitle track.
|
|
*/
|
|
static int demux_ogg_sub_reverse_id(demuxer_t *demuxer, int id) {
|
|
ogg_demuxer_t *ogg_d = (ogg_demuxer_t *)demuxer->priv;
|
|
int i;
|
|
for (i = 0; i < ogg_d->n_text; i++)
|
|
if (ogg_d->text_ids[i] == id) return i;
|
|
return -1;
|
|
}
|
|
|
|
/** \brief Lookup the subtitle language by the subtitle number. Returns NULL on out-of-bounds input.
|
|
* \param demuxer The demuxer about whose subtitles we are inquiring.
|
|
* \param index The subtitle number.
|
|
*/
|
|
char *demux_ogg_sub_lang(demuxer_t *demuxer, int index) {
|
|
ogg_demuxer_t *ogg_d = (ogg_demuxer_t *)demuxer->priv;
|
|
return (index < 0) ? NULL : (index >= ogg_d->n_text) ? NULL : ogg_d->text_langs[index];
|
|
}
|
|
|
|
void demux_close_ogg(demuxer_t* demuxer);
|
|
|
|
/// Open an ogg physical stream
|
|
int demux_ogg_open(demuxer_t* demuxer) {
|
|
ogg_demuxer_t* ogg_d;
|
|
stream_t *s;
|
|
char* buf;
|
|
int np,s_no, n_audio = 0, n_video = 0;
|
|
int audio_id = -1, video_id = -1, text_id = -1;
|
|
ogg_sync_state* sync;
|
|
ogg_page* page;
|
|
ogg_packet pack;
|
|
sh_audio_t* sh_a;
|
|
sh_video_t* sh_v;
|
|
|
|
#ifdef USE_ICONV
|
|
subcp_open(NULL);
|
|
#endif
|
|
|
|
clear_sub = -1;
|
|
s = demuxer->stream;
|
|
|
|
ogg_d = (ogg_demuxer_t*)calloc(1,sizeof(ogg_demuxer_t));
|
|
sync = &ogg_d->sync;
|
|
page = &ogg_d->page;
|
|
|
|
ogg_sync_init(sync);
|
|
|
|
while(1) {
|
|
/// Try to get a page
|
|
ogg_d->pos += ogg_d->last_size;
|
|
np = ogg_sync_pageseek(sync,page);
|
|
/// Error
|
|
if(np < 0) {
|
|
mp_msg(MSGT_DEMUX,MSGL_DBG2,"Ogg demuxer : Bad page sync\n");
|
|
goto err_out;
|
|
}
|
|
/// Need some more data
|
|
if(np == 0) {
|
|
int len;
|
|
buf = ogg_sync_buffer(sync,BLOCK_SIZE);
|
|
len = stream_read(s,buf,BLOCK_SIZE);
|
|
if(len == 0 && s->eof) {
|
|
goto err_out;
|
|
}
|
|
ogg_sync_wrote(sync,len);
|
|
continue;
|
|
}
|
|
ogg_d->last_size = np;
|
|
// We got one page now
|
|
|
|
if( ! ogg_page_bos(page) ) { // It's not a begining page
|
|
// Header parsing end here, we need to get the page otherwise it will be lost
|
|
int id = demux_ogg_get_page_stream(ogg_d,NULL);
|
|
if(id >= 0)
|
|
ogg_stream_pagein(&ogg_d->subs[id].stream,page);
|
|
else
|
|
mp_msg(MSGT_DEMUX,MSGL_ERR,"Ogg : Warning found none bos page from unknown stream %d\n",ogg_page_serialno(page));
|
|
break;
|
|
}
|
|
|
|
/// Init the data structure needed for a logical stream
|
|
ogg_d->subs = (ogg_stream_t*)realloc(ogg_d->subs,(ogg_d->num_sub+1)*sizeof(ogg_stream_t));
|
|
memset(&ogg_d->subs[ogg_d->num_sub],0,sizeof(ogg_stream_t));
|
|
/// Get the stream serial number
|
|
s_no = ogg_page_serialno(page);
|
|
ogg_stream_init(&ogg_d->subs[ogg_d->num_sub].stream,s_no);
|
|
mp_msg(MSGT_DEMUX,MSGL_DBG2,"Ogg : Found a stream with serial=%d\n",s_no);
|
|
// Take the first page
|
|
ogg_stream_pagein(&ogg_d->subs[ogg_d->num_sub].stream,page);
|
|
// Get first packet of the page
|
|
ogg_stream_packetout(&ogg_d->subs[ogg_d->num_sub].stream,&pack);
|
|
|
|
// Reset our vars
|
|
sh_a = NULL;
|
|
sh_v = NULL;
|
|
|
|
demux_aid_vid_mismatch = 1; // don't identify in new_sh_* since ids don't match
|
|
|
|
// Check for Vorbis
|
|
if(pack.bytes >= 7 && ! strncmp(&pack.packet[1],"vorbis", 6) ) {
|
|
sh_a = new_sh_audio(demuxer,ogg_d->num_sub);
|
|
sh_a->format = FOURCC_VORBIS;
|
|
ogg_d->subs[ogg_d->num_sub].vorbis = 1;
|
|
if (identify)
|
|
mp_msg(MSGT_GLOBAL, MSGL_INFO, "ID_AUDIO_ID=%d\n", n_audio);
|
|
ogg_d->subs[ogg_d->num_sub].id = n_audio;
|
|
n_audio++;
|
|
mp_msg(MSGT_DEMUX,MSGL_INFO,"[Ogg] stream %d: audio (Vorbis), -aid %d\n",ogg_d->num_sub,n_audio-1);
|
|
|
|
