2007-06-05 01:10:54 +03:00
/*
2008-08-05 03:42:43 +03:00
* Matroska file demuxer
2008-08-05 11:28:57 +03:00
* Copyright ( c ) 2003 - 2008 The FFmpeg Project
2007-06-05 01:10:54 +03:00
*
* This file is part of FFmpeg .
*
* FFmpeg is free software ; you can redistribute it and / or
* modify it under the terms of the GNU Lesser General Public
* License as published by the Free Software Foundation ; either
* version 2.1 of the License , or ( at your option ) any later version .
*
* FFmpeg 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
* Lesser General Public License for more details .
*
* You should have received a copy of the GNU Lesser General Public
* License along with FFmpeg ; if not , write to the Free Software
* Foundation , Inc . , 51 Franklin Street , Fifth Floor , Boston , MA 02110 - 1301 USA
*/
/**
2010-04-20 17:45:34 +03:00
* @ file
2007-06-05 01:10:54 +03:00
* Matroska file demuxer
2011-10-30 21:10:50 +03:00
* @ author Ronald Bultje < rbultje @ ronald . bitfreak . net >
* @ author with a little help from Moritz Bunkus < moritz @ bunkus . org >
* @ author totally reworked by Aurelien Jacobs < aurel @ gnuage . org >
* @ see specs available on the Matroska project page : http : //www.matroska.org/
2007-06-05 01:10:54 +03:00
*/
2014-03-01 19:28:15 +03:00
# include "config.h"
2014-03-10 17:35:59 +03:00
# include <inttypes.h>
2008-09-05 02:26:12 +03:00
# include <stdio.h>
2014-03-01 19:28:15 +03:00
# include "libavutil/avstring.h"
2014-03-08 01:46:37 +03:00
# include "libavutil/base64.h"
2014-03-01 19:28:15 +03:00
# include "libavutil/dict.h"
# include "libavutil/intfloat.h"
# include "libavutil/intreadwrite.h"
# include "libavutil/lzo.h"
2016-03-01 02:41:04 +02:00
# include "libavutil/mastering_display_metadata.h"
2014-04-29 13:03:13 +03:00
# include "libavutil/mathematics.h"
2015-04-01 01:51:57 +02:00
# include "libavutil/opt.h"
2014-11-02 20:19:07 +02:00
# include "libavutil/time_internal.h"
2016-12-06 19:48:45 +02:00
# include "libavutil/spherical.h"
2014-03-01 19:28:15 +03:00
# include "libavcodec/bytestream.h"
2014-05-25 15:05:51 +03:00
# include "libavcodec/flac.h"
2014-03-01 19:28:15 +03:00
# include "libavcodec/mpeg4audio.h"
2020-08-17 17:03:50 +02:00
# include "libavcodec/packet_internal.h"
2014-03-01 19:28:15 +03:00
2007-06-05 01:10:54 +03:00
# include "avformat.h"
2011-02-20 12:04:13 +02:00
# include "avio_internal.h"
2014-03-01 19:28:15 +03:00
# include "internal.h"
# include "isom.h"
# include "matroska.h"
2014-05-26 13:48:56 +03:00
# include "oggdec.h"
2009-06-23 02:09:34 +03:00
/* For ff_codec_get_id(). */
2007-06-05 01:10:54 +03:00
# include "riff.h"
2012-10-15 00:41:12 +03:00
# include "rmsipr.h"
2007-06-05 01:10:54 +03:00
2014-11-27 10:00:36 +02:00
# if CONFIG_BZLIB
# include <bzlib.h>
# endif
# if CONFIG_ZLIB
# include <zlib.h>
# endif
2015-12-27 22:28:09 +02:00
# include "qtpalette.h"
2019-02-23 12:14:33 +02:00
# define EBML_UNKNOWN_LENGTH UINT64_MAX /* EBML unknown length, in uint64_t */
2019-06-24 01:42:30 +02:00
# define NEEDS_CHECKING 2 / * Indicates that some error checks
* still need to be performed */
2019-06-24 01:46:54 +02:00
# define LEVEL_ENDED 3 / * return value of ebml_parse when the
* syntax level used for parsing ended . */
2019-05-17 00:30:14 +02:00
# define SKIP_THRESHOLD 1024 * 1024 / * In non-seekable mode, if more than SKIP_THRESHOLD
* of unkown , potentially damaged data is encountered ,
* it is considered an error . */
# define UNKNOWN_EQUIV 50 * 1024 / * An unknown element is considered equivalent
* to this many bytes of unknown data for the
* SKIP_THRESHOLD check . */
2019-02-23 12:14:33 +02:00
2008-08-05 03:40:00 +03:00
typedef enum {
EBML_NONE ,
EBML_UINT ,
2019-05-17 00:30:13 +02:00
EBML_SINT ,
2008-08-05 03:40:00 +03:00
EBML_FLOAT ,
EBML_STR ,
EBML_UTF8 ,
EBML_BIN ,
EBML_NEST ,
2015-02-09 21:39:00 +02:00
EBML_LEVEL1 ,
2008-08-05 03:40:00 +03:00
EBML_STOP ,
2011-02-06 12:32:03 +02:00
EBML_TYPE_COUNT
2008-08-05 03:40:00 +03:00
} EbmlType ;
avformat/matroskadec: Allow to count the number of element occurences
Up until now, the generic EBML reader used by the Matroska demuxer did
not have the capability to record whether an element was actually
present or not; instead, in cases where it mattered one typically added
an invalid default value and checked whether the value is valid (in
which case it is guaranteed to be present). This worked pretty well so
far, yet the EBML specifications have evolved: It is now legal to use
zero-length elements for floats, ints, uints and strings (both ASCII and
UTF-8); the value of these elements is the default value of the element
(if it has one) or zero for scalar types and an empty string for
strings. Furthermore, having a default value does no longer imply that
the element may be presumed to be present (with its default value) if it
is absent; this is only true if the element is mandatory, too.
These rules are designed to allow size savings as follows: Consider the
newly added FlagOriginal: It being zero means the track is not in its
original language, it being one means it is. For backward compatibility
reasons, neither of the two values may be inferred automatically in the
absence of the element. But one can still save a byte when one wants to
write the element with a value of zero, as one can write the integer with
a length of zero: 0x55AE 80 instead of 0x55AE 81 00. In the former case,
a parser has to infer the value of the element to be zero (which is the
element's default value).
When encountering an element with length zero, our parser always infers
a value of zero (or an empty string); this is wrong for values with
a different default value. It needs to infer the default value (or zero
in its absence) and this precludes using an invalid default value for
elements like FlagOriginal. Ergo one needs to be able to record whether
an element is present or not by other means. This patch allows to use a
simple counter for this. While just at it, some invalid and unnecessary
default values have been removed (mastering metadata elements used
default values of -1.0, despite these elements only being used if they
are > 0).
Reviewed-by: Ridley Combs <rcombs@rcombs.me>
Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@gmail.com>
2021-02-16 15:33:51 +02:00
typedef struct CountedElement {
union {
uint64_t u ;
int64_t i ;
double f ;
char * s ;
} el ;
unsigned count ;
} CountedElement ;
2008-08-05 03:40:00 +03:00
typedef const struct EbmlSyntax {
uint32_t id ;
avformat/matroskadec: Allow to count the number of element occurences
Up until now, the generic EBML reader used by the Matroska demuxer did
not have the capability to record whether an element was actually
present or not; instead, in cases where it mattered one typically added
an invalid default value and checked whether the value is valid (in
which case it is guaranteed to be present). This worked pretty well so
far, yet the EBML specifications have evolved: It is now legal to use
zero-length elements for floats, ints, uints and strings (both ASCII and
UTF-8); the value of these elements is the default value of the element
(if it has one) or zero for scalar types and an empty string for
strings. Furthermore, having a default value does no longer imply that
the element may be presumed to be present (with its default value) if it
is absent; this is only true if the element is mandatory, too.
These rules are designed to allow size savings as follows: Consider the
newly added FlagOriginal: It being zero means the track is not in its
original language, it being one means it is. For backward compatibility
reasons, neither of the two values may be inferred automatically in the
absence of the element. But one can still save a byte when one wants to
write the element with a value of zero, as one can write the integer with
a length of zero: 0x55AE 80 instead of 0x55AE 81 00. In the former case,
a parser has to infer the value of the element to be zero (which is the
element's default value).
When encountering an element with length zero, our parser always infers
a value of zero (or an empty string); this is wrong for values with
a different default value. It needs to infer the default value (or zero
in its absence) and this precludes using an invalid default value for
elements like FlagOriginal. Ergo one needs to be able to record whether
an element is present or not by other means. This patch allows to use a
simple counter for this. While just at it, some invalid and unnecessary
default values have been removed (mastering metadata elements used
default values of -1.0, despite these elements only being used if they
are > 0).
Reviewed-by: Ridley Combs <rcombs@rcombs.me>
Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@gmail.com>
2021-02-16 15:33:51 +02:00
uint8_t type ;
uint8_t is_counted ;
2019-09-03 23:52:51 +02:00
size_t list_elem_size ;
size_t data_offset ;
2008-08-05 03:40:00 +03:00
union {
2017-02-04 00:42:44 +02:00
int64_t i ;
2008-08-05 03:40:00 +03:00
uint64_t u ;
double f ;
const char * s ;
const struct EbmlSyntax * n ;
} def ;
} EbmlSyntax ;
2014-09-22 10:19:33 +03:00
typedef struct EbmlList {
2008-08-05 03:40:00 +03:00
int nb_elem ;
2019-09-03 23:45:04 +02:00
unsigned int alloc_elem_size ;
2008-08-05 03:40:00 +03:00
void * elem ;
} EbmlList ;
2014-09-22 10:19:33 +03:00
typedef struct EbmlBin {
2008-08-05 03:40:00 +03:00
int size ;
2018-04-04 19:12:54 +02:00
AVBufferRef * buf ;
2008-08-05 03:40:00 +03:00
uint8_t * data ;
int64_t pos ;
} EbmlBin ;
2014-09-22 10:19:33 +03:00
typedef struct Ebml {
2008-08-05 03:40:02 +03:00
uint64_t version ;
uint64_t max_size ;
uint64_t id_length ;
char * doctype ;
uint64_t doctype_version ;
} Ebml ;
2014-09-22 10:19:33 +03:00
typedef struct MatroskaTrackCompression {
2008-08-05 03:40:31 +03:00
uint64_t algo ;
EbmlBin settings ;
} MatroskaTrackCompression ;
2007-06-05 01:10:54 +03:00
2015-02-14 22:07:40 +02:00
typedef struct MatroskaTrackEncryption {
2013-03-07 19:11:38 +03:00
uint64_t algo ;
EbmlBin key_id ;
} MatroskaTrackEncryption ;
2014-09-22 10:19:33 +03:00
typedef struct MatroskaTrackEncoding {
2008-08-05 03:40:31 +03:00
uint64_t scope ;
uint64_t type ;
MatroskaTrackCompression compression ;
2013-03-07 19:11:38 +03:00
MatroskaTrackEncryption encryption ;
2008-08-05 03:40:31 +03:00
} MatroskaTrackEncoding ;
2007-06-05 01:10:54 +03:00
2016-03-01 02:41:04 +02:00
typedef struct MatroskaMasteringMeta {
double r_x ;
double r_y ;
double g_x ;
double g_y ;
double b_x ;
double b_y ;
double white_x ;
double white_y ;
double max_luminance ;
2021-02-16 16:34:06 +02:00
CountedElement min_luminance ;
2016-03-01 02:41:04 +02:00
} MatroskaMasteringMeta ;
typedef struct MatroskaTrackVideoColor {
uint64_t matrix_coefficients ;
uint64_t bits_per_channel ;
uint64_t chroma_sub_horz ;
uint64_t chroma_sub_vert ;
uint64_t cb_sub_horz ;
uint64_t cb_sub_vert ;
uint64_t chroma_siting_horz ;
uint64_t chroma_siting_vert ;
uint64_t range ;
uint64_t transfer_characteristics ;
uint64_t primaries ;
uint64_t max_cll ;
uint64_t max_fall ;
MatroskaMasteringMeta mastering_meta ;
} MatroskaTrackVideoColor ;
2016-12-06 19:48:45 +02:00
typedef struct MatroskaTrackVideoProjection {
uint64_t type ;
EbmlBin private ;
double yaw ;
double pitch ;
double roll ;
} MatroskaTrackVideoProjection ;
2014-09-22 10:19:33 +03:00
typedef struct MatroskaTrackVideo {
2008-08-05 03:40:31 +03:00
double frame_rate ;
uint64_t display_width ;
uint64_t display_height ;
uint64_t pixel_width ;
uint64_t pixel_height ;
2019-05-17 00:30:13 +02:00
EbmlBin color_space ;
2016-10-15 23:01:50 +02:00
uint64_t display_unit ;
2016-03-28 20:29:54 +02:00
uint64_t interlaced ;
uint64_t field_order ;
2011-05-24 02:09:24 +03:00
uint64_t stereo_mode ;
2013-02-05 02:17:52 +03:00
uint64_t alpha_mode ;
2016-12-05 04:22:39 +02:00
EbmlList color ;
2016-12-06 19:48:45 +02:00
MatroskaTrackVideoProjection projection ;
2008-08-05 03:40:31 +03:00
} MatroskaTrackVideo ;
2007-06-05 01:10:54 +03:00
2014-09-22 10:19:33 +03:00
typedef struct MatroskaTrackAudio {
2008-08-05 03:40:31 +03:00
double samplerate ;
double out_samplerate ;
uint64_t bitdepth ;
uint64_t channels ;
/* real audio header (extracted from extradata) */
int coded_framesize ;
int sub_packet_h ;
int frame_size ;
int sub_packet_size ;
int sub_packet_cnt ;
int pkt_cnt ;
2011-02-26 13:52:01 +02:00
uint64_t buf_timecode ;
2008-08-05 03:40:31 +03:00
uint8_t * buf ;
} MatroskaTrackAudio ;
2007-06-05 01:10:54 +03:00
2015-02-14 22:07:40 +02:00
typedef struct MatroskaTrackPlane {
2011-05-21 18:14:14 +03:00
uint64_t uid ;
uint64_t type ;
} MatroskaTrackPlane ;
2015-02-14 22:07:40 +02:00
typedef struct MatroskaTrackOperation {
2011-05-21 18:14:14 +03:00
EbmlList combine_planes ;
} MatroskaTrackOperation ;
2014-09-22 10:19:33 +03:00
typedef struct MatroskaTrack {
2008-08-05 03:40:31 +03:00
uint64_t num ;
2009-02-15 17:34:22 +02:00
uint64_t uid ;
2008-08-05 03:40:31 +03:00
uint64_t type ;
2009-02-15 17:29:09 +02:00
char * name ;
2008-08-05 03:40:31 +03:00
char * codec_id ;
EbmlBin codec_priv ;
char * language ;
2008-06-01 16:54:11 +03:00
double time_scale ;
2007-06-05 01:10:54 +03:00
uint64_t default_duration ;
2008-08-05 03:39:53 +03:00
uint64_t flag_default ;
2010-07-02 19:38:44 +03:00
uint64_t flag_forced ;
2021-02-16 19:23:59 +02:00
uint64_t flag_comment ;
2021-02-16 20:08:51 +02:00
uint64_t flag_hearingimpaired ;
uint64_t flag_visualimpaired ;
2021-02-16 20:38:19 +02:00
uint64_t flag_textdescriptions ;
2021-02-16 19:41:30 +02:00
CountedElement flag_original ;
2013-10-14 20:42:08 +03:00
uint64_t seek_preroll ;
2008-08-05 03:40:31 +03:00
MatroskaTrackVideo video ;
MatroskaTrackAudio audio ;
2011-05-21 18:14:14 +03:00
MatroskaTrackOperation operation ;
2008-08-05 03:40:31 +03:00
EbmlList encodings ;
2014-04-29 13:03:13 +03:00
uint64_t codec_delay ;
2016-06-06 04:23:16 +02:00
uint64_t codec_delay_in_track_tb ;
2008-08-05 03:40:06 +03:00
AVStream * stream ;
2008-09-09 15:07:10 +03:00
int64_t end_timecode ;
2010-03-03 23:46:43 +02:00
int ms_compat ;
2019-12-04 18:54:45 +02:00
int needs_decoding ;
2013-02-14 00:51:48 +03:00
uint64_t max_block_additional_id ;
2016-02-24 19:14:05 +02:00
uint32_t palette [ AVPALETTE_COUNT ] ;
int has_palette ;
2007-06-05 01:10:54 +03:00
} MatroskaTrack ;
2014-09-22 10:19:33 +03:00
typedef struct MatroskaAttachment {
2009-02-15 17:34:22 +02:00
uint64_t uid ;
2008-08-05 03:40:24 +03:00
char * filename ;
2020-05-06 15:24:33 +02:00
char * description ;
2008-08-05 03:40:24 +03:00
char * mime ;
EbmlBin bin ;
2009-02-15 17:53:55 +02:00
AVStream * stream ;
2014-03-07 15:54:18 +03:00
} MatroskaAttachment ;
2008-08-05 03:40:24 +03:00
2014-09-22 10:19:33 +03:00
typedef struct MatroskaChapter {
2008-08-05 03:40:21 +03:00
uint64_t start ;
uint64_t end ;
uint64_t uid ;
char * title ;
2009-02-15 17:25:14 +02:00
AVChapter * chapter ;
2008-08-05 03:40:21 +03:00
} MatroskaChapter ;
2014-09-22 10:19:33 +03:00
typedef struct MatroskaIndexPos {
2008-08-05 03:40:15 +03:00
uint64_t track ;
uint64_t pos ;
} MatroskaIndexPos ;
2014-09-22 10:19:33 +03:00
typedef struct MatroskaIndex {
2008-08-05 03:40:15 +03:00
uint64_t time ;
EbmlList pos ;
} MatroskaIndex ;
2014-09-22 10:19:33 +03:00
typedef struct MatroskaTag {
2008-08-09 02:50:38 +03:00
char * name ;
char * string ;
2009-02-15 18:05:37 +02:00
char * lang ;
uint64_t def ;
2008-08-09 02:50:38 +03:00
EbmlList sub ;
} MatroskaTag ;
2014-09-22 10:19:33 +03:00
typedef struct MatroskaTagTarget {
2009-02-15 17:53:55 +02:00
char * type ;
uint64_t typevalue ;
uint64_t trackuid ;
uint64_t chapteruid ;
uint64_t attachuid ;
} MatroskaTagTarget ;
2014-09-22 10:19:33 +03:00
typedef struct MatroskaTags {
2009-02-15 17:53:55 +02:00
MatroskaTagTarget target ;
EbmlList tag ;
} MatroskaTags ;
2014-09-22 10:19:33 +03:00
typedef struct MatroskaSeekhead {
2008-08-05 03:40:36 +03:00
uint64_t id ;
uint64_t pos ;
} MatroskaSeekhead ;
2014-09-22 10:19:33 +03:00
typedef struct MatroskaLevel {
2007-06-05 01:35:16 +03:00
uint64_t start ;
uint64_t length ;
2007-06-05 01:10:54 +03:00
} MatroskaLevel ;
2019-05-17 00:29:59 +02:00
typedef struct MatroskaBlock {
uint64_t duration ;
2021-02-16 17:18:02 +02:00
CountedElement reference ;
2019-05-17 00:29:59 +02:00
uint64_t non_simple ;
EbmlBin bin ;
uint64_t additional_id ;
EbmlBin additional ;
2019-05-17 00:30:13 +02:00
int64_t discard_padding ;
2019-05-17 00:29:59 +02:00
} MatroskaBlock ;
2014-09-22 10:19:33 +03:00
typedef struct MatroskaCluster {
2019-05-17 00:29:59 +02:00
MatroskaBlock block ;
2012-04-19 21:12:24 +03:00
uint64_t timecode ;
2019-05-17 00:29:59 +02:00
int64_t pos ;
2012-04-19 21:12:24 +03:00
} MatroskaCluster ;
2015-02-14 22:07:40 +02:00
typedef struct MatroskaLevel1Element {
2019-05-17 00:30:20 +02:00
int64_t pos ;
2019-05-17 00:29:47 +02:00
uint32_t id ;
2015-02-09 21:39:00 +02:00
int parsed ;
} MatroskaLevel1Element ;
2014-09-22 10:19:33 +03:00
typedef struct MatroskaDemuxContext {
2015-04-01 01:51:57 +02:00
const AVClass * class ;
2007-06-05 01:10:54 +03:00
AVFormatContext * ctx ;
2008-08-05 11:28:57 +03:00
/* EBML stuff */
2007-06-05 01:10:54 +03:00
MatroskaLevel levels [ EBML_MAX_DEPTH ] ;
2019-05-17 00:30:13 +02:00
int num_levels ;
2010-06-11 19:34:01 +03:00
uint32_t current_id ;
avformat/matroskadec: Introduce a "last known good" position
Currently, resyncing during reading packets works as follows:
The current position is recorded, then a call to matroska_parse_cluster
is made and if said call fails, the demuxer tries to resync from the
earlier position. If the call doesn't fail, but also doesn't deliver a
packet, then this is looped.
There are two problems with this approach:
1. The Matroska file format aims to be forward-compatible; to achieve
this, a demuxer should simply ignore and skip elements it doesn't
know about. But it is not possible to reliably distinguish unknown
elements from junk. If matroska_parse_cluster encounters an unknown
element, it can therefore not simply error out; instead it returns zero
and the loop is iterated which includes an update of the position that
is intended to be used in case of errors, i.e. the element that is
skipped is not searched for level 1 element ids to resync to at all if
later calls to matroska_parse_cluster return an error.
Notice that in case that sync has been lost there can be a chain of
several unknown/possibly junk elements before an error is detected.
2. Even if a call to matroska_parse_cluster delivers a packet, this does
not mean that everything is fine. E.g. it might be that some of the
block's data is missing and that the data that was presumed to be from
the block just read actually contains the beginning of the next element.
This will only be apparent at the next call of matroska_read_packet,
which uses the (false) end of the earlier block as resync position so
that in the (not unlikely) case that the call to matroska_parse_cluster
fails, the data believed to be part of the earlier block is not searched
for a level 1 element to resync to.
To counter this, a "last known good" position is introduced. When an
element id that is known to be allowed at this position in the hierarchy
(according to the syntax currently in use for parsing) is read and some
further checks (regarding the length of the element and its containing
master element) are passed, then the beginning of the current element is
treated as a "good" position and recorded as such in the
MatroskaDemuxContext. Because of 2., only the start of the element is
treated as a "good" position, not the whole element. If an error occurs
later during parsing of clusters, the resync process starts at the last
known good position.
Given that when the header is damaged the subsequent resync never skips over
data and is therefore unaffected by both issues, the "last known good"
concept is not used there.
Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@gmail.com>
2019-05-17 00:30:05 +02:00
int64_t resync_pos ;
2019-05-17 00:30:14 +02:00
int unknown_count ;
2007-06-05 01:10:54 +03:00
2008-08-05 03:40:27 +03:00
uint64_t time_scale ;
double duration ;
char * title ;
2013-11-24 11:31:48 +03:00
char * muxingapp ;
2019-05-17 00:30:13 +02:00
EbmlBin date_utc ;
2008-08-05 03:40:31 +03:00
EbmlList tracks ;
2008-08-05 03:40:24 +03:00
EbmlList attachments ;
2008-08-05 03:40:21 +03:00
EbmlList chapters ;
2008-08-05 03:40:15 +03:00
EbmlList index ;
2008-08-09 02:50:38 +03:00
EbmlList tags ;
2008-08-05 03:40:36 +03:00
EbmlList seekhead ;
2007-06-05 01:10:54 +03:00
/* byte position of the segment inside the stream */
2008-10-03 13:16:29 +03:00
int64_t segment_start ;
2007-06-05 01:10:54 +03:00
2021-03-18 07:04:17 +02:00
/* This packet coincides with AVFormatInternal.parse_pkt
* and is not owned by us . */
2021-01-29 15:42:48 +02:00
AVPacket * pkt ;
2008-08-05 11:28:57 +03:00
/* the packet queue */
2021-03-05 16:26:24 +02:00
PacketList * queue ;
PacketList * queue_end ;
2007-06-05 01:10:54 +03:00
2007-06-05 01:35:16 +03:00
int done ;
2007-06-05 01:10:54 +03:00
/* What to skip before effectively reading a packet. */
int skip_to_keyframe ;
2008-09-09 15:01:51 +03:00
uint64_t skip_to_timecode ;
2011-07-09 08:48:43 +03:00
/* File has a CUES element, but we defer parsing until it is needed. */
int cues_parsing_deferred ;
2012-04-19 21:12:24 +03:00
2015-02-09 21:39:00 +02:00
/* Level1 elements and whether they were read yet */
MatroskaLevel1Element level1_elems [ 64 ] ;
int num_level1_elems ;
2012-04-19 21:12:24 +03:00
MatroskaCluster current_cluster ;
2017-04-17 19:59:31 +02:00
/* WebM DASH Manifest live flag */
2015-04-01 01:51:57 +02:00
int is_live ;
2017-04-12 06:33:28 +02:00
/* Bandwidth value for WebM DASH Manifest */
int bandwidth ;
2007-06-05 01:10:54 +03:00
} MatroskaDemuxContext ;
2019-05-17 00:30:06 +02:00
# define CHILD_OF(parent) { .def = { .n = parent } }
2019-07-17 05:29:40 +02:00
// The following forward declarations need their size because
// a tentative definition with internal linkage must not be an
// incomplete type (6.7.2 in C90, 6.9.2 in C99).
// Removing the sizes breaks MSVC.
2019-07-18 21:07:20 +02:00
static EbmlSyntax ebml_syntax [ 3 ] , matroska_segment [ 9 ] , matroska_track_video_color [ 15 ] , matroska_track_video [ 19 ] ,
2021-02-16 19:41:30 +02:00
matroska_track [ 32 ] , matroska_track_encoding [ 6 ] , matroska_track_encodings [ 2 ] ,
2019-07-18 21:07:20 +02:00
matroska_track_combine_planes [ 2 ] , matroska_track_operation [ 2 ] , matroska_tracks [ 2 ] ,
matroska_attachments [ 2 ] , matroska_chapter_entry [ 9 ] , matroska_chapter [ 6 ] , matroska_chapters [ 2 ] ,
matroska_index_entry [ 3 ] , matroska_index [ 2 ] , matroska_tag [ 3 ] , matroska_tags [ 2 ] , matroska_seekhead [ 2 ] ,
matroska_blockadditions [ 2 ] , matroska_blockgroup [ 8 ] , matroska_cluster_parsing [ 8 ] ;
static EbmlSyntax ebml_header [ ] = {
avformat/matroskadec: Allow to count the number of element occurences
Up until now, the generic EBML reader used by the Matroska demuxer did
not have the capability to record whether an element was actually
present or not; instead, in cases where it mattered one typically added
an invalid default value and checked whether the value is valid (in
which case it is guaranteed to be present). This worked pretty well so
far, yet the EBML specifications have evolved: It is now legal to use
zero-length elements for floats, ints, uints and strings (both ASCII and
UTF-8); the value of these elements is the default value of the element
(if it has one) or zero for scalar types and an empty string for
strings. Furthermore, having a default value does no longer imply that
the element may be presumed to be present (with its default value) if it
is absent; this is only true if the element is mandatory, too.
These rules are designed to allow size savings as follows: Consider the
newly added FlagOriginal: It being zero means the track is not in its
original language, it being one means it is. For backward compatibility
reasons, neither of the two values may be inferred automatically in the
absence of the element. But one can still save a byte when one wants to
write the element with a value of zero, as one can write the integer with
a length of zero: 0x55AE 80 instead of 0x55AE 81 00. In the former case,
a parser has to infer the value of the element to be zero (which is the
element's default value).
When encountering an element with length zero, our parser always infers
a value of zero (or an empty string); this is wrong for values with
a different default value. It needs to infer the default value (or zero
in its absence) and this precludes using an invalid default value for
elements like FlagOriginal. Ergo one needs to be able to record whether
an element is present or not by other means. This patch allows to use a
simple counter for this. While just at it, some invalid and unnecessary
default values have been removed (mastering metadata elements used
default values of -1.0, despite these elements only being used if they
are > 0).
Reviewed-by: Ridley Combs <rcombs@rcombs.me>
Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@gmail.com>
2021-02-16 15:33:51 +02:00
{ EBML_ID_EBMLREADVERSION , EBML_UINT , 0 , 0 , offsetof ( Ebml , version ) , { . u = EBML_VERSION } } ,
{ EBML_ID_EBMLMAXSIZELENGTH , EBML_UINT , 0 , 0 , offsetof ( Ebml , max_size ) , { . u = 8 } } ,
{ EBML_ID_EBMLMAXIDLENGTH , EBML_UINT , 0 , 0 , offsetof ( Ebml , id_length ) , { . u = 4 } } ,
{ EBML_ID_DOCTYPE , EBML_STR , 0 , 0 , offsetof ( Ebml , doctype ) , { . s = " (none) " } } ,
{ EBML_ID_DOCTYPEREADVERSION , EBML_UINT , 0 , 0 , offsetof ( Ebml , doctype_version ) , { . u = 1 } } ,
2014-03-01 19:28:15 +03:00
{ EBML_ID_EBMLVERSION , EBML_NONE } ,
{ EBML_ID_DOCTYPEVERSION , EBML_NONE } ,
2019-05-17 00:30:06 +02:00
CHILD_OF ( ebml_syntax )
2008-08-05 03:40:02 +03:00
} ;
2019-07-18 21:07:20 +02:00
static EbmlSyntax ebml_syntax [ ] = {
avformat/matroskadec: Allow to count the number of element occurences
Up until now, the generic EBML reader used by the Matroska demuxer did
not have the capability to record whether an element was actually
present or not; instead, in cases where it mattered one typically added
an invalid default value and checked whether the value is valid (in
which case it is guaranteed to be present). This worked pretty well so
far, yet the EBML specifications have evolved: It is now legal to use
zero-length elements for floats, ints, uints and strings (both ASCII and
UTF-8); the value of these elements is the default value of the element
(if it has one) or zero for scalar types and an empty string for
strings. Furthermore, having a default value does no longer imply that
the element may be presumed to be present (with its default value) if it
is absent; this is only true if the element is mandatory, too.
These rules are designed to allow size savings as follows: Consider the
newly added FlagOriginal: It being zero means the track is not in its
original language, it being one means it is. For backward compatibility
reasons, neither of the two values may be inferred automatically in the
absence of the element. But one can still save a byte when one wants to
write the element with a value of zero, as one can write the integer with
a length of zero: 0x55AE 80 instead of 0x55AE 81 00. In the former case,
a parser has to infer the value of the element to be zero (which is the
element's default value).
When encountering an element with length zero, our parser always infers
a value of zero (or an empty string); this is wrong for values with
a different default value. It needs to infer the default value (or zero
in its absence) and this precludes using an invalid default value for
elements like FlagOriginal. Ergo one needs to be able to record whether
an element is present or not by other means. This patch allows to use a
simple counter for this. While just at it, some invalid and unnecessary
default values have been removed (mastering metadata elements used
default values of -1.0, despite these elements only being used if they
are > 0).
Reviewed-by: Ridley Combs <rcombs@rcombs.me>
Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@gmail.com>
2021-02-16 15:33:51 +02:00
{ EBML_ID_HEADER , EBML_NEST , 0 , 0 , 0 , { . n = ebml_header } } ,
2019-05-17 00:30:06 +02:00
{ MATROSKA_ID_SEGMENT , EBML_STOP } ,
2008-08-05 03:40:02 +03:00
{ 0 }
} ;
2019-07-18 21:07:20 +02:00
static EbmlSyntax matroska_info [ ] = {
avformat/matroskadec: Allow to count the number of element occurences
Up until now, the generic EBML reader used by the Matroska demuxer did
not have the capability to record whether an element was actually
present or not; instead, in cases where it mattered one typically added
an invalid default value and checked whether the value is valid (in
which case it is guaranteed to be present). This worked pretty well so
far, yet the EBML specifications have evolved: It is now legal to use
zero-length elements for floats, ints, uints and strings (both ASCII and
UTF-8); the value of these elements is the default value of the element
(if it has one) or zero for scalar types and an empty string for
strings. Furthermore, having a default value does no longer imply that
the element may be presumed to be present (with its default value) if it
is absent; this is only true if the element is mandatory, too.
These rules are designed to allow size savings as follows: Consider the
newly added FlagOriginal: It being zero means the track is not in its
original language, it being one means it is. For backward compatibility
reasons, neither of the two values may be inferred automatically in the
absence of the element. But one can still save a byte when one wants to
write the element with a value of zero, as one can write the integer with
a length of zero: 0x55AE 80 instead of 0x55AE 81 00. In the former case,
a parser has to infer the value of the element to be zero (which is the
element's default value).
When encountering an element with length zero, our parser always infers
a value of zero (or an empty string); this is wrong for values with
a different default value. It needs to infer the default value (or zero
in its absence) and this precludes using an invalid default value for
elements like FlagOriginal. Ergo one needs to be able to record whether
an element is present or not by other means. This patch allows to use a
simple counter for this. While just at it, some invalid and unnecessary
default values have been removed (mastering metadata elements used
default values of -1.0, despite these elements only being used if they
are > 0).
Reviewed-by: Ridley Combs <rcombs@rcombs.me>
Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@gmail.com>
2021-02-16 15:33:51 +02:00
{ MATROSKA_ID_TIMECODESCALE , EBML_UINT , 0 , 0 , offsetof ( MatroskaDemuxContext , time_scale ) , { . u = 1000000 } } ,
{ MATROSKA_ID_DURATION , EBML_FLOAT , 0 , 0 , offsetof ( MatroskaDemuxContext , duration ) } ,
{ MATROSKA_ID_TITLE , EBML_UTF8 , 0 , 0 , offsetof ( MatroskaDemuxContext , title ) } ,
2014-03-01 19:28:15 +03:00
{ MATROSKA_ID_WRITINGAPP , EBML_NONE } ,
avformat/matroskadec: Allow to count the number of element occurences
Up until now, the generic EBML reader used by the Matroska demuxer did
not have the capability to record whether an element was actually
present or not; instead, in cases where it mattered one typically added
an invalid default value and checked whether the value is valid (in
which case it is guaranteed to be present). This worked pretty well so
far, yet the EBML specifications have evolved: It is now legal to use
zero-length elements for floats, ints, uints and strings (both ASCII and
UTF-8); the value of these elements is the default value of the element
(if it has one) or zero for scalar types and an empty string for
strings. Furthermore, having a default value does no longer imply that
the element may be presumed to be present (with its default value) if it
is absent; this is only true if the element is mandatory, too.
These rules are designed to allow size savings as follows: Consider the
newly added FlagOriginal: It being zero means the track is not in its
original language, it being one means it is. For backward compatibility
reasons, neither of the two values may be inferred automatically in the
absence of the element. But one can still save a byte when one wants to
write the element with a value of zero, as one can write the integer with
a length of zero: 0x55AE 80 instead of 0x55AE 81 00. In the former case,
a parser has to infer the value of the element to be zero (which is the
element's default value).
When encountering an element with length zero, our parser always infers
a value of zero (or an empty string); this is wrong for values with
a different default value. It needs to infer the default value (or zero
in its absence) and this precludes using an invalid default value for
elements like FlagOriginal. Ergo one needs to be able to record whether
an element is present or not by other means. This patch allows to use a
simple counter for this. While just at it, some invalid and unnecessary
default values have been removed (mastering metadata elements used
default values of -1.0, despite these elements only being used if they
are > 0).
Reviewed-by: Ridley Combs <rcombs@rcombs.me>
Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@gmail.com>
2021-02-16 15:33:51 +02:00
{ MATROSKA_ID_MUXINGAPP , EBML_UTF8 , 0 , 0 , offsetof ( MatroskaDemuxContext , muxingapp ) } ,
{ MATROSKA_ID_DATEUTC , EBML_BIN , 0 , 0 , offsetof ( MatroskaDemuxContext , date_utc ) } ,
2014-03-01 19:28:15 +03:00
{ MATROSKA_ID_SEGMENTUID , EBML_NONE } ,
2019-05-17 00:30:06 +02:00
CHILD_OF ( matroska_segment )
2008-08-05 03:40:27 +03:00
} ;
2019-07-18 21:07:20 +02:00
static EbmlSyntax matroska_mastering_meta [ ] = {
avformat/matroskadec: Allow to count the number of element occurences
Up until now, the generic EBML reader used by the Matroska demuxer did
not have the capability to record whether an element was actually
present or not; instead, in cases where it mattered one typically added
an invalid default value and checked whether the value is valid (in
which case it is guaranteed to be present). This worked pretty well so
far, yet the EBML specifications have evolved: It is now legal to use
zero-length elements for floats, ints, uints and strings (both ASCII and
UTF-8); the value of these elements is the default value of the element
(if it has one) or zero for scalar types and an empty string for
strings. Furthermore, having a default value does no longer imply that
the element may be presumed to be present (with its default value) if it
is absent; this is only true if the element is mandatory, too.
These rules are designed to allow size savings as follows: Consider the
newly added FlagOriginal: It being zero means the track is not in its
original language, it being one means it is. For backward compatibility
reasons, neither of the two values may be inferred automatically in the
absence of the element. But one can still save a byte when one wants to
write the element with a value of zero, as one can write the integer with
a length of zero: 0x55AE 80 instead of 0x55AE 81 00. In the former case,
a parser has to infer the value of the element to be zero (which is the
element's default value).
When encountering an element with length zero, our parser always infers
a value of zero (or an empty string); this is wrong for values with
a different default value. It needs to infer the default value (or zero
in its absence) and this precludes using an invalid default value for
elements like FlagOriginal. Ergo one needs to be able to record whether
an element is present or not by other means. This patch allows to use a
simple counter for this. While just at it, some invalid and unnecessary
default values have been removed (mastering metadata elements used
default values of -1.0, despite these elements only being used if they
are > 0).
Reviewed-by: Ridley Combs <rcombs@rcombs.me>
Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@gmail.com>
2021-02-16 15:33:51 +02:00
{ MATROSKA_ID_VIDEOCOLOR_RX , EBML_FLOAT , 0 , 0 , offsetof ( MatroskaMasteringMeta , r_x ) } ,
{ MATROSKA_ID_VIDEOCOLOR_RY , EBML_FLOAT , 0 , 0 , offsetof ( MatroskaMasteringMeta , r_y ) } ,
{ MATROSKA_ID_VIDEOCOLOR_GX , EBML_FLOAT , 0 , 0 , offsetof ( MatroskaMasteringMeta , g_x ) } ,
{ MATROSKA_ID_VIDEOCOLOR_GY , EBML_FLOAT , 0 , 0 , offsetof ( MatroskaMasteringMeta , g_y ) } ,
{ MATROSKA_ID_VIDEOCOLOR_BX , EBML_FLOAT , 0 , 0 , offsetof ( MatroskaMasteringMeta , b_x ) } ,
{ MATROSKA_ID_VIDEOCOLOR_BY , EBML_FLOAT , 0 , 0 , offsetof ( MatroskaMasteringMeta , b_y ) } ,
{ MATROSKA_ID_VIDEOCOLOR_WHITEX , EBML_FLOAT , 0 , 0 , offsetof ( MatroskaMasteringMeta , white_x ) } ,
{ MATROSKA_ID_VIDEOCOLOR_WHITEY , EBML_FLOAT , 0 , 0 , offsetof ( MatroskaMasteringMeta , white_y ) } ,
2021-02-16 16:34:06 +02:00
{ MATROSKA_ID_VIDEOCOLOR_LUMINANCEMIN , EBML_FLOAT , 1 , 0 , offsetof ( MatroskaMasteringMeta , min_luminance ) } ,
avformat/matroskadec: Allow to count the number of element occurences
Up until now, the generic EBML reader used by the Matroska demuxer did
not have the capability to record whether an element was actually
present or not; instead, in cases where it mattered one typically added
an invalid default value and checked whether the value is valid (in
which case it is guaranteed to be present). This worked pretty well so
far, yet the EBML specifications have evolved: It is now legal to use
zero-length elements for floats, ints, uints and strings (both ASCII and
UTF-8); the value of these elements is the default value of the element
(if it has one) or zero for scalar types and an empty string for
strings. Furthermore, having a default value does no longer imply that
the element may be presumed to be present (with its default value) if it
is absent; this is only true if the element is mandatory, too.
These rules are designed to allow size savings as follows: Consider the
newly added FlagOriginal: It being zero means the track is not in its
original language, it being one means it is. For backward compatibility
reasons, neither of the two values may be inferred automatically in the
absence of the element. But one can still save a byte when one wants to
write the element with a value of zero, as one can write the integer with
a length of zero: 0x55AE 80 instead of 0x55AE 81 00. In the former case,
a parser has to infer the value of the element to be zero (which is the
element's default value).
When encountering an element with length zero, our parser always infers
a value of zero (or an empty string); this is wrong for values with
a different default value. It needs to infer the default value (or zero
in its absence) and this precludes using an invalid default value for
elements like FlagOriginal. Ergo one needs to be able to record whether
an element is present or not by other means. This patch allows to use a
simple counter for this. While just at it, some invalid and unnecessary
default values have been removed (mastering metadata elements used
default values of -1.0, despite these elements only being used if they
are > 0).
Reviewed-by: Ridley Combs <rcombs@rcombs.me>
Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@gmail.com>
2021-02-16 15:33:51 +02:00
{ MATROSKA_ID_VIDEOCOLOR_LUMINANCEMAX , EBML_FLOAT , 0 , 0 , offsetof ( MatroskaMasteringMeta , max_luminance ) } ,
2019-05-17 00:30:06 +02:00
CHILD_OF ( matroska_track_video_color )
2016-03-01 02:41:04 +02:00
} ;
2019-07-18 21:07:20 +02:00
static EbmlSyntax matroska_track_video_color [ ] = {
avformat/matroskadec: Allow to count the number of element occurences
Up until now, the generic EBML reader used by the Matroska demuxer did
not have the capability to record whether an element was actually
present or not; instead, in cases where it mattered one typically added
an invalid default value and checked whether the value is valid (in
which case it is guaranteed to be present). This worked pretty well so
far, yet the EBML specifications have evolved: It is now legal to use
zero-length elements for floats, ints, uints and strings (both ASCII and
UTF-8); the value of these elements is the default value of the element
(if it has one) or zero for scalar types and an empty string for
strings. Furthermore, having a default value does no longer imply that
the element may be presumed to be present (with its default value) if it
is absent; this is only true if the element is mandatory, too.
These rules are designed to allow size savings as follows: Consider the
newly added FlagOriginal: It being zero means the track is not in its
original language, it being one means it is. For backward compatibility
reasons, neither of the two values may be inferred automatically in the
absence of the element. But one can still save a byte when one wants to
write the element with a value of zero, as one can write the integer with
a length of zero: 0x55AE 80 instead of 0x55AE 81 00. In the former case,
a parser has to infer the value of the element to be zero (which is the
element's default value).
When encountering an element with length zero, our parser always infers
a value of zero (or an empty string); this is wrong for values with
a different default value. It needs to infer the default value (or zero
in its absence) and this precludes using an invalid default value for
elements like FlagOriginal. Ergo one needs to be able to record whether
an element is present or not by other means. This patch allows to use a
simple counter for this. While just at it, some invalid and unnecessary
default values have been removed (mastering metadata elements used
default values of -1.0, despite these elements only being used if they
are > 0).
Reviewed-by: Ridley Combs <rcombs@rcombs.me>
Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@gmail.com>
2021-02-16 15:33:51 +02:00
{ MATROSKA_ID_VIDEOCOLORMATRIXCOEFF , EBML_UINT , 0 , 0 , offsetof ( MatroskaTrackVideoColor , matrix_coefficients ) , { . u = AVCOL_SPC_UNSPECIFIED } } ,
{ MATROSKA_ID_VIDEOCOLORBITSPERCHANNEL , EBML_UINT , 0 , 0 , offsetof ( MatroskaTrackVideoColor , bits_per_channel ) , { . u = 0 } } ,
{ MATROSKA_ID_VIDEOCOLORCHROMASUBHORZ , EBML_UINT , 0 , 0 , offsetof ( MatroskaTrackVideoColor , chroma_sub_horz ) } ,
{ MATROSKA_ID_VIDEOCOLORCHROMASUBVERT , EBML_UINT , 0 , 0 , offsetof ( MatroskaTrackVideoColor , chroma_sub_vert ) } ,
{ MATROSKA_ID_VIDEOCOLORCBSUBHORZ , EBML_UINT , 0 , 0 , offsetof ( MatroskaTrackVideoColor , cb_sub_horz ) } ,
{ MATROSKA_ID_VIDEOCOLORCBSUBVERT , EBML_UINT , 0 , 0 , offsetof ( MatroskaTrackVideoColor , cb_sub_vert ) } ,
{ MATROSKA_ID_VIDEOCOLORCHROMASITINGHORZ , EBML_UINT , 0 , 0 , offsetof ( MatroskaTrackVideoColor , chroma_siting_horz ) , { . u = MATROSKA_COLOUR_CHROMASITINGHORZ_UNDETERMINED } } ,
{ MATROSKA_ID_VIDEOCOLORCHROMASITINGVERT , EBML_UINT , 0 , 0 , offsetof ( MatroskaTrackVideoColor , chroma_siting_vert ) , { . u = MATROSKA_COLOUR_CHROMASITINGVERT_UNDETERMINED } } ,
{ MATROSKA_ID_VIDEOCOLORRANGE , EBML_UINT , 0 , 0 , offsetof ( MatroskaTrackVideoColor , range ) , { . u = AVCOL_RANGE_UNSPECIFIED } } ,
{ MATROSKA_ID_VIDEOCOLORTRANSFERCHARACTERISTICS , EBML_UINT , 0 , 0 , offsetof ( MatroskaTrackVideoColor , transfer_characteristics ) , { . u = AVCOL_TRC_UNSPECIFIED } } ,
{ MATROSKA_ID_VIDEOCOLORPRIMARIES , EBML_UINT , 0 , 0 , offsetof ( MatroskaTrackVideoColor , primaries ) , { . u = AVCOL_PRI_UNSPECIFIED } } ,
{ MATROSKA_ID_VIDEOCOLORMAXCLL , EBML_UINT , 0 , 0 , offsetof ( MatroskaTrackVideoColor , max_cll ) } ,
{ MATROSKA_ID_VIDEOCOLORMAXFALL , EBML_UINT , 0 , 0 , offsetof ( MatroskaTrackVideoColor , max_fall ) } ,
{ MATROSKA_ID_VIDEOCOLORMASTERINGMETA , EBML_NEST , 0 , 0 , offsetof ( MatroskaTrackVideoColor , mastering_meta ) , { . n = matroska_mastering_meta } } ,
2019-05-17 00:30:06 +02:00
CHILD_OF ( matroska_track_video )
2016-03-01 02:41:04 +02:00
} ;
2019-07-18 21:07:20 +02:00
static EbmlSyntax matroska_track_video_projection [ ] = {
avformat/matroskadec: Allow to count the number of element occurences
Up until now, the generic EBML reader used by the Matroska demuxer did
not have the capability to record whether an element was actually
present or not; instead, in cases where it mattered one typically added
an invalid default value and checked whether the value is valid (in
which case it is guaranteed to be present). This worked pretty well so
far, yet the EBML specifications have evolved: It is now legal to use
zero-length elements for floats, ints, uints and strings (both ASCII and
UTF-8); the value of these elements is the default value of the element
(if it has one) or zero for scalar types and an empty string for
strings. Furthermore, having a default value does no longer imply that
the element may be presumed to be present (with its default value) if it
is absent; this is only true if the element is mandatory, too.
These rules are designed to allow size savings as follows: Consider the
newly added FlagOriginal: It being zero means the track is not in its
original language, it being one means it is. For backward compatibility
reasons, neither of the two values may be inferred automatically in the
absence of the element. But one can still save a byte when one wants to
write the element with a value of zero, as one can write the integer with
a length of zero: 0x55AE 80 instead of 0x55AE 81 00. In the former case,
a parser has to infer the value of the element to be zero (which is the
element's default value).
When encountering an element with length zero, our parser always infers
a value of zero (or an empty string); this is wrong for values with
a different default value. It needs to infer the default value (or zero
in its absence) and this precludes using an invalid default value for
elements like FlagOriginal. Ergo one needs to be able to record whether
an element is present or not by other means. This patch allows to use a
simple counter for this. While just at it, some invalid and unnecessary
default values have been removed (mastering metadata elements used
default values of -1.0, despite these elements only being used if they
are > 0).
Reviewed-by: Ridley Combs <rcombs@rcombs.me>
Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@gmail.com>
2021-02-16 15:33:51 +02:00
{ MATROSKA_ID_VIDEOPROJECTIONTYPE , EBML_UINT , 0 , 0 , offsetof ( MatroskaTrackVideoProjection , type ) , { . u = MATROSKA_VIDEO_PROJECTION_TYPE_RECTANGULAR } } ,
{ MATROSKA_ID_VIDEOPROJECTIONPRIVATE , EBML_BIN , 0 , 0 , offsetof ( MatroskaTrackVideoProjection , private ) } ,
{ MATROSKA_ID_VIDEOPROJECTIONPOSEYAW , EBML_FLOAT , 0 , 0 , offsetof ( MatroskaTrackVideoProjection , yaw ) , { . f = 0.0 } } ,
{ MATROSKA_ID_VIDEOPROJECTIONPOSEPITCH , EBML_FLOAT , 0 , 0 , offsetof ( MatroskaTrackVideoProjection , pitch ) , { . f = 0.0 } } ,
{ MATROSKA_ID_VIDEOPROJECTIONPOSEROLL , EBML_FLOAT , 0 , 0 , offsetof ( MatroskaTrackVideoProjection , roll ) , { . f = 0.0 } } ,
2019-05-17 00:30:06 +02:00
CHILD_OF ( matroska_track_video )
2016-12-06 19:48:45 +02:00
} ;
2019-07-18 21:07:20 +02:00
static EbmlSyntax matroska_track_video [ ] = {
avformat/matroskadec: Allow to count the number of element occurences
Up until now, the generic EBML reader used by the Matroska demuxer did
not have the capability to record whether an element was actually
present or not; instead, in cases where it mattered one typically added
an invalid default value and checked whether the value is valid (in
which case it is guaranteed to be present). This worked pretty well so
far, yet the EBML specifications have evolved: It is now legal to use
zero-length elements for floats, ints, uints and strings (both ASCII and
UTF-8); the value of these elements is the default value of the element
(if it has one) or zero for scalar types and an empty string for
strings. Furthermore, having a default value does no longer imply that
the element may be presumed to be present (with its default value) if it
is absent; this is only true if the element is mandatory, too.
These rules are designed to allow size savings as follows: Consider the
newly added FlagOriginal: It being zero means the track is not in its
original language, it being one means it is. For backward compatibility
reasons, neither of the two values may be inferred automatically in the
absence of the element. But one can still save a byte when one wants to
write the element with a value of zero, as one can write the integer with
a length of zero: 0x55AE 80 instead of 0x55AE 81 00. In the former case,
a parser has to infer the value of the element to be zero (which is the
element's default value).
When encountering an element with length zero, our parser always infers
a value of zero (or an empty string); this is wrong for values with
a different default value. It needs to infer the default value (or zero
in its absence) and this precludes using an invalid default value for
elements like FlagOriginal. Ergo one needs to be able to record whether
an element is present or not by other means. This patch allows to use a
simple counter for this. While just at it, some invalid and unnecessary
default values have been removed (mastering metadata elements used
default values of -1.0, despite these elements only being used if they
are > 0).
Reviewed-by: Ridley Combs <rcombs@rcombs.me>
Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@gmail.com>
2021-02-16 15:33:51 +02:00
{ MATROSKA_ID_VIDEOFRAMERATE , EBML_FLOAT , 0 , 0 , offsetof ( MatroskaTrackVideo , frame_rate ) } ,
{ MATROSKA_ID_VIDEODISPLAYWIDTH , EBML_UINT , 0 , 0 , offsetof ( MatroskaTrackVideo , display_width ) , { . u = - 1 } } ,
{ MATROSKA_ID_VIDEODISPLAYHEIGHT , EBML_UINT , 0 , 0 , offsetof ( MatroskaTrackVideo , display_height ) , { . u = - 1 } } ,
{ MATROSKA_ID_VIDEOPIXELWIDTH , EBML_UINT , 0 , 0 , offsetof ( MatroskaTrackVideo , pixel_width ) } ,
{ MATROSKA_ID_VIDEOPIXELHEIGHT , EBML_UINT , 0 , 0 , offsetof ( MatroskaTrackVideo , pixel_height ) } ,
{ MATROSKA_ID_VIDEOCOLORSPACE , EBML_BIN , 0 , 0 , offsetof ( MatroskaTrackVideo , color_space ) } ,
{ MATROSKA_ID_VIDEOALPHAMODE , EBML_UINT , 0 , 0 , offsetof ( MatroskaTrackVideo , alpha_mode ) , { . u = 0 } } ,
{ MATROSKA_ID_VIDEOCOLOR , EBML_NEST , 0 , sizeof ( MatroskaTrackVideoColor ) , offsetof ( MatroskaTrackVideo , color ) , { . n = matroska_track_video_color } } ,
{ MATROSKA_ID_VIDEOPROJECTION , EBML_NEST , 0 , 0 , offsetof ( MatroskaTrackVideo , projection ) , { . n = matroska_track_video_projection } } ,
2014-03-01 19:28:15 +03:00
{ MATROSKA_ID_VIDEOPIXELCROPB , EBML_NONE } ,
{ MATROSKA_ID_VIDEOPIXELCROPT , EBML_NONE } ,
{ MATROSKA_ID_VIDEOPIXELCROPL , EBML_NONE } ,
{ MATROSKA_ID_VIDEOPIXELCROPR , EBML_NONE } ,
avformat/matroskadec: Allow to count the number of element occurences
Up until now, the generic EBML reader used by the Matroska demuxer did
not have the capability to record whether an element was actually
present or not; instead, in cases where it mattered one typically added
an invalid default value and checked whether the value is valid (in
which case it is guaranteed to be present). This worked pretty well so
far, yet the EBML specifications have evolved: It is now legal to use
zero-length elements for floats, ints, uints and strings (both ASCII and
UTF-8); the value of these elements is the default value of the element
(if it has one) or zero for scalar types and an empty string for
strings. Furthermore, having a default value does no longer imply that
the element may be presumed to be present (with its default value) if it
is absent; this is only true if the element is mandatory, too.
These rules are designed to allow size savings as follows: Consider the
newly added FlagOriginal: It being zero means the track is not in its
original language, it being one means it is. For backward compatibility
reasons, neither of the two values may be inferred automatically in the
absence of the element. But one can still save a byte when one wants to
write the element with a value of zero, as one can write the integer with
a length of zero: 0x55AE 80 instead of 0x55AE 81 00. In the former case,
a parser has to infer the value of the element to be zero (which is the
element's default value).
When encountering an element with length zero, our parser always infers
a value of zero (or an empty string); this is wrong for values with
a different default value. It needs to infer the default value (or zero
in its absence) and this precludes using an invalid default value for
elements like FlagOriginal. Ergo one needs to be able to record whether
an element is present or not by other means. This patch allows to use a
simple counter for this. While just at it, some invalid and unnecessary
default values have been removed (mastering metadata elements used
default values of -1.0, despite these elements only being used if they
are > 0).
Reviewed-by: Ridley Combs <rcombs@rcombs.me>
Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@gmail.com>
2021-02-16 15:33:51 +02:00
{ MATROSKA_ID_VIDEODISPLAYUNIT , EBML_UINT , 0 , 0 , offsetof ( MatroskaTrackVideo , display_unit ) , { . u = MATROSKA_VIDEO_DISPLAYUNIT_PIXELS } } ,
{ MATROSKA_ID_VIDEOFLAGINTERLACED , EBML_UINT , 0 , 0 , offsetof ( MatroskaTrackVideo , interlaced ) , { . u = MATROSKA_VIDEO_INTERLACE_FLAG_UNDETERMINED } } ,
{ MATROSKA_ID_VIDEOFIELDORDER , EBML_UINT , 0 , 0 , offsetof ( MatroskaTrackVideo , field_order ) , { . u = MATROSKA_VIDEO_FIELDORDER_UNDETERMINED } } ,
{ MATROSKA_ID_VIDEOSTEREOMODE , EBML_UINT , 0 , 0 , offsetof ( MatroskaTrackVideo , stereo_mode ) , { . u = MATROSKA_VIDEO_STEREOMODE_TYPE_NB } } ,
2014-03-01 19:28:15 +03:00
{ MATROSKA_ID_VIDEOASPECTRATIO , EBML_NONE } ,
2019-05-17 00:30:06 +02:00
CHILD_OF ( matroska_track )
2008-08-05 03:40:31 +03:00
} ;
2019-07-18 21:07:20 +02:00
static EbmlSyntax matroska_track_audio [ ] = {
avformat/matroskadec: Allow to count the number of element occurences
Up until now, the generic EBML reader used by the Matroska demuxer did
not have the capability to record whether an element was actually
present or not; instead, in cases where it mattered one typically added
an invalid default value and checked whether the value is valid (in
which case it is guaranteed to be present). This worked pretty well so
far, yet the EBML specifications have evolved: It is now legal to use
zero-length elements for floats, ints, uints and strings (both ASCII and
UTF-8); the value of these elements is the default value of the element
(if it has one) or zero for scalar types and an empty string for
strings. Furthermore, having a default value does no longer imply that
the element may be presumed to be present (with its default value) if it
is absent; this is only true if the element is mandatory, too.
These rules are designed to allow size savings as follows: Consider the
newly added FlagOriginal: It being zero means the track is not in its
original language, it being one means it is. For backward compatibility
reasons, neither of the two values may be inferred automatically in the
absence of the element. But one can still save a byte when one wants to
write the element with a value of zero, as one can write the integer with
a length of zero: 0x55AE 80 instead of 0x55AE 81 00. In the former case,
a parser has to infer the value of the element to be zero (which is the
element's default value).
When encountering an element with length zero, our parser always infers
a value of zero (or an empty string); this is wrong for values with
a different default value. It needs to infer the default value (or zero
in its absence) and this precludes using an invalid default value for
elements like FlagOriginal. Ergo one needs to be able to record whether
an element is present or not by other means. This patch allows to use a
simple counter for this. While just at it, some invalid and unnecessary
default values have been removed (mastering metadata elements used
default values of -1.0, despite these elements only being used if they
are > 0).
Reviewed-by: Ridley Combs <rcombs@rcombs.me>
Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@gmail.com>
2021-02-16 15:33:51 +02:00
{ MATROSKA_ID_AUDIOSAMPLINGFREQ , EBML_FLOAT , 0 , 0 , offsetof ( MatroskaTrackAudio , samplerate ) , { . f = 8000.0 } } ,
{ MATROSKA_ID_AUDIOOUTSAMPLINGFREQ , EBML_FLOAT , 0 , 0 , offsetof ( MatroskaTrackAudio , out_samplerate ) } ,
{ MATROSKA_ID_AUDIOBITDEPTH , EBML_UINT , 0 , 0 , offsetof ( MatroskaTrackAudio , bitdepth ) } ,
{ MATROSKA_ID_AUDIOCHANNELS , EBML_UINT , 0 , 0 , offsetof ( MatroskaTrackAudio , channels ) , { . u = 1 } } ,
2019-05-17 00:30:06 +02:00
CHILD_OF ( matroska_track )
2008-08-05 03:40:31 +03:00
} ;
2019-07-18 21:07:20 +02:00
static EbmlSyntax matroska_track_encoding_compression [ ] = {
avformat/matroskadec: Allow to count the number of element occurences
Up until now, the generic EBML reader used by the Matroska demuxer did
not have the capability to record whether an element was actually
present or not; instead, in cases where it mattered one typically added
an invalid default value and checked whether the value is valid (in
which case it is guaranteed to be present). This worked pretty well so
far, yet the EBML specifications have evolved: It is now legal to use
zero-length elements for floats, ints, uints and strings (both ASCII and
UTF-8); the value of these elements is the default value of the element
(if it has one) or zero for scalar types and an empty string for
strings. Furthermore, having a default value does no longer imply that
the element may be presumed to be present (with its default value) if it
is absent; this is only true if the element is mandatory, too.
These rules are designed to allow size savings as follows: Consider the
newly added FlagOriginal: It being zero means the track is not in its
original language, it being one means it is. For backward compatibility
reasons, neither of the two values may be inferred automatically in the
absence of the element. But one can still save a byte when one wants to
write the element with a value of zero, as one can write the integer with
a length of zero: 0x55AE 80 instead of 0x55AE 81 00. In the former case,
a parser has to infer the value of the element to be zero (which is the
element's default value).
When encountering an element with length zero, our parser always infers
a value of zero (or an empty string); this is wrong for values with
a different default value. It needs to infer the default value (or zero
in its absence) and this precludes using an invalid default value for
elements like FlagOriginal. Ergo one needs to be able to record whether
an element is present or not by other means. This patch allows to use a
simple counter for this. While just at it, some invalid and unnecessary
default values have been removed (mastering metadata elements used
default values of -1.0, despite these elements only being used if they
are > 0).
Reviewed-by: Ridley Combs <rcombs@rcombs.me>
Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@gmail.com>
2021-02-16 15:33:51 +02:00
{ MATROSKA_ID_ENCODINGCOMPALGO , EBML_UINT , 0 , 0 , offsetof ( MatroskaTrackCompression , algo ) , { . u = MATROSKA_TRACK_ENCODING_COMP_ZLIB } } ,
{ MATROSKA_ID_ENCODINGCOMPSETTINGS , EBML_BIN , 0 , 0 , offsetof ( MatroskaTrackCompression , settings ) } ,
2019-05-17 00:30:06 +02:00
CHILD_OF ( matroska_track_encoding )
2008-08-05 03:40:31 +03:00
} ;
2019-07-18 21:07:20 +02:00
static EbmlSyntax matroska_track_encoding_encryption [ ] = {
avformat/matroskadec: Allow to count the number of element occurences
Up until now, the generic EBML reader used by the Matroska demuxer did
not have the capability to record whether an element was actually
present or not; instead, in cases where it mattered one typically added
an invalid default value and checked whether the value is valid (in
which case it is guaranteed to be present). This worked pretty well so
far, yet the EBML specifications have evolved: It is now legal to use
zero-length elements for floats, ints, uints and strings (both ASCII and
UTF-8); the value of these elements is the default value of the element
(if it has one) or zero for scalar types and an empty string for
strings. Furthermore, having a default value does no longer imply that
the element may be presumed to be present (with its default value) if it
is absent; this is only true if the element is mandatory, too.
These rules are designed to allow size savings as follows: Consider the
newly added FlagOriginal: It being zero means the track is not in its
original language, it being one means it is. For backward compatibility
reasons, neither of the two values may be inferred automatically in the
absence of the element. But one can still save a byte when one wants to
write the element with a value of zero, as one can write the integer with
a length of zero: 0x55AE 80 instead of 0x55AE 81 00. In the former case,
a parser has to infer the value of the element to be zero (which is the
element's default value).
When encountering an element with length zero, our parser always infers
a value of zero (or an empty string); this is wrong for values with
a different default value. It needs to infer the default value (or zero
in its absence) and this precludes using an invalid default value for
elements like FlagOriginal. Ergo one needs to be able to record whether
an element is present or not by other means. This patch allows to use a
simple counter for this. While just at it, some invalid and unnecessary
default values have been removed (mastering metadata elements used
default values of -1.0, despite these elements only being used if they
are > 0).
Reviewed-by: Ridley Combs <rcombs@rcombs.me>
Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@gmail.com>
2021-02-16 15:33:51 +02:00
{ MATROSKA_ID_ENCODINGENCALGO , EBML_UINT , 0 , 0 , offsetof ( MatroskaTrackEncryption , algo ) , { . u = 0 } } ,
{ MATROSKA_ID_ENCODINGENCKEYID , EBML_BIN , 0 , 0 , offsetof ( MatroskaTrackEncryption , key_id ) } ,
2013-03-07 19:11:38 +03:00
{ MATROSKA_ID_ENCODINGENCAESSETTINGS , EBML_NONE } ,
{ MATROSKA_ID_ENCODINGSIGALGO , EBML_NONE } ,
{ MATROSKA_ID_ENCODINGSIGHASHALGO , EBML_NONE } ,
{ MATROSKA_ID_ENCODINGSIGKEYID , EBML_NONE } ,
{ MATROSKA_ID_ENCODINGSIGNATURE , EBML_NONE } ,
2019-05-17 00:30:06 +02:00
CHILD_OF ( matroska_track_encoding )
2013-03-07 19:11:38 +03:00
} ;
2019-07-18 21:07:20 +02:00
static EbmlSyntax matroska_track_encoding [ ] = {
avformat/matroskadec: Allow to count the number of element occurences
Up until now, the generic EBML reader used by the Matroska demuxer did
not have the capability to record whether an element was actually
present or not; instead, in cases where it mattered one typically added
an invalid default value and checked whether the value is valid (in
which case it is guaranteed to be present). This worked pretty well so
far, yet the EBML specifications have evolved: It is now legal to use
zero-length elements for floats, ints, uints and strings (both ASCII and
UTF-8); the value of these elements is the default value of the element
(if it has one) or zero for scalar types and an empty string for
strings. Furthermore, having a default value does no longer imply that
the element may be presumed to be present (with its default value) if it
is absent; this is only true if the element is mandatory, too.
These rules are designed to allow size savings as follows: Consider the
newly added FlagOriginal: It being zero means the track is not in its
original language, it being one means it is. For backward compatibility
reasons, neither of the two values may be inferred automatically in the
absence of the element. But one can still save a byte when one wants to
write the element with a value of zero, as one can write the integer with
a length of zero: 0x55AE 80 instead of 0x55AE 81 00. In the former case,
a parser has to infer the value of the element to be zero (which is the
element's default value).
When encountering an element with length zero, our parser always infers
a value of zero (or an empty string); this is wrong for values with
a different default value. It needs to infer the default value (or zero
in its absence) and this precludes using an invalid default value for
elements like FlagOriginal. Ergo one needs to be able to record whether
an element is present or not by other means. This patch allows to use a
simple counter for this. While just at it, some invalid and unnecessary
default values have been removed (mastering metadata elements used
default values of -1.0, despite these elements only being used if they
are > 0).
Reviewed-by: Ridley Combs <rcombs@rcombs.me>
Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@gmail.com>
2021-02-16 15:33:51 +02:00
{ MATROSKA_ID_ENCODINGSCOPE , EBML_UINT , 0 , 0 , offsetof ( MatroskaTrackEncoding , scope ) , { . u = 1 } } ,
{ MATROSKA_ID_ENCODINGTYPE , EBML_UINT , 0 , 0 , offsetof ( MatroskaTrackEncoding , type ) , { . u = 0 } } ,
{ MATROSKA_ID_ENCODINGCOMPRESSION , EBML_NEST , 0 , 0 , offsetof ( MatroskaTrackEncoding , compression ) , { . n = matroska_track_encoding_compression } } ,
{ MATROSKA_ID_ENCODINGENCRYPTION , EBML_NEST , 0 , 0 , offsetof ( MatroskaTrackEncoding , encryption ) , { . n = matroska_track_encoding_encryption } } ,
2014-03-01 19:28:15 +03:00
{ MATROSKA_ID_ENCODINGORDER , EBML_NONE } ,
2019-05-17 00:30:06 +02:00
CHILD_OF ( matroska_track_encodings )
2008-08-05 03:40:31 +03:00
} ;
2019-07-18 21:07:20 +02:00
static EbmlSyntax matroska_track_encodings [ ] = {
avformat/matroskadec: Allow to count the number of element occurences
Up until now, the generic EBML reader used by the Matroska demuxer did
not have the capability to record whether an element was actually
present or not; instead, in cases where it mattered one typically added
an invalid default value and checked whether the value is valid (in
which case it is guaranteed to be present). This worked pretty well so
far, yet the EBML specifications have evolved: It is now legal to use
zero-length elements for floats, ints, uints and strings (both ASCII and
UTF-8); the value of these elements is the default value of the element
(if it has one) or zero for scalar types and an empty string for
strings. Furthermore, having a default value does no longer imply that
the element may be presumed to be present (with its default value) if it
is absent; this is only true if the element is mandatory, too.
These rules are designed to allow size savings as follows: Consider the
newly added FlagOriginal: It being zero means the track is not in its
original language, it being one means it is. For backward compatibility
reasons, neither of the two values may be inferred automatically in the
absence of the element. But one can still save a byte when one wants to
write the element with a value of zero, as one can write the integer with
a length of zero: 0x55AE 80 instead of 0x55AE 81 00. In the former case,
a parser has to infer the value of the element to be zero (which is the
element's default value).
When encountering an element with length zero, our parser always infers
a value of zero (or an empty string); this is wrong for values with
a different default value. It needs to infer the default value (or zero
in its absence) and this precludes using an invalid default value for
elements like FlagOriginal. Ergo one needs to be able to record whether
an element is present or not by other means. This patch allows to use a
simple counter for this. While just at it, some invalid and unnecessary
default values have been removed (mastering metadata elements used
default values of -1.0, despite these elements only being used if they
are > 0).
Reviewed-by: Ridley Combs <rcombs@rcombs.me>
Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@gmail.com>
2021-02-16 15:33:51 +02:00
{ MATROSKA_ID_TRACKCONTENTENCODING , EBML_NEST , 0 , sizeof ( MatroskaTrackEncoding ) , offsetof ( MatroskaTrack , encodings ) , { . n = matroska_track_encoding } } ,
2019-05-17 00:30:06 +02:00
CHILD_OF ( matroska_track )
2008-08-05 03:40:31 +03:00
} ;
2019-07-18 21:07:20 +02:00
static EbmlSyntax matroska_track_plane [ ] = {
avformat/matroskadec: Allow to count the number of element occurences
Up until now, the generic EBML reader used by the Matroska demuxer did
not have the capability to record whether an element was actually
present or not; instead, in cases where it mattered one typically added
an invalid default value and checked whether the value is valid (in
which case it is guaranteed to be present). This worked pretty well so
far, yet the EBML specifications have evolved: It is now legal to use
zero-length elements for floats, ints, uints and strings (both ASCII and
UTF-8); the value of these elements is the default value of the element
(if it has one) or zero for scalar types and an empty string for
strings. Furthermore, having a default value does no longer imply that
the element may be presumed to be present (with its default value) if it
is absent; this is only true if the element is mandatory, too.
These rules are designed to allow size savings as follows: Consider the
newly added FlagOriginal: It being zero means the track is not in its
original language, it being one means it is. For backward compatibility
reasons, neither of the two values may be inferred automatically in the
absence of the element. But one can still save a byte when one wants to
write the element with a value of zero, as one can write the integer with
a length of zero: 0x55AE 80 instead of 0x55AE 81 00. In the former case,
a parser has to infer the value of the element to be zero (which is the
element's default value).
When encountering an element with length zero, our parser always infers
a value of zero (or an empty string); this is wrong for values with
a different default value. It needs to infer the default value (or zero
in its absence) and this precludes using an invalid default value for
elements like FlagOriginal. Ergo one needs to be able to record whether
an element is present or not by other means. This patch allows to use a
simple counter for this. While just at it, some invalid and unnecessary
default values have been removed (mastering metadata elements used
default values of -1.0, despite these elements only being used if they
are > 0).
Reviewed-by: Ridley Combs <rcombs@rcombs.me>
Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@gmail.com>
2021-02-16 15:33:51 +02:00
{ MATROSKA_ID_TRACKPLANEUID , EBML_UINT , 0 , 0 , offsetof ( MatroskaTrackPlane , uid ) } ,
{ MATROSKA_ID_TRACKPLANETYPE , EBML_UINT , 0 , 0 , offsetof ( MatroskaTrackPlane , type ) } ,
2019-05-17 00:30:06 +02:00
CHILD_OF ( matroska_track_combine_planes )
2011-05-21 18:14:14 +03:00
} ;
2019-07-18 21:07:20 +02:00
static EbmlSyntax matroska_track_combine_planes [ ] = {
avformat/matroskadec: Allow to count the number of element occurences
Up until now, the generic EBML reader used by the Matroska demuxer did
not have the capability to record whether an element was actually
present or not; instead, in cases where it mattered one typically added
an invalid default value and checked whether the value is valid (in
which case it is guaranteed to be present). This worked pretty well so
far, yet the EBML specifications have evolved: It is now legal to use
zero-length elements for floats, ints, uints and strings (both ASCII and
UTF-8); the value of these elements is the default value of the element
(if it has one) or zero for scalar types and an empty string for
strings. Furthermore, having a default value does no longer imply that
the element may be presumed to be present (with its default value) if it
is absent; this is only true if the element is mandatory, too.
These rules are designed to allow size savings as follows: Consider the
newly added FlagOriginal: It being zero means the track is not in its
original language, it being one means it is. For backward compatibility
reasons, neither of the two values may be inferred automatically in the
absence of the element. But one can still save a byte when one wants to
write the element with a value of zero, as one can write the integer with
a length of zero: 0x55AE 80 instead of 0x55AE 81 00. In the former case,
a parser has to infer the value of the element to be zero (which is the
element's default value).
When encountering an element with length zero, our parser always infers
a value of zero (or an empty string); this is wrong for values with
a different default value. It needs to infer the default value (or zero
in its absence) and this precludes using an invalid default value for
elements like FlagOriginal. Ergo one needs to be able to record whether
an element is present or not by other means. This patch allows to use a
simple counter for this. While just at it, some invalid and unnecessary
default values have been removed (mastering metadata elements used
default values of -1.0, despite these elements only being used if they
are > 0).
Reviewed-by: Ridley Combs <rcombs@rcombs.me>
Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@gmail.com>
2021-02-16 15:33:51 +02:00
{ MATROSKA_ID_TRACKPLANE , EBML_NEST , 0 , sizeof ( MatroskaTrackPlane ) , offsetof ( MatroskaTrackOperation , combine_planes ) , { . n = matroska_track_plane } } ,
2019-05-17 00:30:06 +02:00
CHILD_OF ( matroska_track_operation )
2011-05-21 18:14:14 +03:00
} ;
2019-07-18 21:07:20 +02:00
static EbmlSyntax matroska_track_operation [ ] = {
avformat/matroskadec: Allow to count the number of element occurences
Up until now, the generic EBML reader used by the Matroska demuxer did
not have the capability to record whether an element was actually
present or not; instead, in cases where it mattered one typically added
an invalid default value and checked whether the value is valid (in
which case it is guaranteed to be present). This worked pretty well so
far, yet the EBML specifications have evolved: It is now legal to use
zero-length elements for floats, ints, uints and strings (both ASCII and
UTF-8); the value of these elements is the default value of the element
(if it has one) or zero for scalar types and an empty string for
strings. Furthermore, having a default value does no longer imply that
the element may be presumed to be present (with its default value) if it
is absent; this is only true if the element is mandatory, too.
These rules are designed to allow size savings as follows: Consider the
newly added FlagOriginal: It being zero means the track is not in its
original language, it being one means it is. For backward compatibility
reasons, neither of the two values may be inferred automatically in the
absence of the element. But one can still save a byte when one wants to
write the element with a value of zero, as one can write the integer with
a length of zero: 0x55AE 80 instead of 0x55AE 81 00. In the former case,
a parser has to infer the value of the element to be zero (which is the
element's default value).
When encountering an element with length zero, our parser always infers
a value of zero (or an empty string); this is wrong for values with
a different default value. It needs to infer the default value (or zero
in its absence) and this precludes using an invalid default value for
elements like FlagOriginal. Ergo one needs to be able to record whether
an element is present or not by other means. This patch allows to use a
simple counter for this. While just at it, some invalid and unnecessary
default values have been removed (mastering metadata elements used
default values of -1.0, despite these elements only being used if they
are > 0).
Reviewed-by: Ridley Combs <rcombs@rcombs.me>
Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@gmail.com>
2021-02-16 15:33:51 +02:00
{ MATROSKA_ID_TRACKCOMBINEPLANES , EBML_NEST , 0 , 0 , 0 , { . n = matroska_track_combine_planes } } ,
2019-05-17 00:30:06 +02:00
CHILD_OF ( matroska_track )
2011-05-21 18:14:14 +03:00
} ;
2019-07-18 21:07:20 +02:00
static EbmlSyntax matroska_track [ ] = {
avformat/matroskadec: Allow to count the number of element occurences
Up until now, the generic EBML reader used by the Matroska demuxer did
not have the capability to record whether an element was actually
present or not; instead, in cases where it mattered one typically added
an invalid default value and checked whether the value is valid (in
which case it is guaranteed to be present). This worked pretty well so
far, yet the EBML specifications have evolved: It is now legal to use
zero-length elements for floats, ints, uints and strings (both ASCII and
UTF-8); the value of these elements is the default value of the element
(if it has one) or zero for scalar types and an empty string for
strings. Furthermore, having a default value does no longer imply that
the element may be presumed to be present (with its default value) if it
is absent; this is only true if the element is mandatory, too.
These rules are designed to allow size savings as follows: Consider the
newly added FlagOriginal: It being zero means the track is not in its
original language, it being one means it is. For backward compatibility
reasons, neither of the two values may be inferred automatically in the
absence of the element. But one can still save a byte when one wants to
write the element with a value of zero, as one can write the integer with
a length of zero: 0x55AE 80 instead of 0x55AE 81 00. In the former case,
a parser has to infer the value of the element to be zero (which is the
element's default value).
When encountering an element with length zero, our parser always infers
a value of zero (or an empty string); this is wrong for values with
a different default value. It needs to infer the default value (or zero
in its absence) and this precludes using an invalid default value for
elements like FlagOriginal. Ergo one needs to be able to record whether
an element is present or not by other means. This patch allows to use a
simple counter for this. While just at it, some invalid and unnecessary
default values have been removed (mastering metadata elements used
default values of -1.0, despite these elements only being used if they
are > 0).
Reviewed-by: Ridley Combs <rcombs@rcombs.me>
Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@gmail.com>
2021-02-16 15:33:51 +02:00
{ MATROSKA_ID_TRACKNUMBER , EBML_UINT , 0 , 0 , offsetof ( MatroskaTrack , num ) } ,
{ MATROSKA_ID_TRACKNAME , EBML_UTF8 , 0 , 0 , offsetof ( MatroskaTrack , name ) } ,
{ MATROSKA_ID_TRACKUID , EBML_UINT , 0 , 0 , offsetof ( MatroskaTrack , uid ) } ,
{ MATROSKA_ID_TRACKTYPE , EBML_UINT , 0 , 0 , offsetof ( MatroskaTrack , type ) } ,
{ MATROSKA_ID_CODECID , EBML_STR , 0 , 0 , offsetof ( MatroskaTrack , codec_id ) } ,
{ MATROSKA_ID_CODECPRIVATE , EBML_BIN , 0 , 0 , offsetof ( MatroskaTrack , codec_priv ) } ,
{ MATROSKA_ID_CODECDELAY , EBML_UINT , 0 , 0 , offsetof ( MatroskaTrack , codec_delay ) , { . u = 0 } } ,
{ MATROSKA_ID_TRACKLANGUAGE , EBML_STR , 0 , 0 , offsetof ( MatroskaTrack , language ) , { . s = " eng " } } ,
{ MATROSKA_ID_TRACKDEFAULTDURATION , EBML_UINT , 0 , 0 , offsetof ( MatroskaTrack , default_duration ) } ,
{ MATROSKA_ID_TRACKTIMECODESCALE , EBML_FLOAT , 0 , 0 , offsetof ( MatroskaTrack , time_scale ) , { . f = 1.0 } } ,
2021-02-16 19:23:59 +02:00
{ MATROSKA_ID_TRACKFLAGCOMMENTARY , EBML_UINT , 0 , 0 , offsetof ( MatroskaTrack , flag_comment ) , { . u = 0 } } ,
avformat/matroskadec: Allow to count the number of element occurences
Up until now, the generic EBML reader used by the Matroska demuxer did
not have the capability to record whether an element was actually
present or not; instead, in cases where it mattered one typically added
an invalid default value and checked whether the value is valid (in
which case it is guaranteed to be present). This worked pretty well so
far, yet the EBML specifications have evolved: It is now legal to use
zero-length elements for floats, ints, uints and strings (both ASCII and
UTF-8); the value of these elements is the default value of the element
(if it has one) or zero for scalar types and an empty string for
strings. Furthermore, having a default value does no longer imply that
the element may be presumed to be present (with its default value) if it
is absent; this is only true if the element is mandatory, too.
These rules are designed to allow size savings as follows: Consider the
newly added FlagOriginal: It being zero means the track is not in its
original language, it being one means it is. For backward compatibility
reasons, neither of the two values may be inferred automatically in the
absence of the element. But one can still save a byte when one wants to
write the element with a value of zero, as one can write the integer with
a length of zero: 0x55AE 80 instead of 0x55AE 81 00. In the former case,
a parser has to infer the value of the element to be zero (which is the
element's default value).
When encountering an element with length zero, our parser always infers
a value of zero (or an empty string); this is wrong for values with
a different default value. It needs to infer the default value (or zero
in its absence) and this precludes using an invalid default value for
elements like FlagOriginal. Ergo one needs to be able to record whether
an element is present or not by other means. This patch allows to use a
simple counter for this. While just at it, some invalid and unnecessary
default values have been removed (mastering metadata elements used
default values of -1.0, despite these elements only being used if they
are > 0).
Reviewed-by: Ridley Combs <rcombs@rcombs.me>
Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@gmail.com>
2021-02-16 15:33:51 +02:00
{ MATROSKA_ID_TRACKFLAGDEFAULT , EBML_UINT , 0 , 0 , offsetof ( MatroskaTrack , flag_default ) , { . u = 1 } } ,
{ MATROSKA_ID_TRACKFLAGFORCED , EBML_UINT , 0 , 0 , offsetof ( MatroskaTrack , flag_forced ) , { . u = 0 } } ,
2021-02-16 20:08:51 +02:00
{ MATROSKA_ID_TRACKFLAGHEARINGIMPAIRED , EBML_UINT , 0 , 0 , offsetof ( MatroskaTrack , flag_hearingimpaired ) , { . u = 0 } } ,
{ MATROSKA_ID_TRACKFLAGVISUALIMPAIRED , EBML_UINT , 0 , 0 , offsetof ( MatroskaTrack , flag_visualimpaired ) , { . u = 0 } } ,
2021-02-16 20:38:19 +02:00
{ MATROSKA_ID_TRACKFLAGTEXTDESCRIPTIONS , EBML_UINT , 0 , 0 , offsetof ( MatroskaTrack , flag_textdescriptions ) , { . u = 0 } } ,
2021-02-16 19:41:30 +02:00
{ MATROSKA_ID_TRACKFLAGORIGINAL , EBML_UINT , 1 , 0 , offsetof ( MatroskaTrack , flag_original ) , { . u = 0 } } ,
avformat/matroskadec: Allow to count the number of element occurences
Up until now, the generic EBML reader used by the Matroska demuxer did
not have the capability to record whether an element was actually
present or not; instead, in cases where it mattered one typically added
an invalid default value and checked whether the value is valid (in
which case it is guaranteed to be present). This worked pretty well so
far, yet the EBML specifications have evolved: It is now legal to use
zero-length elements for floats, ints, uints and strings (both ASCII and
UTF-8); the value of these elements is the default value of the element
(if it has one) or zero for scalar types and an empty string for
strings. Furthermore, having a default value does no longer imply that
the element may be presumed to be present (with its default value) if it
is absent; this is only true if the element is mandatory, too.
These rules are designed to allow size savings as follows: Consider the
newly added FlagOriginal: It being zero means the track is not in its
original language, it being one means it is. For backward compatibility
reasons, neither of the two values may be inferred automatically in the
absence of the element. But one can still save a byte when one wants to
write the element with a value of zero, as one can write the integer with
a length of zero: 0x55AE 80 instead of 0x55AE 81 00. In the former case,
a parser has to infer the value of the element to be zero (which is the
element's default value).
When encountering an element with length zero, our parser always infers
a value of zero (or an empty string); this is wrong for values with
a different default value. It needs to infer the default value (or zero
in its absence) and this precludes using an invalid default value for
elements like FlagOriginal. Ergo one needs to be able to record whether
an element is present or not by other means. This patch allows to use a
simple counter for this. While just at it, some invalid and unnecessary
default values have been removed (mastering metadata elements used
default values of -1.0, despite these elements only being used if they
are > 0).
Reviewed-by: Ridley Combs <rcombs@rcombs.me>
Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@gmail.com>
2021-02-16 15:33:51 +02:00
{ MATROSKA_ID_TRACKVIDEO , EBML_NEST , 0 , 0 , offsetof ( MatroskaTrack , video ) , { . n = matroska_track_video } } ,
{ MATROSKA_ID_TRACKAUDIO , EBML_NEST , 0 , 0 , offsetof ( MatroskaTrack , audio ) , { . n = matroska_track_audio } } ,
{ MATROSKA_ID_TRACKOPERATION , EBML_NEST , 0 , 0 , offsetof ( MatroskaTrack , operation ) , { . n = matroska_track_operation } } ,
{ MATROSKA_ID_TRACKCONTENTENCODINGS , EBML_NEST , 0 , 0 , 0 , { . n = matroska_track_encodings } } ,
{ MATROSKA_ID_TRACKMAXBLKADDID , EBML_UINT , 0 , 0 , offsetof ( MatroskaTrack , max_block_additional_id ) , { . u = 0 } } ,
{ MATROSKA_ID_SEEKPREROLL , EBML_UINT , 0 , 0 , offsetof ( MatroskaTrack , seek_preroll ) , { . u = 0 } } ,
2014-03-01 19:28:15 +03:00
{ MATROSKA_ID_TRACKFLAGENABLED , EBML_NONE } ,
{ MATROSKA_ID_TRACKFLAGLACING , EBML_NONE } ,
{ MATROSKA_ID_CODECNAME , EBML_NONE } ,
{ MATROSKA_ID_CODECDECODEALL , EBML_NONE } ,
{ MATROSKA_ID_CODECINFOURL , EBML_NONE } ,
{ MATROSKA_ID_CODECDOWNLOADURL , EBML_NONE } ,
{ MATROSKA_ID_TRACKMINCACHE , EBML_NONE } ,
{ MATROSKA_ID_TRACKMAXCACHE , EBML_NONE } ,
2019-05-17 00:30:06 +02:00
CHILD_OF ( matroska_tracks )
2008-08-05 03:40:31 +03:00
} ;
2019-07-18 21:07:20 +02:00
static EbmlSyntax matroska_tracks [ ] = {
avformat/matroskadec: Allow to count the number of element occurences
Up until now, the generic EBML reader used by the Matroska demuxer did
not have the capability to record whether an element was actually
present or not; instead, in cases where it mattered one typically added
an invalid default value and checked whether the value is valid (in
which case it is guaranteed to be present). This worked pretty well so
far, yet the EBML specifications have evolved: It is now legal to use
zero-length elements for floats, ints, uints and strings (both ASCII and
UTF-8); the value of these elements is the default value of the element
(if it has one) or zero for scalar types and an empty string for
strings. Furthermore, having a default value does no longer imply that
the element may be presumed to be present (with its default value) if it
is absent; this is only true if the element is mandatory, too.
These rules are designed to allow size savings as follows: Consider the
newly added FlagOriginal: It being zero means the track is not in its
original language, it being one means it is. For backward compatibility
reasons, neither of the two values may be inferred automatically in the
absence of the element. But one can still save a byte when one wants to
write the element with a value of zero, as one can write the integer with
a length of zero: 0x55AE 80 instead of 0x55AE 81 00. In the former case,
a parser has to infer the value of the element to be zero (which is the
element's default value).
When encountering an element with length zero, our parser always infers
a value of zero (or an empty string); this is wrong for values with
a different default value. It needs to infer the default value (or zero
in its absence) and this precludes using an invalid default value for
elements like FlagOriginal. Ergo one needs to be able to record whether
an element is present or not by other means. This patch allows to use a
simple counter for this. While just at it, some invalid and unnecessary
default values have been removed (mastering metadata elements used
default values of -1.0, despite these elements only being used if they
are > 0).
Reviewed-by: Ridley Combs <rcombs@rcombs.me>
Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@gmail.com>
2021-02-16 15:33:51 +02:00
{ MATROSKA_ID_TRACKENTRY , EBML_NEST , 0 , sizeof ( MatroskaTrack ) , offsetof ( MatroskaDemuxContext , tracks ) , { . n = matroska_track } } ,
2019-05-17 00:30:06 +02:00
CHILD_OF ( matroska_segment )
2008-08-05 03:40:31 +03:00
} ;
2019-07-18 21:07:20 +02:00
static EbmlSyntax matroska_attachment [ ] = {
avformat/matroskadec: Allow to count the number of element occurences
Up until now, the generic EBML reader used by the Matroska demuxer did
not have the capability to record whether an element was actually
present or not; instead, in cases where it mattered one typically added
an invalid default value and checked whether the value is valid (in
which case it is guaranteed to be present). This worked pretty well so
far, yet the EBML specifications have evolved: It is now legal to use
zero-length elements for floats, ints, uints and strings (both ASCII and
UTF-8); the value of these elements is the default value of the element
(if it has one) or zero for scalar types and an empty string for
strings. Furthermore, having a default value does no longer imply that
the element may be presumed to be present (with its default value) if it
is absent; this is only true if the element is mandatory, too.