// check for Theora
|
|
# ifdef HAVE_OGGTHEORA
|
|
} else if (pack.bytes >= 7 && !strncmp (&pack.packet[1], "theora", 6)) {
|
|
int errorCode = 0;
|
|
theora_info inf;
|
|
theora_comment cc;
|
|
|
|
theora_info_init (&inf);
|
|
theora_comment_init (&cc);
|
|
|
|
errorCode = theora_decode_header (&inf, &cc, &pack);
|
|
if (errorCode)
|
|
mp_msg(MSGT_DEMUX,MSGL_ERR,"Theora header parsing failed: %i \n",
|
|
errorCode);
|
|
else
|
|
{
|
|
sh_v = new_sh_video(demuxer,ogg_d->num_sub);
|
|
|
|
sh_v->context = NULL;
|
|
sh_v->bih = (BITMAPINFOHEADER*)calloc(1,sizeof(BITMAPINFOHEADER));
|
|
sh_v->bih->biSize=sizeof(BITMAPINFOHEADER);
|
|
sh_v->bih->biCompression= sh_v->format = FOURCC_THEORA;
|
|
sh_v->fps = ((double)inf.fps_numerator)/
|
|
(double)inf.fps_denominator;
|
|
sh_v->frametime = ((double)inf.fps_denominator)/
|
|
(double)inf.fps_numerator;
|
|
sh_v->disp_w = sh_v->bih->biWidth = inf.frame_width;
|
|
sh_v->disp_h = sh_v->bih->biHeight = inf.frame_height;
|
|
sh_v->bih->biBitCount = 24;
|
|
sh_v->bih->biPlanes = 3;
|
|
sh_v->bih->biSizeImage = ((sh_v->bih->biBitCount/8) *
|
|
sh_v->bih->biWidth*sh_v->bih->biHeight);
|
|
ogg_d->subs[ogg_d->num_sub].samplerate = sh_v->fps;
|
|
ogg_d->subs[ogg_d->num_sub].theora = 1;
|
|
if (identify)
|
|
mp_msg(MSGT_GLOBAL, MSGL_INFO, "ID_VIDEO_ID=%d\n", n_video);
|
|
ogg_d->subs[ogg_d->num_sub].id = n_video;
|
|
n_video++;
|
|
mp_msg(MSGT_DEMUX,MSGL_INFO,
|
|
"[Ogg] stream %d: video (Theora v%d.%d.%d), -vid %d\n",
|
|
ogg_d->num_sub,
|
|
(int)inf.version_major,
|
|
(int)inf.version_minor,
|
|
(int)inf.version_subminor,
|
|
n_video - 1);
|
|
if(verbose>0) print_video_header(sh_v->bih);
|
|
}
|
|
# endif /* HAVE_OGGTHEORA */
|
|
# ifdef HAVE_FLAC
|
|
} else if (pack.bytes >= 4 && !strncmp (&pack.packet[0], "fLaC", 4)) {
|
|
sh_a = new_sh_audio(demuxer,ogg_d->num_sub);
|
|
sh_a->format = mmioFOURCC('f', 'L', 'a', 'C');
|
|
if (identify)
|
|
mp_msg(MSGT_GLOBAL, MSGL_INFO, "ID_AUDIO_ID=%d\n", n_audio);
|
|
ogg_d->subs[ogg_d->num_sub].id = n_audio;
|
|
n_audio++;
|
|
ogg_d->subs[ogg_d->num_sub].flac = 1;
|
|
sh_a->wf = NULL;
|
|
mp_msg(MSGT_DEMUX,MSGL_INFO,"[Ogg] stream %d: audio (FLAC), -aid %d\n",ogg_d->num_sub,n_audio-1);
|
|
# endif /* HAVE_FLAC */
|
|
|
|
/// Check for old header
|
|
} else if(pack.bytes >= 142 && ! strncmp(&pack.packet[1],"Direct Show Samples embedded in Ogg",35) ) {
|
|
|
|
// Old video header
|
|
if(get_uint32 (pack.packet+96) == 0x05589f80 && pack.bytes >= 184) {
|
|
sh_v = new_sh_video(demuxer,ogg_d->num_sub);
|
|
sh_v->bih = (BITMAPINFOHEADER*)calloc(1,sizeof(BITMAPINFOHEADER));
|
|
sh_v->bih->biSize=sizeof(BITMAPINFOHEADER);
|
|
sh_v->bih->biCompression=
|
|
sh_v->format = mmioFOURCC(pack.packet[68],pack.packet[69],
|
|
pack.packet[70],pack.packet[71]);
|
|
sh_v->frametime = get_uint64(pack.packet+164)*0.0000001;
|
|
sh_v->fps = 1/sh_v->frametime;
|
|
sh_v->disp_w = sh_v->bih->biWidth = get_uint32(pack.packet+176);
|
|
sh_v->disp_h = sh_v->bih->biHeight = get_uint32(pack.packet+180);
|
|
sh_v->bih->biBitCount = get_uint16(pack.packet+182);
|
|
if(!sh_v->bih->biBitCount) sh_v->bih->biBitCount=24; // hack, FIXME
|
|
sh_v->bih->biPlanes=1;
|
|
sh_v->bih->biSizeImage=(sh_v->bih->biBitCount>>3)*sh_v->bih->biWidth*sh_v->bih->biHeight;
|
|
|
|
ogg_d->subs[ogg_d->num_sub].samplerate = sh_v->fps;
|
|
if (identify)
|
|
mp_msg(MSGT_GLOBAL, MSGL_INFO, "ID_VIDEO_ID=%d\n", n_video);
|
|
ogg_d->subs[ogg_d->num_sub].id = n_video;
|
|
n_video++;
|
|
mp_msg(MSGT_DEMUX,MSGL_INFO,"[Ogg] stream %d: video (FOURCC %c%c%c%c), -vid %d\n",
|
|