These rules are designed to allow size savings as follows: Consider the
newly added FlagOriginal: It being zero means the track is not in its
original language, it being one means it is. For backward compatibility
reasons, neither of the two values may be inferred automatically in the
absence of the element. But one can still save a byte when one wants to
write the element with a value of zero, as one can write the integer with
a length of zero: 0x55AE 80 instead of 0x55AE 81 00. In the former case,
a parser has to infer the value of the element to be zero (which is the
element's default value).
When encountering an element with length zero, our parser always infers
a value of zero (or an empty string); this is wrong for values with
a different default value. It needs to infer the default value (or zero
in its absence) and this precludes using an invalid default value for
elements like FlagOriginal. Ergo one needs to be able to record whether
an element is present or not by other means. This patch allows to use a
simple counter for this. While just at it, some invalid and unnecessary
default values have been removed (mastering metadata elements used
default values of -1.0, despite these elements only being used if they
are > 0).
Reviewed-by: Ridley Combs <rcombs@rcombs.me>
Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@gmail.com>
2021-02-16 15:33:51 +02:00
{ MATROSKA_ID_FILEUID , EBML_UINT , 0 , 0 , offsetof ( MatroskaAttachment , uid ) } ,
{ MATROSKA_ID_FILENAME , EBML_UTF8 , 0 , 0 , offsetof ( MatroskaAttachment , filename ) } ,
{ MATROSKA_ID_FILEMIMETYPE , EBML_STR , 0 , 0 , offsetof ( MatroskaAttachment , mime ) } ,
{ MATROSKA_ID_FILEDATA , EBML_BIN , 0 , 0 , offsetof ( MatroskaAttachment , bin ) } ,
{ MATROSKA_ID_FILEDESC , EBML_UTF8 , 0 , 0 , offsetof ( MatroskaAttachment , description ) } ,
2019-05-17 00:30:06 +02:00
CHILD_OF ( matroska_attachments )
2008-08-05 03:40:24 +03:00
} ;
2019-07-18 21:07:20 +02:00
static EbmlSyntax matroska_attachments [ ] = {
avformat/matroskadec: Allow to count the number of element occurences
Up until now, the generic EBML reader used by the Matroska demuxer did
not have the capability to record whether an element was actually
present or not; instead, in cases where it mattered one typically added
an invalid default value and checked whether the value is valid (in
which case it is guaranteed to be present). This worked pretty well so
far, yet the EBML specifications have evolved: It is now legal to use
zero-length elements for floats, ints, uints and strings (both ASCII and
UTF-8); the value of these elements is the default value of the element
(if it has one) or zero for scalar types and an empty string for
strings. Furthermore, having a default value does no longer imply that
the element may be presumed to be present (with its default value) if it
is absent; this is only true if the element is mandatory, too.
These rules are designed to allow size savings as follows: Consider the
newly added FlagOriginal: It being zero means the track is not in its
original language, it being one means it is. For backward compatibility
reasons, neither of the two values may be inferred automatically in the
absence of the element. But one can still save a byte when one wants to
write the element with a value of zero, as one can write the integer with
a length of zero: 0x55AE 80 instead of 0x55AE 81 00. In the former case,
a parser has to infer the value of the element to be zero (which is the
element's default value).
When encountering an element with length zero, our parser always infers
a value of zero (or an empty string); this is wrong for values with
a different default value. It needs to infer the default value (or zero
in its absence) and this precludes using an invalid default value for
elements like FlagOriginal. Ergo one needs to be able to record whether
an element is present or not by other means. This patch allows to use a
simple counter for this. While just at it, some invalid and unnecessary
default values have been removed (mastering metadata elements used
default values of -1.0, despite these elements only being used if they
are > 0).
Reviewed-by: Ridley Combs <rcombs@rcombs.me>
Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@gmail.com>
2021-02-16 15:33:51 +02:00
{ MATROSKA_ID_ATTACHEDFILE , EBML_NEST , 0 , sizeof ( MatroskaAttachment ) , offsetof ( MatroskaDemuxContext , attachments ) , { . n = matroska_attachment } } ,
2019-05-17 00:30:06 +02:00
CHILD_OF ( matroska_segment )
2008-08-05 03:40:24 +03:00
} ;
2019-07-18 21:07:20 +02:00
static EbmlSyntax matroska_chapter_display [ ] = {
avformat/matroskadec: Allow to count the number of element occurences
Up until now, the generic EBML reader used by the Matroska demuxer did
not have the capability to record whether an element was actually
present or not; instead, in cases where it mattered one typically added
an invalid default value and checked whether the value is valid (in
which case it is guaranteed to be present). This worked pretty well so
far, yet the EBML specifications have evolved: It is now legal to use
zero-length elements for floats, ints, uints and strings (both ASCII and
UTF-8); the value of these elements is the default value of the element
(if it has one) or zero for scalar types and an empty string for
strings. Furthermore, having a default value does no longer imply that
the element may be presumed to be present (with its default value) if it
is absent; this is only true if the element is mandatory, too.
These rules are designed to allow size savings as follows: Consider the
newly added FlagOriginal: It being zero means the track is not in its
original language, it being one means it is. For backward compatibility
reasons, neither of the two values may be inferred automatically in the
absence of the element. But one can still save a byte when one wants to
write the element with a value of zero, as one can write the integer with
a length of zero: 0x55AE 80 instead of 0x55AE 81 00. In the former case,
a parser has to infer the value of the element to be zero (which is the
element's default value).
When encountering an element with length zero, our parser always infers
a value of zero (or an empty string); this is wrong for values with
a different default value. It needs to infer the default value (or zero
in its absence) and this precludes using an invalid default value for
elements like FlagOriginal. Ergo one needs to be able to record whether
an element is present or not by other means. This patch allows to use a
simple counter for this. While just at it, some invalid and unnecessary
default values have been removed (mastering metadata elements used
default values of -1.0, despite these elements only being used if they
are > 0).
Reviewed-by: Ridley Combs <rcombs@rcombs.me>
Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@gmail.com>
2021-02-16 15:33:51 +02:00
{ MATROSKA_ID_CHAPSTRING , EBML_UTF8 , 0 , 0 , offsetof ( MatroskaChapter , title ) } ,
2015-09-20 16:34:05 +02:00
{ MATROSKA_ID_CHAPLANG , EBML_NONE } ,
{ MATROSKA_ID_CHAPCOUNTRY , EBML_NONE } ,
2019-05-17 00:30:06 +02:00
CHILD_OF ( matroska_chapter_entry )
2008-08-05 03:40:21 +03:00
} ;
2019-07-18 21:07:20 +02:00
static EbmlSyntax matroska_chapter_entry [ ] = {
avformat/matroskadec: Allow to count the number of element occurences
Up until now, the generic EBML reader used by the Matroska demuxer did
not have the capability to record whether an element was actually
present or not; instead, in cases where it mattered one typically added
an invalid default value and checked whether the value is valid (in
which case it is guaranteed to be present). This worked pretty well so
far, yet the EBML specifications have evolved: It is now legal to use
zero-length elements for floats, ints, uints and strings (both ASCII and
UTF-8); the value of these elements is the default value of the element
(if it has one) or zero for scalar types and an empty string for
strings. Furthermore, having a default value does no longer imply that
the element may be presumed to be present (with its default value) if it
is absent; this is only true if the element is mandatory, too.
These rules are designed to allow size savings as follows: Consider the
newly added FlagOriginal: It being zero means the track is not in its
original language, it being one means it is. For backward compatibility
reasons, neither of the two values may be inferred automatically in the
absence of the element. But one can still save a byte when one wants to
write the element with a value of zero, as one can write the integer with
a length of zero: 0x55AE 80 instead of 0x55AE 81 00. In the former case,
a parser has to infer the value of the element to be zero (which is the
element's default value).
When encountering an element with length zero, our parser always infers
a value of zero (or an empty string); this is wrong for values with
a different default value. It needs to infer the default value (or zero
in its absence) and this precludes using an invalid default value for
elements like FlagOriginal. Ergo one needs to be able to record whether
an element is present or not by other means. This patch allows to use a
simple counter for this. While just at it, some invalid and unnecessary
default values have been removed (mastering metadata elements used
default values of -1.0, despite these elements only being used if they
are > 0).
Reviewed-by: Ridley Combs <rcombs@rcombs.me>
Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@gmail.com>
2021-02-16 15:33:51 +02:00
{ MATROSKA_ID_CHAPTERTIMESTART , EBML_UINT , 0 , 0 , offsetof ( MatroskaChapter , start ) , { . u = AV_NOPTS_VALUE } } ,
{ MATROSKA_ID_CHAPTERTIMEEND , EBML_UINT , 0 , 0 , offsetof ( MatroskaChapter , end ) , { . u = AV_NOPTS_VALUE } } ,
{ MATROSKA_ID_CHAPTERUID , EBML_UINT , 0 , 0 , offsetof ( MatroskaChapter , uid ) } ,
{ MATROSKA_ID_CHAPTERDISPLAY , EBML_NEST , 0 , 0 , 0 , { . n = matroska_chapter_display } } ,
2008-08-05 03:40:21 +03:00
{ MATROSKA_ID_CHAPTERFLAGHIDDEN , EBML_NONE } ,
2008-08-14 00:15:15 +03:00
{ MATROSKA_ID_CHAPTERFLAGENABLED , EBML_NONE } ,
{ MATROSKA_ID_CHAPTERPHYSEQUIV , EBML_NONE } ,
{ MATROSKA_ID_CHAPTERATOM , EBML_NONE } ,
2019-05-17 00:30:06 +02:00
CHILD_OF ( matroska_chapter )
2008-08-05 03:40:21 +03:00
} ;
2019-07-18 21:07:20 +02:00
static EbmlSyntax matroska_chapter [ ] = {
avformat/matroskadec: Allow to count the number of element occurences
Up until now, the generic EBML reader used by the Matroska demuxer did
not have the capability to record whether an element was actually
present or not; instead, in cases where it mattered one typically added
an invalid default value and checked whether the value is valid (in
which case it is guaranteed to be present). This worked pretty well so
far, yet the EBML specifications have evolved: It is now legal to use
zero-length elements for floats, ints, uints and strings (both ASCII and
UTF-8); the value of these elements is the default value of the element
(if it has one) or zero for scalar types and an empty string for
strings. Furthermore, having a default value does no longer imply that
the element may be presumed to be present (with its default value) if it
is absent; this is only true if the element is mandatory, too.
These rules are designed to allow size savings as follows: Consider the
newly added FlagOriginal: It being zero means the track is not in its
original language, it being one means it is. For backward compatibility
reasons, neither of the two values may be inferred automatically in the
absence of the element. But one can still save a byte when one wants to
write the element with a value of zero, as one can write the integer with
a length of zero: 0x55AE 80 instead of 0x55AE 81 00. In the former case,
a parser has to infer the value of the element to be zero (which is the
element's default value).
When encountering an element with length zero, our parser always infers
a value of zero (or an empty string); this is wrong for values with
a different default value. It needs to infer the default value (or zero
in its absence) and this precludes using an invalid default value for
elements like FlagOriginal. Ergo one needs to be able to record whether
an element is present or not by other means. This patch allows to use a
simple counter for this. While just at it, some invalid and unnecessary
default values have been removed (mastering metadata elements used
default values of -1.0, despite these elements only being used if they
are > 0).
Reviewed-by: Ridley Combs <rcombs@rcombs.me>
Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@gmail.com>
2021-02-16 15:33:51 +02:00
{ MATROSKA_ID_CHAPTERATOM , EBML_NEST , 0 , sizeof ( MatroskaChapter ) , offsetof ( MatroskaDemuxContext , chapters ) , { . n = matroska_chapter_entry } } ,
2008-08-05 03:40:21 +03:00
{ MATROSKA_ID_EDITIONUID , EBML_NONE } ,
{ MATROSKA_ID_EDITIONFLAGHIDDEN , EBML_NONE } ,
{ MATROSKA_ID_EDITIONFLAGDEFAULT , EBML_NONE } ,
2008-08-14 00:15:15 +03:00
{ MATROSKA_ID_EDITIONFLAGORDERED , EBML_NONE } ,
2019-05-17 00:30:06 +02:00
CHILD_OF ( matroska_chapters )
2008-08-05 03:40:21 +03:00
} ;
2019-07-18 21:07:20 +02:00
static EbmlSyntax matroska_chapters [ ] = {
avformat/matroskadec: Allow to count the number of element occurences
Up until now, the generic EBML reader used by the Matroska demuxer did
not have the capability to record whether an element was actually
present or not; instead, in cases where it mattered one typically added
an invalid default value and checked whether the value is valid (in
which case it is guaranteed to be present). This worked pretty well so
far, yet the EBML specifications have evolved: It is now legal to use
zero-length elements for floats, ints, uints and strings (both ASCII and
UTF-8); the value of these elements is the default value of the element
(if it has one) or zero for scalar types and an empty string for
strings. Furthermore, having a default value does no longer imply that
the element may be presumed to be present (with its default value) if it
is absent; this is only true if the element is mandatory, too.
These rules are designed to allow size savings as follows: Consider the
newly added FlagOriginal: It being zero means the track is not in its
original language, it being one means it is. For backward compatibility
reasons, neither of the two values may be inferred automatically in the
absence of the element. But one can still save a byte when one wants to
write the element with a value of zero, as one can write the integer with
a length of zero: 0x55AE 80 instead of 0x55AE 81 00. In the former case,
a parser has to infer the value of the element to be zero (which is the
element's default value).
When encountering an element with length zero, our parser always infers
a value of zero (or an empty string); this is wrong for values with
a different default value. It needs to infer the default value (or zero
in its absence) and this precludes using an invalid default value for
elements like FlagOriginal. Ergo one needs to be able to record whether
an element is present or not by other means. This patch allows to use a
simple counter for this. While just at it, some invalid and unnecessary
default values have been removed (mastering metadata elements used
default values of -1.0, despite these elements only being used if they
are > 0).
Reviewed-by: Ridley Combs <rcombs@rcombs.me>
Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@gmail.com>
2021-02-16 15:33:51 +02:00
{ MATROSKA_ID_EDITIONENTRY , EBML_NEST , 0 , 0 , 0 , { . n = matroska_chapter } } ,
2019-05-17 00:30:06 +02:00
CHILD_OF ( matroska_segment )
2008-08-05 03:40:21 +03:00
} ;
2019-07-18 21:07:20 +02:00
static EbmlSyntax matroska_index_pos [ ] = {
avformat/matroskadec: Allow to count the number of element occurences
Up until now, the generic EBML reader used by the Matroska demuxer did
not have the capability to record whether an element was actually
present or not; instead, in cases where it mattered one typically added
an invalid default value and checked whether the value is valid (in
which case it is guaranteed to be present). This worked pretty well so
far, yet the EBML specifications have evolved: It is now legal to use
zero-length elements for floats, ints, uints and strings (both ASCII and
UTF-8); the value of these elements is the default value of the element
(if it has one) or zero for scalar types and an empty string for
strings. Furthermore, having a default value does no longer imply that
the element may be presumed to be present (with its default value) if it
is absent; this is only true if the element is mandatory, too.
These rules are designed to allow size savings as follows: Consider the
newly added FlagOriginal: It being zero means the track is not in its
original language, it being one means it is. For backward compatibility
reasons, neither of the two values may be inferred automatically in the
absence of the element. But one can still save a byte when one wants to
write the element with a value of zero, as one can write the integer with
a length of zero: 0x55AE 80 instead of 0x55AE 81 00. In the former case,
a parser has to infer the value of the element to be zero (which is the
element's default value).
When encountering an element with length zero, our parser always infers
a value of zero (or an empty string); this is wrong for values with
a different default value. It needs to infer the default value (or zero
in its absence) and this precludes using an invalid default value for
elements like FlagOriginal. Ergo one needs to be able to record whether
an element is present or not by other means. This patch allows to use a
simple counter for this. While just at it, some invalid and unnecessary
default values have been removed (mastering metadata elements used
default values of -1.0, despite these elements only being used if they
are > 0).
Reviewed-by: Ridley Combs <rcombs@rcombs.me>
Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@gmail.com>
2021-02-16 15:33:51 +02:00
{ MATROSKA_ID_CUETRACK , EBML_UINT , 0 , 0 , offsetof ( MatroskaIndexPos , track ) } ,
{ MATROSKA_ID_CUECLUSTERPOSITION , EBML_UINT , 0 , 0 , offsetof ( MatroskaIndexPos , pos ) } ,
2013-09-15 12:32:36 +03:00
{ MATROSKA_ID_CUERELATIVEPOSITION , EBML_NONE } ,
2013-09-21 04:15:49 +03:00
{ MATROSKA_ID_CUEDURATION , EBML_NONE } ,
2008-08-14 00:15:15 +03:00
{ MATROSKA_ID_CUEBLOCKNUMBER , EBML_NONE } ,
2019-05-17 00:30:06 +02:00
CHILD_OF ( matroska_index_entry )
2008-08-05 03:40:15 +03:00
} ;
2019-07-18 21:07:20 +02:00
static EbmlSyntax matroska_index_entry [ ] = {
avformat/matroskadec: Allow to count the number of element occurences
Up until now, the generic EBML reader used by the Matroska demuxer did
not have the capability to record whether an element was actually
present or not; instead, in cases where it mattered one typically added
an invalid default value and checked whether the value is valid (in
which case it is guaranteed to be present). This worked pretty well so
far, yet the EBML specifications have evolved: It is now legal to use
zero-length elements for floats, ints, uints and strings (both ASCII and
UTF-8); the value of these elements is the default value of the element
(if it has one) or zero for scalar types and an empty string for
strings. Furthermore, having a default value does no longer imply that
the element may be presumed to be present (with its default value) if it
is absent; this is only true if the element is mandatory, too.
These rules are designed to allow size savings as follows: Consider the
newly added FlagOriginal: It being zero means the track is not in its
original language, it being one means it is. For backward compatibility
reasons, neither of the two values may be inferred automatically in the
absence of the element. But one can still save a byte when one wants to
write the element with a value of zero, as one can write the integer with
a length of zero: 0x55AE 80 instead of 0x55AE 81 00. In the former case,
a parser has to infer the value of the element to be zero (which is the
element's default value).
When encountering an element with length zero, our parser always infers
a value of zero (or an empty string); this is wrong for values with
a different default value. It needs to infer the default value (or zero
in its absence) and this precludes using an invalid default value for
elements like FlagOriginal. Ergo one needs to be able to record whether
an element is present or not by other means. This patch allows to use a
simple counter for this. While just at it, some invalid and unnecessary
default values have been removed (mastering metadata elements used
default values of -1.0, despite these elements only being used if they
are > 0).
Reviewed-by: Ridley Combs <rcombs@rcombs.me>
Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@gmail.com>
2021-02-16 15:33:51 +02:00
{ MATROSKA_ID_CUETIME , EBML_UINT , 0 , 0 , offsetof ( MatroskaIndex , time ) } ,
{ MATROSKA_ID_CUETRACKPOSITION , EBML_NEST , 0 , sizeof ( MatroskaIndexPos ) , offsetof ( MatroskaIndex , pos ) , { . n = matroska_index_pos } } ,
2019-05-17 00:30:06 +02:00
CHILD_OF ( matroska_index )
2008-08-05 03:40:15 +03:00
} ;
2019-07-18 21:07:20 +02:00
static EbmlSyntax matroska_index [ ] = {
avformat/matroskadec: Allow to count the number of element occurences
Up until now, the generic EBML reader used by the Matroska demuxer did
not have the capability to record whether an element was actually
present or not; instead, in cases where it mattered one typically added
an invalid default value and checked whether the value is valid (in
which case it is guaranteed to be present). This worked pretty well so
far, yet the EBML specifications have evolved: It is now legal to use
zero-length elements for floats, ints, uints and strings (both ASCII and
UTF-8); the value of these elements is the default value of the element
(if it has one) or zero for scalar types and an empty string for
strings. Furthermore, having a default value does no longer imply that
the element may be presumed to be present (with its default value) if it
is absent; this is only true if the element is mandatory, too.
These rules are designed to allow size savings as follows: Consider the
newly added FlagOriginal: It being zero means the track is not in its
original language, it being one means it is. For backward compatibility
reasons, neither of the two values may be inferred automatically in the
absence of the element. But one can still save a byte when one wants to
write the element with a value of zero, as one can write the integer with
a length of zero: 0x55AE 80 instead of 0x55AE 81 00. In the former case,
a parser has to infer the value of the element to be zero (which is the
element's default value).
When encountering an element with length zero, our parser always infers
a value of zero (or an empty string); this is wrong for values with
a different default value. It needs to infer the default value (or zero
in its absence) and this precludes using an invalid default value for
elements like FlagOriginal. Ergo one needs to be able to record whether
an element is present or not by other means. This patch allows to use a
simple counter for this. While just at it, some invalid and unnecessary
default values have been removed (mastering metadata elements used
default values of -1.0, despite these elements only being used if they
are > 0).
Reviewed-by: Ridley Combs <rcombs@rcombs.me>
Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@gmail.com>
2021-02-16 15:33:51 +02:00
{ MATROSKA_ID_POINTENTRY , EBML_NEST , 0 , sizeof ( MatroskaIndex ) , offsetof ( MatroskaDemuxContext , index ) , { . n = matroska_index_entry } } ,
2019-05-17 00:30:06 +02:00
CHILD_OF ( matroska_segment )
2008-08-05 03:40:15 +03:00
} ;
2019-07-18 21:07:20 +02:00
static EbmlSyntax matroska_simpletag [ ] = {
avformat/matroskadec: Allow to count the number of element occurences
Up until now, the generic EBML reader used by the Matroska demuxer did
not have the capability to record whether an element was actually
present or not; instead, in cases where it mattered one typically added
an invalid default value and checked whether the value is valid (in
which case it is guaranteed to be present). This worked pretty well so
far, yet the EBML specifications have evolved: It is now legal to use
zero-length elements for floats, ints, uints and strings (both ASCII and
UTF-8); the value of these elements is the default value of the element
(if it has one) or zero for scalar types and an empty string for
strings. Furthermore, having a default value does no longer imply that
the element may be presumed to be present (with its default value) if it
is absent; this is only true if the element is mandatory, too.
These rules are designed to allow size savings as follows: Consider the
newly added FlagOriginal: It being zero means the track is not in its
original language, it being one means it is. For backward compatibility
reasons, neither of the two values may be inferred automatically in the
absence of the element. But one can still save a byte when one wants to
write the element with a value of zero, as one can write the integer with
a length of zero: 0x55AE 80 instead of 0x55AE 81 00. In the former case,
a parser has to infer the value of the element to be zero (which is the
element's default value).
When encountering an element with length zero, our parser always infers
a value of zero (or an empty string); this is wrong for values with
a different default value. It needs to infer the default value (or zero
in its absence) and this precludes using an invalid default value for
elements like FlagOriginal. Ergo one needs to be able to record whether
an element is present or not by other means. This patch allows to use a
simple counter for this. While just at it, some invalid and unnecessary
default values have been removed (mastering metadata elements used
default values of -1.0, despite these elements only being used if they
are > 0).
Reviewed-by: Ridley Combs <rcombs@rcombs.me>
Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@gmail.com>
2021-02-16 15:33:51 +02:00
{ MATROSKA_ID_TAGNAME , EBML_UTF8 , 0 , 0 , offsetof ( MatroskaTag , name ) } ,
{ MATROSKA_ID_TAGSTRING , EBML_UTF8 , 0 , 0 , offsetof ( MatroskaTag , string ) } ,
{ MATROSKA_ID_TAGLANG , EBML_STR , 0 , 0 , offsetof ( MatroskaTag , lang ) , { . s = " und " } } ,
{ MATROSKA_ID_TAGDEFAULT , EBML_UINT , 0 , 0 , offsetof ( MatroskaTag , def ) } ,
{ MATROSKA_ID_TAGDEFAULT_BUG , EBML_UINT , 0 , 0 , offsetof ( MatroskaTag , def ) } ,
{ MATROSKA_ID_SIMPLETAG , EBML_NEST , 0 , sizeof ( MatroskaTag ) , offsetof ( MatroskaTag , sub ) , { . n = matroska_simpletag } } ,
2019-05-17 00:30:06 +02:00
CHILD_OF ( matroska_tag )
2008-08-09 02:50:38 +03:00
} ;
2019-07-18 21:07:20 +02:00
static EbmlSyntax matroska_tagtargets [ ] = {
avformat/matroskadec: Allow to count the number of element occurences
Up until now, the generic EBML reader used by the Matroska demuxer did
not have the capability to record whether an element was actually
present or not; instead, in cases where it mattered one typically added
an invalid default value and checked whether the value is valid (in
which case it is guaranteed to be present). This worked pretty well so
far, yet the EBML specifications have evolved: It is now legal to use
zero-length elements for floats, ints, uints and strings (both ASCII and
UTF-8); the value of these elements is the default value of the element
(if it has one) or zero for scalar types and an empty string for
strings. Furthermore, having a default value does no longer imply that
the element may be presumed to be present (with its default value) if it
is absent; this is only true if the element is mandatory, too.
These rules are designed to allow size savings as follows: Consider the
newly added FlagOriginal: It being zero means the track is not in its
original language, it being one means it is. For backward compatibility
reasons, neither of the two values may be inferred automatically in the
absence of the element. But one can still save a byte when one wants to
write the element with a value of zero, as one can write the integer with
a length of zero: 0x55AE 80 instead of 0x55AE 81 00. In the former case,
a parser has to infer the value of the element to be zero (which is the
element's default value).
When encountering an element with length zero, our parser always infers
a value of zero (or an empty string); this is wrong for values with
a different default value. It needs to infer the default value (or zero
in its absence) and this precludes using an invalid default value for
elements like FlagOriginal. Ergo one needs to be able to record whether
an element is present or not by other means. This patch allows to use a
simple counter for this. While just at it, some invalid and unnecessary
default values have been removed (mastering metadata elements used
default values of -1.0, despite these elements only being used if they
are > 0).
Reviewed-by: Ridley Combs <rcombs@rcombs.me>
Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@gmail.com>
2021-02-16 15:33:51 +02:00
{ MATROSKA_ID_TAGTARGETS_TYPE , EBML_STR , 0 , 0 , offsetof ( MatroskaTagTarget , type ) } ,
{ MATROSKA_ID_TAGTARGETS_TYPEVALUE , EBML_UINT , 0 , 0 , offsetof ( MatroskaTagTarget , typevalue ) , { . u = 50 } } ,
{ MATROSKA_ID_TAGTARGETS_TRACKUID , EBML_UINT , 0 , 0 , offsetof ( MatroskaTagTarget , trackuid ) , { . u = 0 } } ,
{ MATROSKA_ID_TAGTARGETS_CHAPTERUID , EBML_UINT , 0 , 0 , offsetof ( MatroskaTagTarget , chapteruid ) , { . u = 0 } } ,
{ MATROSKA_ID_TAGTARGETS_ATTACHUID , EBML_UINT , 0 , 0 , offsetof ( MatroskaTagTarget , attachuid ) , { . u = 0 } } ,
2019-05-17 00:30:06 +02:00
CHILD_OF ( matroska_tag )
2009-02-15 17:53:55 +02:00
} ;
2019-07-18 21:07:20 +02:00
static EbmlSyntax matroska_tag [ ] = {
avformat/matroskadec: Allow to count the number of element occurences
Up until now, the generic EBML reader used by the Matroska demuxer did
not have the capability to record whether an element was actually
present or not; instead, in cases where it mattered one typically added
an invalid default value and checked whether the value is valid (in
which case it is guaranteed to be present). This worked pretty well so
far, yet the EBML specifications have evolved: It is now legal to use
zero-length elements for floats, ints, uints and strings (both ASCII and
UTF-8); the value of these elements is the default value of the element
(if it has one) or zero for scalar types and an empty string for
strings. Furthermore, having a default value does no longer imply that
the element may be presumed to be present (with its default value) if it
is absent; this is only true if the element is mandatory, too.
These rules are designed to allow size savings as follows: Consider the
newly added FlagOriginal: It being zero means the track is not in its
original language, it being one means it is. For backward compatibility
reasons, neither of the two values may be inferred automatically in the
absence of the element. But one can still save a byte when one wants to
write the element with a value of zero, as one can write the integer with
a length of zero: 0x55AE 80 instead of 0x55AE 81 00. In the former case,
a parser has to infer the value of the element to be zero (which is the
element's default value).
When encountering an element with length zero, our parser always infers
a value of zero (or an empty string); this is wrong for values with
a different default value. It needs to infer the default value (or zero
in its absence) and this precludes using an invalid default value for
elements like FlagOriginal. Ergo one needs to be able to record whether
an element is present or not by other means. This patch allows to use a
simple counter for this. While just at it, some invalid and unnecessary
default values have been removed (mastering metadata elements used
default values of -1.0, despite these elements only being used if they
are > 0).
Reviewed-by: Ridley Combs <rcombs@rcombs.me>
Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@gmail.com>
2021-02-16 15:33:51 +02:00
{ MATROSKA_ID_SIMPLETAG , EBML_NEST , 0 , sizeof ( MatroskaTag ) , offsetof ( MatroskaTags , tag ) , { . n = matroska_simpletag } } ,
{ MATROSKA_ID_TAGTARGETS , EBML_NEST , 0 , 0 , offsetof ( MatroskaTags , target ) , { . n = matroska_tagtargets } } ,
2019-05-17 00:30:06 +02:00
CHILD_OF ( matroska_tags )
2008-08-09 02:50:38 +03:00
} ;
2019-07-18 21:07:20 +02:00
static EbmlSyntax matroska_tags [ ] = {
avformat/matroskadec: Allow to count the number of element occurences
Up until now, the generic EBML reader used by the Matroska demuxer did
not have the capability to record whether an element was actually
present or not; instead, in cases where it mattered one typically added
an invalid default value and checked whether the value is valid (in
which case it is guaranteed to be present). This worked pretty well so
far, yet the EBML specifications have evolved: It is now legal to use
zero-length elements for floats, ints, uints and strings (both ASCII and
UTF-8); the value of these elements is the default value of the element
(if it has one) or zero for scalar types and an empty string for
strings. Furthermore, having a default value does no longer imply that
the element may be presumed to be present (with its default value) if it
is absent; this is only true if the element is mandatory, too.
These rules are designed to allow size savings as follows: Consider the
newly added FlagOriginal: It being zero means the track is not in its
original language, it being one means it is. For backward compatibility
reasons, neither of the two values may be inferred automatically in the
absence of the element. But one can still save a byte when one wants to
write the element with a value of zero, as one can write the integer with
a length of zero: 0x55AE 80 instead of 0x55AE 81 00. In the former case,
a parser has to infer the value of the element to be zero (which is the
element's default value).
When encountering an element with length zero, our parser always infers
a value of zero (or an empty string); this is wrong for values with
a different default value. It needs to infer the default value (or zero
in its absence) and this precludes using an invalid default value for
elements like FlagOriginal. Ergo one needs to be able to record whether
an element is present or not by other means. This patch allows to use a
simple counter for this. While just at it, some invalid and unnecessary
default values have been removed (mastering metadata elements used
default values of -1.0, despite these elements only being used if they
are > 0).
Reviewed-by: Ridley Combs <rcombs@rcombs.me>
Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@gmail.com>
2021-02-16 15:33:51 +02:00
{ MATROSKA_ID_TAG , EBML_NEST , 0 , sizeof ( MatroskaTags ) , offsetof ( MatroskaDemuxContext , tags ) , { . n = matroska_tag } } ,
2019-05-17 00:30:06 +02:00
CHILD_OF ( matroska_segment )
2008-08-05 03:40:18 +03:00
} ;
2019-07-18 21:07:20 +02:00
static EbmlSyntax matroska_seekhead_entry [ ] = {
avformat/matroskadec: Allow to count the number of element occurences
Up until now, the generic EBML reader used by the Matroska demuxer did
not have the capability to record whether an element was actually
present or not; instead, in cases where it mattered one typically added
an invalid default value and checked whether the value is valid (in
which case it is guaranteed to be present). This worked pretty well so
far, yet the EBML specifications have evolved: It is now legal to use
zero-length elements for floats, ints, uints and strings (both ASCII and
UTF-8); the value of these elements is the default value of the element
(if it has one) or zero for scalar types and an empty string for
strings. Furthermore, having a default value does no longer imply that
the element may be presumed to be present (with its default value) if it
is absent; this is only true if the element is mandatory, too.
These rules are designed to allow size savings as follows: Consider the
newly added FlagOriginal: It being zero means the track is not in its
original language, it being one means it is. For backward compatibility
reasons, neither of the two values may be inferred automatically in the
absence of the element. But one can still save a byte when one wants to
write the element with a value of zero, as one can write the integer with
a length of zero: 0x55AE 80 instead of 0x55AE 81 00. In the former case,
a parser has to infer the value of the element to be zero (which is the
element's default value).
When encountering an element with length zero, our parser always infers
a value of zero (or an empty string); this is wrong for values with
a different default value. It needs to infer the default value (or zero
in its absence) and this precludes using an invalid default value for
elements like FlagOriginal. Ergo one needs to be able to record whether
an element is present or not by other means. This patch allows to use a
simple counter for this. While just at it, some invalid and unnecessary
default values have been removed (mastering metadata elements used
default values of -1.0, despite these elements only being used if they
are > 0).
Reviewed-by: Ridley Combs <rcombs@rcombs.me>
Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@gmail.com>
2021-02-16 15:33:51 +02:00
{ MATROSKA_ID_SEEKID , EBML_UINT , 0 , 0 , offsetof ( MatroskaSeekhead , id ) } ,
{ MATROSKA_ID_SEEKPOSITION , EBML_UINT , 0 , 0 , offsetof ( MatroskaSeekhead , pos ) , { . u = - 1 } } ,
2019-05-17 00:30:06 +02:00
CHILD_OF ( matroska_seekhead )
2008-08-05 03:40:36 +03:00
} ;
2019-07-18 21:07:20 +02:00
static EbmlSyntax matroska_seekhead [ ] = {
avformat/matroskadec: Allow to count the number of element occurences
Up until now, the generic EBML reader used by the Matroska demuxer did
not have the capability to record whether an element was actually
present or not; instead, in cases where it mattered one typically added
an invalid default value and checked whether the value is valid (in
which case it is guaranteed to be present). This worked pretty well so
far, yet the EBML specifications have evolved: It is now legal to use
zero-length elements for floats, ints, uints and strings (both ASCII and
UTF-8); the value of these elements is the default value of the element
(if it has one) or zero for scalar types and an empty string for
strings. Furthermore, having a default value does no longer imply that
the element may be presumed to be present (with its default value) if it
is absent; this is only true if the element is mandatory, too.
These rules are designed to allow size savings as follows: Consider the
newly added FlagOriginal: It being zero means the track is not in its
original language, it being one means it is. For backward compatibility
reasons, neither of the two values may be inferred automatically in the
absence of the element. But one can still save a byte when one wants to
write the element with a value of zero, as one can write the integer with
a length of zero: 0x55AE 80 instead of 0x55AE 81 00. In the former case,
a parser has to infer the value of the element to be zero (which is the
element's default value).
When encountering an element with length zero, our parser always infers
a value of zero (or an empty string); this is wrong for values with
a different default value. It needs to infer the default value (or zero
in its absence) and this precludes using an invalid default value for
elements like FlagOriginal. Ergo one needs to be able to record whether
an element is present or not by other means. This patch allows to use a
simple counter for this. While just at it, some invalid and unnecessary
default values have been removed (mastering metadata elements used
default values of -1.0, despite these elements only being used if they
are > 0).
Reviewed-by: Ridley Combs <rcombs@rcombs.me>
Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@gmail.com>
2021-02-16 15:33:51 +02:00
{ MATROSKA_ID_SEEKENTRY , EBML_NEST , 0 , sizeof ( MatroskaSeekhead ) , offsetof ( MatroskaDemuxContext , seekhead ) , { . n = matroska_seekhead_entry } } ,
2019-05-17 00:30:06 +02:00
CHILD_OF ( matroska_segment )
2008-08-05 03:40:36 +03:00
} ;
2019-07-18 21:07:20 +02:00
static EbmlSyntax matroska_segment [ ] = {
2019-05-17 00:30:18 +02:00
{ MATROSKA_ID_CLUSTER , EBML_STOP } ,
avformat/matroskadec: Allow to count the number of element occurences
Up until now, the generic EBML reader used by the Matroska demuxer did
not have the capability to record whether an element was actually
present or not; instead, in cases where it mattered one typically added
an invalid default value and checked whether the value is valid (in
which case it is guaranteed to be present). This worked pretty well so
far, yet the EBML specifications have evolved: It is now legal to use
zero-length elements for floats, ints, uints and strings (both ASCII and
UTF-8); the value of these elements is the default value of the element
(if it has one) or zero for scalar types and an empty string for
strings. Furthermore, having a default value does no longer imply that
the element may be presumed to be present (with its default value) if it
is absent; this is only true if the element is mandatory, too.
These rules are designed to allow size savings as follows: Consider the
newly added FlagOriginal: It being zero means the track is not in its
original language, it being one means it is. For backward compatibility
reasons, neither of the two values may be inferred automatically in the
absence of the element. But one can still save a byte when one wants to
write the element with a value of zero, as one can write the integer with
a length of zero: 0x55AE 80 instead of 0x55AE 81 00. In the former case,
a parser has to infer the value of the element to be zero (which is the
element's default value).
When encountering an element with length zero, our parser always infers
a value of zero (or an empty string); this is wrong for values with
a different default value. It needs to infer the default value (or zero
in its absence) and this precludes using an invalid default value for
elements like FlagOriginal. Ergo one needs to be able to record whether
an element is present or not by other means. This patch allows to use a
simple counter for this. While just at it, some invalid and unnecessary
default values have been removed (mastering metadata elements used
default values of -1.0, despite these elements only being used if they
are > 0).
Reviewed-by: Ridley Combs <rcombs@rcombs.me>
Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@gmail.com>
2021-02-16 15:33:51 +02:00
{ MATROSKA_ID_INFO , EBML_LEVEL1 , 0 , 0 , 0 , { . n = matroska_info } } ,
{ MATROSKA_ID_TRACKS , EBML_LEVEL1 , 0 , 0 , 0 , { . n = matroska_tracks } } ,
{ MATROSKA_ID_ATTACHMENTS , EBML_LEVEL1 , 0 , 0 , 0 , { . n = matroska_attachments } } ,
{ MATROSKA_ID_CHAPTERS , EBML_LEVEL1 , 0 , 0 , 0 , { . n = matroska_chapters } } ,
{ MATROSKA_ID_CUES , EBML_LEVEL1 , 0 , 0 , 0 , { . n = matroska_index } } ,
{ MATROSKA_ID_TAGS , EBML_LEVEL1 , 0 , 0 , 0 , { . n = matroska_tags } } ,
{ MATROSKA_ID_SEEKHEAD , EBML_LEVEL1 , 0 , 0 , 0 , { . n = matroska_seekhead } } ,
2019-05-17 00:30:06 +02:00
{ 0 } /* We don't want to go back to level 0, so don't add the parent. */
2008-08-05 03:40:58 +03:00
} ;
2019-07-18 21:07:20 +02:00
static EbmlSyntax matroska_segments [ ] = {
avformat/matroskadec: Allow to count the number of element occurences
Up until now, the generic EBML reader used by the Matroska demuxer did
not have the capability to record whether an element was actually
present or not; instead, in cases where it mattered one typically added
an invalid default value and checked whether the value is valid (in
which case it is guaranteed to be present). This worked pretty well so
far, yet the EBML specifications have evolved: It is now legal to use
zero-length elements for floats, ints, uints and strings (both ASCII and
UTF-8); the value of these elements is the default value of the element
(if it has one) or zero for scalar types and an empty string for
strings. Furthermore, having a default value does no longer imply that
the element may be presumed to be present (with its default value) if it
is absent; this is only true if the element is mandatory, too.
These rules are designed to allow size savings as follows: Consider the
newly added FlagOriginal: It being zero means the track is not in its
original language, it being one means it is. For backward compatibility
reasons, neither of the two values may be inferred automatically in the
absence of the element. But one can still save a byte when one wants to
write the element with a value of zero, as one can write the integer with
a length of zero: 0x55AE 80 instead of 0x55AE 81 00. In the former case,
a parser has to infer the value of the element to be zero (which is the
element's default value).
When encountering an element with length zero, our parser always infers
a value of zero (or an empty string); this is wrong for values with
a different default value. It needs to infer the default value (or zero
in its absence) and this precludes using an invalid default value for
elements like FlagOriginal. Ergo one needs to be able to record whether
an element is present or not by other means. This patch allows to use a
simple counter for this. While just at it, some invalid and unnecessary
default values have been removed (mastering metadata elements used
default values of -1.0, despite these elements only being used if they
are > 0).
Reviewed-by: Ridley Combs <rcombs@rcombs.me>
Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@gmail.com>
2021-02-16 15:33:51 +02:00
{ MATROSKA_ID_SEGMENT , EBML_NEST , 0 , 0 , 0 , { . n = matroska_segment } } ,
2008-08-05 03:40:58 +03:00
{ 0 }
} ;
2019-07-18 21:07:20 +02:00
static EbmlSyntax matroska_blockmore [ ] = {
avformat/matroskadec: Allow to count the number of element occurences
Up until now, the generic EBML reader used by the Matroska demuxer did
not have the capability to record whether an element was actually
present or not; instead, in cases where it mattered one typically added
an invalid default value and checked whether the value is valid (in
which case it is guaranteed to be present). This worked pretty well so
far, yet the EBML specifications have evolved: It is now legal to use
zero-length elements for floats, ints, uints and strings (both ASCII and
UTF-8); the value of these elements is the default value of the element
(if it has one) or zero for scalar types and an empty string for
strings. Furthermore, having a default value does no longer imply that
the element may be presumed to be present (with its default value) if it
is absent; this is only true if the element is mandatory, too.
These rules are designed to allow size savings as follows: Consider the
newly added FlagOriginal: It being zero means the track is not in its
original language, it being one means it is. For backward compatibility
reasons, neither of the two values may be inferred automatically in the
absence of the element. But one can still save a byte when one wants to
write the element with a value of zero, as one can write the integer with
a length of zero: 0x55AE 80 instead of 0x55AE 81 00. In the former case,
a parser has to infer the value of the element to be zero (which is the
element's default value).
When encountering an element with length zero, our parser always infers
a value of zero (or an empty string); this is wrong for values with
a different default value. It needs to infer the default value (or zero
in its absence) and this precludes using an invalid default value for
elements like FlagOriginal. Ergo one needs to be able to record whether
an element is present or not by other means. This patch allows to use a
simple counter for this. While just at it, some invalid and unnecessary
default values have been removed (mastering metadata elements used
default values of -1.0, despite these elements only being used if they
are > 0).
Reviewed-by: Ridley Combs <rcombs@rcombs.me>
Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@gmail.com>
2021-02-16 15:33:51 +02:00
{ MATROSKA_ID_BLOCKADDID , EBML_UINT , 0 , 0 , offsetof ( MatroskaBlock , additional_id ) , { . u = 1 } } ,
{ MATROSKA_ID_BLOCKADDITIONAL , EBML_BIN , 0 , 0 , offsetof ( MatroskaBlock , additional ) } ,
2019-05-17 00:30:06 +02:00
CHILD_OF ( matroska_blockadditions )
2013-02-14 00:51:48 +03:00
} ;
2019-07-18 21:07:20 +02:00
static EbmlSyntax matroska_blockadditions [ ] = {
avformat/matroskadec: Allow to count the number of element occurences
Up until now, the generic EBML reader used by the Matroska demuxer did
not have the capability to record whether an element was actually
present or not; instead, in cases where it mattered one typically added
an invalid default value and checked whether the value is valid (in
which case it is guaranteed to be present). This worked pretty well so
far, yet the EBML specifications have evolved: It is now legal to use
zero-length elements for floats, ints, uints and strings (both ASCII and
UTF-8); the value of these elements is the default value of the element
(if it has one) or zero for scalar types and an empty string for
strings. Furthermore, having a default value does no longer imply that
the element may be presumed to be present (with its default value) if it
is absent; this is only true if the element is mandatory, too.
These rules are designed to allow size savings as follows: Consider the
newly added FlagOriginal: It being zero means the track is not in its
original language, it being one means it is. For backward compatibility
reasons, neither of the two values may be inferred automatically in the
absence of the element. But one can still save a byte when one wants to
write the element with a value of zero, as one can write the integer with
a length of zero: 0x55AE 80 instead of 0x55AE 81 00. In the former case,
a parser has to infer the value of the element to be zero (which is the
element's default value).
When encountering an element with length zero, our parser always infers
a value of zero (or an empty string); this is wrong for values with
a different default value. It needs to infer the default value (or zero
in its absence) and this precludes using an invalid default value for
elements like FlagOriginal. Ergo one needs to be able to record whether
an element is present or not by other means. This patch allows to use a
simple counter for this. While just at it, some invalid and unnecessary
default values have been removed (mastering metadata elements used
default values of -1.0, despite these elements only being used if they
are > 0).
Reviewed-by: Ridley Combs <rcombs@rcombs.me>
Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@gmail.com>
2021-02-16 15:33:51 +02:00
{ MATROSKA_ID_BLOCKMORE , EBML_NEST , 0 , 0 , 0 , { . n = matroska_blockmore } } ,
2019-05-17 00:30:06 +02:00
CHILD_OF ( matroska_blockgroup )
2013-02-14 00:51:48 +03:00
} ;
2019-07-18 21:07:20 +02:00
static EbmlSyntax matroska_blockgroup [ ] = {
avformat/matroskadec: Allow to count the number of element occurences
Up until now, the generic EBML reader used by the Matroska demuxer did
not have the capability to record whether an element was actually
present or not; instead, in cases where it mattered one typically added
an invalid default value and checked whether the value is valid (in
which case it is guaranteed to be present). This worked pretty well so
far, yet the EBML specifications have evolved: It is now legal to use
zero-length elements for floats, ints, uints and strings (both ASCII and
UTF-8); the value of these elements is the default value of the element
(if it has one) or zero for scalar types and an empty string for
strings. Furthermore, having a default value does no longer imply that
the element may be presumed to be present (with its default value) if it
is absent; this is only true if the element is mandatory, too.
These rules are designed to allow size savings as follows: Consider the
newly added FlagOriginal: It being zero means the track is not in its
original language, it being one means it is. For backward compatibility
reasons, neither of the two values may be inferred automatically in the
absence of the element. But one can still save a byte when one wants to
write the element with a value of zero, as one can write the integer with
a length of zero: 0x55AE 80 instead of 0x55AE 81 00. In the former case,
a parser has to infer the value of the element to be zero (which is the
element's default value).
When encountering an element with length zero, our parser always infers
a value of zero (or an empty string); this is wrong for values with
a different default value. It needs to infer the default value (or zero
in its absence) and this precludes using an invalid default value for
elements like FlagOriginal. Ergo one needs to be able to record whether
an element is present or not by other means. This patch allows to use a
simple counter for this. While just at it, some invalid and unnecessary
default values have been removed (mastering metadata elements used
default values of -1.0, despite these elements only being used if they
are > 0).
Reviewed-by: Ridley Combs <rcombs@rcombs.me>
Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@gmail.com>
2021-02-16 15:33:51 +02:00
{ MATROSKA_ID_BLOCK , EBML_BIN , 0 , 0 , offsetof ( MatroskaBlock , bin ) } ,
{ MATROSKA_ID_BLOCKADDITIONS , EBML_NEST , 0 , 0 , 0 , { . n = matroska_blockadditions } } ,
{ MATROSKA_ID_BLOCKDURATION , EBML_UINT , 0 , 0 , offsetof ( MatroskaBlock , duration ) } ,
{ MATROSKA_ID_DISCARDPADDING , EBML_SINT , 0 , 0 , offsetof ( MatroskaBlock , discard_padding ) } ,
2021-02-16 17:18:02 +02:00
{ MATROSKA_ID_BLOCKREFERENCE , EBML_SINT , 1 , 0 , offsetof ( MatroskaBlock , reference ) } ,
2013-05-15 16:48:15 +03:00
{ MATROSKA_ID_CODECSTATE , EBML_NONE } ,
avformat/matroskadec: Allow to count the number of element occurences
Up until now, the generic EBML reader used by the Matroska demuxer did
not have the capability to record whether an element was actually
present or not; instead, in cases where it mattered one typically added
an invalid default value and checked whether the value is valid (in
which case it is guaranteed to be present). This worked pretty well so
far, yet the EBML specifications have evolved: It is now legal to use
zero-length elements for floats, ints, uints and strings (both ASCII and
UTF-8); the value of these elements is the default value of the element
(if it has one) or zero for scalar types and an empty string for
strings. Furthermore, having a default value does no longer imply that
the element may be presumed to be present (with its default value) if it
is absent; this is only true if the element is mandatory, too.
These rules are designed to allow size savings as follows: Consider the
newly added FlagOriginal: It being zero means the track is not in its
original language, it being one means it is. For backward compatibility
reasons, neither of the two values may be inferred automatically in the
absence of the element. But one can still save a byte when one wants to
write the element with a value of zero, as one can write the integer with
a length of zero: 0x55AE 80 instead of 0x55AE 81 00. In the former case,
a parser has to infer the value of the element to be zero (which is the
element's default value).
When encountering an element with length zero, our parser always infers
a value of zero (or an empty string); this is wrong for values with
a different default value. It needs to infer the default value (or zero
in its absence) and this precludes using an invalid default value for
elements like FlagOriginal. Ergo one needs to be able to record whether
an element is present or not by other means. This patch allows to use a
simple counter for this. While just at it, some invalid and unnecessary
default values have been removed (mastering metadata elements used
default values of -1.0, despite these elements only being used if they
are > 0).
Reviewed-by: Ridley Combs <rcombs@rcombs.me>
Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@gmail.com>
2021-02-16 15:33:51 +02:00
{ 1 , EBML_UINT , 0 , 0 , offsetof ( MatroskaBlock , non_simple ) , { . u = 1 } } ,
2019-05-17 00:30:06 +02:00
CHILD_OF ( matroska_cluster_parsing )
2008-08-05 03:41:05 +03:00
} ;
2019-05-17 00:30:10 +02:00
// The following array contains SimpleBlock and BlockGroup twice
// in order to reuse the other values for matroska_cluster_enter.
2019-07-18 21:07:20 +02:00
static EbmlSyntax matroska_cluster_parsing [ ] = {
avformat/matroskadec: Allow to count the number of element occurences
Up until now, the generic EBML reader used by the Matroska demuxer did
not have the capability to record whether an element was actually
present or not; instead, in cases where it mattered one typically added
an invalid default value and checked whether the value is valid (in
which case it is guaranteed to be present). This worked pretty well so
far, yet the EBML specifications have evolved: It is now legal to use
zero-length elements for floats, ints, uints and strings (both ASCII and
UTF-8); the value of these elements is the default value of the element
(if it has one) or zero for scalar types and an empty string for
strings. Furthermore, having a default value does no longer imply that
the element may be presumed to be present (with its default value) if it
is absent; this is only true if the element is mandatory, too.
These rules are designed to allow size savings as follows: Consider the
newly added FlagOriginal: It being zero means the track is not in its
original language, it being one means it is. For backward compatibility
reasons, neither of the two values may be inferred automatically in the
absence of the element. But one can still save a byte when one wants to
write the element with a value of zero, as one can write the integer with
a length of zero: 0x55AE 80 instead of 0x55AE 81 00. In the former case,
a parser has to infer the value of the element to be zero (which is the
element's default value).
When encountering an element with length zero, our parser always infers
a value of zero (or an empty string); this is wrong for values with
a different default value. It needs to infer the default value (or zero
in its absence) and this precludes using an invalid default value for
elements like FlagOriginal. Ergo one needs to be able to record whether
an element is present or not by other means. This patch allows to use a
simple counter for this. While just at it, some invalid and unnecessary
default values have been removed (mastering metadata elements used
default values of -1.0, despite these elements only being used if they
are > 0).
Reviewed-by: Ridley Combs <rcombs@rcombs.me>
Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@gmail.com>
2021-02-16 15:33:51 +02:00
{ MATROSKA_ID_SIMPLEBLOCK , EBML_BIN , 0 , 0 , offsetof ( MatroskaBlock , bin ) } ,
{ MATROSKA_ID_BLOCKGROUP , EBML_NEST , 0 , 0 , 0 , { . n = matroska_blockgroup } } ,
{ MATROSKA_ID_CLUSTERTIMECODE , EBML_UINT , 0 , 0 , offsetof ( MatroskaCluster , timecode ) } ,
2014-03-01 19:28:15 +03:00
{ MATROSKA_ID_SIMPLEBLOCK , EBML_STOP } ,
2019-05-17 00:30:10 +02:00
{ MATROSKA_ID_BLOCKGROUP , EBML_STOP } ,
2014-03-01 19:28:15 +03:00
{ MATROSKA_ID_CLUSTERPOSITION , EBML_NONE } ,
{ MATROSKA_ID_CLUSTERPREVSIZE , EBML_NONE } ,
2019-05-17 00:30:06 +02:00
CHILD_OF ( matroska_segment )
2012-04-19 21:12:24 +03:00
} ;
2019-07-18 21:07:20 +02:00
static EbmlSyntax matroska_cluster_enter [ ] = {
avformat/matroskadec: Allow to count the number of element occurences
Up until now, the generic EBML reader used by the Matroska demuxer did
not have the capability to record whether an element was actually
present or not; instead, in cases where it mattered one typically added
an invalid default value and checked whether the value is valid (in
which case it is guaranteed to be present). This worked pretty well so
far, yet the EBML specifications have evolved: It is now legal to use
zero-length elements for floats, ints, uints and strings (both ASCII and
UTF-8); the value of these elements is the default value of the element
(if it has one) or zero for scalar types and an empty string for
strings. Furthermore, having a default value does no longer imply that
the element may be presumed to be present (with its default value) if it
is absent; this is only true if the element is mandatory, too.
These rules are designed to allow size savings as follows: Consider the
newly added FlagOriginal: It being zero means the track is not in its
original language, it being one means it is. For backward compatibility
reasons, neither of the two values may be inferred automatically in the
absence of the element. But one can still save a byte when one wants to
write the element with a value of zero, as one can write the integer with
a length of zero: 0x55AE 80 instead of 0x55AE 81 00. In the former case,
a parser has to infer the value of the element to be zero (which is the
element's default value).
When encountering an element with length zero, our parser always infers
a value of zero (or an empty string); this is wrong for values with
a different default value. It needs to infer the default value (or zero
in its absence) and this precludes using an invalid default value for
elements like FlagOriginal. Ergo one needs to be able to record whether
an element is present or not by other means. This patch allows to use a
simple counter for this. While just at it, some invalid and unnecessary
default values have been removed (mastering metadata elements used
default values of -1.0, despite these elements only being used if they
are > 0).
Reviewed-by: Ridley Combs <rcombs@rcombs.me>
Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@gmail.com>
2021-02-16 15:33:51 +02:00
{ MATROSKA_ID_CLUSTER , EBML_NEST , 0 , 0 , 0 , { . n = & matroska_cluster_parsing [ 2 ] } } ,
2019-05-17 00:30:08 +02:00
{ 0 }
} ;
2019-05-17 00:30:06 +02:00
# undef CHILD_OF
2012-04-19 21:12:24 +03:00
2020-04-16 03:39:05 +02:00
static const CodecMime mkv_image_mime_tags [ ] = {
{ " image/gif " , AV_CODEC_ID_GIF } ,
{ " image/jpeg " , AV_CODEC_ID_MJPEG } ,
{ " image/png " , AV_CODEC_ID_PNG } ,
{ " image/tiff " , AV_CODEC_ID_TIFF } ,
{ " " , AV_CODEC_ID_NONE }
} ;
static const CodecMime mkv_mime_tags [ ] = {
{ " text/plain " , AV_CODEC_ID_TEXT } ,
{ " application/x-truetype-font " , AV_CODEC_ID_TTF } ,
{ " application/x-font " , AV_CODEC_ID_TTF } ,
{ " application/vnd.ms-opentype " , AV_CODEC_ID_OTF } ,
{ " binary " , AV_CODEC_ID_BIN_DATA } ,
{ " " , AV_CODEC_ID_NONE }
} ;
2012-02-20 11:42:33 +03:00
static const char * const matroska_doctypes [ ] = { " matroska " , " webm " } ;
2010-05-22 04:41:32 +03:00
2016-06-24 18:26:46 +02:00
static int matroska_read_close ( AVFormatContext * s ) ;
2019-05-17 00:30:02 +02:00
/*
* This function prepares the status for parsing of level 1 elements .
*/
static int matroska_reset_status ( MatroskaDemuxContext * matroska ,
uint32_t id , int64_t position )
{
if ( position > = 0 ) {
2019-08-17 02:27:51 +02:00
int64_t err = avio_seek ( matroska - > ctx - > pb , position , SEEK_SET ) ;
2019-05-17 00:30:02 +02:00
if ( err < 0 )
return err ;
}
2019-05-17 00:30:14 +02:00
matroska - > current_id = id ;
matroska - > num_levels = 1 ;
matroska - > unknown_count = 0 ;
avformat/matroskadec: Introduce a "last known good" position
Currently, resyncing during reading packets works as follows:
The current position is recorded, then a call to matroska_parse_cluster
is made and if said call fails, the demuxer tries to resync from the
earlier position. If the call doesn't fail, but also doesn't deliver a
packet, then this is looped.
There are two problems with this approach:
1. The Matroska file format aims to be forward-compatible; to achieve
this, a demuxer should simply ignore and skip elements it doesn't
know about. But it is not possible to reliably distinguish unknown
elements from junk. If matroska_parse_cluster encounters an unknown
element, it can therefore not simply error out; instead it returns zero
and the loop is iterated which includes an update of the position that
is intended to be used in case of errors, i.e. the element that is
skipped is not searched for level 1 element ids to resync to at all if
later calls to matroska_parse_cluster return an error.
Notice that in case that sync has been lost there can be a chain of
several unknown/possibly junk elements before an error is detected.
2. Even if a call to matroska_parse_cluster delivers a packet, this does
not mean that everything is fine. E.g. it might be that some of the
block's data is missing and that the data that was presumed to be from
the block just read actually contains the beginning of the next element.
This will only be apparent at the next call of matroska_read_packet,
which uses the (false) end of the earlier block as resync position so
that in the (not unlikely) case that the call to matroska_parse_cluster
fails, the data believed to be part of the earlier block is not searched
for a level 1 element to resync to.
To counter this, a "last known good" position is introduced. When an
element id that is known to be allowed at this position in the hierarchy
(according to the syntax currently in use for parsing) is read and some
further checks (regarding the length of the element and its containing
master element) are passed, then the beginning of the current element is
treated as a "good" position and recorded as such in the
MatroskaDemuxContext. Because of 2., only the start of the element is
treated as a "good" position, not the whole element. If an error occurs
later during parsing of clusters, the resync process starts at the last
known good position.
Given that when the header is damaged the subsequent resync never skips over
data and is therefore unaffected by both issues, the "last known good"
concept is not used there.
Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@gmail.com>
2019-05-17 00:30:05 +02:00
matroska - > resync_pos = avio_tell ( matroska - > ctx - > pb ) ;
if ( id )
matroska - > resync_pos - = ( av_log2 ( id ) + 7 ) / 8 ;
2019-05-17 00:30:02 +02:00
return 0 ;
}
2012-02-14 01:06:19 +03:00
static int matroska_resync ( MatroskaDemuxContext * matroska , int64_t last_pos )
{
AVIOContext * pb = matroska - > ctx - > pb ;
uint32_t id ;
2019-05-17 00:30:01 +02:00
/* Try to seek to the last position to resync from. If this doesn't work,
* we resync from the earliest position available : The start of the buffer . */
if ( last_pos < avio_tell ( pb ) & & avio_seek ( pb , last_pos + 1 , SEEK_SET ) < 0 ) {
av_log ( matroska - > ctx , AV_LOG_WARNING ,
" Seek to desired resync point failed. Seeking to "
" earliest point available instead. \n " ) ;
avio_seek ( pb , FFMAX ( avio_tell ( pb ) + ( pb - > buffer - pb - > buf_ptr ) ,
last_pos + 1 ) , SEEK_SET ) ;
2016-09-27 21:00:29 +02:00
}
2012-02-14 01:06:19 +03:00
id = avio_rb32 ( pb ) ;
// try to find a toplevel element
2014-08-07 23:12:41 +03:00
while ( ! avio_feof ( pb ) ) {
2013-05-28 01:11:50 +03:00
if ( id = = MATROSKA_ID_INFO | | id = = MATROSKA_ID_TRACKS | |
id = = MATROSKA_ID_CUES | | id = = MATROSKA_ID_TAGS | |
2012-02-14 01:06:19 +03:00
id = = MATROSKA_ID_SEEKHEAD | | id = = MATROSKA_ID_ATTACHMENTS | |
2013-05-28 01:11:50 +03:00
id = = MATROSKA_ID_CLUSTER | | id = = MATROSKA_ID_CHAPTERS ) {
2019-05-17 00:30:03 +02:00
/* Prepare the context for parsing of a level 1 element. */
matroska_reset_status ( matroska , id , - 1 ) ;
2019-08-28 19:04:42 +02:00
/* Given that we are here means that an error has occurred,
2019-05-17 00:30:03 +02:00
* so treat the segment as unknown length in order not to
* discard valid data that happens to be beyond the designated
* end of the segment . */
matroska - > levels [ 0 ] . length = EBML_UNKNOWN_LENGTH ;
2014-03-01 19:28:15 +03:00
return 0 ;
2012-02-14 01:06:19 +03:00
}
id = ( id < < 8 ) | avio_r8 ( pb ) ;
}
2014-03-01 19:28:15 +03:00
2012-02-14 01:06:19 +03:00
matroska - > done = 1 ;
2019-05-17 00:30:01 +02:00
return pb - > error ? pb - > error : AVERROR_EOF ;
2012-02-14 01:06:19 +03:00
}
2007-06-05 01:10:54 +03:00
/*
* Read : an " EBML number " , which is defined as a variable - length
* array of bytes . The first byte indicates the length by giving a
* number of 0 - bits followed by a one . The position of the first
* " one " bit inside the first byte indicates the length of this
* number .
2008-08-05 11:28:57 +03:00
* Returns : number of bytes read , < 0 on error
2007-06-05 01:10:54 +03:00
*/
2011-02-20 12:04:12 +02:00
static int ebml_read_num ( MatroskaDemuxContext * matroska , AVIOContext * pb ,
2019-06-25 03:08:56 +02:00
int max_size , uint64_t * number , int eof_forbidden )
2007-06-05 01:10:54 +03:00
{
2019-06-25 03:08:56 +02:00
int read , n = 1 ;
uint64_t total ;
int64_t pos ;
2007-06-05 01:10:54 +03:00
2019-06-25 03:08:56 +02:00
/* The first byte tells us the length in bytes - except when it is zero. */
total = avio_r8 ( pb ) ;
if ( pb - > eof_reached )
goto err ;
2007-06-05 01:10:54 +03:00
/* get the length of the EBML number */
2010-09-02 22:17:46 +03:00
read = 8 - ff_log2_tab [ total ] ;
2019-06-25 03:08:56 +02:00
if ( ! total | | read > max_size ) {
pos = avio_tell ( pb ) - 1 ;
if ( ! total ) {
av_log ( matroska - > ctx , AV_LOG_ERROR ,
" 0x00 at pos % " PRId64 " (0x% " PRIx64 " ) invalid as first byte "
" of an EBML number \n " , pos , pos ) ;
} else {
av_log ( matroska - > ctx , AV_LOG_ERROR ,
" Length %d indicated by an EBML number's first byte 0x%02x "
" at pos % " PRId64 " (0x% " PRIx64 " ) exceeds max length %d. \n " ,
read , ( uint8_t ) total , pos , pos , max_size ) ;
}
2007-06-05 01:10:54 +03:00
return AVERROR_INVALIDDATA ;
}
/* read out length */
2010-09-02 22:17:46 +03:00
total ^ = 1 < < ff_log2_tab [ total ] ;
2007-06-05 01:10:54 +03:00
while ( n + + < read )
2011-02-21 17:43:01 +02:00
total = ( total < < 8 ) | avio_r8 ( pb ) ;
2007-06-05 01:10:54 +03:00
2019-06-25 03:08:56 +02:00
if ( pb - > eof_reached ) {
eof_forbidden = 1 ;
goto err ;
}
2007-06-05 01:10:54 +03:00
* number = total ;
return read ;
2019-06-25 03:08:56 +02:00
err :
pos = avio_tell ( pb ) ;
if ( pb - > error ) {
av_log ( matroska - > ctx , AV_LOG_ERROR ,
" Read error at pos. % " PRIu64 " (0x% " PRIx64 " ) \n " ,
pos , pos ) ;
return pb - > error ;
}
if ( eof_forbidden ) {
av_log ( matroska - > ctx , AV_LOG_ERROR , " File ended prematurely "
" at pos. % " PRIu64 " (0x% " PRIx64 " ) \n " , pos , pos ) ;
return AVERROR ( EIO ) ;
}
return AVERROR_EOF ;
2007-06-05 01:10:54 +03:00
}
2010-09-06 20:51:44 +03:00
/**
* Read a EBML length value .
* This needs special handling for the " unknown length " case which has multiple
* encodings .
*/
2011-02-20 12:04:12 +02:00
static int ebml_read_length ( MatroskaDemuxContext * matroska , AVIOContext * pb ,
2010-09-06 20:51:44 +03:00
uint64_t * number )
{
2019-06-25 03:08:56 +02:00
int res = ebml_read_num ( matroska , pb , 8 , number , 1 ) ;
2010-09-06 20:51:44 +03:00
if ( res > 0 & & * number + 1 = = 1ULL < < ( 7 * res ) )
2019-02-23 12:14:33 +02:00
* number = EBML_UNKNOWN_LENGTH ;
2010-09-06 20:51:44 +03:00
return res ;
}
2007-06-05 01:10:54 +03:00
/*
* Read the next element as an unsigned int .
2021-02-16 17:51:40 +02:00
* Returns NEEDS_CHECKING unless size = = 0.
2007-06-05 01:10:54 +03:00
*/
2021-02-16 17:51:40 +02:00
static int ebml_read_uint ( AVIOContext * pb , int size ,
uint64_t default_value , uint64_t * num )
2007-06-05 01:10:54 +03:00
{
2008-08-05 03:41:55 +03:00
int n = 0 ;
2007-06-05 01:10:54 +03:00
2021-02-16 17:51:40 +02:00
if ( size = = 0 ) {
* num = default_value ;
return 0 ;
}
2008-08-05 11:28:57 +03:00
/* big-endian ordering; build up number */
2007-06-05 01:10:54 +03:00
* num = 0 ;
while ( n + + < size )
2011-02-21 17:43:01 +02:00
* num = ( * num < < 8 ) | avio_r8 ( pb ) ;
2007-06-05 01:10:54 +03:00
2019-06-24 01:42:30 +02:00
return NEEDS_CHECKING ;
2007-06-05 01:10:54 +03:00
}
2013-11-15 21:00:37 +03:00
/*
* Read the next element as a signed int .
2021-02-16 17:51:40 +02:00
* Returns NEEDS_CHECKING unless size = = 0.
2013-11-15 21:00:37 +03:00
*/
2021-02-16 17:51:40 +02:00
static int ebml_read_sint ( AVIOContext * pb , int size ,
int64_t default_value , int64_t * num )
2013-11-15 21:00:37 +03:00
{
int n = 1 ;
if ( size = = 0 ) {
2021-02-16 17:51:40 +02:00
* num = default_value ;
return 0 ;
2013-11-15 21:00:37 +03:00
} else {
2013-11-15 23:30:30 +03:00
* num = sign_extend ( avio_r8 ( pb ) , 8 ) ;
2013-11-15 21:00:37 +03:00
/* big-endian ordering; build up number */
while ( n + + < size )
2015-07-01 11:59:57 +02:00
* num = ( ( uint64_t ) * num < < 8 ) | avio_r8 ( pb ) ;
2013-11-15 21:00:37 +03:00
}
2019-06-24 01:42:30 +02:00
return NEEDS_CHECKING ;
2013-11-15 21:00:37 +03:00
}
2007-06-05 01:10:54 +03:00
/*
* Read the next element as a float .
2021-02-16 17:51:40 +02:00
* Returns 0 if size = = 0 , NEEDS_CHECKING or < 0 on obvious failure .
2007-06-05 01:10:54 +03:00
*/
2021-02-16 17:51:40 +02:00
static int ebml_read_float ( AVIOContext * pb , int size ,
double default_value , double * num )
2007-06-05 01:10:54 +03:00
{
2021-02-16 17:51:40 +02:00
if ( size = = 0 ) {
* num = default_value ;
return 0 ;
} else if ( size = = 4 ) {
2011-11-27 17:04:16 +03:00
* num = av_int2float ( avio_rb32 ( pb ) ) ;
2021-02-16 17:51:40 +02:00
} else if ( size = = 8 ) {
2011-11-27 17:04:16 +03:00
* num = av_int2double ( avio_rb64 ( pb ) ) ;
2021-02-16 17:51:40 +02:00
} else
2007-06-05 01:10:54 +03:00
return AVERROR_INVALIDDATA ;
2019-06-24 01:42:30 +02:00
return NEEDS_CHECKING ;
2007-06-05 01:10:54 +03:00
}
/*
* Read the next element as an ASCII string .
2019-06-24 01:42:30 +02:00
* 0 is success , < 0 or NEEDS_CHECKING is failure .
2007-06-05 01:10:54 +03:00
*/
2021-02-16 17:51:40 +02:00
static int ebml_read_ascii ( AVIOContext * pb , int size ,
const char * default_value , char * * str )
2007-06-05 01:10:54 +03:00
{
2012-02-25 03:12:18 +03:00
char * res ;
2019-06-24 01:42:30 +02:00
int ret ;
2012-02-25 03:12:18 +03:00
2021-02-16 17:51:40 +02:00
if ( size = = 0 & & default_value ) {
res = av_strdup ( default_value ) ;
if ( ! res )
return AVERROR ( ENOMEM ) ;
} else {
2021-02-16 18:14:03 +02:00
/* EBML strings are usually not 0-terminated, so we allocate one
* byte more , read the string and NUL - terminate it ourselves . */
if ( ! ( res = av_malloc ( size + 1 ) ) )
return AVERROR ( ENOMEM ) ;
if ( ( ret = avio_read ( pb , ( uint8_t * ) res , size ) ) ! = size ) {
av_free ( res ) ;
return ret < 0 ? ret : NEEDS_CHECKING ;
}
( res ) [ size ] = ' \0 ' ;
2021-02-16 17:51:40 +02:00
}
2012-02-25 03:12:18 +03:00
av_free ( * str ) ;
* str = res ;
2007-06-05 01:10:54 +03:00
return 0 ;
}
2008-08-05 03:42:39 +03:00
/*
* Read the next element as binary data .
2019-06-24 01:42:30 +02:00
* 0 is success , < 0 or NEEDS_CHECKING is failure .
2008-08-05 03:42:39 +03:00
*/
2019-05-17 00:30:12 +02:00
static int ebml_read_binary ( AVIOContext * pb , int length ,
int64_t pos , EbmlBin * bin )
2008-08-05 03:42:39 +03:00
{
2018-04-04 19:12:54 +02:00
int ret ;
ret = av_buffer_realloc ( & bin - > buf , length + AV_INPUT_BUFFER_PADDING_SIZE ) ;
if ( ret < 0 )
return ret ;
memset ( bin - > buf - > data + length , 0 , AV_INPUT_BUFFER_PADDING_SIZE ) ;
2008-08-05 03:42:39 +03:00
2018-04-04 19:12:54 +02:00
bin - > data = bin - > buf - > data ;
2008-08-05 03:42:39 +03:00
bin - > size = length ;
2019-05-17 00:30:12 +02:00
bin - > pos = pos ;
2019-06-24 01:42:30 +02:00
if ( ( ret = avio_read ( pb , bin - > data , length ) ) ! = length ) {
2018-04-04 19:12:54 +02:00
av_buffer_unref ( & bin - > buf ) ;
bin - > data = NULL ;
2012-12-04 05:30:40 +03:00
bin - > size = 0 ;
2019-06-24 01:42:30 +02:00
return ret < 0 ? ret : NEEDS_CHECKING ;
2010-05-19 00:21:37 +03:00
}
2008-08-05 03:42:39 +03:00
return 0 ;
}
2007-06-05 01:10:54 +03:00
/*
* Read the next element , but only the header . The contents
* are supposed to be sub - elements which can be read separately .
* 0 is success , < 0 is failure .
*/
2019-05-17 00:30:12 +02:00
static int ebml_read_master ( MatroskaDemuxContext * matroska ,
uint64_t length , int64_t pos )
2007-06-05 01:10:54 +03:00
{
MatroskaLevel * level ;
if ( matroska - > num_levels > = EBML_MAX_DEPTH ) {
av_log ( matroska - > ctx , AV_LOG_ERROR ,
" File moves beyond max. allowed depth (%d) \n " , EBML_MAX_DEPTH ) ;
2007-07-19 18:38:33 +03:00
return AVERROR ( ENOSYS ) ;
2007-06-05 01:10:54 +03:00
}
2014-03-01 19:28:15 +03:00
level = & matroska - > levels [ matroska - > num_levels + + ] ;
2019-05-17 00:30:12 +02:00
level - > start = pos ;
2007-06-05 01:10:54 +03:00
level - > length = length ;
return 0 ;
}
/*
2019-12-03 19:09:10 +02:00
* Read a signed " EBML number "
2008-08-05 11:28:57 +03:00
* Return : number of bytes processed , < 0 on error
2007-06-05 01:10:54 +03:00
*/
2008-08-05 03:42:52 +03:00
static int matroska_ebmlnum_sint ( MatroskaDemuxContext * matroska ,
2019-12-03 19:09:10 +02:00
AVIOContext * pb , int64_t * num )
2007-06-05 01:10:54 +03:00
{
uint64_t unum ;
int res ;
/* read as unsigned number first */
2019-12-03 19:09:10 +02:00
if ( ( res = ebml_read_num ( matroska , pb , 8 , & unum , 1 ) ) < 0 )
2007-06-05 01:10:54 +03:00
return res ;
/* make signed (weird way) */
2014-03-01 19:28:15 +03:00
* num = unum - ( ( 1LL < < ( 7 * res - 1 ) ) - 1 ) ;
2007-06-05 01:10:54 +03:00
return res ;
}
2019-05-17 00:30:04 +02:00
static int ebml_parse ( MatroskaDemuxContext * matroska ,
EbmlSyntax * syntax , void * data ) ;
2007-06-05 01:10:54 +03:00
2019-05-17 00:30:04 +02:00
static EbmlSyntax * ebml_parse_id ( EbmlSyntax * syntax , uint32_t id )
2007-06-05 01:10:54 +03:00
{
int i ;
2019-05-17 00:30:10 +02:00
// Whoever touches this should be aware of the duplication
// existing in matroska_cluster_parsing.
2014-03-01 19:28:15 +03:00
for ( i = 0 ; syntax [ i ] . id ; i + + )
2008-08-05 03:42:39 +03:00
if ( id = = syntax [ i ] . id )
break ;
2007-06-05 01:10:54 +03:00
2019-05-17 00:30:04 +02:00
return & syntax [ i ] ;
2007-06-05 01:10:54 +03:00
}
2008-08-05 03:42:39 +03:00
static int ebml_parse_nest ( MatroskaDemuxContext * matroska , EbmlSyntax * syntax ,
void * data )
2007-06-05 01:10:54 +03:00
{
2019-05-17 00:30:19 +02:00
int res ;
2007-06-05 01:10:54 +03:00
2019-05-17 00:30:18 +02:00
if ( data ) {
2021-02-16 18:21:14 +02:00
for ( int i = 0 ; syntax [ i ] . id ; i + + ) {
void * dst = ( char * ) data + syntax [ i ] . data_offset ;
2019-05-17 00:30:19 +02:00
switch ( syntax [ i ] . type ) {
case EBML_UINT :
2021-02-16 18:21:14 +02:00
* ( uint64_t * ) dst = syntax [ i ] . def . u ;
2019-05-17 00:30:19 +02:00
break ;
case EBML_SINT :
2021-02-16 18:21:14 +02:00
* ( int64_t * ) dst = syntax [ i ] . def . i ;
2019-05-17 00:30:19 +02:00
break ;
case EBML_FLOAT :
2021-02-16 18:21:14 +02:00
* ( double * ) dst = syntax [ i ] . def . f ;
2019-05-17 00:30:19 +02:00
break ;
case EBML_STR :
case EBML_UTF8 :
// the default may be NULL
if ( syntax [ i ] . def . s ) {
2021-02-16 18:21:14 +02:00
* ( char * * ) dst = av_strdup ( syntax [ i ] . def . s ) ;
if ( ! * ( char * * ) dst )
2019-05-17 00:30:19 +02:00
return AVERROR ( ENOMEM ) ;
}
break ;
2013-09-04 09:55:17 +03:00
}
2021-02-16 18:21:14 +02:00
}
2007-06-05 01:10:54 +03:00
2019-05-17 00:30:19 +02:00
if ( ! matroska - > levels [ matroska - > num_levels - 1 ] . length ) {
matroska - > num_levels - - ;
return 0 ;
}
2019-05-17 00:30:18 +02:00
}
2019-06-24 01:46:54 +02:00
do {
2008-08-05 03:42:39 +03:00
res = ebml_parse ( matroska , syntax , data ) ;
2019-06-24 01:46:54 +02:00
} while ( ! res ) ;
2007-06-05 01:10:54 +03:00
2019-06-24 01:46:54 +02:00
return res = = LEVEL_ENDED ? 0 : res ;
2007-06-05 01:10:54 +03:00
}
2015-06-12 13:11:41 +02:00
static int is_ebml_id_valid ( uint32_t id )
{
// Due to endian nonsense in Matroska, the highest byte with any bits set
// will contain the leading length bit. This bit in turn identifies the
// total byte length of the element by its position within the byte.
unsigned int bits = av_log2 ( id ) ;
return id & & ( bits + 7 ) / 8 = = ( 8 - bits % 8 ) ;
}
2015-02-09 21:39:00 +02:00
/*
* Allocate and return the entry for the level1 element with the given ID . If
* an entry already exists , return the existing entry .
*/
static MatroskaLevel1Element * matroska_find_level1_elem ( MatroskaDemuxContext * matroska ,
2020-04-30 23:16:44 +02:00
uint32_t id , int64_t pos )
2015-02-09 21:39:00 +02:00
{
int i ;
MatroskaLevel1Element * elem ;
2015-06-12 13:11:41 +02:00
if ( ! is_ebml_id_valid ( id ) )
return NULL ;
2015-02-09 21:39:00 +02:00
// Some files link to all clusters; useless.
if ( id = = MATROSKA_ID_CLUSTER )
return NULL ;
2020-05-01 00:49:49 +02:00
// There can be multiple SeekHeads and Tags.
2020-04-30 23:16:44 +02:00
for ( i = 0 ; i < matroska - > num_level1_elems ; i + + ) {
if ( matroska - > level1_elems [ i ] . id = = id ) {
if ( matroska - > level1_elems [ i ] . pos = = pos | |
2020-05-01 00:49:49 +02:00
id ! = MATROSKA_ID_SEEKHEAD & & id ! = MATROSKA_ID_TAGS )
2015-02-09 21:39:00 +02:00
return & matroska - > level1_elems [ i ] ;
}
}
// Only a completely broken file would have more elements.
if ( matroska - > num_level1_elems > = FF_ARRAY_ELEMS ( matroska - > level1_elems ) ) {
2020-04-30 23:16:44 +02:00
av_log ( matroska - > ctx , AV_LOG_ERROR , " Too many level1 elements. \n " ) ;
2015-02-09 21:39:00 +02:00
return NULL ;
}
elem = & matroska - > level1_elems [ matroska - > num_level1_elems + + ] ;
* elem = ( MatroskaLevel1Element ) { . id = id } ;
return elem ;
}
2019-05-17 00:30:04 +02:00
static int ebml_parse ( MatroskaDemuxContext * matroska ,
EbmlSyntax * syntax , void * data )
2008-08-05 03:40:00 +03:00
{
2011-02-06 12:32:03 +02:00
static const uint64_t max_lengths [ EBML_TYPE_COUNT ] = {
2019-05-17 00:30:15 +02:00
// Forbid unknown-length EBML_NONE elements.
[ EBML_NONE ] = EBML_UNKNOWN_LENGTH - 1 ,
2011-02-06 12:32:03 +02:00
[ EBML_UINT ] = 8 ,
2019-05-17 00:29:50 +02:00
[ EBML_SINT ] = 8 ,
2011-02-06 12:32:03 +02:00
[ EBML_FLOAT ] = 8 ,
// max. 16 MB for strings
[ EBML_STR ] = 0x1000000 ,
[ EBML_UTF8 ] = 0x1000000 ,
// max. 256 MB for binary data
[ EBML_BIN ] = 0x10000000 ,
// no limits for anything else
} ;
2011-02-20 12:04:12 +02:00
AVIOContext * pb = matroska - > ctx - > pb ;
2019-05-17 00:30:04 +02:00
uint32_t id ;
2008-08-05 03:41:55 +03:00
uint64_t length ;
2019-05-17 00:30:12 +02:00
int64_t pos = avio_tell ( pb ) , pos_alt ;
2019-06-24 01:46:54 +02:00
int res , update_pos = 1 , level_check ;
2015-02-09 21:39:00 +02:00
MatroskaLevel1Element * level1_elem ;
2019-06-24 01:46:54 +02:00
MatroskaLevel * level = matroska - > num_levels ? & matroska - > levels [ matroska - > num_levels - 1 ] : NULL ;
2008-08-05 03:40:00 +03:00
2019-05-17 00:30:04 +02:00
if ( ! matroska - > current_id ) {
uint64_t id ;
res = ebml_read_num ( matroska , pb , 4 , & id , 0 ) ;
if ( res < 0 ) {
2019-06-24 01:46:54 +02:00
if ( pb - > eof_reached & & res = = AVERROR_EOF ) {
if ( matroska - > is_live )
// in live mode, finish parsing if EOF is reached.
return 1 ;
2019-05-17 00:30:11 +02:00
if ( level & & pos = = avio_tell ( pb ) ) {
if ( level - > length = = EBML_UNKNOWN_LENGTH ) {
// Unknown-length levels automatically end at EOF.
matroska - > num_levels - - ;
return LEVEL_ENDED ;
} else {
av_log ( matroska - > ctx , AV_LOG_ERROR , " File ended prematurely "
" at pos. % " PRIu64 " (0x% " PRIx64 " ) \n " , pos , pos ) ;
}
2019-06-24 01:46:54 +02:00
}
}
return res ;
2019-05-17 00:30:04 +02:00
}
matroska - > current_id = id | 1 < < 7 * res ;
2019-05-17 00:30:12 +02:00
pos_alt = pos + res ;
} else {
pos_alt = pos ;
pos - = ( av_log2 ( matroska - > current_id ) + 7 ) / 8 ;
}
2019-05-17 00:30:04 +02:00
id = matroska - > current_id ;
syntax = ebml_parse_id ( syntax , id ) ;
if ( ! syntax - > id & & id ! = EBML_ID_VOID & & id ! = EBML_ID_CRC32 ) {
2019-06-24 01:46:54 +02:00
if ( level & & level - > length = = EBML_UNKNOWN_LENGTH ) {
// Unknown-length levels end when an element from an upper level
// in the hierarchy is encountered.
while ( syntax - > def . n ) {
syntax = ebml_parse_id ( syntax - > def . n , id ) ;
if ( syntax - > id ) {
matroska - > num_levels - - ;
return LEVEL_ENDED ;
}
} ;
}
av_log ( matroska - > ctx , AV_LOG_DEBUG , " Unknown entry 0x% " PRIX32 " at pos. "
" % " PRId64 " \n " , id , pos ) ;
update_pos = 0 ; /* Don't update resync_pos as an error might have happened. */
2019-05-17 00:30:04 +02:00
}
2019-05-17 00:30:18 +02:00
if ( data ) {
2019-05-17 00:30:19 +02:00
data = ( char * ) data + syntax - > data_offset ;
if ( syntax - > list_elem_size ) {
EbmlList * list = data ;
2019-09-03 23:45:04 +02:00
void * newelem ;
if ( ( unsigned ) list - > nb_elem + 1 > = UINT_MAX / syntax - > list_elem_size )
return AVERROR ( ENOMEM ) ;
newelem = av_fast_realloc ( list - > elem ,
& list - > alloc_elem_size ,
( list - > nb_elem + 1 ) * syntax - > list_elem_size ) ;
2019-05-17 00:30:19 +02:00
if ( ! newelem )
return AVERROR ( ENOMEM ) ;
list - > elem = newelem ;
data = ( char * ) list - > elem + list - > nb_elem * syntax - > list_elem_size ;
memset ( data , 0 , syntax - > list_elem_size ) ;
list - > nb_elem + + ;
}
2019-05-17 00:30:18 +02:00
}
2008-08-05 03:40:00 +03:00
2019-05-17 00:30:00 +02:00
if ( syntax - > type ! = EBML_STOP ) {
2010-06-11 19:34:01 +03:00
matroska - > current_id = 0 ;
2010-09-06 20:51:44 +03:00
if ( ( res = ebml_read_length ( matroska , pb , & length ) ) < 0 )
2008-08-05 03:41:55 +03:00
return res ;
2019-05-17 00:30:12 +02:00
pos_alt + = res ;
2019-02-13 14:20:45 +02:00
if ( matroska - > num_levels > 0 ) {
2019-03-27 13:18:44 +02:00
if ( length ! = EBML_UNKNOWN_LENGTH & &
level - > length ! = EBML_UNKNOWN_LENGTH ) {
2019-05-17 00:30:12 +02:00
uint64_t elem_end = pos_alt + length ,
2019-03-27 13:18:44 +02:00
level_end = level - > start + level - > length ;
2019-06-24 01:46:54 +02:00
if ( elem_end < level_end ) {
level_check = 0 ;
} else if ( elem_end = = level_end ) {
level_check = LEVEL_ENDED ;
} else {
2019-03-27 13:18:44 +02:00
av_log ( matroska - > ctx , AV_LOG_ERROR ,
" Element at 0x% " PRIx64 " ending at 0x% " PRIx64 " exceeds "
" containing master element ending at 0x% " PRIx64 " \n " ,
pos , elem_end , level_end ) ;
return AVERROR_INVALIDDATA ;
}
2019-06-24 01:46:54 +02:00
} else if ( length ! = EBML_UNKNOWN_LENGTH ) {
level_check = 0 ;
2019-03-27 13:18:44 +02:00
} else if ( level - > length ! = EBML_UNKNOWN_LENGTH ) {
av_log ( matroska - > ctx , AV_LOG_ERROR , " Unknown-sized element "
" at 0x% " PRIx64 " inside parent with finite size \n " , pos ) ;
return AVERROR_INVALIDDATA ;
2019-05-17 00:30:16 +02:00
} else {
2019-06-24 01:46:54 +02:00
level_check = 0 ;
2019-05-17 00:30:16 +02:00
if ( id ! = MATROSKA_ID_CLUSTER & & ( syntax - > type = = EBML_LEVEL1
| | syntax - > type = = EBML_NEST ) ) {
// According to the current specifications only clusters and
// segments are allowed to be unknown-length. We also accept
// other unknown-length master elements.
av_log ( matroska - > ctx , AV_LOG_WARNING ,
" Found unknown-length element 0x% " PRIX32 " other than "
" a cluster at 0x% " PRIx64 " . Spec-incompliant, but "
" parsing will nevertheless be attempted. \n " , id , pos ) ;
update_pos = - 1 ;
}
}
2019-06-24 01:46:54 +02:00
} else
level_check = 0 ;
avformat/matroskadec: Introduce a "last known good" position
Currently, resyncing during reading packets works as follows:
The current position is recorded, then a call to matroska_parse_cluster
is made and if said call fails, the demuxer tries to resync from the
earlier position. If the call doesn't fail, but also doesn't deliver a
packet, then this is looped.
There are two problems with this approach:
1. The Matroska file format aims to be forward-compatible; to achieve
this, a demuxer should simply ignore and skip elements it doesn't
know about. But it is not possible to reliably distinguish unknown
elements from junk. If matroska_parse_cluster encounters an unknown
element, it can therefore not simply error out; instead it returns zero
and the loop is iterated which includes an update of the position that
is intended to be used in case of errors, i.e. the element that is
skipped is not searched for level 1 element ids to resync to at all if
later calls to matroska_parse_cluster return an error.
Notice that in case that sync has been lost there can be a chain of
several unknown/possibly junk elements before an error is detected.
2. Even if a call to matroska_parse_cluster delivers a packet, this does
not mean that everything is fine. E.g. it might be that some of the
block's data is missing and that the data that was presumed to be from
the block just read actually contains the beginning of the next element.
This will only be apparent at the next call of matroska_read_packet,
which uses the (false) end of the earlier block as resync position so
that in the (not unlikely) case that the call to matroska_parse_cluster
fails, the data believed to be part of the earlier block is not searched
for a level 1 element to resync to.
To counter this, a "last known good" position is introduced. When an
element id that is known to be allowed at this position in the hierarchy
(according to the syntax currently in use for parsing) is read and some
further checks (regarding the length of the element and its containing
master element) are passed, then the beginning of the current element is
treated as a "good" position and recorded as such in the
MatroskaDemuxContext. Because of 2., only the start of the element is
treated as a "good" position, not the whole element. If an error occurs
later during parsing of clusters, the resync process starts at the last
known good position.