ogg_d->num_sub,pack.packet[68],pack.packet[69],pack.packet[70],pack.packet[71],n_video-1);
|
|
if(verbose>0) print_video_header(sh_v->bih);
|
|
// Old audio header
|
|
} else if(get_uint32(pack.packet+96) == 0x05589F81) {
|
|
unsigned int extra_size;
|
|
sh_a = new_sh_audio(demuxer,ogg_d->num_sub);
|
|
extra_size = get_uint16(pack.packet+140);
|
|
sh_a->wf = (WAVEFORMATEX*)calloc(1,sizeof(WAVEFORMATEX)+extra_size);
|
|
sh_a->format = sh_a->wf->wFormatTag = get_uint16(pack.packet+124);
|
|
sh_a->channels = sh_a->wf->nChannels = get_uint16(pack.packet+126);
|
|
sh_a->samplerate = sh_a->wf->nSamplesPerSec = get_uint32(pack.packet+128);
|
|
sh_a->wf->nAvgBytesPerSec = get_uint32(pack.packet+132);
|
|
sh_a->wf->nBlockAlign = get_uint16(pack.packet+136);
|
|
sh_a->wf->wBitsPerSample = get_uint16(pack.packet+138);
|
|
sh_a->samplesize = (sh_a->wf->wBitsPerSample+7)/8;
|
|
sh_a->wf->cbSize = extra_size;
|
|
if(extra_size > 0)
|
|
memcpy(((char *)sh_a->wf)+sizeof(WAVEFORMATEX),pack.packet+142,extra_size);
|
|
|
|
ogg_d->subs[ogg_d->num_sub].samplerate = sh_a->samplerate; // * sh_a->channels;
|
|
if (identify)
|
|
mp_msg(MSGT_GLOBAL, MSGL_INFO, "ID_AUDIO_ID=%d\n", n_audio);
|
|
ogg_d->subs[ogg_d->num_sub].id = n_audio;
|
|
n_audio++;
|
|
mp_msg(MSGT_DEMUX,MSGL_INFO,"[Ogg] stream %d: audio (format 0x%04x), -aid %d\n",ogg_d->num_sub,sh_a->format,n_audio-1);
|
|
if(verbose>0) print_wave_header(sh_a->wf);
|
|
} else
|
|
mp_msg(MSGT_DEMUX,MSGL_WARN,"Ogg stream %d contains an old header but the header type is unknown\n",ogg_d->num_sub);
|
|
|
|
// Check new header
|
|
} else if ( (*pack.packet & PACKET_TYPE_BITS ) == PACKET_TYPE_HEADER &&
|
|
pack.bytes >= (int)sizeof(stream_header)+1) {
|
|
stream_header *st = (stream_header*)(pack.packet+1);
|
|
/// New video header
|
|
if(strncmp(st->streamtype,"video",5) == 0) {
|
|
sh_v = new_sh_video(demuxer,ogg_d->num_sub);
|
|
sh_v->bih = (BITMAPINFOHEADER*)calloc(1,sizeof(BITMAPINFOHEADER));
|
|
sh_v->bih->biSize=sizeof(BITMAPINFOHEADER);
|
|
sh_v->bih->biCompression=
|
|
sh_v->format = mmioFOURCC(st->subtype[0],st->subtype[1],
|
|
st->subtype[2],st->subtype[3]);
|
|
sh_v->frametime = get_uint64(&st->time_unit)*0.0000001;
|
|
sh_v->fps = 1.0/sh_v->frametime;
|
|
sh_v->bih->biBitCount = get_uint16(&st->bits_per_sample);
|
|
sh_v->disp_w = sh_v->bih->biWidth = get_uint32(&st->sh.video.width);
|
|
sh_v->disp_h = sh_v->bih->biHeight = get_uint32(&st->sh.video.height);
|
|
if(!sh_v->bih->biBitCount) sh_v->bih->biBitCount=24; // hack, FIXME
|
|
sh_v->bih->biPlanes=1;
|
|
sh_v->bih->biSizeImage=(sh_v->bih->biBitCount>>3)*sh_v->bih->biWidth*sh_v->bih->biHeight;
|
|
|
|
ogg_d->subs[ogg_d->num_sub].samplerate= sh_v->fps;
|
|
if (identify)
|
|
mp_msg(MSGT_GLOBAL, MSGL_INFO, "ID_VIDEO_ID=%d\n", n_video);
|
|
ogg_d->subs[ogg_d->num_sub].id = n_video;
|
|
n_video++;
|
|
mp_msg(MSGT_DEMUX,MSGL_INFO,"[Ogg] stream %d: video (FOURCC %c%c%c%c), -vid %d\n",
|
|
ogg_d->num_sub,st->subtype[0],st->subtype[1],st->subtype[2],st->subtype[3],n_video-1);
|
|
if(verbose>0) print_video_header(sh_v->bih);
|
|
/// New audio header
|
|
} else if(strncmp(st->streamtype,"audio",5) == 0) {
|
|
char buffer[5];
|
|
unsigned int extra_size = get_uint32 (&st->size) - sizeof(stream_header);
|
|
memcpy(buffer,st->subtype,4);
|
|
buffer[4] = '\0';
|
|
sh_a = new_sh_audio(demuxer,ogg_d->num_sub);
|
|
sh_a->wf = (WAVEFORMATEX*)calloc(1,sizeof(WAVEFORMATEX)+extra_size);
|
|
sh_a->format = sh_a->wf->wFormatTag = strtol(buffer, NULL, 16);
|
|
sh_a->channels = sh_a->wf->nChannels = get_uint16(&st->sh.audio.channels);
|
|
sh_a->samplerate = sh_a->wf->nSamplesPerSec = get_uint64(&st->samples_per_unit);