Given that when the header is damaged the subsequent resync never skips over
data and is therefore unaffected by both issues, the "last known good"
concept is not used there.
Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@gmail.com>
2019-05-17 00:30:05 +02:00
2019-05-17 00:30:15 +02:00
if ( max_lengths [ syntax - > type ] & & length > max_lengths [ syntax - > type ] ) {
if ( length ! = EBML_UNKNOWN_LENGTH ) {
av_log ( matroska - > ctx , AV_LOG_ERROR ,
" Invalid length 0x% " PRIx64 " > 0x% " PRIx64 " for element "
" with ID 0x% " PRIX32 " at 0x% " PRIx64 " \n " ,
length , max_lengths [ syntax - > type ] , id , pos ) ;
} else if ( syntax - > type ! = EBML_NONE ) {
av_log ( matroska - > ctx , AV_LOG_ERROR ,
" Element with ID 0x% " PRIX32 " at pos. 0x% " PRIx64 " has "
" unknown length, yet the length of an element of its "
" type must be known. \n " , id , pos ) ;
} else {
av_log ( matroska - > ctx , AV_LOG_ERROR ,
" Found unknown-length element with ID 0x% " PRIX32 " at "
" pos. 0x% " PRIx64 " for which no syntax for parsing is "
" available. \n " , id , pos ) ;
}
return AVERROR_INVALIDDATA ;
}
2019-05-17 00:30:14 +02:00
if ( ! ( pb - > seekable & AVIO_SEEKABLE_NORMAL ) ) {
// Loosing sync will likely manifest itself as encountering unknown
// elements which are not reliably distinguishable from elements
// belonging to future extensions of the format.
// We use a heuristic to detect such situations: If the current
// element is not expected at the current syntax level and there
// were only a few unknown elements in a row, then the element is
// skipped or considered defective based upon the length of the
// current element (i.e. how much would be skipped); if there were
// more than a few skipped elements in a row and skipping the current
// element would lead us more than SKIP_THRESHOLD away from the last
2019-08-28 19:04:42 +02:00
// known good position, then it is inferred that an error occurred.
2019-05-17 00:30:14 +02:00
// The dependency on the number of unknown elements in a row exists
// because the distance to the last known good position is
// automatically big if the last parsed element was big.
// In both cases, each unknown element is considered equivalent to
// UNKNOWN_EQUIV of skipped bytes for the check.
// The whole check is only done for non-seekable output, because
// in this situation skipped data can't simply be rechecked later.
// This is especially important when using unkown length elements
// as the check for whether a child exceeds its containing master
// element is not effective in this situation.
if ( update_pos ) {
matroska - > unknown_count = 0 ;
} else {
int64_t dist = length + UNKNOWN_EQUIV * matroska - > unknown_count + + ;
if ( matroska - > unknown_count > 3 )
dist + = pos_alt - matroska - > resync_pos ;
if ( dist > SKIP_THRESHOLD ) {
av_log ( matroska - > ctx , AV_LOG_ERROR ,
" Unknown element % " PRIX32 " at pos. 0x% " PRIx64 " with "
" length 0x% " PRIx64 " considered as invalid data. Last "
" known good position 0x% " PRIx64 " , %d unknown elements "
" in a row \n " , id , pos , length , matroska - > resync_pos ,
matroska - > unknown_count ) ;
return AVERROR_INVALIDDATA ;
}
}
}
2019-05-17 00:30:16 +02:00
if ( update_pos > 0 ) {
avformat/matroskadec: Introduce a "last known good" position
Currently, resyncing during reading packets works as follows:
The current position is recorded, then a call to matroska_parse_cluster
is made and if said call fails, the demuxer tries to resync from the
earlier position. If the call doesn't fail, but also doesn't deliver a
packet, then this is looped.
There are two problems with this approach:
1. The Matroska file format aims to be forward-compatible; to achieve
this, a demuxer should simply ignore and skip elements it doesn't
know about. But it is not possible to reliably distinguish unknown
elements from junk. If matroska_parse_cluster encounters an unknown
element, it can therefore not simply error out; instead it returns zero
and the loop is iterated which includes an update of the position that
is intended to be used in case of errors, i.e. the element that is
skipped is not searched for level 1 element ids to resync to at all if
later calls to matroska_parse_cluster return an error.
Notice that in case that sync has been lost there can be a chain of
several unknown/possibly junk elements before an error is detected.
2. Even if a call to matroska_parse_cluster delivers a packet, this does
not mean that everything is fine. E.g. it might be that some of the
block's data is missing and that the data that was presumed to be from
the block just read actually contains the beginning of the next element.
This will only be apparent at the next call of matroska_read_packet,
which uses the (false) end of the earlier block as resync position so
that in the (not unlikely) case that the call to matroska_parse_cluster
fails, the data believed to be part of the earlier block is not searched
for a level 1 element to resync to.
To counter this, a "last known good" position is introduced. When an
element id that is known to be allowed at this position in the hierarchy
(according to the syntax currently in use for parsing) is read and some
further checks (regarding the length of the element and its containing
master element) are passed, then the beginning of the current element is
treated as a "good" position and recorded as such in the
MatroskaDemuxContext. Because of 2., only the start of the element is
treated as a "good" position, not the whole element. If an error occurs
later during parsing of clusters, the resync process starts at the last
known good position.
Given that when the header is damaged the subsequent resync never skips over
data and is therefore unaffected by both issues, the "last known good"
concept is not used there.
Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@gmail.com>
2019-05-17 00:30:05 +02:00
// We have found an element that is allowed at this place
// in the hierarchy and it passed all checks, so treat the beginning
// of the element as the "last known good" position.
matroska - > resync_pos = pos ;
}
2019-05-17 00:30:18 +02:00
if ( ! data & & length ! = EBML_UNKNOWN_LENGTH )
goto skip ;
2010-06-11 19:34:01 +03:00
}
2008-08-05 03:41:55 +03:00
2008-08-05 03:40:00 +03:00
switch ( syntax - > type ) {
2014-03-01 19:28:15 +03:00
case EBML_UINT :
2021-02-16 17:51:40 +02:00
res = ebml_read_uint ( pb , length , syntax - > def . u , data ) ;
2014-03-01 19:28:15 +03:00
break ;
2014-03-08 01:46:37 +03:00
case EBML_SINT :
2021-02-16 17:51:40 +02:00
res = ebml_read_sint ( pb , length , syntax - > def . i , data ) ;
2014-03-08 01:46:37 +03:00
break ;
2014-03-01 19:28:15 +03:00
case EBML_FLOAT :
2021-02-16 17:51:40 +02:00
res = ebml_read_float ( pb , length , syntax - > def . f , data ) ;
2014-03-01 19:28:15 +03:00
break ;
2008-08-05 03:40:00 +03:00
case EBML_STR :
2014-03-01 19:28:15 +03:00
case EBML_UTF8 :
2021-02-16 17:51:40 +02:00
res = ebml_read_ascii ( pb , length , syntax - > def . s , data ) ;
2014-03-01 19:28:15 +03:00
break ;
case EBML_BIN :
2019-05-17 00:30:12 +02:00
res = ebml_read_binary ( pb , length , pos_alt , data ) ;
2014-03-01 19:28:15 +03:00
break ;
2015-02-09 21:39:00 +02:00
case EBML_LEVEL1 :
2014-03-01 19:28:15 +03:00
case EBML_NEST :
2019-05-17 00:30:12 +02:00
if ( ( res = ebml_read_master ( matroska , length , pos_alt ) ) < 0 )
2014-03-01 19:28:15 +03:00
return res ;
if ( id = = MATROSKA_ID_SEGMENT )
2019-05-17 00:30:12 +02:00
matroska - > segment_start = pos_alt ;
2015-02-09 21:39:00 +02:00
if ( id = = MATROSKA_ID_CUES )
matroska - > cues_parsing_deferred = 0 ;
if ( syntax - > type = = EBML_LEVEL1 & &
2020-04-30 23:16:44 +02:00
( level1_elem = matroska_find_level1_elem ( matroska , syntax - > id , pos ) ) ) {
2019-05-17 00:30:21 +02:00
if ( ! level1_elem - > pos ) {
// Zero is not a valid position for a level 1 element.
level1_elem - > pos = pos ;
} else if ( level1_elem - > pos ! = pos )
2015-02-09 21:39:00 +02:00
av_log ( matroska - > ctx , AV_LOG_ERROR , " Duplicate element \n " ) ;
level1_elem - > parsed = 1 ;
}
2019-06-24 01:46:54 +02:00
if ( res = ebml_parse_nest ( matroska , syntax - > def . n , data ) )
return res ;
break ;
2014-03-01 19:28:15 +03:00
case EBML_STOP :
return 1 ;
2019-05-17 00:30:18 +02:00
skip :
2012-08-04 03:27:51 +03:00
default :
2019-06-24 01:42:31 +02:00
if ( length ) {
2019-07-06 18:59:22 +02:00
int64_t res2 ;
2019-06-24 01:42:31 +02:00
if ( ffio_limit ( pb , length ) ! = length ) {
// ffio_limit emits its own error message,
// so we don't have to.
return AVERROR ( EIO ) ;
}
2019-07-06 18:59:22 +02:00
if ( ( res2 = avio_skip ( pb , length - 1 ) ) > = 0 ) {
2019-06-24 01:42:31 +02:00
// avio_skip might take us past EOF. We check for this
// by skipping only length - 1 bytes, reading a byte and
// checking the error flags. This is done in order to check
// that the element has been properly skipped even when
// no filesize (that ffio_limit relies on) is available.
avio_r8 ( pb ) ;
res = NEEDS_CHECKING ;
2019-07-06 18:59:22 +02:00
} else
res = res2 ;
2019-06-24 01:42:31 +02:00
} else
res = 0 ;
2019-06-24 01:42:30 +02:00
}
if ( res ) {
if ( res = = NEEDS_CHECKING ) {
if ( pb - > eof_reached ) {
if ( pb - > error )
res = pb - > error ;
else
res = AVERROR_EOF ;
} else
2019-06-24 01:46:54 +02:00
goto level_check ;
2019-06-24 01:42:30 +02:00
}
if ( res = = AVERROR_INVALIDDATA )
av_log ( matroska - > ctx , AV_LOG_ERROR , " Invalid element \n " ) ;
else if ( res = = AVERROR ( EIO ) )
av_log ( matroska - > ctx , AV_LOG_ERROR , " Read error \n " ) ;
else if ( res = = AVERROR_EOF ) {
av_log ( matroska - > ctx , AV_LOG_ERROR , " File ended prematurely \n " ) ;
res = AVERROR ( EIO ) ;
}
2019-06-24 01:46:54 +02:00
return res ;
2008-08-05 03:40:00 +03:00
}
2019-06-24 01:46:54 +02:00
level_check :
avformat/matroskadec: Allow to count the number of element occurences
Up until now, the generic EBML reader used by the Matroska demuxer did
not have the capability to record whether an element was actually
present or not; instead, in cases where it mattered one typically added
an invalid default value and checked whether the value is valid (in
which case it is guaranteed to be present). This worked pretty well so
far, yet the EBML specifications have evolved: It is now legal to use
zero-length elements for floats, ints, uints and strings (both ASCII and
UTF-8); the value of these elements is the default value of the element
(if it has one) or zero for scalar types and an empty string for
strings. Furthermore, having a default value does no longer imply that
the element may be presumed to be present (with its default value) if it
is absent; this is only true if the element is mandatory, too.
These rules are designed to allow size savings as follows: Consider the
newly added FlagOriginal: It being zero means the track is not in its
original language, it being one means it is. For backward compatibility
reasons, neither of the two values may be inferred automatically in the
absence of the element. But one can still save a byte when one wants to
write the element with a value of zero, as one can write the integer with
a length of zero: 0x55AE 80 instead of 0x55AE 81 00. In the former case,
a parser has to infer the value of the element to be zero (which is the
element's default value).
When encountering an element with length zero, our parser always infers
a value of zero (or an empty string); this is wrong for values with
a different default value. It needs to infer the default value (or zero
in its absence) and this precludes using an invalid default value for
elements like FlagOriginal. Ergo one needs to be able to record whether
an element is present or not by other means. This patch allows to use a
simple counter for this. While just at it, some invalid and unnecessary
default values have been removed (mastering metadata elements used
default values of -1.0, despite these elements only being used if they
are > 0).
Reviewed-by: Ridley Combs <rcombs@rcombs.me>
Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@gmail.com>
2021-02-16 15:33:51 +02:00
if ( syntax - > is_counted & & data ) {
CountedElement * elem = data ;
if ( elem - > count ! = UINT_MAX )
elem - > count + + ;
}
2019-06-24 01:46:54 +02:00
if ( level_check = = LEVEL_ENDED & & matroska - > num_levels ) {
level = & matroska - > levels [ matroska - > num_levels - 1 ] ;
pos = avio_tell ( pb ) ;
// Given that pos >= level->start no check for
// level->length != EBML_UNKNOWN_LENGTH is necessary.
while ( matroska - > num_levels & & pos = = level - > start + level - > length ) {
matroska - > num_levels - - ;
level - - ;
}
}
return level_check ;
2008-08-05 03:40:00 +03:00
}
static void ebml_free ( EbmlSyntax * syntax , void * data )
{
int i , j ;
2014-03-01 19:28:15 +03:00
for ( i = 0 ; syntax [ i ] . id ; i + + ) {
void * data_off = ( char * ) data + syntax [ i ] . data_offset ;
2008-08-05 03:40:00 +03:00
switch ( syntax [ i ] . type ) {
case EBML_STR :
2014-03-01 19:28:15 +03:00
case EBML_UTF8 :
av_freep ( data_off ) ;
break ;
case EBML_BIN :
2018-04-04 19:12:54 +02:00
av_buffer_unref ( & ( ( EbmlBin * ) data_off ) - > buf ) ;
2014-03-01 19:28:15 +03:00
break ;
2015-02-09 21:39:00 +02:00
case EBML_LEVEL1 :
2008-08-05 03:40:00 +03:00
case EBML_NEST :
if ( syntax [ i ] . list_elem_size ) {
EbmlList * list = data_off ;
char * ptr = list - > elem ;
2014-03-01 19:28:15 +03:00
for ( j = 0 ; j < list - > nb_elem ;
j + + , ptr + = syntax [ i ] . list_elem_size )
2008-08-05 03:40:00 +03:00
ebml_free ( syntax [ i ] . def . n , ptr ) ;
2015-01-06 13:48:38 +02:00
av_freep ( & list - > elem ) ;
2016-11-25 10:36:20 +02:00
list - > nb_elem = 0 ;
2019-09-03 23:45:04 +02:00
list - > alloc_elem_size = 0 ;
2008-08-05 03:40:00 +03:00
} else
ebml_free ( syntax [ i ] . def . n , data_off ) ;
2014-03-01 19:28:15 +03:00
default :
break ;
2008-08-05 03:40:00 +03:00
}
}
}
2008-08-05 03:42:39 +03:00
/*
* Autodetecting . . .
*/
2019-03-21 02:18:37 +02:00
static int matroska_probe ( const AVProbeData * p )
2008-08-05 03:42:39 +03:00
{
uint64_t total = 0 ;
2010-05-22 04:41:32 +03:00
int len_mask = 0x80 , size = 1 , n = 1 , i ;
2008-08-05 03:42:39 +03:00
2008-08-05 11:28:57 +03:00
/* EBML header? */
2008-08-05 03:42:39 +03:00
if ( AV_RB32 ( p - > buf ) ! = EBML_ID_HEADER )
return 0 ;
/* length of header */
total = p - > buf [ 4 ] ;
while ( size < = 8 & & ! ( total & len_mask ) ) {
size + + ;
len_mask > > = 1 ;
}
if ( size > 8 )
2014-03-01 19:28:15 +03:00
return 0 ;
2008-08-05 03:42:39 +03:00
total & = ( len_mask - 1 ) ;
while ( n < size )
total = ( total < < 8 ) | p - > buf [ 4 + n + + ] ;
2019-05-17 00:30:17 +02:00
if ( total + 1 = = 1ULL < < ( 7 * size ) ) {
/* Unknown-length header - simply parse the whole buffer. */
total = p - > buf_size - 4 - size ;
} else {
/* Does the probe data contain the whole header? */
if ( p - > buf_size < 4 + size + total )
return 0 ;
}
2008-08-05 03:42:39 +03:00
2010-05-22 04:41:32 +03:00
/* The header should contain a known document type. For now,
2008-08-05 03:42:39 +03:00
* we don ' t parse the whole header but simply check for the
* availability of that array of characters inside the header .
* Not fully fool - proof , but good enough . */
2010-05-22 04:41:32 +03:00
for ( i = 0 ; i < FF_ARRAY_ELEMS ( matroska_doctypes ) ; i + + ) {
2015-05-11 03:51:17 +02:00
size_t probelen = strlen ( matroska_doctypes [ i ] ) ;
2011-07-20 03:51:48 +03:00
if ( total < probelen )
continue ;
2014-03-01 19:28:15 +03:00
for ( n = 4 + size ; n < = 4 + size + total - probelen ; n + + )
if ( ! memcmp ( p - > buf + n , matroska_doctypes [ i ] , probelen ) )
2010-05-22 04:41:32 +03:00
return AVPROBE_SCORE_MAX ;
}
2008-08-05 03:42:39 +03:00
2010-05-22 04:41:35 +03:00
// probably valid EBML header but no recognized doctype
2013-03-25 18:12:51 +03:00
return AVPROBE_SCORE_EXTENSION ;
2008-08-05 03:42:39 +03:00
}
static MatroskaTrack * matroska_find_track_by_num ( MatroskaDemuxContext * matroska ,
2019-12-26 05:17:24 +02:00
uint64_t num )
2008-08-05 03:42:39 +03:00
{
MatroskaTrack * tracks = matroska - > tracks . elem ;
int i ;
2014-03-01 19:28:15 +03:00
for ( i = 0 ; i < matroska - > tracks . nb_elem ; i + + )
2008-08-05 03:42:39 +03:00
if ( tracks [ i ] . num = = num )
return & tracks [ i ] ;
2019-12-26 05:17:24 +02:00
av_log ( matroska - > ctx , AV_LOG_ERROR , " Invalid track number % " PRIu64 " \n " , num ) ;
2008-08-05 03:42:39 +03:00
return NULL ;
}
2014-03-01 19:28:15 +03:00
static int matroska_decode_buffer ( uint8_t * * buf , int * buf_size ,
2008-08-05 03:42:05 +03:00
MatroskaTrack * track )
2008-06-22 18:49:44 +03:00
{
2008-08-05 03:40:31 +03:00
MatroskaTrackEncoding * encodings = track - > encodings . elem ;
2014-03-01 19:28:15 +03:00
uint8_t * data = * buf ;
2008-06-22 18:49:44 +03:00
int isize = * buf_size ;
2014-03-01 19:28:15 +03:00
uint8_t * pkt_data = NULL ;
2012-02-12 12:58:46 +03:00
uint8_t av_unused * newpktdata ;
2008-06-22 18:49:44 +03:00
int pkt_size = isize ;
int result = 0 ;
int olen ;
2012-06-15 02:29:30 +03:00
if ( pkt_size > = 10000000U )
2012-09-14 21:03:37 +03:00
return AVERROR_INVALIDDATA ;
2010-08-17 17:05:23 +03:00
2008-08-05 03:40:31 +03:00
switch ( encodings [ 0 ] . compression . algo ) {
2014-03-01 19:28:15 +03:00
case MATROSKA_TRACK_ENCODING_COMP_HEADERSTRIP :
{
2012-09-14 19:51:49 +03:00
int header_size = encodings [ 0 ] . compression . settings . size ;
uint8_t * header = encodings [ 0 ] . compression . settings . data ;
2012-09-20 14:30:44 +03:00
if ( header_size & & ! header ) {
2012-11-23 20:10:02 +03:00
av_log ( NULL , AV_LOG_ERROR , " Compression size but no data in headerstrip \n " ) ;
2012-04-17 18:12:22 +03:00
return - 1 ;
}
2012-09-20 14:30:44 +03:00
2012-09-14 19:51:49 +03:00
if ( ! header_size )
return 0 ;
pkt_size = isize + header_size ;
2018-04-04 23:12:53 +02:00
pkt_data = av_malloc ( pkt_size + AV_INPUT_BUFFER_PADDING_SIZE ) ;
2012-09-14 19:51:49 +03:00
if ( ! pkt_data )
return AVERROR ( ENOMEM ) ;
memcpy ( pkt_data , header , header_size ) ;
memcpy ( pkt_data + header_size , data , isize ) ;
break ;
}
2012-10-18 20:48:27 +03:00
# if CONFIG_LZO
2008-06-22 18:49:44 +03:00
case MATROSKA_TRACK_ENCODING_COMP_LZO :
do {
2019-12-28 04:40:33 +02:00
int insize = isize ;
2014-03-01 19:28:15 +03:00
olen = pkt_size * = 3 ;
2018-04-04 23:12:53 +02:00
newpktdata = av_realloc ( pkt_data , pkt_size + AV_LZO_OUTPUT_PADDING
+ AV_INPUT_BUFFER_PADDING_SIZE ) ;
2012-09-14 19:39:58 +03:00
if ( ! newpktdata ) {
2012-09-14 21:03:37 +03:00
result = AVERROR ( ENOMEM ) ;
2012-09-14 19:39:58 +03:00
goto failed ;
}
pkt_data = newpktdata ;
2019-12-28 04:40:33 +02:00
result = av_lzo1x_decode ( pkt_data , & olen , data , & insize ) ;
2014-03-01 19:28:15 +03:00
} while ( result = = AV_LZO_OUTPUT_FULL & & pkt_size < 10000000 ) ;
2012-09-14 21:03:37 +03:00
if ( result ) {
result = AVERROR_INVALIDDATA ;
2008-06-22 18:49:44 +03:00
goto failed ;
2012-09-14 21:03:37 +03:00
}
2008-06-22 18:49:44 +03:00
pkt_size - = olen ;
break ;
2012-10-18 20:48:27 +03:00
# endif
2009-01-14 01:44:16 +02:00
# if CONFIG_ZLIB
2014-03-01 19:28:15 +03:00
case MATROSKA_TRACK_ENCODING_COMP_ZLIB :
{
z_stream zstream = { 0 } ;
2008-06-22 18:49:44 +03:00
if ( inflateInit ( & zstream ) ! = Z_OK )
return - 1 ;
2014-03-01 19:28:15 +03:00
zstream . next_in = data ;
2008-06-22 18:49:44 +03:00
zstream . avail_in = isize ;
do {
2014-03-01 19:28:15 +03:00
pkt_size * = 3 ;
2018-04-04 23:12:53 +02:00
newpktdata = av_realloc ( pkt_data , pkt_size + AV_INPUT_BUFFER_PADDING_SIZE ) ;
2011-07-28 15:59:54 +03:00
if ( ! newpktdata ) {
inflateEnd ( & zstream ) ;
2015-05-12 18:32:12 +02:00
result = AVERROR ( ENOMEM ) ;
2011-07-28 15:59:54 +03:00
goto failed ;
}
2014-03-01 19:28:15 +03:00
pkt_data = newpktdata ;
2008-06-22 18:49:44 +03:00
zstream . avail_out = pkt_size - zstream . total_out ;
2014-03-01 19:28:15 +03:00
zstream . next_out = pkt_data + zstream . total_out ;
2015-05-12 18:32:12 +02:00
result = inflate ( & zstream , Z_NO_FLUSH ) ;
2014-03-01 19:28:15 +03:00
} while ( result = = Z_OK & & pkt_size < 10000000 ) ;
2008-06-22 18:49:44 +03:00
pkt_size = zstream . total_out ;
inflateEnd ( & zstream ) ;
2012-09-14 21:03:37 +03:00
if ( result ! = Z_STREAM_END ) {
if ( result = = Z_MEM_ERROR )
result = AVERROR ( ENOMEM ) ;
else
result = AVERROR_INVALIDDATA ;
2008-06-22 18:49:44 +03:00
goto failed ;
2012-09-14 21:03:37 +03:00
}
2008-06-22 18:49:44 +03:00
break ;
}
# endif
2009-01-14 01:44:16 +02:00
# if CONFIG_BZLIB
2014-03-01 19:28:15 +03:00
case MATROSKA_TRACK_ENCODING_COMP_BZLIB :
{
bz_stream bzstream = { 0 } ;
2008-06-22 18:49:44 +03:00
if ( BZ2_bzDecompressInit ( & bzstream , 0 , 0 ) ! = BZ_OK )
return - 1 ;
2014-03-01 19:28:15 +03:00
bzstream . next_in = data ;
2008-06-22 18:49:44 +03:00
bzstream . avail_in = isize ;
do {
2014-03-01 19:28:15 +03:00
pkt_size * = 3 ;
2018-04-04 23:12:53 +02:00
newpktdata = av_realloc ( pkt_data , pkt_size + AV_INPUT_BUFFER_PADDING_SIZE ) ;
2011-07-28 15:59:54 +03:00
if ( ! newpktdata ) {
BZ2_bzDecompressEnd ( & bzstream ) ;
2015-05-12 18:32:12 +02:00
result = AVERROR ( ENOMEM ) ;
2011-07-28 15:59:54 +03:00
goto failed ;
}
2014-03-01 19:28:15 +03:00
pkt_data = newpktdata ;
2008-06-22 18:49:44 +03:00
bzstream . avail_out = pkt_size - bzstream . total_out_lo32 ;
2014-03-01 19:28:15 +03:00
bzstream . next_out = pkt_data + bzstream . total_out_lo32 ;
2015-05-12 18:32:12 +02:00
result = BZ2_bzDecompress ( & bzstream ) ;
2014-03-01 19:28:15 +03:00
} while ( result = = BZ_OK & & pkt_size < 10000000 ) ;
2008-06-22 18:49:44 +03:00
pkt_size = bzstream . total_out_lo32 ;
BZ2_bzDecompressEnd ( & bzstream ) ;
2012-09-14 21:03:37 +03:00
if ( result ! = BZ_STREAM_END ) {
if ( result = = BZ_MEM_ERROR )
result = AVERROR ( ENOMEM ) ;
else
result = AVERROR_INVALIDDATA ;
2008-06-22 18:49:44 +03:00
goto failed ;
2012-09-14 21:03:37 +03:00
}
2008-06-22 18:49:44 +03:00
break ;
}
# endif
2008-08-21 02:08:07 +03:00
default :
2012-09-14 21:03:37 +03:00
return AVERROR_INVALIDDATA ;
2008-06-22 18:49:44 +03:00
}
2018-04-04 23:12:53 +02:00
memset ( pkt_data + pkt_size , 0 , AV_INPUT_BUFFER_PADDING_SIZE ) ;
2014-03-01 19:28:15 +03:00
* buf = pkt_data ;
2008-06-22 18:49:44 +03:00
* buf_size = pkt_size ;
return 0 ;
2014-03-01 19:28:15 +03:00
failed :
2008-06-22 18:49:44 +03:00
av_free ( pkt_data ) ;
2012-09-14 21:03:37 +03:00
return result ;
2008-06-22 18:49:44 +03:00
}
2009-02-15 17:53:55 +02:00
static void matroska_convert_tag ( AVFormatContext * s , EbmlList * list ,
2011-05-22 13:46:29 +03:00
AVDictionary * * metadata , char * prefix )
2008-08-09 02:50:38 +03:00
{
MatroskaTag * tags = list - > elem ;
2009-02-15 17:53:55 +02:00
char key [ 1024 ] ;
int i ;
2008-08-09 02:50:38 +03:00
2014-03-01 19:28:15 +03:00
for ( i = 0 ; i < list - > nb_elem ; i + + ) {
const char * lang = tags [ i ] . lang & &
strcmp ( tags [ i ] . lang , " und " ) ? tags [ i ] . lang : NULL ;
2010-11-03 08:29:04 +02:00
if ( ! tags [ i ] . name ) {
av_log ( s , AV_LOG_WARNING , " Skipping invalid tag with no TagName. \n " ) ;
continue ;
}
2014-03-01 19:28:15 +03:00
if ( prefix )
snprintf ( key , sizeof ( key ) , " %s/%s " , prefix , tags [ i ] . name ) ;
else
av_strlcpy ( key , tags [ i ] . name , sizeof ( key ) ) ;
2009-02-15 18:05:37 +02:00
if ( tags [ i ] . def | | ! lang ) {
2014-03-01 19:28:15 +03:00
av_dict_set ( metadata , key , tags [ i ] . string , 0 ) ;
if ( tags [ i ] . sub . nb_elem )
matroska_convert_tag ( s , & tags [ i ] . sub , metadata , key ) ;
2009-02-15 18:05:37 +02:00
}
if ( lang ) {
av_strlcat ( key , " - " , sizeof ( key ) ) ;
av_strlcat ( key , lang , sizeof ( key ) ) ;
2011-05-22 13:46:29 +03:00
av_dict_set ( metadata , key , tags [ i ] . string , 0 ) ;
2009-02-15 18:05:37 +02:00
if ( tags [ i ] . sub . nb_elem )
matroska_convert_tag ( s , & tags [ i ] . sub , metadata , key ) ;
}
2009-02-15 17:53:55 +02:00
}
2010-10-16 16:20:41 +03:00
ff_metadata_conv ( metadata , NULL , ff_mkv_metadata_conv ) ;
2009-02-15 17:53:55 +02:00
}
static void matroska_convert_tags ( AVFormatContext * s )
{
MatroskaDemuxContext * matroska = s - > priv_data ;
MatroskaTags * tags = matroska - > tags . elem ;
int i , j ;
2014-03-01 19:28:15 +03:00
for ( i = 0 ; i < matroska - > tags . nb_elem ; i + + ) {
2009-02-15 17:53:55 +02:00
if ( tags [ i ] . target . attachuid ) {
2014-03-07 15:54:18 +03:00
MatroskaAttachment * attachment = matroska - > attachments . elem ;
2015-10-31 20:45:27 +02:00
int found = 0 ;
for ( j = 0 ; j < matroska - > attachments . nb_elem ; j + + ) {
2014-03-01 19:28:15 +03:00
if ( attachment [ j ] . uid = = tags [ i ] . target . attachuid & &
2015-10-31 20:45:27 +02:00
attachment [ j ] . stream ) {
2009-02-15 17:53:55 +02:00
matroska_convert_tag ( s , & tags [ i ] . tag ,
& attachment [ j ] . stream - > metadata , NULL ) ;
2015-10-31 20:45:27 +02:00
found = 1 ;
}
}
if ( ! found ) {
2020-05-18 00:25:48 +02:00
av_log ( s , AV_LOG_WARNING ,
2015-10-31 20:45:27 +02:00
" The tags at index %d refer to a "
" non-existent attachment % " PRId64 " . \n " ,
i , tags [ i ] . target . attachuid ) ;
}
2009-02-15 17:53:55 +02:00
} else if ( tags [ i ] . target . chapteruid ) {
MatroskaChapter * chapter = matroska - > chapters . elem ;
2015-10-31 20:45:27 +02:00
int found = 0 ;
for ( j = 0 ; j < matroska - > chapters . nb_elem ; j + + ) {
2014-03-01 19:28:15 +03:00
if ( chapter [ j ] . uid = = tags [ i ] . target . chapteruid & &
2015-10-31 20:45:27 +02:00
chapter [ j ] . chapter ) {
2009-02-15 17:53:55 +02:00
matroska_convert_tag ( s , & tags [ i ] . tag ,
& chapter [ j ] . chapter - > metadata , NULL ) ;
2015-10-31 20:45:27 +02:00
found = 1 ;
}
}
if ( ! found ) {
2020-05-18 00:25:48 +02:00
av_log ( s , AV_LOG_WARNING ,
2015-10-31 20:45:27 +02:00
" The tags at index %d refer to a non-existent chapter "
" % " PRId64 " . \n " ,
i , tags [ i ] . target . chapteruid ) ;
}
2009-02-15 17:53:55 +02:00
} else if ( tags [ i ] . target . trackuid ) {
MatroskaTrack * track = matroska - > tracks . elem ;
2015-10-31 20:45:27 +02:00
int found = 0 ;
for ( j = 0 ; j < matroska - > tracks . nb_elem ; j + + ) {
if ( track [ j ] . uid = = tags [ i ] . target . trackuid & &
track [ j ] . stream ) {
2009-02-15 17:53:55 +02:00
matroska_convert_tag ( s , & tags [ i ] . tag ,
& track [ j ] . stream - > metadata , NULL ) ;
2015-10-31 20:45:27 +02:00
found = 1 ;
}
}
if ( ! found ) {
2020-05-18 00:25:48 +02:00
av_log ( s , AV_LOG_WARNING ,
2015-10-31 20:45:27 +02:00
" The tags at index %d refer to a non-existent track "
" % " PRId64 " . \n " ,
i , tags [ i ] . target . trackuid ) ;
}
2009-02-15 17:53:55 +02:00
} else {
2009-06-14 01:29:38 +03:00
matroska_convert_tag ( s , & tags [ i ] . tag , & s - > metadata ,
tags [ i ] . target . type ) ;
2009-02-15 17:53:55 +02:00
}
2008-08-09 02:50:38 +03:00
}
}
2014-03-01 19:28:15 +03:00
static int matroska_parse_seekhead_entry ( MatroskaDemuxContext * matroska ,
2019-05-17 00:30:20 +02:00
int64_t pos )
2008-08-05 03:40:36 +03:00
{
2019-05-17 00:30:13 +02:00
uint32_t saved_id = matroska - > current_id ;
2011-03-03 21:11:45 +02:00
int64_t before_pos = avio_tell ( matroska - > ctx - > pb ) ;
2011-07-09 08:48:43 +03:00
int ret = 0 ;
2008-08-05 03:42:33 +03:00
2011-07-09 09:11:30 +03:00
/* seek */
2019-05-17 00:30:20 +02:00
if ( avio_seek ( matroska - > ctx - > pb , pos , SEEK_SET ) = = pos ) {
2008-08-05 11:28:57 +03:00
/* We don't want to lose our seekhead level, so we add
2008-08-05 03:40:43 +03:00
* a dummy . This is a crude hack . */
if ( matroska - > num_levels = = EBML_MAX_DEPTH ) {
av_log ( matroska - > ctx , AV_LOG_INFO ,
" Max EBML element depth (%d) reached, "
" cannot parse further. \n " , EBML_MAX_DEPTH ) ;
2011-07-09 08:48:43 +03:00
ret = AVERROR_INVALIDDATA ;
} else {
2019-05-17 00:30:13 +02:00
matroska - > levels [ matroska - > num_levels ] = ( MatroskaLevel ) { 0 , EBML_UNKNOWN_LENGTH } ;
2011-07-09 09:11:30 +03:00
matroska - > num_levels + + ;
2014-03-01 19:28:15 +03:00
matroska - > current_id = 0 ;
2011-07-09 09:11:30 +03:00
2011-09-09 01:43:32 +03:00
ret = ebml_parse ( matroska , matroska_segment , matroska ) ;
2019-06-24 01:46:54 +02:00
if ( ret = = LEVEL_ENDED ) {
/* This can only happen if the seek brought us beyond EOF. */
ret = AVERROR_EOF ;
}
2008-08-05 03:40:43 +03:00
}
2008-08-05 03:40:36 +03:00
}
2019-05-17 00:30:09 +02:00
/* Seek back - notice that in all instances where this is used
* it is safe to set the level to 1. */
2019-05-17 00:30:03 +02:00
matroska_reset_status ( matroska , saved_id , before_pos ) ;
2011-07-09 08:48:43 +03:00
return ret ;
}
static void matroska_execute_seekhead ( MatroskaDemuxContext * matroska )
{
EbmlList * seekhead_list = & matroska - > seekhead ;
2008-08-05 03:41:01 +03:00
int i ;
2007-06-05 01:10:54 +03:00
2010-06-08 22:31:08 +03:00
// we should not do any seeking in the streaming case
2017-03-21 22:02:30 +02:00
if ( ! ( matroska - > ctx - > pb - > seekable & AVIO_SEEKABLE_NORMAL ) )
2010-06-08 22:31:08 +03:00
return ;
2015-02-09 21:39:00 +02:00
for ( i = 0 ; i < seekhead_list - > nb_elem ; i + + ) {
MatroskaSeekhead * seekheads = seekhead_list - > elem ;
2019-05-17 00:30:20 +02:00
uint32_t id = seekheads [ i ] . id ;
int64_t pos = seekheads [ i ] . pos + matroska - > segment_start ;
2019-08-30 13:20:26 +02:00
MatroskaLevel1Element * elem ;
2014-12-06 17:53:30 +02:00
2019-08-30 13:20:26 +02:00
if ( id ! = seekheads [ i ] . id | | pos < matroska - > segment_start )
continue ;
2020-04-30 23:16:44 +02:00
elem = matroska_find_level1_elem ( matroska , id , pos ) ;
2015-02-09 21:39:00 +02:00
if ( ! elem | | elem - > parsed )
2008-08-05 03:40:36 +03:00
continue ;
2007-06-05 01:10:54 +03:00
2015-02-09 21:39:00 +02:00
elem - > pos = pos ;
2011-07-09 08:48:43 +03:00
// defer cues parsing until we actually need cue data.
2015-02-09 21:39:00 +02:00
if ( id = = MATROSKA_ID_CUES )
2008-08-05 03:40:43 +03:00
continue ;
2015-02-09 21:39:00 +02:00
if ( matroska_parse_seekhead_entry ( matroska , pos ) < 0 ) {
2012-02-12 16:09:03 +03:00
// mark index as broken
matroska - > cues_parsing_deferred = - 1 ;
2008-08-05 03:40:43 +03:00
break ;
2012-02-12 16:09:03 +03:00
}
2015-02-09 21:39:00 +02:00
elem - > parsed = 1 ;
2014-12-07 23:45:34 +02:00
}
2011-07-09 08:48:43 +03:00
}
2008-08-05 03:40:43 +03:00
2014-03-08 01:46:37 +03:00
static void matroska_add_index_entries ( MatroskaDemuxContext * matroska )
2014-03-01 19:28:15 +03:00
{
2011-07-09 08:48:43 +03:00
EbmlList * index_list ;
MatroskaIndex * index ;
2015-05-03 23:07:20 +02:00
uint64_t index_scale = 1 ;
2011-07-09 08:48:43 +03:00
int i , j ;
2008-08-05 03:40:43 +03:00
2015-02-09 21:39:01 +02:00
if ( matroska - > ctx - > flags & AVFMT_FLAG_IGNIDX )
return ;
2011-07-09 08:48:43 +03:00
index_list = & matroska - > index ;
2014-03-01 19:28:15 +03:00
index = index_list - > elem ;
2015-10-08 22:34:59 +02:00
if ( index_list - > nb_elem < 2 )
return ;
if ( index [ 1 ] . time > 1E14 / matroska - > time_scale ) {
av_log ( matroska - > ctx , AV_LOG_WARNING , " Dropping apparently-broken index. \n " ) ;
return ;
2011-07-09 08:48:43 +03:00
}
for ( i = 0 ; i < index_list - > nb_elem ; i + + ) {
2014-03-01 19:28:15 +03:00
EbmlList * pos_list = & index [ i ] . pos ;
2011-07-09 08:48:43 +03:00
MatroskaIndexPos * pos = pos_list - > elem ;
for ( j = 0 ; j < pos_list - > nb_elem ; j + + ) {
2014-03-01 19:28:15 +03:00
MatroskaTrack * track = matroska_find_track_by_num ( matroska ,
pos [ j ] . track ) ;
2011-07-09 08:48:43 +03:00
if ( track & & track - > stream )
av_add_index_entry ( track - > stream ,
pos [ j ] . pos + matroska - > segment_start ,
2014-03-01 19:28:15 +03:00
index [ i ] . time / index_scale , 0 , 0 ,
2011-07-09 08:48:43 +03:00
AVINDEX_KEYFRAME ) ;
2008-08-05 03:40:43 +03:00
}
2008-08-05 03:40:36 +03:00
}
2007-06-05 01:10:54 +03:00
}
2011-09-22 17:51:00 +03:00
static void matroska_parse_cues ( MatroskaDemuxContext * matroska ) {
int i ;
2015-02-09 21:39:01 +02:00
if ( matroska - > ctx - > flags & AVFMT_FLAG_IGNIDX )
return ;
2015-02-09 21:39:00 +02:00
for ( i = 0 ; i < matroska - > num_level1_elems ; i + + ) {
MatroskaLevel1Element * elem = & matroska - > level1_elems [ i ] ;
if ( elem - > id = = MATROSKA_ID_CUES & & ! elem - > parsed ) {
if ( matroska_parse_seekhead_entry ( matroska , elem - > pos ) < 0 )
matroska - > cues_parsing_deferred = - 1 ;
elem - > parsed = 1 ;
2011-09-22 17:51:00 +03:00
break ;
2015-02-09 21:39:00 +02:00
}
}
2011-09-22 17:51:00 +03:00
matroska_add_index_entries ( matroska ) ;
}
2008-08-05 03:42:05 +03:00
static int matroska_aac_profile ( char * codec_id )
2007-06-05 01:10:54 +03:00
{
2014-03-01 19:28:15 +03:00
static const char * const aac_profiles [ ] = { " MAIN " , " LC " , " SSR " } ;
2007-06-05 01:10:54 +03:00
int profile ;
2014-03-01 19:28:15 +03:00
for ( profile = 0 ; profile < FF_ARRAY_ELEMS ( aac_profiles ) ; profile + + )
2007-06-05 01:10:54 +03:00
if ( strstr ( codec_id , aac_profiles [ profile ] ) )
break ;
return profile + 1 ;
}
2008-08-05 03:42:05 +03:00
static int matroska_aac_sri ( int samplerate )
2007-06-05 01:10:54 +03:00
{
int sri ;
2014-03-01 19:28:15 +03:00
for ( sri = 0 ; sri < FF_ARRAY_ELEMS ( avpriv_mpeg4audio_sample_rates ) ; sri + + )
2011-10-17 10:28:53 +03:00
if ( avpriv_mpeg4audio_sample_rates [ sri ] = = samplerate )
2007-06-05 01:10:54 +03:00
break ;
return sri ;
}
2012-03-05 21:02:48 +03:00
static void matroska_metadata_creation_time ( AVDictionary * * metadata , int64_t date_utc )
{
/* Convert to seconds and adjust by number of seconds between 2001-01-01 and Epoch */
2016-06-30 00:55:48 +02:00
avpriv_dict_set_timestamp ( metadata , " creation_time " , date_utc / 1000 + 978307200000000LL ) ;
2012-03-05 21:02:48 +03:00
}
2014-05-25 15:05:51 +03:00
static int matroska_parse_flac ( AVFormatContext * s ,
MatroskaTrack * track ,
int * offset )
{
2014-05-26 13:48:56 +03:00
AVStream * st = track - > stream ;
2014-05-25 15:05:51 +03:00
uint8_t * p = track - > codec_priv . data ;
int size = track - > codec_priv . size ;
if ( size < 8 + FLAC_STREAMINFO_SIZE | | p [ 4 ] & 0x7f ) {
av_log ( s , AV_LOG_WARNING , " Invalid FLAC private data \n " ) ;
track - > codec_priv . size = 0 ;
return 0 ;
}
* offset = 8 ;
track - > codec_priv . size = 8 + FLAC_STREAMINFO_SIZE ;
2014-05-26 13:48:56 +03:00
p + = track - > codec_priv . size ;
size - = track - > codec_priv . size ;
/* parse the remaining metadata blocks if present */
while ( size > = 4 ) {
int block_last , block_type , block_size ;
flac_parse_block_header ( p , & block_last , & block_type , & block_size ) ;
p + = 4 ;
size - = 4 ;
if ( block_size > size )
return 0 ;
/* check for the channel mask */
if ( block_type = = FLAC_METADATA_TYPE_VORBIS_COMMENT ) {
AVDictionary * dict = NULL ;
AVDictionaryEntry * chmask ;
ff_vorbis_comment ( s , & dict , p , block_size , 0 ) ;
chmask = av_dict_get ( dict , " WAVEFORMATEXTENSIBLE_CHANNEL_MASK " , NULL , 0 ) ;
if ( chmask ) {
uint64_t mask = strtol ( chmask - > value , NULL , 0 ) ;
if ( ! mask | | mask & ~ 0x3ffffULL ) {
av_log ( s , AV_LOG_WARNING ,
" Invalid value of WAVEFORMATEXTENSIBLE_CHANNEL_MASK \n " ) ;
} else
lavf: replace AVStream.codec with AVStream.codecpar
Currently, AVStream contains an embedded AVCodecContext instance, which
is used by demuxers to export stream parameters to the caller and by
muxers to receive stream parameters from the caller. It is also used
internally as the codec context that is passed to parsers.