|
|
sh_a->wf->nAvgBytesPerSec = get_uint32(&st->sh.audio.avgbytespersec);
|
|
sh_a->wf->nBlockAlign = get_uint16(&st->sh.audio.blockalign);
|
|
sh_a->wf->wBitsPerSample = get_uint16(&st->bits_per_sample);
|
|
sh_a->samplesize = (sh_a->wf->wBitsPerSample+7)/8;
|
|
sh_a->wf->cbSize = extra_size;
|
|
if(extra_size)
|
|
memcpy(((char *)sh_a->wf)+sizeof(WAVEFORMATEX),st+1,extra_size);
|
|
|
|
ogg_d->subs[ogg_d->num_sub].samplerate = sh_a->samplerate; // * sh_a->channels;
|
|
if (identify)
|
|
mp_msg(MSGT_GLOBAL, MSGL_INFO, "ID_AUDIO_ID=%d\n", n_audio);
|
|
ogg_d->subs[ogg_d->num_sub].id = n_audio;
|
|
n_audio++;
|
|
mp_msg(MSGT_DEMUX,MSGL_INFO,"[Ogg] stream %d: audio (format 0x%04x), -aid %d\n",ogg_d->num_sub,sh_a->format,n_audio-1);
|
|
if(verbose>0) print_wave_header(sh_a->wf);
|
|
|
|
/// Check for text (subtitles) header
|
|
} else if (strncmp(st->streamtype, "text", 4) == 0) {
|
|
mp_msg(MSGT_DEMUX, MSGL_INFO, "[Ogg] stream %d: subtitles (SRT-like text subtitles), -sid %d\n", ogg_d->num_sub, ogg_d->n_text);
|
|
ogg_d->subs[ogg_d->num_sub].samplerate= get_uint64(&st->time_unit)/10;
|
|
ogg_d->subs[ogg_d->num_sub].text = 1;
|
|
if (identify)
|
|
mp_msg(MSGT_GLOBAL, MSGL_INFO, "ID_SUBTITLE_ID=%d\n", ogg_d->n_text);
|
|
ogg_d->subs[ogg_d->num_sub].id = ogg_d->n_text;
|
|
if (demuxer->sub->id == ogg_d->n_text)
|
|
text_id = ogg_d->num_sub;
|
|
ogg_d->n_text++;
|
|
ogg_d->text_ids = (int *)realloc(ogg_d->text_ids, sizeof(int) * ogg_d->n_text);
|
|
ogg_d->text_ids[ogg_d->n_text - 1] = ogg_d->num_sub;
|
|
ogg_d->text_langs = (char **)realloc(ogg_d->text_langs, sizeof(char *) * ogg_d->n_text);
|
|
ogg_d->text_langs[ogg_d->n_text - 1] = NULL;
|
|
demux_ogg_init_sub();
|
|
//// Unknown header type
|
|
} else
|
|
mp_msg(MSGT_DEMUX,MSGL_ERR,"Ogg stream %d has a header marker but is of an unknown type\n",ogg_d->num_sub);
|
|
/// Unknown (invalid ?) header
|
|
} else
|
|
mp_msg(MSGT_DEMUX,MSGL_ERR,"Ogg stream %d is of an unknown type\n",ogg_d->num_sub);
|
|
|
|
if(sh_a || sh_v) {
|
|
demux_stream_t* ds = NULL;
|
|
if(sh_a) {
|
|
// If the audio stream is not defined we took the first one
|
|
if(demuxer->audio->id == -1) {
|
|
demuxer->audio->id = n_audio - 1;
|
|
// if(sh_a->wf) print_wave_header(sh_a->wf);
|
|
}
|
|
/// Is it the stream we want
|
|
if(demuxer->audio->id == (n_audio - 1)) {
|
|
demuxer->audio->sh = sh_a;
|
|
sh_a->ds = demuxer->audio;
|
|
ds = demuxer->audio;
|
|
audio_id = ogg_d->num_sub;
|
|
}
|
|
}
|
|
if(sh_v) {
|
|
/// Also for video
|
|
if(demuxer->video->id == -1) {
|
|
demuxer->video->id = n_video - 1;
|
|
// if(sh_v->bih) print_video_header(sh_v->bih);
|
|
}
|
|
if(demuxer->video->id == (n_video - 1)) {
|
|
demuxer->video->sh = sh_v;
|
|
sh_v->ds = demuxer->video;
|
|
ds = demuxer->video;
|
|
video_id = ogg_d->num_sub;
|
|
}
|
|
}
|
|
/// Add the header packets if the stream isn't seekable
|
|
if(ds && !s->end_pos) {
|
|
/// Finish the page, otherwise packets will be lost
|
|
do {
|
|
demux_ogg_add_packet(ds,&ogg_d->subs[ogg_d->num_sub],ogg_d->num_sub,&pack);
|
|
} while(ogg_stream_packetout(&ogg_d->subs[ogg_d->num_sub].stream,&pack) == 1);
|
|
}
|
|
}
|
|
ogg_d->num_sub++;
|
|
}
|
|
|
|
if(!n_video && !n_audio) {
|
|
goto err_out;
|
|
}
|
|
|
|
/// Finish to setup the demuxer
|
|
demuxer->priv = ogg_d;
|
|
|
|
if(!n_video || (video_id < 0))
|
|
demuxer->video->id = -2;
|
|
else
|
|
demuxer->video->id = video_id;
|
|
if(!n_audio || (audio_id < 0))
|
|
demuxer->audio->id = -2;
|
|
else
|
|
demuxer->audio->id = audio_id;
|
|
/* Disable the subs only if there are no text streams at all.