In addition, it is also widely used by the callers as the decoding (when
demuxer) or encoding (when muxing) context, though this has been
officially discouraged since Libav 11.
There are multiple important problems with this approach:
- the fields in AVCodecContext are in general one of
* stream parameters
* codec options
* codec state
However, it's not clear which ones are which. It is consequently
unclear which fields are a demuxer allowed to set or a muxer allowed to
read. This leads to erratic behaviour depending on whether decoding or
encoding is being performed or not (and whether it uses the AVStream
embedded codec context).
- various synchronization issues arising from the fact that the same
context is used by several different APIs (muxers/demuxers,
parsers, bitstream filters and encoders/decoders) simultaneously, with
there being no clear rules for who can modify what and the different
processes being typically delayed with respect to each other.
- avformat_find_stream_info() making it necessary to support opening
and closing a single codec context multiple times, thus
complicating the semantics of freeing various allocated objects in the
codec context.
Those problems are resolved by replacing the AVStream embedded codec
context with a newly added AVCodecParameters instance, which stores only
the stream parameters exported by the demuxers or read by the muxers.
2014-06-18 21:42:52 +03:00
st - > codecpar - > channel_layout = mask ;
2014-05-26 13:48:56 +03:00
}
av_dict_free ( & dict ) ;
}
p + = block_size ;
size - = block_size ;
}
2014-05-25 15:05:51 +03:00
return 0 ;
}
2016-10-13 19:22:07 +02:00
static int mkv_field_order ( MatroskaDemuxContext * matroska , int64_t field_order )
2016-03-28 20:29:54 +02:00
{
2020-07-20 19:14:38 +02:00
int minor , micro , bttb = 0 ;
2016-10-13 19:22:07 +02:00
/* workaround a bug in our Matroska muxer, introduced in version 57.36 alongside
* this function , and fixed in 57.52 */
2020-07-20 19:14:38 +02:00
if ( matroska - > muxingapp & & sscanf ( matroska - > muxingapp , " Lavf57.%d.%d " , & minor , & micro ) = = 2 )
bttb = ( minor > = 36 & & minor < = 51 & & micro > = 100 ) ;
2016-10-13 19:22:07 +02:00
2016-03-28 20:29:54 +02:00
switch ( field_order ) {
case MATROSKA_VIDEO_FIELDORDER_PROGRESSIVE :
return AV_FIELD_PROGRESSIVE ;
case MATROSKA_VIDEO_FIELDORDER_UNDETERMINED :
return AV_FIELD_UNKNOWN ;
case MATROSKA_VIDEO_FIELDORDER_TT :
return AV_FIELD_TT ;
case MATROSKA_VIDEO_FIELDORDER_BB :
return AV_FIELD_BB ;
case MATROSKA_VIDEO_FIELDORDER_BT :
2016-10-13 19:22:07 +02:00
return bttb ? AV_FIELD_TB : AV_FIELD_BT ;
2016-03-28 20:29:54 +02:00
case MATROSKA_VIDEO_FIELDORDER_TB :
2016-10-13 19:22:07 +02:00
return bttb ? AV_FIELD_BT : AV_FIELD_TB ;
2016-03-28 20:29:54 +02:00
default :
return AV_FIELD_UNKNOWN ;
}
}
2015-12-03 01:13:18 +02:00
static void mkv_stereo_mode_display_mul ( int stereo_mode ,
int * h_width , int * h_height )
2015-11-23 22:06:14 +02:00
{
switch ( stereo_mode ) {
case MATROSKA_VIDEO_STEREOMODE_TYPE_MONO :
case MATROSKA_VIDEO_STEREOMODE_TYPE_CHECKERBOARD_RL :
case MATROSKA_VIDEO_STEREOMODE_TYPE_CHECKERBOARD_LR :
case MATROSKA_VIDEO_STEREOMODE_TYPE_BOTH_EYES_BLOCK_RL :
case MATROSKA_VIDEO_STEREOMODE_TYPE_BOTH_EYES_BLOCK_LR :
break ;
case MATROSKA_VIDEO_STEREOMODE_TYPE_RIGHT_LEFT :
case MATROSKA_VIDEO_STEREOMODE_TYPE_LEFT_RIGHT :
case MATROSKA_VIDEO_STEREOMODE_TYPE_COL_INTERLEAVED_RL :
case MATROSKA_VIDEO_STEREOMODE_TYPE_COL_INTERLEAVED_LR :
* h_width = 2 ;
break ;
case MATROSKA_VIDEO_STEREOMODE_TYPE_BOTTOM_TOP :
case MATROSKA_VIDEO_STEREOMODE_TYPE_TOP_BOTTOM :
case MATROSKA_VIDEO_STEREOMODE_TYPE_ROW_INTERLEAVED_RL :
case MATROSKA_VIDEO_STEREOMODE_TYPE_ROW_INTERLEAVED_LR :
* h_height = 2 ;
break ;
}
}
2016-03-01 02:41:04 +02:00
static int mkv_parse_video_color ( AVStream * st , const MatroskaTrack * track ) {
2016-12-05 04:22:39 +02:00
const MatroskaTrackVideoColor * color = track - > video . color . elem ;
const MatroskaMasteringMeta * mastering_meta ;
int has_mastering_primaries , has_mastering_luminance ;
if ( ! track - > video . color . nb_elem )
return 0 ;
mastering_meta = & color - > mastering_meta ;
2016-03-01 02:41:04 +02:00
// Mastering primaries are CIE 1931 coords, and must be > 0.
2016-12-05 04:22:39 +02:00
has_mastering_primaries =
2016-03-01 02:41:04 +02:00
mastering_meta - > r_x > 0 & & mastering_meta - > r_y > 0 & &
mastering_meta - > g_x > 0 & & mastering_meta - > g_y > 0 & &
mastering_meta - > b_x > 0 & & mastering_meta - > b_y > 0 & &
mastering_meta - > white_x > 0 & & mastering_meta - > white_y > 0 ;
2021-02-16 16:34:06 +02:00
has_mastering_luminance = mastering_meta - > max_luminance >
mastering_meta - > min_luminance . el . f & &
mastering_meta - > min_luminance . el . f > = 0 & &
mastering_meta - > min_luminance . count ;
2016-12-05 04:22:39 +02:00
if ( color - > matrix_coefficients ! = AVCOL_SPC_RESERVED )
st - > codecpar - > color_space = color - > matrix_coefficients ;
if ( color - > primaries ! = AVCOL_PRI_RESERVED & &
color - > primaries ! = AVCOL_PRI_RESERVED0 )
st - > codecpar - > color_primaries = color - > primaries ;
if ( color - > transfer_characteristics ! = AVCOL_TRC_RESERVED & &
color - > transfer_characteristics ! = AVCOL_TRC_RESERVED0 )
st - > codecpar - > color_trc = color - > transfer_characteristics ;
if ( color - > range ! = AVCOL_RANGE_UNSPECIFIED & &
color - > range < = AVCOL_RANGE_JPEG )
st - > codecpar - > color_range = color - > range ;
if ( color - > chroma_siting_horz ! = MATROSKA_COLOUR_CHROMASITINGHORZ_UNDETERMINED & &
color - > chroma_siting_vert ! = MATROSKA_COLOUR_CHROMASITINGVERT_UNDETERMINED & &
color - > chroma_siting_horz < MATROSKA_COLOUR_CHROMASITINGHORZ_NB & &
color - > chroma_siting_vert < MATROSKA_COLOUR_CHROMASITINGVERT_NB ) {
2016-10-18 04:50:14 +02:00
st - > codecpar - > chroma_location =
2016-12-05 04:22:39 +02:00
avcodec_chroma_pos_to_enum ( ( color - > chroma_siting_horz - 1 ) < < 7 ,
( color - > chroma_siting_vert - 1 ) < < 7 ) ;
2016-10-18 04:50:14 +02:00
}
2017-04-25 20:04:00 +02:00
if ( color - > max_cll & & color - > max_fall ) {
size_t size = 0 ;
int ret ;
AVContentLightMetadata * metadata = av_content_light_metadata_alloc ( & size ) ;
if ( ! metadata )
return AVERROR ( ENOMEM ) ;
ret = av_stream_add_side_data ( st , AV_PKT_DATA_CONTENT_LIGHT_LEVEL ,
( uint8_t * ) metadata , size ) ;
if ( ret < 0 ) {
av_freep ( & metadata ) ;
return ret ;
}
metadata - > MaxCLL = color - > max_cll ;
metadata - > MaxFALL = color - > max_fall ;
}
2016-03-01 02:41:04 +02:00
if ( has_mastering_primaries | | has_mastering_luminance ) {
AVMasteringDisplayMetadata * metadata =
( AVMasteringDisplayMetadata * ) av_stream_new_side_data (
st , AV_PKT_DATA_MASTERING_DISPLAY_METADATA ,
sizeof ( AVMasteringDisplayMetadata ) ) ;
if ( ! metadata ) {
return AVERROR ( ENOMEM ) ;
}
memset ( metadata , 0 , sizeof ( AVMasteringDisplayMetadata ) ) ;
if ( has_mastering_primaries ) {
2019-10-04 01:52:20 +02:00
metadata - > display_primaries [ 0 ] [ 0 ] = av_d2q ( mastering_meta - > r_x , INT_MAX ) ;
metadata - > display_primaries [ 0 ] [ 1 ] = av_d2q ( mastering_meta - > r_y , INT_MAX ) ;
metadata - > display_primaries [ 1 ] [ 0 ] = av_d2q ( mastering_meta - > g_x , INT_MAX ) ;
metadata - > display_primaries [ 1 ] [ 1 ] = av_d2q ( mastering_meta - > g_y , INT_MAX ) ;
metadata - > display_primaries [ 2 ] [ 0 ] = av_d2q ( mastering_meta - > b_x , INT_MAX ) ;
metadata - > display_primaries [ 2 ] [ 1 ] = av_d2q ( mastering_meta - > b_y , INT_MAX ) ;
metadata - > white_point [ 0 ] = av_d2q ( mastering_meta - > white_x , INT_MAX ) ;
metadata - > white_point [ 1 ] = av_d2q ( mastering_meta - > white_y , INT_MAX ) ;
2016-03-01 02:41:04 +02:00
metadata - > has_primaries = 1 ;
}
if ( has_mastering_luminance ) {
2019-10-04 01:52:20 +02:00
metadata - > max_luminance = av_d2q ( mastering_meta - > max_luminance , INT_MAX ) ;
2021-02-16 16:34:06 +02:00
metadata - > min_luminance = av_d2q ( mastering_meta - > min_luminance . el . f , INT_MAX ) ;
2016-03-01 02:41:04 +02:00
metadata - > has_luminance = 1 ;
}
}
return 0 ;
}
2020-05-18 00:25:48 +02:00
static int mkv_parse_video_projection ( AVStream * st , const MatroskaTrack * track ,
void * logctx )
{
2016-12-06 19:48:45 +02:00
AVSphericalMapping * spherical ;
2020-07-19 10:16:33 +02:00
const MatroskaTrackVideoProjection * mkv_projection = & track - > video . projection ;
const uint8_t * priv_data = mkv_projection - > private . data ;
2016-12-06 19:48:45 +02:00
enum AVSphericalProjection projection ;
size_t spherical_size ;
2017-03-16 23:20:47 +02:00
uint32_t l = 0 , t = 0 , r = 0 , b = 0 ;
uint32_t padding = 0 ;
2016-12-06 19:48:45 +02:00
int ret ;
2017-02-10 23:02:22 +02:00
2020-07-19 10:16:33 +02:00
if ( mkv_projection - > private . size & & priv_data [ 0 ] ! = 0 ) {
2020-05-18 00:25:48 +02:00
av_log ( logctx , AV_LOG_WARNING , " Unknown spherical metadata \n " ) ;
2017-02-10 23:02:22 +02:00
return 0 ;
}
2016-12-06 19:48:45 +02:00
switch ( track - > video . projection . type ) {
case MATROSKA_VIDEO_PROJECTION_TYPE_EQUIRECTANGULAR :
2017-03-08 00:34:32 +02:00
if ( track - > video . projection . private . size = = 20 ) {
2020-07-19 10:16:33 +02:00
t = AV_RB32 ( priv_data + 4 ) ;
b = AV_RB32 ( priv_data + 8 ) ;
l = AV_RB32 ( priv_data + 12 ) ;
r = AV_RB32 ( priv_data + 16 ) ;
2017-02-10 23:02:22 +02:00
if ( b > = UINT_MAX - t | | r > = UINT_MAX - l ) {
2020-05-18 00:25:48 +02:00
av_log ( logctx , AV_LOG_ERROR ,
2017-02-10 23:02:22 +02:00
" Invalid bounding rectangle coordinates "
2017-03-16 23:20:47 +02:00
" % " PRIu32 " ,% " PRIu32 " ,% " PRIu32 " ,% " PRIu32 " \n " ,
2017-02-10 23:02:22 +02:00
l , t , r , b ) ;
return AVERROR_INVALIDDATA ;
}
2017-03-08 00:34:32 +02:00
} else if ( track - > video . projection . private . size ! = 0 ) {
2020-05-18 00:25:48 +02:00
av_log ( logctx , AV_LOG_ERROR , " Unknown spherical metadata \n " ) ;
2017-02-10 23:02:22 +02:00
return AVERROR_INVALIDDATA ;
}
2017-03-08 00:34:32 +02:00
if ( l | | t | | r | | b )
projection = AV_SPHERICAL_EQUIRECTANGULAR_TILE ;
else
projection = AV_SPHERICAL_EQUIRECTANGULAR ;
2016-12-06 19:48:45 +02:00
break ;
case MATROSKA_VIDEO_PROJECTION_TYPE_CUBEMAP :
2017-02-10 23:02:22 +02:00
if ( track - > video . projection . private . size < 4 ) {
2020-05-18 00:25:48 +02:00
av_log ( logctx , AV_LOG_ERROR , " Missing projection private properties \n " ) ;
2016-12-06 19:48:45 +02:00
return AVERROR_INVALIDDATA ;
2017-02-10 23:02:22 +02:00
} else if ( track - > video . projection . private . size = = 12 ) {
2020-07-19 10:16:33 +02:00
uint32_t layout = AV_RB32 ( priv_data + 4 ) ;
2017-03-08 00:34:32 +02:00
if ( layout ) {
2020-05-18 00:25:48 +02:00
av_log ( logctx , AV_LOG_WARNING ,
2017-02-10 23:02:22 +02:00
" Unknown spherical cubemap layout % " PRIu32 " \n " , layout ) ;
return 0 ;
}
2017-03-08 00:34:32 +02:00
projection = AV_SPHERICAL_CUBEMAP ;
2020-07-19 10:16:33 +02:00
padding = AV_RB32 ( priv_data + 8 ) ;
2017-02-10 23:02:22 +02:00
} else {
2020-05-18 00:25:48 +02:00
av_log ( logctx , AV_LOG_ERROR , " Unknown spherical metadata \n " ) ;
2017-02-10 23:02:22 +02:00
return AVERROR_INVALIDDATA ;
}
2016-12-06 19:48:45 +02:00
break ;
2017-11-03 01:42:05 +02:00
case MATROSKA_VIDEO_PROJECTION_TYPE_RECTANGULAR :
/* No Spherical metadata */
return 0 ;
2016-12-06 19:48:45 +02:00
default :
2020-05-18 00:25:48 +02:00
av_log ( logctx , AV_LOG_WARNING ,
2016-12-06 19:48:45 +02:00
" Unknown spherical metadata type % " PRIu64 " \n " ,
track - > video . projection . type ) ;
2016-12-06 19:48:45 +02:00
return 0 ;
}
spherical = av_spherical_alloc ( & spherical_size ) ;
if ( ! spherical )
return AVERROR ( ENOMEM ) ;
2016-12-06 19:48:45 +02:00
2016-12-06 19:48:45 +02:00
spherical - > projection = projection ;
2016-12-06 19:48:45 +02:00
spherical - > yaw = ( int32_t ) ( track - > video . projection . yaw * ( 1 < < 16 ) ) ;
spherical - > pitch = ( int32_t ) ( track - > video . projection . pitch * ( 1 < < 16 ) ) ;
spherical - > roll = ( int32_t ) ( track - > video . projection . roll * ( 1 < < 16 ) ) ;
2016-12-06 19:48:45 +02:00
2017-02-10 23:02:22 +02:00
spherical - > padding = padding ;
spherical - > bound_left = l ;
spherical - > bound_top = t ;
spherical - > bound_right = r ;
spherical - > bound_bottom = b ;
2016-12-06 19:48:45 +02:00
ret = av_stream_add_side_data ( st , AV_PKT_DATA_SPHERICAL , ( uint8_t * ) spherical ,
spherical_size ) ;
2016-12-08 06:21:11 +02:00
if ( ret < 0 ) {
av_freep ( & spherical ) ;
2016-12-06 19:48:45 +02:00
return ret ;
2016-12-08 06:21:11 +02:00
}
2016-12-06 19:48:45 +02:00
return 0 ;
}
2016-01-10 04:59:21 +02:00
static int get_qt_codec ( MatroskaTrack * track , uint32_t * fourcc , enum AVCodecID * codec_id )
{
const AVCodecTag * codec_tags ;
codec_tags = track - > type = = MATROSKA_TRACK_TYPE_VIDEO ?
ff_codec_movvideo_tags : ff_codec_movaudio_tags ;
/* Normalize noncompliant private data that starts with the fourcc
* by expanding / shifting the data by 4 bytes and storing the data
* size at the start . */
if ( ff_codec_get_id ( codec_tags , AV_RL32 ( track - > codec_priv . data ) ) ) {
2018-04-04 19:12:54 +02:00
int ret = av_buffer_realloc ( & track - > codec_priv . buf ,
track - > codec_priv . size + 4 + AV_INPUT_BUFFER_PADDING_SIZE ) ;
if ( ret < 0 )
return ret ;
track - > codec_priv . data = track - > codec_priv . buf - > data ;
memmove ( track - > codec_priv . data + 4 , track - > codec_priv . data , track - > codec_priv . size ) ;
2016-01-10 04:59:21 +02:00
track - > codec_priv . size + = 4 ;
AV_WB32 ( track - > codec_priv . data , track - > codec_priv . size ) ;
}
* fourcc = AV_RL32 ( track - > codec_priv . data + 4 ) ;
* codec_id = ff_codec_get_id ( codec_tags , * fourcc ) ;
return 0 ;
}
2014-05-25 10:07:32 +03:00
static int matroska_parse_tracks ( AVFormatContext * s )
2007-06-05 01:10:54 +03:00
{
MatroskaDemuxContext * matroska = s - > priv_data ;
2014-05-25 10:07:32 +03:00
MatroskaTrack * tracks = matroska - > tracks . elem ;
2008-08-05 03:40:49 +03:00
AVStream * st ;
2014-05-25 15:05:51 +03:00
int i , j , ret ;
2014-05-28 13:41:35 +03:00
int k ;
2012-03-05 21:02:48 +03:00
2014-03-01 19:28:15 +03:00
for ( i = 0 ; i < matroska - > tracks . nb_elem ; i + + ) {
2008-08-05 03:40:52 +03:00
MatroskaTrack * track = & tracks [ i ] ;
2012-08-05 12:11:04 +03:00
enum AVCodecID codec_id = AV_CODEC_ID_NONE ;
2011-08-18 01:21:21 +03:00
EbmlList * encodings_list = & track - > encodings ;
2008-08-05 03:40:55 +03:00
MatroskaTrackEncoding * encodings = encodings_list - > elem ;
2008-08-05 03:40:52 +03:00
uint8_t * extradata = NULL ;
int extradata_size = 0 ;
int extradata_offset = 0 ;
2011-06-14 02:58:11 +03:00
uint32_t fourcc = 0 ;
2011-02-20 12:04:12 +02:00
AVIOContext b ;
2013-03-07 19:11:38 +03:00
char * key_id_base64 = NULL ;
2014-02-26 02:02:51 +03:00
int bit_depth = - 1 ;
2008-08-05 03:40:52 +03:00
/* Apply some sanity checks. */
2008-08-05 03:40:55 +03:00
if ( track - > type ! = MATROSKA_TRACK_TYPE_VIDEO & &
track - > type ! = MATROSKA_TRACK_TYPE_AUDIO & &
2013-08-09 01:40:03 +03:00
track - > type ! = MATROSKA_TRACK_TYPE_SUBTITLE & &
track - > type ! = MATROSKA_TRACK_TYPE_METADATA ) {
2008-08-05 03:40:55 +03:00
av_log ( matroska - > ctx , AV_LOG_INFO ,
" Unknown or unsupported track type % " PRIu64 " \n " ,
track - > type ) ;
continue ;
}
2014-08-14 23:31:24 +03:00
if ( ! track - > codec_id )
2008-08-05 03:40:52 +03:00
continue ;
2020-12-07 01:37:25 +02:00
if ( track - > type = = MATROSKA_TRACK_TYPE_AUDIO & & track - > codec_id [ 0 ] ! = ' A '
| | track - > type = = MATROSKA_TRACK_TYPE_VIDEO & & track - > codec_id [ 0 ] ! = ' V '
| | track - > type = = MATROSKA_TRACK_TYPE_SUBTITLE & & track - > codec_id [ 0 ] ! = ' D ' & & track - > codec_id [ 0 ] ! = ' S '
| | track - > type = = MATROSKA_TRACK_TYPE_METADATA & & track - > codec_id [ 0 ] ! = ' D ' & & track - > codec_id [ 0 ] ! = ' S '
) {
av_log ( matroska - > ctx , AV_LOG_INFO , " Inconsistent track type \n " ) ;
continue ;
}
2015-06-15 20:59:22 +02:00
if ( track - > audio . samplerate < 0 | | track - > audio . samplerate > INT_MAX | |
isnan ( track - > audio . samplerate ) ) {
av_log ( matroska - > ctx , AV_LOG_WARNING ,
" Invalid sample rate %f, defaulting to 8000 instead. \n " ,
track - > audio . samplerate ) ;
track - > audio . samplerate = 8000 ;
}
2008-08-05 03:40:55 +03:00
if ( track - > type = = MATROSKA_TRACK_TYPE_VIDEO ) {
2018-01-19 01:21:56 +02:00
if ( ! track - > default_duration & & track - > video . frame_rate > 0 ) {
double default_duration = 1000000000 / track - > video . frame_rate ;
if ( default_duration > UINT64_MAX | | default_duration < 0 ) {
av_log ( matroska - > ctx , AV_LOG_WARNING ,
" Invalid frame rate %e. Cannot calculate default duration. \n " ,
track - > video . frame_rate ) ;
} else {
track - > default_duration = default_duration ;
}
}
2013-04-04 00:40:13 +03:00
if ( track - > video . display_width = = - 1 )
2008-08-05 03:40:55 +03:00
track - > video . display_width = track - > video . pixel_width ;
2013-04-04 00:40:13 +03:00
if ( track - > video . display_height = = - 1 )
2008-08-05 03:40:55 +03:00
track - > video . display_height = track - > video . pixel_height ;
2011-06-14 03:00:06 +03:00
if ( track - > video . color_space . size = = 4 )
fourcc = AV_RL32 ( track - > video . color_space . data ) ;
2008-08-05 03:40:55 +03:00
} else if ( track - > type = = MATROSKA_TRACK_TYPE_AUDIO ) {
if ( ! track - > audio . out_samplerate )
track - > audio . out_samplerate = track - > audio . samplerate ;
}
if ( encodings_list - > nb_elem > 1 ) {
av_log ( matroska - > ctx , AV_LOG_ERROR ,
2011-09-15 10:34:38 +03:00
" Multiple combined encodings not supported " ) ;
2008-08-05 03:40:55 +03:00
} else if ( encodings_list - > nb_elem = = 1 ) {
2013-03-07 19:11:38 +03:00
if ( encodings [ 0 ] . type ) {
if ( encodings [ 0 ] . encryption . key_id . size > 0 ) {
/* Save the encryption key id to be stored later as a
metadata tag . */
const int b64_size = AV_BASE64_SIZE ( encodings [ 0 ] . encryption . key_id . size ) ;
key_id_base64 = av_malloc ( b64_size ) ;
if ( key_id_base64 = = NULL )
return AVERROR ( ENOMEM ) ;
av_base64_encode ( key_id_base64 , b64_size ,
encodings [ 0 ] . encryption . key_id . data ,
encodings [ 0 ] . encryption . key_id . size ) ;
} else {
encodings [ 0 ] . scope = 0 ;
av_log ( matroska - > ctx , AV_LOG_ERROR ,
" Unsupported encoding type " ) ;
}
} else if (
2009-01-14 01:44:16 +02:00
# if CONFIG_ZLIB
2014-03-01 19:28:15 +03:00
encodings [ 0 ] . compression . algo ! = MATROSKA_TRACK_ENCODING_COMP_ZLIB & &
2008-08-05 03:40:55 +03:00
# endif
2009-01-14 01:44:16 +02:00
# if CONFIG_BZLIB
2008-08-05 03:40:55 +03:00
encodings [ 0 ] . compression . algo ! = MATROSKA_TRACK_ENCODING_COMP_BZLIB & &
# endif
2012-10-18 20:48:27 +03:00
# if CONFIG_LZO
2014-03-01 19:28:15 +03:00
encodings [ 0 ] . compression . algo ! = MATROSKA_TRACK_ENCODING_COMP_LZO & &
2012-10-18 20:48:27 +03:00
# endif
2013-03-07 19:11:38 +03:00
encodings [ 0 ] . compression . algo ! = MATROSKA_TRACK_ENCODING_COMP_HEADERSTRIP ) {
2008-08-05 03:40:55 +03:00
encodings [ 0 ] . scope = 0 ;
av_log ( matroska - > ctx , AV_LOG_ERROR ,
" Unsupported encoding type " ) ;
2014-03-01 19:28:15 +03:00
} else if ( track - > codec_priv . size & & encodings [ 0 ] . scope & 2 ) {
2008-08-05 03:40:55 +03:00
uint8_t * codec_priv = track - > codec_priv . data ;
2012-09-14 19:51:49 +03:00
int ret = matroska_decode_buffer ( & track - > codec_priv . data ,
& track - > codec_priv . size ,
track ) ;
if ( ret < 0 ) {
2008-08-05 03:40:55 +03:00
track - > codec_priv . data = NULL ;
track - > codec_priv . size = 0 ;
av_log ( matroska - > ctx , AV_LOG_ERROR ,
" Failed to decode codec private data \n " ) ;
}
2012-09-14 19:51:49 +03:00
2018-04-04 19:12:54 +02:00
if ( codec_priv ! = track - > codec_priv . data ) {
av_buffer_unref ( & track - > codec_priv . buf ) ;
if ( track - > codec_priv . data ) {
track - > codec_priv . buf = av_buffer_create ( track - > codec_priv . data ,
track - > codec_priv . size + AV_INPUT_BUFFER_PADDING_SIZE ,
NULL , NULL , 0 ) ;
if ( ! track - > codec_priv . buf ) {
av_freep ( & track - > codec_priv . data ) ;
track - > codec_priv . size = 0 ;
return AVERROR ( ENOMEM ) ;
}
}
}
2008-08-05 03:40:55 +03:00
}
}
2019-12-04 18:54:45 +02:00
track - > needs_decoding = encodings & & ! encodings [ 0 ] . type & &
encodings [ 0 ] . scope & 1 & &
( encodings [ 0 ] . compression . algo ! =
MATROSKA_TRACK_ENCODING_COMP_HEADERSTRIP | |
encodings [ 0 ] . compression . settings . size ) ;
2008-08-05 03:40:55 +03:00
2014-03-01 19:28:15 +03:00
for ( j = 0 ; ff_mkv_codec_tags [ j ] . id ! = AV_CODEC_ID_NONE ; j + + ) {
2021-02-24 08:44:31 +02:00
if ( av_strstart ( track - > codec_id , ff_mkv_codec_tags [ j ] . str , NULL ) ) {
2014-03-01 19:28:15 +03:00
codec_id = ff_mkv_codec_tags [ j ] . id ;
2008-08-05 03:40:52 +03:00
break ;
2007-06-05 01:10:54 +03:00
}
2008-08-05 03:40:52 +03:00
}
2007-06-05 01:10:54 +03:00
2011-06-18 12:43:24 +03:00
st = track - > stream = avformat_new_stream ( s , NULL ) ;
2014-08-15 21:33:21 +03:00
if ( ! st ) {
2013-03-07 19:11:38 +03:00
av_free ( key_id_base64 ) ;
2008-08-05 03:40:52 +03:00
return AVERROR ( ENOMEM ) ;
2013-03-07 19:11:38 +03:00
}
if ( key_id_base64 ) {
/* export encryption key id as base64 metadata tag */
2019-11-10 06:07:31 +02:00
av_dict_set ( & st - > metadata , " enc_key_id " , key_id_base64 ,
AV_DICT_DONT_STRDUP_VAL ) ;
2013-03-07 19:11:38 +03:00
}
2008-08-05 03:40:52 +03:00
2014-03-01 19:28:15 +03:00
if ( ! strcmp ( track - > codec_id , " V_MS/VFW/FOURCC " ) & &
2014-03-08 01:46:37 +03:00
track - > codec_priv . size > = 40 & &
2014-08-14 23:31:25 +03:00
track - > codec_priv . data ) {
2014-03-01 19:28:15 +03:00
track - > ms_compat = 1 ;
2014-03-08 01:46:37 +03:00
bit_depth = AV_RL16 ( track - > codec_priv . data + 14 ) ;
fourcc = AV_RL32 ( track - > codec_priv . data + 16 ) ;
2014-03-01 19:28:15 +03:00
codec_id = ff_codec_get_id ( ff_codec_bmp_tags ,
2014-03-08 01:46:37 +03:00
fourcc ) ;
2014-04-27 10:34:13 +03:00
if ( ! codec_id )
codec_id = ff_codec_get_id ( ff_codec_movvideo_tags ,
fourcc ) ;
2014-03-01 19:28:15 +03:00
extradata_offset = 40 ;
} else if ( ! strcmp ( track - > codec_id , " A_MS/ACM " ) & &
track - > codec_priv . size > = 14 & &
2014-08-14 23:31:25 +03:00
track - > codec_priv . data ) {
2011-04-12 18:44:20 +03:00
int ret ;
2014-03-01 19:28:15 +03:00
ffio_init_context ( & b , track - > codec_priv . data ,
track - > codec_priv . size ,
2013-01-04 15:00:14 +03:00
0 , NULL , NULL , NULL , NULL ) ;
2016-04-10 21:58:15 +02:00
ret = ff_get_wav_header ( s , & b , st - > codecpar , track - > codec_priv . size , 0 ) ;
2011-04-12 18:44:20 +03:00
if ( ret < 0 )
return ret ;
lavf: replace AVStream.codec with AVStream.codecpar
Currently, AVStream contains an embedded AVCodecContext instance, which
is used by demuxers to export stream parameters to the caller and by
muxers to receive stream parameters from the caller. It is also used
internally as the codec context that is passed to parsers.
In addition, it is also widely used by the callers as the decoding (when
demuxer) or encoding (when muxing) context, though this has been
officially discouraged since Libav 11.
There are multiple important problems with this approach:
- the fields in AVCodecContext are in general one of
* stream parameters
* codec options
* codec state
However, it's not clear which ones are which. It is consequently
unclear which fields are a demuxer allowed to set or a muxer allowed to
read. This leads to erratic behaviour depending on whether decoding or
encoding is being performed or not (and whether it uses the AVStream
embedded codec context).
- various synchronization issues arising from the fact that the same
context is used by several different APIs (muxers/demuxers,
parsers, bitstream filters and encoders/decoders) simultaneously, with
there being no clear rules for who can modify what and the different
processes being typically delayed with respect to each other.
- avformat_find_stream_info() making it necessary to support opening
and closing a single codec context multiple times, thus
complicating the semantics of freeing various allocated objects in the
codec context.
Those problems are resolved by replacing the AVStream embedded codec
context with a newly added AVCodecParameters instance, which stores only
the stream parameters exported by the demuxers or read by the muxers.
2014-06-18 21:42:52 +03:00
codec_id = st - > codecpar - > codec_id ;
2016-04-10 21:58:15 +02:00
fourcc = st - > codecpar - > codec_tag ;
2009-10-02 00:14:05 +03:00
extradata_offset = FFMIN ( track - > codec_priv . size , 18 ) ;
2014-01-19 20:35:33 +03:00
} else if ( ! strcmp ( track - > codec_id , " A_QUICKTIME " )
2016-01-10 04:59:21 +02:00
/* Normally 36, but allow noncompliant private data */
& & ( track - > codec_priv . size > = 32 )
2014-08-15 22:31:59 +03:00
& & ( track - > codec_priv . data ) ) {
2016-01-16 00:25:38 +02:00
uint16_t sample_size ;
2016-01-10 04:59:21 +02:00
int ret = get_qt_codec ( track , & fourcc , & codec_id ) ;
if ( ret < 0 )
return ret ;
2016-01-16 00:25:38 +02:00
sample_size = AV_RB16 ( track - > codec_priv . data + 26 ) ;
2016-01-08 13:55:59 +02:00
if ( fourcc = = 0 ) {
2016-01-16 00:25:38 +02:00
if ( sample_size = = 8 ) {
2016-01-08 13:55:59 +02:00
fourcc = MKTAG ( ' r ' , ' a ' , ' w ' , ' ' ) ;
codec_id = ff_codec_get_id ( ff_codec_movaudio_tags , fourcc ) ;
2016-01-16 00:25:38 +02:00
} else if ( sample_size = = 16 ) {
2016-01-08 13:55:59 +02:00
fourcc = MKTAG ( ' t ' , ' w ' , ' o ' , ' s ' ) ;
codec_id = ff_codec_get_id ( ff_codec_movaudio_tags , fourcc ) ;
}
}
2016-01-16 00:25:38 +02:00
if ( ( fourcc = = MKTAG ( ' t ' , ' w ' , ' o ' , ' s ' ) | |
fourcc = = MKTAG ( ' s ' , ' o ' , ' w ' , ' t ' ) ) & &
sample_size = = 8 )
codec_id = AV_CODEC_ID_PCM_S8 ;
2014-03-01 19:28:15 +03:00
} else if ( ! strcmp ( track - > codec_id , " V_QUICKTIME " ) & &
2014-03-08 01:46:37 +03:00
( track - > codec_priv . size > = 21 ) & &
2014-08-14 23:31:25 +03:00
( track - > codec_priv . data ) ) {
2016-01-10 04:59:21 +02:00
int ret = get_qt_codec ( track , & fourcc , & codec_id ) ;
if ( ret < 0 )
return ret ;
2016-01-06 05:16:32 +02:00
if ( codec_id = = AV_CODEC_ID_NONE & & AV_RL32 ( track - > codec_priv . data + 4 ) = = AV_RL32 ( " SMI " ) ) {
fourcc = MKTAG ( ' S ' , ' V ' , ' Q ' , ' 3 ' ) ;
codec_id = ff_codec_get_id ( ff_codec_movvideo_tags , fourcc ) ;
}
2017-03-27 01:31:52 +02:00
if ( codec_id = = AV_CODEC_ID_NONE )
2015-05-11 18:23:43 +02:00
av_log ( matroska - > ctx , AV_LOG_ERROR ,
2017-03-27 01:31:52 +02:00
" mov FourCC not found %s. \n " , av_fourcc2str ( fourcc ) ) ;
2015-12-27 22:28:09 +02:00
if ( track - > codec_priv . size > = 86 ) {
bit_depth = AV_RB16 ( track - > codec_priv . data + 82 ) ;
ffio_init_context ( & b , track - > codec_priv . data ,
track - > codec_priv . size ,
0 , NULL , NULL , NULL , NULL ) ;
2016-02-24 19:14:05 +02:00
if ( ff_get_qtpalette ( codec_id , & b , track - > palette ) ) {
2015-12-27 22:28:09 +02:00
bit_depth & = 0x1F ;
2016-02-24 19:14:05 +02:00
track - > has_palette = 1 ;
2015-12-27 22:28:09 +02:00
}
}
2012-08-05 12:11:04 +03:00
} else if ( codec_id = = AV_CODEC_ID_PCM_S16BE ) {
2008-08-20 03:49:45 +03:00
switch ( track - > audio . bitdepth ) {
2014-03-01 19:28:15 +03:00
case 8 :
codec_id = AV_CODEC_ID_PCM_U8 ;
break ;
case 24 :
codec_id = AV_CODEC_ID_PCM_S24BE ;
break ;
case 32 :
codec_id = AV_CODEC_ID_PCM_S32BE ;
break ;
2008-08-20 03:49:45 +03:00
}
2012-08-05 12:11:04 +03:00
} else if ( codec_id = = AV_CODEC_ID_PCM_S16LE ) {
2008-08-20 03:49:45 +03:00
switch ( track - > audio . bitdepth ) {
2014-03-01 19:28:15 +03:00
case 8 :
codec_id = AV_CODEC_ID_PCM_U8 ;
break ;
case 24 :
codec_id = AV_CODEC_ID_PCM_S24LE ;
break ;
case 32 :
codec_id = AV_CODEC_ID_PCM_S32LE ;
break ;
2008-08-20 03:49:45 +03:00
}
2014-03-01 19:28:15 +03:00
} else if ( codec_id = = AV_CODEC_ID_PCM_F32LE & &
track - > audio . bitdepth = = 64 ) {
2012-08-05 12:11:04 +03:00
codec_id = AV_CODEC_ID_PCM_F64LE ;
} else if ( codec_id = = AV_CODEC_ID_AAC & & ! track - > codec_priv . size ) {
2008-08-05 03:40:52 +03:00
int profile = matroska_aac_profile ( track - > codec_id ) ;
2014-03-01 19:28:15 +03:00
int sri = matroska_aac_sri ( track - > audio . samplerate ) ;
2015-06-29 23:48:34 +02:00
extradata = av_mallocz ( 5 + AV_INPUT_BUFFER_PADDING_SIZE ) ;
2014-08-14 23:31:24 +03:00
if ( ! extradata )
2008-08-05 03:40:52 +03:00
return AVERROR ( ENOMEM ) ;
2014-03-01 19:28:15 +03:00
extradata [ 0 ] = ( profile < < 3 ) | ( ( sri & 0x0E ) > > 1 ) ;
extradata [ 1 ] = ( ( sri & 0x01 ) < < 7 ) | ( track - > audio . channels < < 3 ) ;
2008-08-05 03:40:52 +03:00
if ( strstr ( track - > codec_id , " SBR " ) ) {
2014-03-01 19:28:15 +03:00
sri = matroska_aac_sri ( track - > audio . out_samplerate ) ;
extradata [ 2 ] = 0x56 ;
extradata [ 3 ] = 0xE5 ;
extradata [ 4 ] = 0x80 | ( sri < < 3 ) ;
2008-08-05 03:40:52 +03:00
extradata_size = 5 ;
2008-08-05 03:41:10 +03:00
} else
2008-08-05 03:40:52 +03:00
extradata_size = 2 ;
2015-07-27 22:53:16 +02:00
} else if ( codec_id = = AV_CODEC_ID_ALAC & & track - > codec_priv . size & & track - > codec_priv . size < INT_MAX - 12 - AV_INPUT_BUFFER_PADDING_SIZE ) {
2013-06-07 13:06:15 +03:00
/* Only ALAC's magic cookie is stored in Matroska's track headers.
2014-03-01 19:28:15 +03:00
* Create the " atom size " , " tag " , and " tag version " fields the
* decoder expects manually . */
2013-06-07 13:06:15 +03:00
extradata_size = 12 + track - > codec_priv . size ;
2014-03-01 19:28:15 +03:00
extradata = av_mallocz ( extradata_size +
2015-06-29 23:48:34 +02:00
AV_INPUT_BUFFER_PADDING_SIZE ) ;
2014-08-14 23:31:24 +03:00
if ( ! extradata )
2013-06-07 13:06:15 +03:00
return AVERROR ( ENOMEM ) ;
AV_WB32 ( extradata , extradata_size ) ;
memcpy ( & extradata [ 4 ] , " alac " , 4 ) ;
AV_WB32 ( & extradata [ 8 ] , 0 ) ;
memcpy ( & extradata [ 12 ] , track - > codec_priv . data ,
2014-03-01 19:28:15 +03:00
track - > codec_priv . size ) ;
2012-08-05 12:11:04 +03:00
} else if ( codec_id = = AV_CODEC_ID_TTA ) {
2019-12-03 19:09:05 +02:00
uint8_t * ptr ;
2015-06-15 21:06:51 +02:00
if ( track - > audio . channels > UINT16_MAX | |
track - > audio . bitdepth > UINT16_MAX ) {
av_log ( matroska - > ctx , AV_LOG_WARNING ,
" Too large audio channel number % " PRIu64
" or bitdepth % " PRIu64 " . Skipping track. \n " ,
track - > audio . channels , track - > audio . bitdepth ) ;
if ( matroska - > ctx - > error_recognition & AV_EF_EXPLODE )
return AVERROR_INVALIDDATA ;
else
continue ;
}
2013-08-18 03:20:54 +03:00
if ( track - > audio . out_samplerate < 0 | | track - > audio . out_samplerate > INT_MAX )
return AVERROR_INVALIDDATA ;
2019-12-03 19:09:04 +02:00
extradata_size = 22 ;
2019-12-03 19:09:03 +02:00
extradata = av_mallocz ( extradata_size + AV_INPUT_BUFFER_PADDING_SIZE ) ;
if ( ! extradata )
return AVERROR ( ENOMEM ) ;
2019-12-03 19:09:05 +02:00
ptr = extradata ;
bytestream_put_be32 ( & ptr , AV_RB32 ( " TTA1 " ) ) ;
bytestream_put_le16 ( & ptr , 1 ) ;
bytestream_put_le16 ( & ptr , track - > audio . channels ) ;
bytestream_put_le16 ( & ptr , track - > audio . bitdepth ) ;
bytestream_put_le32 ( & ptr , track - > audio . out_samplerate ) ;
bytestream_put_le32 ( & ptr , av_rescale ( matroska - > duration * matroska - > time_scale ,
track - > audio . out_samplerate ,
AV_TIME_BASE * 1000 ) ) ;
2014-03-01 19:28:15 +03:00
} else if ( codec_id = = AV_CODEC_ID_RV10 | |
codec_id = = AV_CODEC_ID_RV20 | |
codec_id = = AV_CODEC_ID_RV30 | |
codec_id = = AV_CODEC_ID_RV40 ) {
2008-08-05 03:40:52 +03:00
extradata_offset = 26 ;
2012-08-05 12:11:04 +03:00
} else if ( codec_id = = AV_CODEC_ID_RA_144 ) {
2008-08-05 03:40:52 +03:00
track - > audio . out_samplerate = 8000 ;
2014-03-01 19:28:15 +03:00
track - > audio . channels = 1 ;
2014-03-08 01:46:37 +03:00
} else if ( ( codec_id = = AV_CODEC_ID_RA_288 | |
codec_id = = AV_CODEC_ID_COOK | |
codec_id = = AV_CODEC_ID_ATRAC3 | |
codec_id = = AV_CODEC_ID_SIPR )
& & track - > codec_priv . data ) {
2010-03-13 01:49:06 +02:00
int flavor ;
2012-04-14 20:33:24 +03:00
2014-03-01 19:28:15 +03:00
ffio_init_context ( & b , track - > codec_priv . data ,
track - > codec_priv . size ,
0 , NULL , NULL , NULL , NULL ) ;
2011-03-15 10:14:38 +02:00
avio_skip ( & b , 22 ) ;
2011-02-21 17:43:01 +02:00
flavor = avio_rb16 ( & b ) ;
track - > audio . coded_framesize = avio_rb32 ( & b ) ;
2011-03-15 10:14:38 +02:00
avio_skip ( & b , 12 ) ;
2011-02-21 17:43:01 +02:00
track - > audio . sub_packet_h = avio_rb16 ( & b ) ;
track - > audio . frame_size = avio_rb16 ( & b ) ;
track - > audio . sub_packet_size = avio_rb16 ( & b ) ;
2019-12-02 11:41:12 +02:00
if ( track - > audio . coded_framesize < = 0 | |
2014-03-01 19:28:15 +03:00
track - > audio . sub_packet_h < = 0 | |
2020-04-20 07:47:06 +02:00
track - > audio . frame_size < = 0 )
2013-09-16 15:36:24 +03:00
return AVERROR_INVALIDDATA ;
2019-12-02 11:41:12 +02:00
2012-08-05 12:11:04 +03:00
if ( codec_id = = AV_CODEC_ID_RA_288 ) {
2020-04-22 15:17:41 +02:00
if ( track - > audio . sub_packet_h & 1 | | 2 * track - > audio . frame_size
! = ( int64_t ) track - > audio . sub_packet_h * track - > audio . coded_framesize )
2020-04-22 00:15:54 +02:00
return AVERROR_INVALIDDATA ;
lavf: replace AVStream.codec with AVStream.codecpar
Currently, AVStream contains an embedded AVCodecContext instance, which
is used by demuxers to export stream parameters to the caller and by
muxers to receive stream parameters from the caller. It is also used
internally as the codec context that is passed to parsers.
In addition, it is also widely used by the callers as the decoding (when
demuxer) or encoding (when muxing) context, though this has been
officially discouraged since Libav 11.
There are multiple important problems with this approach:
- the fields in AVCodecContext are in general one of
* stream parameters
* codec options
* codec state
However, it's not clear which ones are which. It is consequently
unclear which fields are a demuxer allowed to set or a muxer allowed to
read. This leads to erratic behaviour depending on whether decoding or
encoding is being performed or not (and whether it uses the AVStream
embedded codec context).
- various synchronization issues arising from the fact that the same
context is used by several different APIs (muxers/demuxers,
parsers, bitstream filters and encoders/decoders) simultaneously, with
there being no clear rules for who can modify what and the different
processes being typically delayed with respect to each other.
- avformat_find_stream_info() making it necessary to support opening
and closing a single codec context multiple times, thus
complicating the semantics of freeing various allocated objects in the
codec context.
Those problems are resolved by replacing the AVStream embedded codec
context with a newly added AVCodecParameters instance, which stores only
the stream parameters exported by the demuxers or read by the muxers.
2014-06-18 21:42:52 +03:00
st - > codecpar - > block_align = track - > audio . coded_framesize ;
2008-08-05 03:40:52 +03:00
track - > codec_priv . size = 0 ;
} else {
2019-12-02 11:41:12 +02:00
if ( codec_id = = AV_CODEC_ID_SIPR ) {
2013-08-04 22:18:49 +03:00
static const int sipr_bit_rate [ 4 ] = { 6504 , 8496 , 5000 , 16000 } ;
2019-12-02 11:41:12 +02:00
if ( flavor > 3 )
return AVERROR_INVALIDDATA ;
2010-03-13 01:49:06 +02:00
track - > audio . sub_packet_size = ff_sipr_subpk_size [ flavor ] ;
lavf: replace AVStream.codec with AVStream.codecpar
Currently, AVStream contains an embedded AVCodecContext instance, which
is used by demuxers to export stream parameters to the caller and by
muxers to receive stream parameters from the caller. It is also used
internally as the codec context that is passed to parsers.
In addition, it is also widely used by the callers as the decoding (when
demuxer) or encoding (when muxing) context, though this has been
officially discouraged since Libav 11.
There are multiple important problems with this approach:
- the fields in AVCodecContext are in general one of
* stream parameters
* codec options
* codec state
However, it's not clear which ones are which. It is consequently
unclear which fields are a demuxer allowed to set or a muxer allowed to
read. This leads to erratic behaviour depending on whether decoding or
encoding is being performed or not (and whether it uses the AVStream
embedded codec context).
- various synchronization issues arising from the fact that the same
context is used by several different APIs (muxers/demuxers,
parsers, bitstream filters and encoders/decoders) simultaneously, with
there being no clear rules for who can modify what and the different
processes being typically delayed with respect to each other.
- avformat_find_stream_info() making it necessary to support opening
and closing a single codec context multiple times, thus
complicating the semantics of freeing various allocated objects in the
codec context.
Those problems are resolved by replacing the AVStream embedded codec
context with a newly added AVCodecParameters instance, which stores only
the stream parameters exported by the demuxers or read by the muxers.
2014-06-18 21:42:52 +03:00
st - > codecpar - > bit_rate = sipr_bit_rate [ flavor ] ;
2020-04-20 08:54:23 +02:00
} else if ( track - > audio . sub_packet_size < = 0 | |
track - > audio . frame_size % track - > audio . sub_packet_size )
2020-04-20 07:47:06 +02:00
return AVERROR_INVALIDDATA ;
lavf: replace AVStream.codec with AVStream.codecpar
Currently, AVStream contains an embedded AVCodecContext instance, which
is used by demuxers to export stream parameters to the caller and by
muxers to receive stream parameters from the caller. It is also used
internally as the codec context that is passed to parsers.
In addition, it is also widely used by the callers as the decoding (when
demuxer) or encoding (when muxing) context, though this has been
officially discouraged since Libav 11.
There are multiple important problems with this approach:
- the fields in AVCodecContext are in general one of
* stream parameters
* codec options
* codec state
However, it's not clear which ones are which. It is consequently
unclear which fields are a demuxer allowed to set or a muxer allowed to
read. This leads to erratic behaviour depending on whether decoding or
encoding is being performed or not (and whether it uses the AVStream
embedded codec context).
- various synchronization issues arising from the fact that the same
context is used by several different APIs (muxers/demuxers,
parsers, bitstream filters and encoders/decoders) simultaneously, with
there being no clear rules for who can modify what and the different
processes being typically delayed with respect to each other.
- avformat_find_stream_info() making it necessary to support opening
and closing a single codec context multiple times, thus
complicating the semantics of freeing various allocated objects in the
codec context.
Those problems are resolved by replacing the AVStream embedded codec
context with a newly added AVCodecParameters instance, which stores only
the stream parameters exported by the demuxers or read by the muxers.
2014-06-18 21:42:52 +03:00
st - > codecpar - > block_align = track - > audio . sub_packet_size ;
2014-03-01 19:28:15 +03:00
extradata_offset = 78 ;
2007-06-05 01:10:54 +03:00
}
2019-12-02 11:41:12 +02:00
track - > audio . buf = av_malloc_array ( track - > audio . sub_packet_h ,
track - > audio . frame_size ) ;
if ( ! track - > audio . buf )
return AVERROR ( ENOMEM ) ;
2014-05-25 15:05:51 +03:00
} else if ( codec_id = = AV_CODEC_ID_FLAC & & track - > codec_priv . size ) {
ret = matroska_parse_flac ( s , track , & extradata_offset ) ;
if ( ret < 0 )
return ret ;
2019-12-14 02:36:54 +02:00
} else if ( codec_id = = AV_CODEC_ID_WAVPACK & & track - > codec_priv . size < 2 ) {
av_log ( matroska - > ctx , AV_LOG_INFO , " Assuming WavPack version 4.10 "
" in absence of valid CodecPrivate. \n " ) ;
extradata_size = 2 ;
extradata = av_mallocz ( 2 + AV_INPUT_BUFFER_PADDING_SIZE ) ;
if ( ! extradata )
return AVERROR ( ENOMEM ) ;
AV_WL16 ( extradata , 0x410 ) ;
2014-04-06 21:54:48 +03:00
} else if ( codec_id = = AV_CODEC_ID_PRORES & & track - > codec_priv . size = = 4 ) {
fourcc = AV_RL32 ( track - > codec_priv . data ) ;
2018-01-13 21:04:21 +02:00
} else if ( codec_id = = AV_CODEC_ID_VP9 & & track - > codec_priv . size ) {
/* we don't need any value stored in CodecPrivate.
make sure that it ' s not exported as extradata . */
track - > codec_priv . size = 0 ;
2018-08-16 23:01:44 +02:00
} else if ( codec_id = = AV_CODEC_ID_AV1 & & track - > codec_priv . size ) {
/* For now, propagate only the OBUs, if any. Once libavcodec is
updated to handle isobmff style extradata this can be removed . */
extradata_offset = 4 ;
2008-08-05 03:40:52 +03:00
}
2009-08-24 16:40:30 +03:00
track - > codec_priv . size - = extradata_offset ;
2007-06-05 01:10:54 +03:00
2012-08-05 12:11:04 +03:00
if ( codec_id = = AV_CODEC_ID_NONE )
2008-08-05 03:40:52 +03:00
av_log ( matroska - > ctx , AV_LOG_INFO ,
2012-08-05 12:11:04 +03:00
" Unknown/unsupported AVCodecID %s. \n " , track - > codec_id ) ;
2007-06-05 01:19:17 +03:00
2020-11-15 10:59:47 +02:00
if ( track - > time_scale < 0.01 ) {
av_log ( matroska - > ctx , AV_LOG_WARNING ,
" Track TimestampScale too small %f, assuming 1.0. \n " ,
track - > time_scale ) ;
2008-09-09 14:23:48 +03:00
track - > time_scale = 1.0 ;
2020-11-15 10:59:47 +02:00
}
2014-03-01 19:28:15 +03:00
avpriv_set_pts_info ( st , 64 , matroska - > time_scale * track - > time_scale ,
1000 * 1000 * 1000 ) ; /* 64 bit pts in ns */
2007-06-05 01:19:17 +03:00
2014-04-29 13:03:13 +03:00
/* convert the delay from ns to the track timebase */
2016-06-06 04:23:16 +02:00
track - > codec_delay_in_track_tb = av_rescale_q ( track - > codec_delay ,
2014-04-29 13:03:13 +03:00
( AVRational ) { 1 , 1000000000 } ,
st - > time_base ) ;
lavf: replace AVStream.codec with AVStream.codecpar
Currently, AVStream contains an embedded AVCodecContext instance, which
is used by demuxers to export stream parameters to the caller and by
muxers to receive stream parameters from the caller. It is also used
internally as the codec context that is passed to parsers.
In addition, it is also widely used by the callers as the decoding (when
demuxer) or encoding (when muxing) context, though this has been
officially discouraged since Libav 11.
There are multiple important problems with this approach:
- the fields in AVCodecContext are in general one of
* stream parameters
* codec options
* codec state
However, it's not clear which ones are which. It is consequently
unclear which fields are a demuxer allowed to set or a muxer allowed to
read. This leads to erratic behaviour depending on whether decoding or
encoding is being performed or not (and whether it uses the AVStream
embedded codec context).
- various synchronization issues arising from the fact that the same
context is used by several different APIs (muxers/demuxers,
parsers, bitstream filters and encoders/decoders) simultaneously, with
there being no clear rules for who can modify what and the different
processes being typically delayed with respect to each other.
- avformat_find_stream_info() making it necessary to support opening
and closing a single codec context multiple times, thus
complicating the semantics of freeing various allocated objects in the
codec context.
Those problems are resolved by replacing the AVStream embedded codec
context with a newly added AVCodecParameters instance, which stores only
the stream parameters exported by the demuxers or read by the muxers.
2014-06-18 21:42:52 +03:00
st - > codecpar - > codec_id = codec_id ;
2013-12-23 12:41:35 +03:00
2008-08-05 03:40:52 +03:00
if ( strcmp ( track - > language , " und " ) )
2011-05-22 13:46:29 +03:00
av_dict_set ( & st - > metadata , " language " , track - > language , 0 ) ;
av_dict_set ( & st - > metadata , " title " , track - > name , 0 ) ;
2007-06-05 01:10:54 +03:00
2008-08-05 03:40:52 +03:00
if ( track - > flag_default )
st - > disposition | = AV_DISPOSITION_DEFAULT ;
2010-07-02 19:38:44 +03:00
if ( track - > flag_forced )
st - > disposition | = AV_DISPOSITION_FORCED ;
2021-02-16 19:23:59 +02:00
if ( track - > flag_comment )
st - > disposition | = AV_DISPOSITION_COMMENT ;
2021-02-16 20:08:51 +02:00
if ( track - > flag_hearingimpaired )
st - > disposition | = AV_DISPOSITION_HEARING_IMPAIRED ;
if ( track - > flag_visualimpaired )
st - > disposition | = AV_DISPOSITION_VISUAL_IMPAIRED ;
2021-02-16 19:41:30 +02:00
if ( track - > flag_original . count > 0 )
st - > disposition | = track - > flag_original . el . u ? AV_DISPOSITION_ORIGINAL
: AV_DISPOSITION_DUB ;
2007-06-05 01:10:54 +03:00
lavf: replace AVStream.codec with AVStream.codecpar
Currently, AVStream contains an embedded AVCodecContext instance, which
is used by demuxers to export stream parameters to the caller and by
muxers to receive stream parameters from the caller. It is also used
internally as the codec context that is passed to parsers.
In addition, it is also widely used by the callers as the decoding (when
demuxer) or encoding (when muxing) context, though this has been
officially discouraged since Libav 11.
There are multiple important problems with this approach:
- the fields in AVCodecContext are in general one of
* stream parameters
* codec options
* codec state
However, it's not clear which ones are which. It is consequently
unclear which fields are a demuxer allowed to set or a muxer allowed to
read. This leads to erratic behaviour depending on whether decoding or
encoding is being performed or not (and whether it uses the AVStream
embedded codec context).
- various synchronization issues arising from the fact that the same
context is used by several different APIs (muxers/demuxers,
parsers, bitstream filters and encoders/decoders) simultaneously, with
there being no clear rules for who can modify what and the different
processes being typically delayed with respect to each other.
- avformat_find_stream_info() making it necessary to support opening
and closing a single codec context multiple times, thus
complicating the semantics of freeing various allocated objects in the
codec context.
Those problems are resolved by replacing the AVStream embedded codec
context with a newly added AVCodecParameters instance, which stores only
the stream parameters exported by the demuxers or read by the muxers.
2014-06-18 21:42:52 +03:00
if ( ! st - > codecpar - > extradata ) {
2014-03-01 19:28:15 +03:00
if ( extradata ) {
lavf: replace AVStream.codec with AVStream.codecpar
Currently, AVStream contains an embedded AVCodecContext instance, which
is used by demuxers to export stream parameters to the caller and by
muxers to receive stream parameters from the caller. It is also used
internally as the codec context that is passed to parsers.
In addition, it is also widely used by the callers as the decoding (when
demuxer) or encoding (when muxing) context, though this has been
officially discouraged since Libav 11.
There are multiple important problems with this approach:
- the fields in AVCodecContext are in general one of
* stream parameters
* codec options
* codec state
However, it's not clear which ones are which. It is consequently
unclear which fields are a demuxer allowed to set or a muxer allowed to
read. This leads to erratic behaviour depending on whether decoding or
encoding is being performed or not (and whether it uses the AVStream
embedded codec context).
- various synchronization issues arising from the fact that the same
context is used by several different APIs (muxers/demuxers,
parsers, bitstream filters and encoders/decoders) simultaneously, with
there being no clear rules for who can modify what and the different
processes being typically delayed with respect to each other.
- avformat_find_stream_info() making it necessary to support opening
and closing a single codec context multiple times, thus
complicating the semantics of freeing various allocated objects in the
codec context.
Those problems are resolved by replacing the AVStream embedded codec
context with a newly added AVCodecParameters instance, which stores only
the stream parameters exported by the demuxers or read by the muxers.
2014-06-18 21:42:52 +03:00
st - > codecpar - > extradata = extradata ;
st - > codecpar - > extradata_size = extradata_size ;
2014-03-01 19:28:15 +03:00
} else if ( track - > codec_priv . data & & track - > codec_priv . size > 0 ) {
2016-04-10 21:58:15 +02:00
if ( ff_alloc_extradata ( st - > codecpar , track - > codec_priv . size ) )
2009-10-02 00:15:36 +03:00
return AVERROR ( ENOMEM ) ;
lavf: replace AVStream.codec with AVStream.codecpar
Currently, AVStream contains an embedded AVCodecContext instance, which
is used by demuxers to export stream parameters to the caller and by
muxers to receive stream parameters from the caller. It is also used
internally as the codec context that is passed to parsers.
In addition, it is also widely used by the callers as the decoding (when
demuxer) or encoding (when muxing) context, though this has been
officially discouraged since Libav 11.
There are multiple important problems with this approach:
- the fields in AVCodecContext are in general one of
* stream parameters
* codec options
* codec state
However, it's not clear which ones are which. It is consequently
unclear which fields are a demuxer allowed to set or a muxer allowed to
read. This leads to erratic behaviour depending on whether decoding or
encoding is being performed or not (and whether it uses the AVStream
embedded codec context).
- various synchronization issues arising from the fact that the same
context is used by several different APIs (muxers/demuxers,
parsers, bitstream filters and encoders/decoders) simultaneously, with
there being no clear rules for who can modify what and the different
processes being typically delayed with respect to each other.
- avformat_find_stream_info() making it necessary to support opening
and closing a single codec context multiple times, thus
complicating the semantics of freeing various allocated objects in the
codec context.
Those problems are resolved by replacing the AVStream embedded codec
context with a newly added AVCodecParameters instance, which stores only
the stream parameters exported by the demuxers or read by the muxers.
2014-06-18 21:42:52 +03:00
memcpy ( st - > codecpar - > extradata ,
2009-10-02 00:15:36 +03:00
track - > codec_priv . data + extradata_offset ,
track - > codec_priv . size ) ;
}
2009-10-02 00:14:46 +03:00
}
2008-08-05 03:40:52 +03:00
if ( track - > type = = MATROSKA_TRACK_TYPE_VIDEO ) {
2011-05-24 02:09:24 +03:00
MatroskaTrackPlane * planes = track - > operation . combine_planes . elem ;
2015-12-03 01:13:18 +02:00
int display_width_mul = 1 ;
2015-11-23 22:06:14 +02:00
int display_height_mul = 1 ;
2011-05-24 02:09:24 +03:00
lavf: replace AVStream.codec with AVStream.codecpar
Currently, AVStream contains an embedded AVCodecContext instance, which
is used by demuxers to export stream parameters to the caller and by
muxers to receive stream parameters from the caller. It is also used
internally as the codec context that is passed to parsers.
In addition, it is also widely used by the callers as the decoding (when
demuxer) or encoding (when muxing) context, though this has been
officially discouraged since Libav 11.
There are multiple important problems with this approach:
- the fields in AVCodecContext are in general one of
* stream parameters
* codec options
* codec state
However, it's not clear which ones are which. It is consequently
unclear which fields are a demuxer allowed to set or a muxer allowed to
read. This leads to erratic behaviour depending on whether decoding or
encoding is being performed or not (and whether it uses the AVStream
embedded codec context).
- various synchronization issues arising from the fact that the same
context is used by several different APIs (muxers/demuxers,
parsers, bitstream filters and encoders/decoders) simultaneously, with
there being no clear rules for who can modify what and the different
processes being typically delayed with respect to each other.
- avformat_find_stream_info() making it necessary to support opening
and closing a single codec context multiple times, thus
complicating the semantics of freeing various allocated objects in the
codec context.
Those problems are resolved by replacing the AVStream embedded codec
context with a newly added AVCodecParameters instance, which stores only
the stream parameters exported by the demuxers or read by the muxers.
2014-06-18 21:42:52 +03:00
st - > codecpar - > codec_type = AVMEDIA_TYPE_VIDEO ;
2016-04-10 21:58:15 +02:00
st - > codecpar - > codec_tag = fourcc ;
2014-02-26 02:02:51 +03:00
if ( bit_depth > = 0 )
2016-04-10 21:58:15 +02:00
st - > codecpar - > bits_per_coded_sample = bit_depth ;
lavf: replace AVStream.codec with AVStream.codecpar
Currently, AVStream contains an embedded AVCodecContext instance, which
is used by demuxers to export stream parameters to the caller and by
muxers to receive stream parameters from the caller. It is also used
internally as the codec context that is passed to parsers.
In addition, it is also widely used by the callers as the decoding (when
demuxer) or encoding (when muxing) context, though this has been
officially discouraged since Libav 11.
There are multiple important problems with this approach:
- the fields in AVCodecContext are in general one of
* stream parameters
* codec options
* codec state
However, it's not clear which ones are which. It is consequently
unclear which fields are a demuxer allowed to set or a muxer allowed to
read. This leads to erratic behaviour depending on whether decoding or
encoding is being performed or not (and whether it uses the AVStream
embedded codec context).
- various synchronization issues arising from the fact that the same
context is used by several different APIs (muxers/demuxers,
parsers, bitstream filters and encoders/decoders) simultaneously, with
there being no clear rules for who can modify what and the different
processes being typically delayed with respect to each other.
- avformat_find_stream_info() making it necessary to support opening
and closing a single codec context multiple times, thus
complicating the semantics of freeing various allocated objects in the
codec context.
Those problems are resolved by replacing the AVStream embedded codec
context with a newly added AVCodecParameters instance, which stores only
the stream parameters exported by the demuxers or read by the muxers.
2014-06-18 21:42:52 +03:00
st - > codecpar - > width = track - > video . pixel_width ;
st - > codecpar - > height = track - > video . pixel_height ;
2015-11-23 22:06:14 +02:00
2016-03-28 20:29:54 +02:00
if ( track - > video . interlaced = = MATROSKA_VIDEO_INTERLACE_FLAG_INTERLACED )
2016-10-13 19:22:07 +02:00
st - > codecpar - > field_order = mkv_field_order ( matroska , track - > video . field_order ) ;
2016-09-27 19:48:33 +02:00
else if ( track - > video . interlaced = = MATROSKA_VIDEO_INTERLACE_FLAG_PROGRESSIVE )
st - > codecpar - > field_order = AV_FIELD_PROGRESSIVE ;
2016-03-28 20:29:54 +02:00
2015-11-23 22:06:14 +02:00
if ( track - > video . stereo_mode & & track - > video . stereo_mode < MATROSKA_VIDEO_STEREOMODE_TYPE_NB )
mkv_stereo_mode_display_mul ( track - > video . stereo_mode , & display_width_mul , & display_height_mul ) ;
2016-10-16 15:13:45 +02:00
if ( track - > video . display_unit < MATROSKA_VIDEO_DISPLAYUNIT_UNKNOWN ) {
2016-10-15 23:01:50 +02:00
av_reduce ( & st - > sample_aspect_ratio . num ,
& st - > sample_aspect_ratio . den ,
st - > codecpar - > height * track - > video . display_width * display_width_mul ,
st - > codecpar - > width * track - > video . display_height * display_height_mul ,
255 ) ;
}
2016-04-10 21:58:15 +02:00
if ( st - > codecpar - > codec_id ! = AV_CODEC_ID_HEVC )
2013-10-12 12:55:46 +03:00
st - > need_parsing = AVSTREAM_PARSE_HEADERS ;
2013-11-01 19:57:57 +03:00
2012-04-18 02:32:07 +03:00
if ( track - > default_duration ) {
2012-06-26 14:10:01 +03:00
av_reduce ( & st - > avg_frame_rate . num , & st - > avg_frame_rate . den ,
2012-04-18 02:32:07 +03:00
1000000000 , track - > default_duration , 30000 ) ;
2012-06-26 14:10:01 +03:00
# if FF_API_R_FRAME_RATE
2015-02-01 20:34:52 +02:00
if ( st - > avg_frame_rate . num < st - > avg_frame_rate . den * 1000LL
& & st - > avg_frame_rate . num > st - > avg_frame_rate . den * 5LL )
2013-10-06 01:07:28 +03:00
st - > r_frame_rate = st - > avg_frame_rate ;
2012-06-26 14:10:01 +03:00
# endif
2012-04-18 02:32:07 +03:00
}
2011-05-21 18:14:14 +03:00
2011-05-24 02:09:24 +03:00
/* export stereo mode flag as metadata tag */
2014-08-29 02:26:52 +03:00
if ( track - > video . stereo_mode & & track - > video . stereo_mode < MATROSKA_VIDEO_STEREOMODE_TYPE_NB )
2012-09-16 03:58:40 +03:00
av_dict_set ( & st - > metadata , " stereo_mode " , ff_matroska_video_stereo_mode [ track - > video . stereo_mode ] , 0 ) ;
2011-05-24 02:09:24 +03:00
2013-02-05 02:17:52 +03:00
/* export alpha mode flag as metadata tag */
if ( track - > video . alpha_mode )
av_dict_set ( & st - > metadata , " alpha_mode " , " 1 " , 0 ) ;
2011-05-24 02:09:24 +03:00
/* if we have virtual track, mark the real tracks */
for ( j = 0 ; j < track - > operation . combine_planes . nb_elem ; j + + ) {
char buf [ 32 ] ;
2011-05-25 00:26:24 +03:00
if ( planes [ j ] . type > = MATROSKA_VIDEO_STEREO_PLANE_COUNT )
2011-05-24 02:09:24 +03:00
continue ;
snprintf ( buf , sizeof ( buf ) , " %s_%d " ,
2012-09-16 03:58:40 +03:00
ff_matroska_video_stereo_plane [ planes [ j ] . type ] , i ) ;
2011-05-24 02:09:24 +03:00
for ( k = 0 ; k < matroska - > tracks . nb_elem ; k + + )
2015-05-04 15:47:54 +02:00
if ( planes [ j ] . uid = = tracks [ k ] . uid & & tracks [ k ] . stream ) {
av_dict_set ( & tracks [ k ] . stream - > metadata ,
2011-06-13 20:02:50 +03:00
" stereo_mode " , buf , 0 ) ;
2011-05-21 18:14:14 +03:00
break ;
}
}
2014-08-13 00:28:49 +03:00
// add stream level stereo3d side data if it is a supported format
if ( track - > video . stereo_mode < MATROSKA_VIDEO_STEREOMODE_TYPE_NB & &
track - > video . stereo_mode ! = 10 & & track - > video . stereo_mode ! = 12 ) {
int ret = ff_mkv_stereo3d_conv ( st , track - > video . stereo_mode ) ;
if ( ret < 0 )
return ret ;
}
2016-03-01 02:41:04 +02:00
2016-12-05 04:27:54 +02:00
ret = mkv_parse_video_color ( st , track ) ;
if ( ret < 0 )
return ret ;
2020-05-18 00:25:48 +02:00
ret = mkv_parse_video_projection ( st , track , matroska - > ctx ) ;
2016-12-06 19:48:45 +02:00
if ( ret < 0 )
return ret ;
2008-08-05 03:40:52 +03:00
} else if ( track - > type = = MATROSKA_TRACK_TYPE_AUDIO ) {
lavf: replace AVStream.codec with AVStream.codecpar
Currently, AVStream contains an embedded AVCodecContext instance, which
is used by demuxers to export stream parameters to the caller and by
muxers to receive stream parameters from the caller. It is also used
internally as the codec context that is passed to parsers.
In addition, it is also widely used by the callers as the decoding (when
demuxer) or encoding (when muxing) context, though this has been
officially discouraged since Libav 11.
There are multiple important problems with this approach:
- the fields in AVCodecContext are in general one of
* stream parameters
* codec options
* codec state
However, it's not clear which ones are which. It is consequently
unclear which fields are a demuxer allowed to set or a muxer allowed to
read. This leads to erratic behaviour depending on whether decoding or
encoding is being performed or not (and whether it uses the AVStream
embedded codec context).
- various synchronization issues arising from the fact that the same
context is used by several different APIs (muxers/demuxers,
parsers, bitstream filters and encoders/decoders) simultaneously, with
there being no clear rules for who can modify what and the different
processes being typically delayed with respect to each other.
- avformat_find_stream_info() making it necessary to support opening
and closing a single codec context multiple times, thus
complicating the semantics of freeing various allocated objects in the
codec context.
Those problems are resolved by replacing the AVStream embedded codec
context with a newly added AVCodecParameters instance, which stores only
the stream parameters exported by the demuxers or read by the muxers.
2014-06-18 21:42:52 +03:00
st - > codecpar - > codec_type = AVMEDIA_TYPE_AUDIO ;
2016-04-10 21:58:15 +02:00
st - > codecpar - > codec_tag = fourcc ;
lavf: replace AVStream.codec with AVStream.codecpar
Currently, AVStream contains an embedded AVCodecContext instance, which
is used by demuxers to export stream parameters to the caller and by
muxers to receive stream parameters from the caller. It is also used
internally as the codec context that is passed to parsers.
In addition, it is also widely used by the callers as the decoding (when
demuxer) or encoding (when muxing) context, though this has been
officially discouraged since Libav 11.
There are multiple important problems with this approach:
- the fields in AVCodecContext are in general one of
* stream parameters
* codec options
* codec state
However, it's not clear which ones are which. It is consequently
unclear which fields are a demuxer allowed to set or a muxer allowed to
read. This leads to erratic behaviour depending on whether decoding or
encoding is being performed or not (and whether it uses the AVStream
embedded codec context).
- various synchronization issues arising from the fact that the same
context is used by several different APIs (muxers/demuxers,
parsers, bitstream filters and encoders/decoders) simultaneously, with
there being no clear rules for who can modify what and the different
processes being typically delayed with respect to each other.
- avformat_find_stream_info() making it necessary to support opening
and closing a single codec context multiple times, thus
complicating the semantics of freeing various allocated objects in the
codec context.
Those problems are resolved by replacing the AVStream embedded codec
context with a newly added AVCodecParameters instance, which stores only
the stream parameters exported by the demuxers or read by the muxers.
2014-06-18 21:42:52 +03:00
st - > codecpar - > sample_rate = track - > audio . out_samplerate ;
st - > codecpar - > channels = track - > audio . channels ;
2016-04-10 21:58:15 +02:00
if ( ! st - > codecpar - > bits_per_coded_sample )
st - > codecpar - > bits_per_coded_sample = track - > audio . bitdepth ;
2018-01-30 03:59:52 +02:00
if ( st - > codecpar - > codec_id = = AV_CODEC_ID_MP3 | |
st - > codecpar - > codec_id = = AV_CODEC_ID_MLP | |
st - > codecpar - > codec_id = = AV_CODEC_ID_TRUEHD )
2015-08-16 10:06:04 +02:00
st - > need_parsing = AVSTREAM_PARSE_FULL ;
2016-04-10 21:58:15 +02:00
else if ( st - > codecpar - > codec_id ! = AV_CODEC_ID_AAC )
2014-03-01 19:28:15 +03:00
st - > need_parsing = AVSTREAM_PARSE_HEADERS ;
2013-10-14 20:42:08 +03:00
if ( track - > codec_delay > 0 ) {
2016-04-10 21:58:15 +02:00
st - > codecpar - > initial_padding = av_rescale_q ( track - > codec_delay ,
2016-06-06 04:23:16 +02:00
( AVRational ) { 1 , 1000000000 } ,
2016-06-06 06:14:11 +02:00
( AVRational ) { 1 , st - > codecpar - > codec_id = = AV_CODEC_ID_OPUS ?
48000 : st - > codecpar - > sample_rate } ) ;
2013-10-14 20:42:08 +03:00
}
if ( track - > seek_preroll > 0 ) {
2016-04-10 21:58:15 +02:00
st - > codecpar - > seek_preroll = av_rescale_q ( track - > seek_preroll ,
( AVRational ) { 1 , 1000000000 } ,
( AVRational ) { 1 , st - > codecpar - > sample_rate } ) ;
2013-10-14 20:42:08 +03:00
}
2013-08-09 01:40:03 +03:00
} else if ( codec_id = = AV_CODEC_ID_WEBVTT ) {
2016-04-10 21:58:15 +02:00
st - > codecpar - > codec_type = AVMEDIA_TYPE_SUBTITLE ;
2013-08-09 01:40:03 +03:00
if ( ! strcmp ( track - > codec_id , " D_WEBVTT/CAPTIONS " ) ) {
st - > disposition | = AV_DISPOSITION_CAPTIONS ;
} else if ( ! strcmp ( track - > codec_id , " D_WEBVTT/DESCRIPTIONS " ) ) {
st - > disposition | = AV_DISPOSITION_DESCRIPTIONS ;
} else if ( ! strcmp ( track - > codec_id , " D_WEBVTT/METADATA " ) ) {
st - > disposition | = AV_DISPOSITION_METADATA ;
}
2008-08-05 03:40:52 +03:00
} else if ( track - > type = = MATROSKA_TRACK_TYPE_SUBTITLE ) {
lavf: replace AVStream.codec with AVStream.codecpar
Currently, AVStream contains an embedded AVCodecContext instance, which
is used by demuxers to export stream parameters to the caller and by
muxers to receive stream parameters from the caller. It is also used
internally as the codec context that is passed to parsers.
In addition, it is also widely used by the callers as the decoding (when
demuxer) or encoding (when muxing) context, though this has been
officially discouraged since Libav 11.
There are multiple important problems with this approach:
- the fields in AVCodecContext are in general one of
* stream parameters
* codec options
* codec state
However, it's not clear which ones are which. It is consequently
unclear which fields are a demuxer allowed to set or a muxer allowed to
read. This leads to erratic behaviour depending on whether decoding or
encoding is being performed or not (and whether it uses the AVStream
embedded codec context).
- various synchronization issues arising from the fact that the same
context is used by several different APIs (muxers/demuxers,
parsers, bitstream filters and encoders/decoders) simultaneously, with
there being no clear rules for who can modify what and the different
processes being typically delayed with respect to each other.
- avformat_find_stream_info() making it necessary to support opening
and closing a single codec context multiple times, thus
complicating the semantics of freeing various allocated objects in the
codec context.
Those problems are resolved by replacing the AVStream embedded codec
context with a newly added AVCodecParameters instance, which stores only
the stream parameters exported by the demuxers or read by the muxers.
2014-06-18 21:42:52 +03:00
st - > codecpar - > codec_type = AVMEDIA_TYPE_SUBTITLE ;
2021-02-16 20:38:19 +02:00
if ( track - > flag_textdescriptions )
st - > disposition | = AV_DISPOSITION_DESCRIPTIONS ;
2008-08-05 03:40:52 +03:00
}
2007-06-05 01:10:54 +03:00
}
2014-05-25 10:07:32 +03:00
return 0 ;
}
static int matroska_read_header ( AVFormatContext * s )
{
MatroskaDemuxContext * matroska = s - > priv_data ;
EbmlList * attachments_list = & matroska - > attachments ;
EbmlList * chapters_list = & matroska - > chapters ;
MatroskaAttachment * attachments ;
MatroskaChapter * chapters ;
uint64_t max_start = 0 ;
int64_t pos ;
Ebml ebml = { 0 } ;
int i , j , res ;
matroska - > ctx = s ;
2015-02-09 21:39:00 +02:00
matroska - > cues_parsing_deferred = 1 ;
2014-05-25 10:07:32 +03:00
/* First read the EBML header. */
2015-04-10 19:04:51 +02:00
if ( ebml_parse ( matroska , ebml_syntax , & ebml ) | | ! ebml . doctype ) {
av_log ( matroska - > ctx , AV_LOG_ERROR , " EBML header parsing failed \n " ) ;
ebml_free ( ebml_syntax , & ebml ) ;
return AVERROR_INVALIDDATA ;
}
if ( ebml . version > EBML_VERSION | |
2014-05-25 10:07:32 +03:00
ebml . max_size > sizeof ( uint64_t ) | |
ebml . id_length > sizeof ( uint32_t ) | |
2015-04-19 16:45:24 +02:00
ebml . doctype_version > 3 ) {
2015-12-16 19:01:34 +02:00
avpriv_report_missing_feature ( matroska - > ctx ,
" EBML version % " PRIu64 " , doctype %s, doc version % " PRIu64 ,
ebml . version , ebml . doctype , ebml . doctype_version ) ;
2014-05-25 10:07:32 +03:00
ebml_free ( ebml_syntax , & ebml ) ;
return AVERROR_PATCHWELCOME ;
2014-05-28 13:41:35 +03:00
} else if ( ebml . doctype_version = = 3 ) {
av_log ( matroska - > ctx , AV_LOG_WARNING ,
" EBML header using unsupported features \n "
" (EBML version % " PRIu64 " , doctype %s, doc version % " PRIu64 " ) \n " ,
ebml . version , ebml . doctype , ebml . doctype_version ) ;
2014-05-25 10:07:32 +03:00
}
for ( i = 0 ; i < FF_ARRAY_ELEMS ( matroska_doctypes ) ; i + + )
if ( ! strcmp ( ebml . doctype , matroska_doctypes [ i ] ) )
break ;
if ( i > = FF_ARRAY_ELEMS ( matroska_doctypes ) ) {
av_log ( s , AV_LOG_WARNING , " Unknown EBML doctype '%s' \n " , ebml . doctype ) ;
if ( matroska - > ctx - > error_recognition & AV_EF_EXPLODE ) {
ebml_free ( ebml_syntax , & ebml ) ;
return AVERROR_INVALIDDATA ;
}
}
ebml_free ( ebml_syntax , & ebml ) ;
2021-03-18 07:04:17 +02:00
matroska - > pkt = s - > internal - > parse_pkt ;
2021-01-29 15:42:48 +02:00
2014-05-25 10:07:32 +03:00
/* The next thing is a segment. */
pos = avio_tell ( matroska - > ctx - > pb ) ;
res = ebml_parse ( matroska , matroska_segments , matroska ) ;
2019-05-17 00:30:13 +02:00
// Try resyncing until we find an EBML_STOP type element.
2014-05-25 10:07:32 +03:00
while ( res ! = 1 ) {
res = matroska_resync ( matroska , pos ) ;
if ( res < 0 )
2016-06-24 18:26:46 +02:00
goto fail ;
2014-05-25 10:07:32 +03:00
pos = avio_tell ( matroska - > ctx - > pb ) ;
res = ebml_parse ( matroska , matroska_segment , matroska ) ;
2021-01-29 22:18:36 +02:00
if ( res = = AVERROR ( EIO ) ) // EOF is translated to EIO, this exists the loop on EOF
goto fail ;
2014-05-25 10:07:32 +03:00
}
2019-05-17 00:29:51 +02:00
/* Set data_offset as it might be needed later by seek_frame_generic. */
if ( matroska - > current_id = = MATROSKA_ID_CLUSTER )
s - > internal - > data_offset = avio_tell ( matroska - > ctx - > pb ) - 4 ;
2014-05-25 10:07:32 +03:00
matroska_execute_seekhead ( matroska ) ;
if ( ! matroska - > time_scale )
matroska - > time_scale = 1000000 ;
if ( matroska - > duration )
matroska - > ctx - > duration = matroska - > duration * matroska - > time_scale *
1000 / AV_TIME_BASE ;
av_dict_set ( & s - > metadata , " title " , matroska - > title , 0 ) ;
2014-05-28 13:41:35 +03:00
av_dict_set ( & s - > metadata , " encoder " , matroska - > muxingapp , 0 ) ;
if ( matroska - > date_utc . size = = 8 )
matroska_metadata_creation_time ( & s - > metadata , AV_RB64 ( matroska - > date_utc . data ) ) ;
2014-05-25 10:07:32 +03:00
res = matroska_parse_tracks ( s ) ;
if ( res < 0 )
2016-06-24 18:26:46 +02:00
goto fail ;
2014-05-25 10:07:32 +03:00
2014-03-07 15:54:18 +03:00
attachments = attachments_list - > elem ;
for ( j = 0 ; j < attachments_list - > nb_elem ; j + + ) {
if ( ! ( attachments [ j ] . filename & & attachments [ j ] . mime & &
attachments [ j ] . bin . data & & attachments [ j ] . bin . size > 0 ) ) {
2008-08-05 03:40:55 +03:00
av_log ( matroska - > ctx , AV_LOG_ERROR , " incomplete attachment \n " ) ;
} else {
2011-06-18 12:43:24 +03:00
AVStream * st = avformat_new_stream ( s , NULL ) ;
2014-08-14 23:31:24 +03:00
if ( ! st )
2008-08-05 03:40:55 +03:00
break ;
2014-03-07 15:54:18 +03:00
av_dict_set ( & st - > metadata , " filename " , attachments [ j ] . filename , 0 ) ;
av_dict_set ( & st - > metadata , " mimetype " , attachments [ j ] . mime , 0 ) ;
2020-05-06 15:24:33 +02:00
if ( attachments [ j ] . description )
av_dict_set ( & st - > metadata , " title " , attachments [ j ] . description , 0 ) ;
lavf: replace AVStream.codec with AVStream.codecpar
Currently, AVStream contains an embedded AVCodecContext instance, which
is used by demuxers to export stream parameters to the caller and by
muxers to receive stream parameters from the caller. It is also used
internally as the codec context that is passed to parsers.
In addition, it is also widely used by the callers as the decoding (when
demuxer) or encoding (when muxing) context, though this has been
officially discouraged since Libav 11.
There are multiple important problems with this approach:
- the fields in AVCodecContext are in general one of
* stream parameters
* codec options
* codec state
However, it's not clear which ones are which. It is consequently
unclear which fields are a demuxer allowed to set or a muxer allowed to
read. This leads to erratic behaviour depending on whether decoding or
encoding is being performed or not (and whether it uses the AVStream
embedded codec context).
- various synchronization issues arising from the fact that the same
context is used by several different APIs (muxers/demuxers,
parsers, bitstream filters and encoders/decoders) simultaneously, with
there being no clear rules for who can modify what and the different
processes being typically delayed with respect to each other.
- avformat_find_stream_info() making it necessary to support opening
and closing a single codec context multiple times, thus
complicating the semantics of freeing various allocated objects in the
codec context.
Those problems are resolved by replacing the AVStream embedded codec
context with a newly added AVCodecParameters instance, which stores only
the stream parameters exported by the demuxers or read by the muxers.
2014-06-18 21:42:52 +03:00
st - > codecpar - > codec_id = AV_CODEC_ID_NONE ;
2008-08-05 03:40:55 +03:00
2020-04-16 03:39:05 +02:00
for ( i = 0 ; mkv_image_mime_tags [ i ] . id ! = AV_CODEC_ID_NONE ; i + + ) {
2021-02-24 08:44:31 +02:00
if ( av_strstart ( attachments [ j ] . mime , mkv_image_mime_tags [ i ] . str , NULL ) ) {
2020-04-16 03:39:05 +02:00
st - > codecpar - > codec_id = mkv_image_mime_tags [ i ] . id ;
2015-04-03 16:11:53 +02:00
break ;
}
}
2014-03-07 15:54:18 +03:00
attachments [ j ] . stream = st ;
2015-04-03 16:11:53 +02:00
lavf: replace AVStream.codec with AVStream.codecpar
Currently, AVStream contains an embedded AVCodecContext instance, which
is used by demuxers to export stream parameters to the caller and by
muxers to receive stream parameters from the caller. It is also used
internally as the codec context that is passed to parsers.
In addition, it is also widely used by the callers as the decoding (when
demuxer) or encoding (when muxing) context, though this has been
officially discouraged since Libav 11.
There are multiple important problems with this approach:
- the fields in AVCodecContext are in general one of
* stream parameters
* codec options
* codec state
However, it's not clear which ones are which. It is consequently
unclear which fields are a demuxer allowed to set or a muxer allowed to
read. This leads to erratic behaviour depending on whether decoding or
encoding is being performed or not (and whether it uses the AVStream
embedded codec context).
- various synchronization issues arising from the fact that the same
context is used by several different APIs (muxers/demuxers,
parsers, bitstream filters and encoders/decoders) simultaneously, with
there being no clear rules for who can modify what and the different
processes being typically delayed with respect to each other.
- avformat_find_stream_info() making it necessary to support opening
and closing a single codec context multiple times, thus
complicating the semantics of freeing various allocated objects in the
codec context.
Those problems are resolved by replacing the AVStream embedded codec
context with a newly added AVCodecParameters instance, which stores only
the stream parameters exported by the demuxers or read by the muxers.
2014-06-18 21:42:52 +03:00
if ( st - > codecpar - > codec_id ! = AV_CODEC_ID_NONE ) {
2021-03-29 07:58:56 +02:00
res = ff_add_attached_pic ( s , st , NULL , & attachments [ j ] . bin . buf , 0 ) ;
if ( res < 0 )
goto fail ;
2015-04-03 16:11:53 +02:00
} else {
lavf: replace AVStream.codec with AVStream.codecpar
Currently, AVStream contains an embedded AVCodecContext instance, which
is used by demuxers to export stream parameters to the caller and by
muxers to receive stream parameters from the caller. It is also used
internally as the codec context that is passed to parsers.
In addition, it is also widely used by the callers as the decoding (when
demuxer) or encoding (when muxing) context, though this has been
officially discouraged since Libav 11.
There are multiple important problems with this approach:
- the fields in AVCodecContext are in general one of
* stream parameters
* codec options
* codec state
However, it's not clear which ones are which. It is consequently
unclear which fields are a demuxer allowed to set or a muxer allowed to
read. This leads to erratic behaviour depending on whether decoding or
encoding is being performed or not (and whether it uses the AVStream
embedded codec context).
- various synchronization issues arising from the fact that the same
context is used by several different APIs (muxers/demuxers,
parsers, bitstream filters and encoders/decoders) simultaneously, with
there being no clear rules for who can modify what and the different
processes being typically delayed with respect to each other.
- avformat_find_stream_info() making it necessary to support opening
and closing a single codec context multiple times, thus
complicating the semantics of freeing various allocated objects in the
codec context.
Those problems are resolved by replacing the AVStream embedded codec
context with a newly added AVCodecParameters instance, which stores only
the stream parameters exported by the demuxers or read by the muxers.
2014-06-18 21:42:52 +03:00
st - > codecpar - > codec_type = AVMEDIA_TYPE_ATTACHMENT ;
2016-04-10 21:58:15 +02:00
if ( ff_alloc_extradata ( st - > codecpar , attachments [ j ] . bin . size ) )
2008-08-05 03:40:55 +03:00
break ;
lavf: replace AVStream.codec with AVStream.codecpar
Currently, AVStream contains an embedded AVCodecContext instance, which
is used by demuxers to export stream parameters to the caller and by
muxers to receive stream parameters from the caller. It is also used
internally as the codec context that is passed to parsers.