|
|
Otherwise the stream to display might be chosen later when the comment
|
|
packet is encountered and the user used -slang instead of -sid. */
|
|
if(!ogg_d->n_text)
|
|
demuxer->sub->id = -2;
|
|
else if (text_id >= 0) {
|
|
demuxer->sub->id = text_id;
|
|
mp_msg(MSGT_DEMUX, MSGL_V, "Ogg demuxer: Displaying subtitle stream id %d\n", text_id);
|
|
}
|
|
|
|
ogg_d->final_granulepos=0;
|
|
if(!s->end_pos)
|
|
demuxer->seekable = 0;
|
|
else {
|
|
demuxer->movi_start = s->start_pos; // Needed for XCD (Ogg written in MODE2)
|
|
demuxer->movi_end = s->end_pos;
|
|
demuxer->seekable = 1;
|
|
demux_ogg_scan_stream(demuxer);
|
|
}
|
|
|
|
mp_msg(MSGT_DEMUX,MSGL_V,"Ogg demuxer : found %d audio stream%s, %d video stream%s and %d text stream%s\n",n_audio,n_audio>1?"s":"",n_video,n_video>1?"s":"",ogg_d->n_text,ogg_d->n_text>1?"s":"");
|
|
|
|
return 1;
|
|
|
|
err_out:
|
|
demux_close_ogg(demuxer);
|
|
return 0;
|
|
}
|
|
|
|
|
|
int demux_ogg_fill_buffer(demuxer_t *d) {
|
|
ogg_demuxer_t* ogg_d;
|
|
stream_t *s;
|
|
demux_stream_t *ds;
|
|
ogg_sync_state* sync;
|
|
ogg_stream_state* os;
|
|
ogg_page* page;
|
|
ogg_packet pack;
|
|
int np = 0, id=0;
|
|
|
|
s = d->stream;
|
|
ogg_d = d->priv;
|
|
sync = &ogg_d->sync;
|
|
page = &ogg_d->page;
|
|
|
|
/// Find the stream we are working on
|
|
if ( (id = demux_ogg_get_page_stream(ogg_d,&os)) < 0) {
|
|
mp_msg(MSGT_DEMUX,MSGL_ERR,"Ogg demuxer : can't get current stream\n");
|
|
return 0;
|
|
}
|
|
|
|
while(1) {
|
|
np = 0;
|
|
ds = NULL;
|
|
/// Try to get some packet from the current page
|
|
while( (np = ogg_stream_packetout(os,&pack)) != 1) {
|
|
/// No packet we go the next page
|
|
if(np == 0) {
|
|
while(1) {
|
|
int pa,len;
|
|
char *buf;
|
|
ogg_d->pos += ogg_d->last_size;
|
|
/// Get the next page from the physical stream
|
|
while( (pa = ogg_sync_pageseek(sync,page)) <= 0) {
|
|
/// Error : we skip some bytes
|
|
if(pa < 0) {
|
|
mp_msg(MSGT_DEMUX,MSGL_WARN,"Ogg : Page out not synced, we skip some bytes\n");
|
|
ogg_d->pos -= pa;
|
|
continue;
|
|
}
|
|
/// We need more data
|
|
buf = ogg_sync_buffer(sync,BLOCK_SIZE);
|
|
len = stream_read(s,buf,BLOCK_SIZE);
|
|
if(len == 0 && s->eof) {
|
|
mp_msg(MSGT_DEMUX,MSGL_DBG2,"Ogg : Stream EOF !!!!\n");
|
|
return 0;
|
|
}
|
|
ogg_sync_wrote(sync,len);
|
|
} /// Page loop
|
|
ogg_d->last_size = pa;
|
|
/// Find the page's logical stream
|
|
if( (id = demux_ogg_get_page_stream(ogg_d,&os)) < 0) {
|
|
mp_msg(MSGT_DEMUX,MSGL_ERR,"Ogg demuxer error : we met an unknown stream\n");
|
|
return 0;
|
|
}
|
|
/// Take the page
|
|
if(ogg_stream_pagein(os,page) == 0)
|
|
break;
|
|
/// Page was invalid => retry
|
|
mp_msg(MSGT_DEMUX,MSGL_WARN,"Ogg demuxer : got invalid page !!!!!\n");
|
|
ogg_d->pos += ogg_d->last_size;
|
|
}
|
|
} else /// Packet was corrupted
|
|
mp_msg(MSGT_DEMUX,MSGL_WARN,"Ogg : bad packet in stream %d\n",id);
|
|
} /// Packet loop
|
|
|
|
/// Is the actual logical stream in use ?
|
|
if(id == d->audio->id)
|
|
ds = d->audio;
|
|
else if(id == d->video->id)
|
|
ds = d->video;
|
|
else if (ogg_d->subs[id].text)
|
|
ds = d->sub;
|
|
|
|
if(ds) {
|
|
if(!demux_ogg_add_packet(ds,&ogg_d->subs[id],id,&pack))
|
|
continue; /// Unuseful packet, get another
|
|
d->filepos = ogg_d->pos;
|
|
return 1;
|
|
}
|
|
|
|
} /// while(1)
|
|
|
|
}
|
|
|
|
/// For avi with Ogg audio stream we have to create an ogg demuxer for this
|
|
// stream, then we join the avi and ogg demuxer with a demuxers demuxer
|
|
demuxer_t* init_avi_with_ogg(demuxer_t* demuxer) {
|
|
demuxer_t *od;
|
|
ogg_demuxer_t *ogg_d;
|
|
stream_t* s;
|
|
uint32_t hdrsizes[3];
|
|
demux_packet_t *dp;
|
|
sh_audio_t *sh_audio = demuxer->audio->sh;
|
|
int np;
|
|
unsigned char *p = NULL,*buf;
|
|
int plen;
|
|
|
|
/// Check that the cbSize is enouth big for the following reads
|
|
if(sh_audio->wf->cbSize < 22+3*sizeof(uint32_t)) {
|
|
mp_msg(MSGT_DEMUX,MSGL_ERR,"AVI Ogg : Initial audio header is too small !!!!!\n");
|
|
goto fallback;
|
|
}
|
|
/// Get the size of the 3 header packet
|
|
memcpy(hdrsizes, ((unsigned char*)sh_audio->wf)+22+sizeof(WAVEFORMATEX), 3*sizeof(uint32_t));
|
|
// printf("\n!!!!!! hdr sizes: %d %d %d \n",hdrsizes[0],hdrsizes[1],hdrsizes[2]);
|
|
|
|
/// Check the size
|
|
if(sh_audio->wf->cbSize < 22+3*sizeof(uint32_t)+hdrsizes[0]+hdrsizes[1] + hdrsizes[2]) {
|
|
mp_msg(MSGT_DEMUX,MSGL_ERR,"AVI Ogg : Audio header is too small !!!!!\n");
|
|
goto fallback;
|
|
}
|
|
|
|
// Build the ogg demuxer private datas
|
|
ogg_d = (ogg_demuxer_t*)calloc(1,sizeof(ogg_demuxer_t));
|
|
ogg_d->num_sub = 1;
|
|
ogg_d->subs = (ogg_stream_t*)malloc(sizeof(ogg_stream_t));
|
|
ogg_d->subs[0].vorbis = 1;
|
|
|
|
// Init the ogg physical stream
|
|
ogg_sync_init(&ogg_d->sync);
|
|
|
|
// Get the first page of the stream : we assume there only 1 logical stream
|
|
while((np = ogg_sync_pageout(&ogg_d->sync,&ogg_d->page)) <= 0 ) {
|
|
if(np < 0) {
|
|
mp_msg(MSGT_DEMUX,MSGL_ERR,"AVI Ogg error : Can't init using first stream packets\n");
|
|
free(ogg_d);
|
|
goto fallback;
|
|
}
|
|
// Add some data
|
|
plen = ds_get_packet(demuxer->audio,&p);
|
|
buf = ogg_sync_buffer(&ogg_d->sync,plen);
|
|
memcpy(buf,p,plen);
|
|
ogg_sync_wrote(&ogg_d->sync,plen);
|
|
}
|
|
// Init the logical stream
|
|
mp_msg(MSGT_DEMUX,MSGL_DBG2,"AVI Ogg found page with serial %d\n",ogg_page_serialno(&ogg_d->page));
|
|
ogg_stream_init(&ogg_d->subs[0].stream,ogg_page_serialno(&ogg_d->page));
|
|
// Write the page
|
|
ogg_stream_pagein(&ogg_d->subs[0].stream,&ogg_d->page);
|
|
|
|
// Create the ds_stream and the ogg demuxer
|
|
s = new_ds_stream(demuxer->audio);
|
|
od = new_demuxer(s,DEMUXER_TYPE_OGG,0,-2,-2);
|
|
|
|
/// Add the header packets in the ogg demuxer audio stream
|
|
// Initial header
|
|
dp = new_demux_packet(hdrsizes[0]);
|
|
memcpy(dp->buffer,((unsigned char*)sh_audio->wf)+22+sizeof(WAVEFORMATEX)+3*sizeof(uint32_t),hdrsizes[0]);
|
|
ds_add_packet(od->audio,dp);
|
|
/// Comments
|
|
dp = new_demux_packet(hdrsizes[1]);
|
|
memcpy(dp->buffer,((unsigned char*)sh_audio->wf)+22+sizeof(WAVEFORMATEX)+3*sizeof(uint32_t)+hdrsizes[0],hdrsizes[1]);
|
|
ds_add_packet(od->audio,dp);
|
|
/// Code book
|
|
dp = new_demux_packet(hdrsizes[2]);
|
|
memcpy(dp->buffer,((unsigned char*)sh_audio->wf)+22+sizeof(WAVEFORMATEX)+3*sizeof(uint32_t)+hdrsizes[0]+hdrsizes[1],hdrsizes[2]);
|
|
ds_add_packet(od->audio,dp);
|
|
|
|
// Finish setting up the ogg demuxer
|
|
od->priv = ogg_d;
|
|
sh_audio = new_sh_audio(od,0);
|
|
od->audio->id = 0;
|
|
od->video->id = -2;
|
|
od->audio->sh = sh_audio;
|
|
sh_audio->ds = od->audio;
|
|
sh_audio->format = FOURCC_VORBIS;
|
|
|
|
/// Return the joined demuxers
|
|
return new_demuxers_demuxer(demuxer,od,demuxer);
|
|
|
|
fallback:
|
|
demuxer->audio->id = -2;
|
|
return demuxer;
|
|
|
|
}
|
|
|
|
extern void resync_audio_stream(sh_audio_t *sh_audio);
|
|
|
|
void demux_ogg_seek(demuxer_t *demuxer,float rel_seek_secs,int flags) {
|
|
ogg_demuxer_t* ogg_d = demuxer->priv;
|
|
ogg_sync_state* sync = &ogg_d->sync;
|
|
ogg_page* page= &ogg_d->page;
|
|
ogg_stream_state* oss;
|
|
ogg_stream_t* os;
|
|
demux_stream_t* ds;
|
|
sh_audio_t* sh_audio = demuxer->audio->sh;
|
|
ogg_packet op;
|
|
float rate;
|
|
int i,sp,first,precision=1,do_seek=1;
|
|
vorbis_info* vi = NULL;
|
|
int64_t gp = 0, old_gp;
|
|
void *context = NULL;
|
|
off_t pos, old_pos;
|
|
int np;
|
|
int is_gp_valid;
|
|
float pts;
|
|
int is_keyframe;
|
|
int samplesize=1;
|
|
|
|
if(demuxer->video->id >= 0) {
|
|
ds = demuxer->video;
|
|
/* demux_ogg_read_packet needs decoder context for Theora streams */
|
|
if (((sh_video_t*)demuxer->video->sh)->format == FOURCC_THEORA)
|
|
context = ((sh_video_t*)demuxer->video->sh)->context;
|
|
rate = ogg_d->subs[ds->id].samplerate;
|
|
} else {
|
|
ds = demuxer->audio;
|
|
/* demux_ogg_read_packet needs decoder context for Vorbis streams */
|
|
if(((sh_audio_t*)demuxer->audio->sh)->format == FOURCC_VORBIS)
|
|
context = ((sh_audio_t*)demuxer->audio->sh)->context;
|
|
vi = &((ov_struct_t*)((sh_audio_t*)ds->sh)->context)->vi;
|
|
rate = (float)vi->rate;
|
|
samplesize = ((sh_audio_t*)ds->sh)->samplesize;
|
|
}
|
|
|
|
os = &ogg_d->subs[ds->id];
|
|
oss = &os->stream;
|
|
|
|
old_gp = os->lastpos;
|
|
old_pos = ogg_d->pos;
|
|
|
|
//calculate the granulepos to seek to
|
|
gp = flags & 1 ? 0 : os->lastpos;
|
|
if(flags & 2) {
|
|
if (ogg_d->final_granulepos > 0)
|
|
gp += ogg_d->final_granulepos * rel_seek_secs;
|
|
else
|
|
gp += rel_seek_secs * (demuxer->movi_end - demuxer->movi_start) * os->lastpos / ogg_d->pos;
|
|
} else
|
|
gp += rel_seek_secs * rate;
|
|
if (gp < 0) gp = 0;
|
|
|
|
//calculate the filepos to seek to
|
|
if(ogg_d->syncpoints) {
|
|
for(sp = 0; sp < ogg_d->num_syncpoint ; sp++) {
|
|
if(ogg_d->syncpoints[sp].granulepos >= gp) break;
|
|
}
|
|
|
|
if(sp >= ogg_d->num_syncpoint) return;
|
|
if (sp > 0 && ogg_d->syncpoints[sp].granulepos - gp > gp - ogg_d->syncpoints[sp-1].granulepos)
|
|
sp--;
|
|
if (ogg_d->syncpoints[sp].granulepos == os->lastpos) {
|
|
if (sp > 0 && gp < os->lastpos) sp--;
|
|
if (sp < ogg_d->num_syncpoint-1 && gp > os->lastpos) sp++;
|
|
}
|
|
pos = ogg_d->syncpoints[sp].page_pos;
|
|
precision = 0;
|
|
} else {
|
|
pos = flags & 1 ? 0 : ogg_d->pos;
|
|
if(flags & 2)
|
|
pos += (demuxer->movi_end - demuxer->movi_start) * rel_seek_secs;
|
|
else {
|
|
if (ogg_d->final_granulepos > 0) {
|
|
pos += rel_seek_secs * (demuxer->movi_end - demuxer->movi_start) / (ogg_d->final_granulepos / rate);
|
|
} else if (os->lastpos > 0) {
|
|
pos += rel_seek_secs * ogg_d->pos / (os->lastpos / rate);
|
|
}
|
|
}
|
|
if (pos < 0) pos = 0;
|
|
if (pos > (demuxer->movi_end - demuxer->movi_start)) return;
|
|
} // if(ogg_d->syncpoints)
|
|
|
|
while(1) {
|
|
if (do_seek) {
|
|
stream_seek(demuxer->stream,pos+demuxer->movi_start);
|
|
ogg_sync_reset(sync);
|
|
for(i = 0 ; i < ogg_d->num_sub ; i++) {
|
|
ogg_stream_reset(&ogg_d->subs[i].stream);
|
|
ogg_d->subs[i].lastpos = ogg_d->subs[i].lastsize = 0;
|
|
}
|
|
ogg_d->pos = pos;
|
|
ogg_d->last_size = 0;
|
|
/* we just guess that we reached correct granulepos, in case a
|
|
subsequent search occurs before we read a valid granulepos */
|
|
os->lastpos = gp;
|
|
first = !(ogg_d->syncpoints);
|
|
do_seek=0;
|
|
}
|
|
ogg_d->pos += ogg_d->last_size;
|
|
ogg_d->last_size = 0;
|
|
np = ogg_sync_pageseek(sync,page);
|
|
|
|
if(np < 0)
|
|
ogg_d->pos -= np;
|
|
if(np <= 0) { // We need more data
|
|
char* buf = ogg_sync_buffer(sync,BLOCK_SIZE);
|
|
int len = stream_read(demuxer->stream,buf,BLOCK_SIZE);
|
|
if(len == 0 && demuxer->stream->eof) {
|
|
mp_msg(MSGT_DEMUX,MSGL_ERR,"EOF while trying to seek !!!!\n");
|
|
break;
|
|
}
|
|
ogg_sync_wrote(sync,len);
|
|
continue;
|
|
}
|
|
ogg_d->last_size = np;
|
|
if(ogg_page_serialno(page) != oss->serialno)
|
|
continue;
|
|
|
|
if(ogg_stream_pagein(oss,page) != 0)
|
|
continue;
|
|
|
|
while(1) {
|
|
np = ogg_stream_packetout(oss,&op);
|
|
if(np < 0)
|
|
continue;
|
|
else if(np == 0)
|
|
break;
|
|
if (first) { /* Discard the first packet as it's probably broken,
|
|
and we don't have any other means to decide whether it is
|
|
complete or not. */
|
|
first = 0;
|
|
break;
|
|
}
|
|
is_gp_valid = (op.granulepos >= 0);
|
|
demux_ogg_read_packet(os,&op,context,&pts,&is_keyframe,samplesize);
|
|
if (precision && is_gp_valid) {
|
|
precision--;
|
|
if (abs(gp - op.granulepos) > rate && (op.granulepos != old_gp)) {
|
|
//prepare another seek because we are off by more than 1s
|
|
pos += (gp - op.granulepos) * (pos - old_pos) / (op.granulepos - old_gp);
|
|
if (pos < 0) pos = 0;
|
|
if (pos < (demuxer->movi_end - demuxer->movi_start)) {
|
|
do_seek=1;
|
|
break;
|
|
}
|
|
}
|
|
}
|
|
if (is_gp_valid && (pos > 0) && (old_gp > gp)
|
|
&& (2 * (old_gp - op.granulepos) < old_gp - gp)) {
|
|
/* prepare another seek because looking for a syncpoint
|
|
destroyed the backward search */
|
|
pos = old_pos - 1.5 * (old_pos - pos);
|
|
if (pos < 0) pos = 0;
|
|
if (pos < (demuxer->movi_end - demuxer->movi_start)) {
|
|
do_seek=1;
|
|
break;
|
|
}
|
|
}
|
|
if(!precision && (is_keyframe || os->vorbis) ) {
|
|
ogg_sub.lines = 0;
|
|
vo_sub = &ogg_sub;
|
|
vo_osd_changed(OSDTYPE_SUBTITLE);
|
|
clear_sub = -1;
|
|
demux_ogg_add_packet(ds,os,ds->id,&op);
|
|
if(sh_audio)
|
|
resync_audio_stream(sh_audio);
|
|
return;
|
|
}
|
|
}
|
|
}
|
|
|
|
mp_msg(MSGT_DEMUX,MSGL_ERR,"Can't find the good packet :(\n");
|
|
|
|
}
|
|
|
|
void demux_close_ogg(demuxer_t* demuxer) {
|
|
ogg_demuxer_t* ogg_d = demuxer->priv;
|
|
int i;
|
|
|
|
if(!ogg_d)
|
|
return;
|
|
|
|
#ifdef USE_ICONV
|
|
subcp_close();
|
|
#endif
|
|
|
|
ogg_sync_clear(&ogg_d->sync);
|
|
if(ogg_d->subs)
|
|
{
|
|
for (i = 0; i < ogg_d->num_sub; i++)
|
|
ogg_stream_clear(&ogg_d->subs[i].stream);
|
|
free(ogg_d->subs);
|
|
}
|
|
if(ogg_d->syncpoints)
|
|
free(ogg_d->syncpoints);
|
|
if (ogg_d->text_ids)
|
|
free(ogg_d->text_ids);
|
|
if (ogg_d->text_langs) {
|
|
for (i = 0; i < ogg_d->n_text; i++)
|
|
if (ogg_d->text_langs[i]) free(ogg_d->text_langs[i]);
|
|
free(ogg_d->text_langs);
|
|
}
|
|
free(ogg_d);
|
|
}
|
|
|
|
int demux_ogg_control(demuxer_t *demuxer,int cmd, void *arg){
|
|
ogg_demuxer_t* ogg_d = demuxer->priv;
|
|
ogg_stream_t* os;
|
|
float rate;
|
|
|
|
if(demuxer->video->id >= 0) {
|
|
os = &ogg_d->subs[demuxer->video->id];
|
|
rate = os->samplerate;
|
|
} else {
|
|
os = &ogg_d->subs[demuxer->audio->id];
|
|
rate = (float)((ov_struct_t*)((sh_audio_t*)demuxer->audio->sh)->context)->vi.rate;
|
|
}
|
|
|
|
|
|
switch(cmd) {
|
|
case DEMUXER_CTRL_GET_TIME_LENGTH:
|
|
if (ogg_d->final_granulepos<=0) return DEMUXER_CTRL_DONTKNOW;
|
|
*((unsigned long *)arg)=ogg_d->final_granulepos / rate;
|
|
return DEMUXER_CTRL_GUESS;
|
|
|
|
case DEMUXER_CTRL_GET_PERCENT_POS:
|
|
if (ogg_d->final_granulepos<=0) return DEMUXER_CTRL_DONTKNOW;
|
|
*((int *)arg)=(int)( (os->lastpos*100) / ogg_d->final_granulepos);
|
|
return DEMUXER_CTRL_OK;
|
|
|
|
default:
|
|
return DEMUXER_CTRL_NOTIMPL;
|
|
}
|
|
}
|
|
|
|
#endif
|