In addition, it is also widely used by the callers as the decoding (when
demuxer) or encoding (when muxing) context, though this has been
officially discouraged since Libav 11.
There are multiple important problems with this approach:
- the fields in AVCodecContext are in general one of
* stream parameters
* codec options
* codec state
However, it's not clear which ones are which. It is consequently
unclear which fields are a demuxer allowed to set or a muxer allowed to
read. This leads to erratic behaviour depending on whether decoding or
encoding is being performed or not (and whether it uses the AVStream
embedded codec context).
- various synchronization issues arising from the fact that the same
context is used by several different APIs (muxers/demuxers,
parsers, bitstream filters and encoders/decoders) simultaneously, with
there being no clear rules for who can modify what and the different
processes being typically delayed with respect to each other.
- avformat_find_stream_info() making it necessary to support opening
and closing a single codec context multiple times, thus
complicating the semantics of freeing various allocated objects in the
codec context.
Those problems are resolved by replacing the AVStream embedded codec
context with a newly added AVCodecParameters instance, which stores only
the stream parameters exported by the demuxers or read by the muxers.
2014-06-18 21:42:52 +03:00
memcpy ( st - > codecpar - > extradata , attachments [ j ] . bin . data ,
2015-04-03 16:11:53 +02:00
attachments [ j ] . bin . size ) ;
2015-04-03 16:11:53 +02:00
2020-04-16 03:39:05 +02:00
for ( i = 0 ; mkv_mime_tags [ i ] . id ! = AV_CODEC_ID_NONE ; i + + ) {
2021-02-24 08:44:31 +02:00
if ( av_strstart ( attachments [ j ] . mime , mkv_mime_tags [ i ] . str , NULL ) ) {
2020-04-16 03:39:05 +02:00
st - > codecpar - > codec_id = mkv_mime_tags [ i ] . id ;
2015-04-03 16:11:53 +02:00
break ;
}
2008-08-05 03:40:55 +03:00
}
}
}
}
chapters = chapters_list - > elem ;
2014-03-01 19:28:15 +03:00
for ( i = 0 ; i < chapters_list - > nb_elem ; i + + )
if ( chapters [ i ] . start ! = AV_NOPTS_VALUE & & chapters [ i ] . uid & &
( max_start = = 0 | | chapters [ i ] . start > max_start ) ) {
2009-02-15 17:25:14 +02:00
chapters [ i ] . chapter =
2014-03-01 19:28:15 +03:00
avpriv_new_chapter ( s , chapters [ i ] . uid ,
( AVRational ) { 1 , 1000000000 } ,
chapters [ i ] . start , chapters [ i ] . end ,
chapters [ i ] . title ) ;
2009-01-15 02:42:57 +02:00
max_start = chapters [ i ] . start ;
}
2008-08-05 03:40:55 +03:00
2011-09-22 17:51:00 +03:00
matroska_add_index_entries ( matroska ) ;
2009-02-15 17:53:55 +02:00
matroska_convert_tags ( s ) ;
2008-08-05 03:40:58 +03:00
return 0 ;
2016-06-24 18:26:46 +02:00
fail :
matroska_read_close ( s ) ;
return res ;
2007-06-05 01:10:54 +03:00
}
2008-08-05 03:42:39 +03:00
/*
* Put one packet in an application - supplied AVPacket struct .
* Returns 0 on success or - 1 on failure .
*/
static int matroska_deliver_packet ( MatroskaDemuxContext * matroska ,
AVPacket * pkt )
{
2018-03-26 20:02:37 +02:00
if ( matroska - > queue ) {
2016-02-24 19:14:05 +02:00
MatroskaTrack * tracks = matroska - > tracks . elem ;
MatroskaTrack * track ;
2018-03-26 20:02:37 +02:00
2020-08-17 17:03:50 +02:00
avpriv_packet_list_get ( & matroska - > queue , & matroska - > queue_end , pkt ) ;
2016-02-24 19:14:05 +02:00
track = & tracks [ pkt - > stream_index ] ;
if ( track - > has_palette ) {
2015-12-27 22:28:09 +02:00
uint8_t * pal = av_packet_new_side_data ( pkt , AV_PKT_DATA_PALETTE , AVPALETTE_SIZE ) ;
if ( ! pal ) {
av_log ( matroska - > ctx , AV_LOG_ERROR , " Cannot append palette to packet \n " ) ;
} else {
2016-02-24 19:14:05 +02:00
memcpy ( pal , track - > palette , AVPALETTE_SIZE ) ;
2015-12-27 22:28:09 +02:00
}
2016-02-24 19:14:05 +02:00
track - > has_palette = 0 ;
2015-12-27 22:28:09 +02:00
}
2008-08-05 03:42:39 +03:00
return 0 ;
}
return - 1 ;
}
/*
* Free all packets in our internal queue .
*/
static void matroska_clear_queue ( MatroskaDemuxContext * matroska )
{
2020-08-17 17:03:50 +02:00
avpriv_packet_list_free ( & matroska - > queue , & matroska - > queue_end ) ;
2008-08-05 03:42:39 +03:00
}
2012-09-17 02:58:32 +03:00
static int matroska_parse_laces ( MatroskaDemuxContext * matroska , uint8_t * * buf ,
2019-12-03 19:09:10 +02:00
int size , int type , AVIOContext * pb ,
2019-12-03 19:09:06 +02:00
uint32_t lace_size [ 256 ] , int * laces )
2012-09-17 02:58:32 +03:00
{
2019-12-03 19:09:08 +02:00
int n ;
2012-09-17 02:58:32 +03:00
uint8_t * data = * buf ;
if ( ! type ) {
2014-03-01 19:28:15 +03:00
* laces = 1 ;
2019-12-03 19:09:06 +02:00
lace_size [ 0 ] = size ;
2012-09-17 02:58:32 +03:00
return 0 ;
}
2020-03-25 07:00:53 +02:00
if ( size < = 0 )
return AVERROR_INVALIDDATA ;
2020-05-20 21:24:34 +02:00
* laces = * data + 1 ;
data + = 1 ;
size - = 1 ;
2012-09-17 02:58:32 +03:00
switch ( type ) {
2014-03-01 19:28:15 +03:00
case 0x1 : /* Xiph lacing */
{
2012-09-17 02:58:32 +03:00
uint8_t temp ;
uint32_t total = 0 ;
2019-12-03 19:09:07 +02:00
for ( n = 0 ; n < * laces - 1 ; n + + ) {
2019-05-17 00:29:48 +02:00
lace_size [ n ] = 0 ;
2020-05-20 21:24:34 +02:00
do {
if ( size < = total )
2019-12-03 19:09:07 +02:00
return AVERROR_INVALIDDATA ;
2014-03-01 19:28:15 +03:00
temp = * data ;
2014-03-08 01:46:37 +03:00
total + = temp ;
2012-09-17 02:58:32 +03:00
lace_size [ n ] + = temp ;
2014-03-01 19:28:15 +03:00
data + = 1 ;
size - = 1 ;
2020-05-20 21:24:34 +02:00
} while ( temp = = 0xff ) ;
2012-09-17 02:58:32 +03:00
}
2020-05-20 21:24:34 +02:00
if ( size < total )
2019-12-03 19:09:07 +02:00
return AVERROR_INVALIDDATA ;
2012-09-17 02:58:32 +03:00
lace_size [ n ] = size - total ;
break ;
}
case 0x2 : /* fixed-size lacing */
2020-05-20 21:24:34 +02:00
if ( size % ( * laces ) )
2019-12-03 19:09:07 +02:00
return AVERROR_INVALIDDATA ;
2012-09-17 02:58:32 +03:00
for ( n = 0 ; n < * laces ; n + + )
lace_size [ n ] = size / * laces ;
break ;
2014-03-01 19:28:15 +03:00
case 0x3 : /* EBML lacing */
{
2012-09-17 02:58:32 +03:00
uint64_t num ;
2013-03-28 13:52:52 +03:00
uint64_t total ;
2019-12-03 19:09:10 +02:00
int offset ;
avio_skip ( pb , 4 ) ;
n = ebml_read_num ( matroska , pb , 8 , & num , 1 ) ;
2019-12-03 19:09:09 +02:00
if ( n < 0 )
return n ;
if ( num > INT_MAX )
return AVERROR_INVALIDDATA ;
2019-12-03 19:09:10 +02:00
2012-09-17 02:58:32 +03:00
total = lace_size [ 0 ] = num ;
2019-12-03 19:09:10 +02:00
offset = n ;
2019-12-03 19:09:07 +02:00
for ( n = 1 ; n < * laces - 1 ; n + + ) {
2012-09-17 02:58:32 +03:00
int64_t snum ;
int r ;
2019-12-03 19:09:10 +02:00
r = matroska_ebmlnum_sint ( matroska , pb , & snum ) ;
2019-12-03 19:09:09 +02:00
if ( r < 0 )
return r ;
if ( lace_size [ n - 1 ] + snum > ( uint64_t ) INT_MAX )
return AVERROR_INVALIDDATA ;
2019-12-03 19:09:10 +02:00
2012-09-17 02:58:32 +03:00
lace_size [ n ] = lace_size [ n - 1 ] + snum ;
2014-03-01 19:28:15 +03:00
total + = lace_size [ n ] ;
2019-12-03 19:09:10 +02:00
offset + = r ;
2012-09-17 02:58:32 +03:00
}
2019-12-03 19:09:10 +02:00
data + = offset ;
size - = offset ;
2020-05-20 21:24:34 +02:00
if ( size < total )
2019-12-03 19:09:07 +02:00
return AVERROR_INVALIDDATA ;
2020-05-20 21:24:34 +02:00
2012-09-17 02:58:32 +03:00
lace_size [ * laces - 1 ] = size - total ;
break ;
}
}
2020-05-20 21:24:34 +02:00
* buf = data ;
2012-09-17 02:58:32 +03:00
2019-12-03 19:09:07 +02:00
return 0 ;
2012-09-17 02:58:32 +03:00
}
2012-09-17 02:28:13 +03:00
static int matroska_parse_rm_audio ( MatroskaDemuxContext * matroska ,
2014-03-01 19:28:15 +03:00
MatroskaTrack * track , AVStream * st ,
uint8_t * data , int size , uint64_t timecode ,
2012-09-17 02:28:13 +03:00
int64_t pos )
{
2019-12-07 01:11:01 +02:00
const int a = st - > codecpar - > block_align ;
const int sps = track - > audio . sub_packet_size ;
const int cfs = track - > audio . coded_framesize ;
const int h = track - > audio . sub_packet_h ;
const int w = track - > audio . frame_size ;
2014-03-01 19:28:15 +03:00
int y = track - > audio . sub_packet_cnt ;
2012-09-17 02:28:13 +03:00
int x ;
if ( ! track - > audio . pkt_cnt ) {
if ( track - > audio . sub_packet_cnt = = 0 )
track - > audio . buf_timecode = timecode ;
lavf: replace AVStream.codec with AVStream.codecpar
Currently, AVStream contains an embedded AVCodecContext instance, which
is used by demuxers to export stream parameters to the caller and by
muxers to receive stream parameters from the caller. It is also used
internally as the codec context that is passed to parsers.
In addition, it is also widely used by the callers as the decoding (when
demuxer) or encoding (when muxing) context, though this has been
officially discouraged since Libav 11.
There are multiple important problems with this approach:
- the fields in AVCodecContext are in general one of
* stream parameters
* codec options
* codec state
However, it's not clear which ones are which. It is consequently
unclear which fields are a demuxer allowed to set or a muxer allowed to
read. This leads to erratic behaviour depending on whether decoding or
encoding is being performed or not (and whether it uses the AVStream
embedded codec context).
- various synchronization issues arising from the fact that the same
context is used by several different APIs (muxers/demuxers,
parsers, bitstream filters and encoders/decoders) simultaneously, with
there being no clear rules for who can modify what and the different
processes being typically delayed with respect to each other.
- avformat_find_stream_info() making it necessary to support opening
and closing a single codec context multiple times, thus
complicating the semantics of freeing various allocated objects in the
codec context.
Those problems are resolved by replacing the AVStream embedded codec
context with a newly added AVCodecParameters instance, which stores only
the stream parameters exported by the demuxers or read by the muxers.
2014-06-18 21:42:52 +03:00
if ( st - > codecpar - > codec_id = = AV_CODEC_ID_RA_288 ) {
2012-09-17 02:28:13 +03:00
if ( size < cfs * h / 2 ) {
av_log ( matroska - > ctx , AV_LOG_ERROR ,
" Corrupt int4 RM-style audio packet size \n " ) ;
return AVERROR_INVALIDDATA ;
}
2014-03-01 19:28:15 +03:00
for ( x = 0 ; x < h / 2 ; x + + )
memcpy ( track - > audio . buf + x * 2 * w + y * cfs ,
data + x * cfs , cfs ) ;
lavf: replace AVStream.codec with AVStream.codecpar
Currently, AVStream contains an embedded AVCodecContext instance, which
is used by demuxers to export stream parameters to the caller and by
muxers to receive stream parameters from the caller. It is also used
internally as the codec context that is passed to parsers.
In addition, it is also widely used by the callers as the decoding (when
demuxer) or encoding (when muxing) context, though this has been
officially discouraged since Libav 11.
There are multiple important problems with this approach:
- the fields in AVCodecContext are in general one of
* stream parameters
* codec options
* codec state
However, it's not clear which ones are which. It is consequently
unclear which fields are a demuxer allowed to set or a muxer allowed to
read. This leads to erratic behaviour depending on whether decoding or
encoding is being performed or not (and whether it uses the AVStream
embedded codec context).
- various synchronization issues arising from the fact that the same
context is used by several different APIs (muxers/demuxers,
parsers, bitstream filters and encoders/decoders) simultaneously, with
there being no clear rules for who can modify what and the different
processes being typically delayed with respect to each other.
- avformat_find_stream_info() making it necessary to support opening
and closing a single codec context multiple times, thus
complicating the semantics of freeing various allocated objects in the
codec context.
Those problems are resolved by replacing the AVStream embedded codec
context with a newly added AVCodecParameters instance, which stores only
the stream parameters exported by the demuxers or read by the muxers.
2014-06-18 21:42:52 +03:00
} else if ( st - > codecpar - > codec_id = = AV_CODEC_ID_SIPR ) {
2012-09-17 02:28:13 +03:00
if ( size < w ) {
av_log ( matroska - > ctx , AV_LOG_ERROR ,
" Corrupt sipr RM-style audio packet size \n " ) ;
return AVERROR_INVALIDDATA ;
}
2014-03-01 19:28:15 +03:00
memcpy ( track - > audio . buf + y * w , data , w ) ;
2012-09-17 02:28:13 +03:00
} else {
2020-04-20 08:54:23 +02:00
if ( size < w ) {
2012-09-17 02:28:13 +03:00
av_log ( matroska - > ctx , AV_LOG_ERROR ,
" Corrupt generic RM-style audio packet size \n " ) ;
return AVERROR_INVALIDDATA ;
}
2014-03-01 19:28:15 +03:00
for ( x = 0 ; x < w / sps ; x + + )
memcpy ( track - > audio . buf +
sps * ( h * x + ( ( h + 1 ) / 2 ) * ( y & 1 ) + ( y > > 1 ) ) ,
data + x * sps , sps ) ;
2012-09-17 02:28:13 +03:00
}
if ( + + track - > audio . sub_packet_cnt > = h ) {
lavf: replace AVStream.codec with AVStream.codecpar
Currently, AVStream contains an embedded AVCodecContext instance, which
is used by demuxers to export stream parameters to the caller and by
muxers to receive stream parameters from the caller. It is also used
internally as the codec context that is passed to parsers.
In addition, it is also widely used by the callers as the decoding (when
demuxer) or encoding (when muxing) context, though this has been
officially discouraged since Libav 11.
There are multiple important problems with this approach:
- the fields in AVCodecContext are in general one of
* stream parameters
* codec options
* codec state
However, it's not clear which ones are which. It is consequently
unclear which fields are a demuxer allowed to set or a muxer allowed to
read. This leads to erratic behaviour depending on whether decoding or
encoding is being performed or not (and whether it uses the AVStream
embedded codec context).
- various synchronization issues arising from the fact that the same
context is used by several different APIs (muxers/demuxers,
parsers, bitstream filters and encoders/decoders) simultaneously, with
there being no clear rules for who can modify what and the different
processes being typically delayed with respect to each other.
- avformat_find_stream_info() making it necessary to support opening
and closing a single codec context multiple times, thus
complicating the semantics of freeing various allocated objects in the
codec context.
Those problems are resolved by replacing the AVStream embedded codec
context with a newly added AVCodecParameters instance, which stores only
the stream parameters exported by the demuxers or read by the muxers.
2014-06-18 21:42:52 +03:00
if ( st - > codecpar - > codec_id = = AV_CODEC_ID_SIPR )
2012-09-17 02:28:13 +03:00
ff_rm_reorder_sipr_data ( track - > audio . buf , h , w ) ;
track - > audio . sub_packet_cnt = 0 ;
2014-03-01 19:28:15 +03:00
track - > audio . pkt_cnt = h * w / a ;
2012-09-17 02:28:13 +03:00
}
}
while ( track - > audio . pkt_cnt ) {
2014-10-24 02:05:53 +03:00
int ret ;
2021-01-29 15:42:48 +02:00
AVPacket * pkt = matroska - > pkt ;
2014-10-24 02:05:53 +03:00
ret = av_new_packet ( pkt , a ) ;
if ( ret < 0 ) {
return ret ;
2012-10-20 17:55:45 +03:00
}
2014-03-01 19:28:15 +03:00
memcpy ( pkt - > data ,
track - > audio . buf + a * ( h * w / a - track - > audio . pkt_cnt - - ) ,
a ) ;
pkt - > pts = track - > audio . buf_timecode ;
2012-09-17 02:28:13 +03:00
track - > audio . buf_timecode = AV_NOPTS_VALUE ;
2014-03-01 19:28:15 +03:00
pkt - > pos = pos ;
pkt - > stream_index = st - > index ;
2020-08-17 17:03:50 +02:00
ret = avpriv_packet_list_put ( & matroska - > queue , & matroska - > queue_end , pkt , NULL , 0 ) ;
2018-03-26 20:02:37 +02:00
if ( ret < 0 ) {
2018-04-04 18:53:12 +02:00
av_packet_unref ( pkt ) ;
2018-03-26 20:02:37 +02:00
return AVERROR ( ENOMEM ) ;
}
2012-09-17 02:28:13 +03:00
}
return 0 ;
}
2013-05-27 10:44:27 +03:00
/* reconstruct full wavpack blocks from mangled matroska ones */
2019-12-06 10:53:34 +02:00
static int matroska_parse_wavpack ( MatroskaTrack * track ,
uint8_t * * data , int * size )
2013-05-27 10:44:27 +03:00
{
uint8_t * dst = NULL ;
2019-12-06 10:53:34 +02:00
uint8_t * src = * data ;
2013-05-27 10:44:27 +03:00
int dstlen = 0 ;
int srclen = * size ;
uint32_t samples ;
uint16_t ver ;
int ret , offset = 0 ;
2019-12-14 02:36:54 +02:00
if ( srclen < 12 )
2013-05-27 10:44:27 +03:00
return AVERROR_INVALIDDATA ;
2019-12-14 02:36:54 +02:00
av_assert1 ( track - > stream - > codecpar - > extradata_size > = 2 ) ;
lavf: replace AVStream.codec with AVStream.codecpar
Currently, AVStream contains an embedded AVCodecContext instance, which
is used by demuxers to export stream parameters to the caller and by
muxers to receive stream parameters from the caller. It is also used
internally as the codec context that is passed to parsers.
In addition, it is also widely used by the callers as the decoding (when
demuxer) or encoding (when muxing) context, though this has been
officially discouraged since Libav 11.
There are multiple important problems with this approach:
- the fields in AVCodecContext are in general one of
* stream parameters
* codec options
* codec state
However, it's not clear which ones are which. It is consequently
unclear which fields are a demuxer allowed to set or a muxer allowed to
read. This leads to erratic behaviour depending on whether decoding or
encoding is being performed or not (and whether it uses the AVStream
embedded codec context).
- various synchronization issues arising from the fact that the same
context is used by several different APIs (muxers/demuxers,
parsers, bitstream filters and encoders/decoders) simultaneously, with
there being no clear rules for who can modify what and the different
processes being typically delayed with respect to each other.
- avformat_find_stream_info() making it necessary to support opening
and closing a single codec context multiple times, thus
complicating the semantics of freeing various allocated objects in the
codec context.
Those problems are resolved by replacing the AVStream embedded codec
context with a newly added AVCodecParameters instance, which stores only
the stream parameters exported by the demuxers or read by the muxers.
2014-06-18 21:42:52 +03:00
ver = AV_RL16 ( track - > stream - > codecpar - > extradata ) ;
2013-05-27 10:44:27 +03:00
samples = AV_RL32 ( src ) ;
src + = 4 ;
srclen - = 4 ;
while ( srclen > = 8 ) {
int multiblock ;
uint32_t blocksize ;
uint8_t * tmp ;
uint32_t flags = AV_RL32 ( src ) ;
uint32_t crc = AV_RL32 ( src + 4 ) ;
src + = 8 ;
srclen - = 8 ;
multiblock = ( flags & 0x1800 ) ! = 0x1800 ;
if ( multiblock ) {
if ( srclen < 4 ) {
ret = AVERROR_INVALIDDATA ;
goto fail ;
}
blocksize = AV_RL32 ( src ) ;
2014-03-01 19:28:15 +03:00
src + = 4 ;
srclen - = 4 ;
2013-05-27 10:44:27 +03:00
} else
blocksize = srclen ;
if ( blocksize > srclen ) {
ret = AVERROR_INVALIDDATA ;
goto fail ;
}
2018-04-04 23:12:53 +02:00
tmp = av_realloc ( dst , dstlen + blocksize + 32 + AV_INPUT_BUFFER_PADDING_SIZE ) ;
2013-05-27 10:44:27 +03:00
if ( ! tmp ) {
ret = AVERROR ( ENOMEM ) ;
goto fail ;
}
dst = tmp ;
dstlen + = blocksize + 32 ;
2014-03-01 19:28:15 +03:00
AV_WL32 ( dst + offset , MKTAG ( ' w ' , ' v ' , ' p ' , ' k ' ) ) ; // tag
AV_WL32 ( dst + offset + 4 , blocksize + 24 ) ; // blocksize - 8
AV_WL16 ( dst + offset + 8 , ver ) ; // version
AV_WL16 ( dst + offset + 10 , 0 ) ; // track/index_no
AV_WL32 ( dst + offset + 12 , 0 ) ; // total samples
AV_WL32 ( dst + offset + 16 , 0 ) ; // block index
AV_WL32 ( dst + offset + 20 , samples ) ; // number of samples
AV_WL32 ( dst + offset + 24 , flags ) ; // flags
AV_WL32 ( dst + offset + 28 , crc ) ; // crc
memcpy ( dst + offset + 32 , src , blocksize ) ; // block data
2013-05-27 10:44:27 +03:00
src + = blocksize ;
srclen - = blocksize ;
offset + = blocksize + 32 ;
}
2018-04-04 23:12:53 +02:00
memset ( dst + dstlen , 0 , AV_INPUT_BUFFER_PADDING_SIZE ) ;
2019-12-06 10:53:34 +02:00
* data = dst ;
2013-05-27 10:44:27 +03:00
* size = dstlen ;
return 0 ;
fail :
av_freep ( & dst ) ;
return ret ;
}
2019-12-06 10:53:34 +02:00
static int matroska_parse_prores ( MatroskaTrack * track ,
uint8_t * * data , int * size )
2018-04-04 21:07:36 +02:00
{
2019-12-07 01:16:19 +02:00
uint8_t * dst ;
int dstlen = * size + 8 ;
2019-09-28 19:54:25 +02:00
2019-12-07 01:11:01 +02:00
dst = av_malloc ( dstlen + AV_INPUT_BUFFER_PADDING_SIZE ) ;
if ( ! dst )
return AVERROR ( ENOMEM ) ;
2018-04-04 21:07:36 +02:00
2019-12-07 01:11:01 +02:00
AV_WB32 ( dst , dstlen ) ;
AV_WB32 ( dst + 4 , MKBETAG ( ' i ' , ' c ' , ' p ' , ' f ' ) ) ;
memcpy ( dst + 8 , * data , dstlen - 8 ) ;
memset ( dst + dstlen , 0 , AV_INPUT_BUFFER_PADDING_SIZE ) ;
2018-04-04 21:07:36 +02:00
2019-12-06 10:53:34 +02:00
* data = dst ;
2018-04-04 21:07:36 +02:00
* size = dstlen ;
return 0 ;
}
2013-08-09 01:40:03 +03:00
static int matroska_parse_webvtt ( MatroskaDemuxContext * matroska ,
MatroskaTrack * track ,
AVStream * st ,
uint8_t * data , int data_len ,
uint64_t timecode ,
uint64_t duration ,
int64_t pos )
{
2021-01-29 15:42:48 +02:00
AVPacket * pkt = matroska - > pkt ;
2013-08-09 01:40:03 +03:00
uint8_t * id , * settings , * text , * buf ;
int id_len , settings_len , text_len ;
uint8_t * p , * q ;
int err ;
if ( data_len < = 0 )
return AVERROR_INVALIDDATA ;
p = data ;
q = data + data_len ;
id = p ;
id_len = - 1 ;
while ( p < q ) {
if ( * p = = ' \r ' | | * p = = ' \n ' ) {
id_len = p - id ;
if ( * p = = ' \r ' )
p + + ;
break ;
}
p + + ;
}
if ( p > = q | | * p ! = ' \n ' )
return AVERROR_INVALIDDATA ;
p + + ;
settings = p ;
settings_len = - 1 ;
while ( p < q ) {
if ( * p = = ' \r ' | | * p = = ' \n ' ) {
settings_len = p - settings ;
if ( * p = = ' \r ' )
p + + ;
break ;
}
p + + ;
}
if ( p > = q | | * p ! = ' \n ' )
return AVERROR_INVALIDDATA ;
p + + ;
text = p ;
text_len = q - p ;
while ( text_len > 0 ) {
const int len = text_len - 1 ;
const uint8_t c = p [ len ] ;
if ( c ! = ' \r ' & & c ! = ' \n ' )
break ;
text_len = len ;
}
if ( text_len < = 0 )
return AVERROR_INVALIDDATA ;
err = av_new_packet ( pkt , text_len ) ;
if ( err < 0 ) {
2018-02-20 15:26:21 +02:00
return err ;
2013-08-09 01:40:03 +03:00
}
memcpy ( pkt - > data , text , text_len ) ;
if ( id_len > 0 ) {
buf = av_packet_new_side_data ( pkt ,
AV_PKT_DATA_WEBVTT_IDENTIFIER ,
id_len ) ;
2014-08-15 21:33:21 +03:00
if ( ! buf ) {
2018-02-20 15:25:54 +02:00
av_packet_unref ( pkt ) ;
2013-08-09 01:40:03 +03:00
return AVERROR ( ENOMEM ) ;
}
memcpy ( buf , id , id_len ) ;
}
if ( settings_len > 0 ) {
buf = av_packet_new_side_data ( pkt ,
AV_PKT_DATA_WEBVTT_SETTINGS ,
settings_len ) ;
2014-08-15 21:33:21 +03:00
if ( ! buf ) {
2018-02-20 15:25:54 +02:00
av_packet_unref ( pkt ) ;
2013-08-09 01:40:03 +03:00
return AVERROR ( ENOMEM ) ;
}
memcpy ( buf , settings , settings_len ) ;
}
// Do we need this for subtitles?
// pkt->flags = AV_PKT_FLAG_KEY;
pkt - > stream_index = st - > index ;
pkt - > pts = timecode ;
// Do we need this for subtitles?
// pkt->dts = timecode;
pkt - > duration = duration ;
pkt - > pos = pos ;
2020-08-17 17:03:50 +02:00
err = avpriv_packet_list_put ( & matroska - > queue , & matroska - > queue_end , pkt , NULL , 0 ) ;
2018-03-26 20:02:37 +02:00
if ( err < 0 ) {
2018-04-04 18:53:12 +02:00
av_packet_unref ( pkt ) ;
2018-03-26 20:02:37 +02:00
return AVERROR ( ENOMEM ) ;
}
2013-08-09 01:40:03 +03:00
return 0 ;
}
2012-09-17 02:28:13 +03:00
static int matroska_parse_frame ( MatroskaDemuxContext * matroska ,
2014-03-01 19:28:15 +03:00
MatroskaTrack * track , AVStream * st ,
avformat/matroskadec: Free right buffer on error
Since commit 979b5b89594c7628bd846c63198cb64ef9d81d16, reverting the
Matroska ContentCompression is no longer done inside
matroska_parse_frame() (the function that creates AVPackets out of the
parsed data (unless we are dealing with certain codecs that need special
handling)), but instead in matroska_parse_block(). As a consequence,
the data that matroska_parse_frame() receives is no longer always owned
by an AVBuffer; it is owned by an AVBuffer iff no ContentCompression needed
to be reversed; otherwise the data is independently allocated and needs
to be freed on error.
Whether the data is owned by an AVBuffer or not is indicated by a variable
buf of type AVBufferRef *: If it is NULL, the data is independently
allocated, if not it is owned by the underlying AVBuffer (and is used to
avoid copying the data when creating the AVPackets).
Because the allocation of the buffer holding the uncompressed data happens
outside of matroska_parse_frame() (if a ContentCompression needs to be
reversed), the data is passed as uint8_t ** in order to not leave any
dangling pointers behind in matroska_parse_block() should the data need to
be freed: In case of errors, said uint8_t ** would be av_freep()'ed in
case buf indicated the data to be independently allocated.
Yet there is a problem with this: Some codecs (namely WavPack and
ProRes) need special handling: Their packets are only stored in
Matroska in a stripped form to save space and the demuxer reconstructs
full packets. This involved allocating a new, enlarged buffer. And if
an error happens when trying to wrap this new buffer into an AVBuffer,
this buffer needs to be freed; yet instead the given uint8_t ** (holding
the uncompressed, yet still stripped form of the data) would be freed
(av_freep()'ed) which certainly leads to a memleak of the new buffer;
even worse, in case the track does not use ContentCompression the given
uint8_t ** must not be freed as the actual data is owned by an AVBuffer
and the data given to matroska_parse_frame() is not the start of the
actual allocated buffer at all.
Both of these issues are fixed by always freeing the current data in
case it is independently allocated. Furthermore, while it would be
possible to track whether the pointer from matroska_parse_block() needs
to be reset or not, there is no gain in doing so, as the pointer is not
used at all afterwards and the sematics are clear: If the data passed
to matroska_parse_frame() is independently allocated, then ownership
of the data passes to matroska_parse_frame(). So don't pass the data
via uint8_t **.
Fixes Coverity ID 1462661 (the issue as described by Coverity is btw
a false positive: It thinks that this error can be triggered by ProRes
with a size of zero after reconstructing the original packets, but the
reconstructed packets can't have a size of zero).
Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@gmail.com>
2020-05-03 08:35:25 +02:00
AVBufferRef * buf , uint8_t * data , int pkt_size ,
2012-09-20 21:37:26 +03:00
uint64_t timecode , uint64_t lace_duration ,
2013-02-14 00:51:48 +03:00
int64_t pos , int is_keyframe ,
2013-09-10 21:12:21 +03:00
uint8_t * additional , uint64_t additional_id , int additional_size ,
2013-11-14 14:58:28 +03:00
int64_t discard_padding )
2012-09-17 02:28:13 +03:00
{
avformat/matroskadec: Free right buffer on error
Since commit 979b5b89594c7628bd846c63198cb64ef9d81d16, reverting the
Matroska ContentCompression is no longer done inside
matroska_parse_frame() (the function that creates AVPackets out of the
parsed data (unless we are dealing with certain codecs that need special
handling)), but instead in matroska_parse_block(). As a consequence,
the data that matroska_parse_frame() receives is no longer always owned
by an AVBuffer; it is owned by an AVBuffer iff no ContentCompression needed
to be reversed; otherwise the data is independently allocated and needs
to be freed on error.
Whether the data is owned by an AVBuffer or not is indicated by a variable
buf of type AVBufferRef *: If it is NULL, the data is independently
allocated, if not it is owned by the underlying AVBuffer (and is used to
avoid copying the data when creating the AVPackets).
Because the allocation of the buffer holding the uncompressed data happens
outside of matroska_parse_frame() (if a ContentCompression needs to be
reversed), the data is passed as uint8_t ** in order to not leave any
dangling pointers behind in matroska_parse_block() should the data need to
be freed: In case of errors, said uint8_t ** would be av_freep()'ed in
case buf indicated the data to be independently allocated.
Yet there is a problem with this: Some codecs (namely WavPack and
ProRes) need special handling: Their packets are only stored in
Matroska in a stripped form to save space and the demuxer reconstructs
full packets. This involved allocating a new, enlarged buffer. And if
an error happens when trying to wrap this new buffer into an AVBuffer,
this buffer needs to be freed; yet instead the given uint8_t ** (holding
the uncompressed, yet still stripped form of the data) would be freed
(av_freep()'ed) which certainly leads to a memleak of the new buffer;
even worse, in case the track does not use ContentCompression the given
uint8_t ** must not be freed as the actual data is owned by an AVBuffer
and the data given to matroska_parse_frame() is not the start of the
actual allocated buffer at all.
Both of these issues are fixed by always freeing the current data in
case it is independently allocated. Furthermore, while it would be
possible to track whether the pointer from matroska_parse_block() needs
to be reset or not, there is no gain in doing so, as the pointer is not
used at all afterwards and the sematics are clear: If the data passed
to matroska_parse_frame() is independently allocated, then ownership
of the data passes to matroska_parse_frame(). So don't pass the data
via uint8_t **.
Fixes Coverity ID 1462661 (the issue as described by Coverity is btw
a false positive: It thinks that this error can be triggered by ProRes
with a size of zero after reconstructing the original packets, but the
reconstructed packets can't have a size of zero).
Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@gmail.com>
2020-05-03 08:35:25 +02:00
uint8_t * pkt_data = data ;
2020-03-25 07:00:53 +02:00
int res = 0 ;
2021-01-29 15:42:48 +02:00
AVPacket * pkt = matroska - > pkt ;
2012-09-17 02:28:13 +03:00
lavf: replace AVStream.codec with AVStream.codecpar
Currently, AVStream contains an embedded AVCodecContext instance, which
is used by demuxers to export stream parameters to the caller and by
muxers to receive stream parameters from the caller. It is also used
internally as the codec context that is passed to parsers.
In addition, it is also widely used by the callers as the decoding (when
demuxer) or encoding (when muxing) context, though this has been
officially discouraged since Libav 11.
There are multiple important problems with this approach:
- the fields in AVCodecContext are in general one of
* stream parameters
* codec options
* codec state
However, it's not clear which ones are which. It is consequently
unclear which fields are a demuxer allowed to set or a muxer allowed to
read. This leads to erratic behaviour depending on whether decoding or
encoding is being performed or not (and whether it uses the AVStream
embedded codec context).
- various synchronization issues arising from the fact that the same
context is used by several different APIs (muxers/demuxers,
parsers, bitstream filters and encoders/decoders) simultaneously, with
there being no clear rules for who can modify what and the different
processes being typically delayed with respect to each other.
- avformat_find_stream_info() making it necessary to support opening
and closing a single codec context multiple times, thus
complicating the semantics of freeing various allocated objects in the
codec context.
Those problems are resolved by replacing the AVStream embedded codec
context with a newly added AVCodecParameters instance, which stores only
the stream parameters exported by the demuxers or read by the muxers.
2014-06-18 21:42:52 +03:00
if ( st - > codecpar - > codec_id = = AV_CODEC_ID_WAVPACK ) {
2019-12-06 10:53:34 +02:00
res = matroska_parse_wavpack ( track , & pkt_data , & pkt_size ) ;
2013-05-27 10:44:27 +03:00
if ( res < 0 ) {
2014-03-01 19:28:15 +03:00
av_log ( matroska - > ctx , AV_LOG_ERROR ,
" Error parsing a wavpack block. \n " ) ;
2013-05-27 10:44:27 +03:00
goto fail ;
}
2019-12-06 10:53:34 +02:00
if ( ! buf )
avformat/matroskadec: Free right buffer on error
Since commit 979b5b89594c7628bd846c63198cb64ef9d81d16, reverting the
Matroska ContentCompression is no longer done inside
matroska_parse_frame() (the function that creates AVPackets out of the
parsed data (unless we are dealing with certain codecs that need special
handling)), but instead in matroska_parse_block(). As a consequence,
the data that matroska_parse_frame() receives is no longer always owned
by an AVBuffer; it is owned by an AVBuffer iff no ContentCompression needed
to be reversed; otherwise the data is independently allocated and needs
to be freed on error.
Whether the data is owned by an AVBuffer or not is indicated by a variable
buf of type AVBufferRef *: If it is NULL, the data is independently
allocated, if not it is owned by the underlying AVBuffer (and is used to
avoid copying the data when creating the AVPackets).
Because the allocation of the buffer holding the uncompressed data happens
outside of matroska_parse_frame() (if a ContentCompression needs to be
reversed), the data is passed as uint8_t ** in order to not leave any
dangling pointers behind in matroska_parse_block() should the data need to
be freed: In case of errors, said uint8_t ** would be av_freep()'ed in
case buf indicated the data to be independently allocated.
Yet there is a problem with this: Some codecs (namely WavPack and
ProRes) need special handling: Their packets are only stored in
Matroska in a stripped form to save space and the demuxer reconstructs
full packets. This involved allocating a new, enlarged buffer. And if
an error happens when trying to wrap this new buffer into an AVBuffer,
this buffer needs to be freed; yet instead the given uint8_t ** (holding
the uncompressed, yet still stripped form of the data) would be freed
(av_freep()'ed) which certainly leads to a memleak of the new buffer;
even worse, in case the track does not use ContentCompression the given
uint8_t ** must not be freed as the actual data is owned by an AVBuffer
and the data given to matroska_parse_frame() is not the start of the
actual allocated buffer at all.
Both of these issues are fixed by always freeing the current data in
case it is independently allocated. Furthermore, while it would be
possible to track whether the pointer from matroska_parse_block() needs
to be reset or not, there is no gain in doing so, as the pointer is not
used at all afterwards and the sematics are clear: If the data passed
to matroska_parse_frame() is independently allocated, then ownership
of the data passes to matroska_parse_frame(). So don't pass the data
via uint8_t **.
Fixes Coverity ID 1462661 (the issue as described by Coverity is btw
a false positive: It thinks that this error can be triggered by ProRes
with a size of zero after reconstructing the original packets, but the
reconstructed packets can't have a size of zero).
Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@gmail.com>
2020-05-03 08:35:25 +02:00
av_freep ( & data ) ;
2019-12-06 10:53:34 +02:00
buf = NULL ;
2013-05-27 10:44:27 +03:00
}
2019-12-07 01:16:19 +02:00
if ( st - > codecpar - > codec_id = = AV_CODEC_ID_PRORES & &
AV_RB32 ( pkt_data + 4 ) ! = MKBETAG ( ' i ' , ' c ' , ' p ' , ' f ' ) ) {
2019-12-06 10:53:34 +02:00
res = matroska_parse_prores ( track , & pkt_data , & pkt_size ) ;
2018-04-04 21:07:36 +02:00
if ( res < 0 ) {
av_log ( matroska - > ctx , AV_LOG_ERROR ,
" Error parsing a prores block. \n " ) ;
goto fail ;
}
2019-12-06 10:53:34 +02:00
if ( ! buf )
avformat/matroskadec: Free right buffer on error
Since commit 979b5b89594c7628bd846c63198cb64ef9d81d16, reverting the
Matroska ContentCompression is no longer done inside
matroska_parse_frame() (the function that creates AVPackets out of the
parsed data (unless we are dealing with certain codecs that need special
handling)), but instead in matroska_parse_block(). As a consequence,
the data that matroska_parse_frame() receives is no longer always owned
by an AVBuffer; it is owned by an AVBuffer iff no ContentCompression needed
to be reversed; otherwise the data is independently allocated and needs
to be freed on error.
Whether the data is owned by an AVBuffer or not is indicated by a variable
buf of type AVBufferRef *: If it is NULL, the data is independently
allocated, if not it is owned by the underlying AVBuffer (and is used to
avoid copying the data when creating the AVPackets).
Because the allocation of the buffer holding the uncompressed data happens
outside of matroska_parse_frame() (if a ContentCompression needs to be
reversed), the data is passed as uint8_t ** in order to not leave any
dangling pointers behind in matroska_parse_block() should the data need to
be freed: In case of errors, said uint8_t ** would be av_freep()'ed in
case buf indicated the data to be independently allocated.
Yet there is a problem with this: Some codecs (namely WavPack and
ProRes) need special handling: Their packets are only stored in
Matroska in a stripped form to save space and the demuxer reconstructs
full packets. This involved allocating a new, enlarged buffer. And if
an error happens when trying to wrap this new buffer into an AVBuffer,
this buffer needs to be freed; yet instead the given uint8_t ** (holding
the uncompressed, yet still stripped form of the data) would be freed
(av_freep()'ed) which certainly leads to a memleak of the new buffer;
even worse, in case the track does not use ContentCompression the given
uint8_t ** must not be freed as the actual data is owned by an AVBuffer
and the data given to matroska_parse_frame() is not the start of the
actual allocated buffer at all.
Both of these issues are fixed by always freeing the current data in
case it is independently allocated. Furthermore, while it would be
possible to track whether the pointer from matroska_parse_block() needs
to be reset or not, there is no gain in doing so, as the pointer is not
used at all afterwards and the sematics are clear: If the data passed
to matroska_parse_frame() is independently allocated, then ownership
of the data passes to matroska_parse_frame(). So don't pass the data
via uint8_t **.
Fixes Coverity ID 1462661 (the issue as described by Coverity is btw
a false positive: It thinks that this error can be triggered by ProRes
with a size of zero after reconstructing the original packets, but the
reconstructed packets can't have a size of zero).
Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@gmail.com>
2020-05-03 08:35:25 +02:00
av_freep ( & data ) ;
2019-12-06 10:53:34 +02:00
buf = NULL ;
2018-04-04 21:07:36 +02:00
}
2012-09-17 02:28:13 +03:00
2020-03-25 07:00:53 +02:00
if ( ! pkt_size & & ! additional_size )
goto no_output ;
2019-12-06 10:53:34 +02:00
if ( ! buf )
2018-04-04 23:12:53 +02:00
pkt - > buf = av_buffer_create ( pkt_data , pkt_size + AV_INPUT_BUFFER_PADDING_SIZE ,
NULL , NULL , 0 ) ;
else
pkt - > buf = av_buffer_ref ( buf ) ;
if ( ! pkt - > buf ) {
2013-06-03 17:07:06 +03:00
res = AVERROR ( ENOMEM ) ;
goto fail ;
2012-09-17 02:28:13 +03:00
}
2018-04-04 23:12:53 +02:00
pkt - > data = pkt_data ;
pkt - > size = pkt_size ;
2014-03-01 19:28:15 +03:00
pkt - > flags = is_keyframe ;
2012-09-17 02:28:13 +03:00
pkt - > stream_index = st - > index ;
2013-02-14 00:51:48 +03:00
if ( additional_size > 0 ) {
uint8_t * side_data = av_packet_new_side_data ( pkt ,
AV_PKT_DATA_MATROSKA_BLOCKADDITIONAL ,
2013-02-14 02:55:25 +03:00
additional_size + 8 ) ;
2014-08-15 21:33:21 +03:00
if ( ! side_data ) {
2015-10-27 15:35:30 +02:00
av_packet_unref ( pkt ) ;
2013-02-14 00:51:48 +03:00
return AVERROR ( ENOMEM ) ;
}
2013-02-14 02:55:25 +03:00
AV_WB64 ( side_data , additional_id ) ;
2013-02-14 00:51:48 +03:00
memcpy ( side_data + 8 , additional , additional_size ) ;
}
2013-09-10 21:12:21 +03:00
if ( discard_padding ) {
uint8_t * side_data = av_packet_new_side_data ( pkt ,
AV_PKT_DATA_SKIP_SAMPLES ,
10 ) ;
2014-08-15 21:33:21 +03:00
if ( ! side_data ) {
2015-10-27 15:35:30 +02:00
av_packet_unref ( pkt ) ;
2013-09-10 21:12:21 +03:00
return AVERROR ( ENOMEM ) ;
}
2016-11-05 22:46:52 +02:00
discard_padding = av_rescale_q ( discard_padding ,
2013-09-10 21:12:21 +03:00
( AVRational ) { 1 , 1000000000 } ,
2016-11-05 22:46:52 +02:00
( AVRational ) { 1 , st - > codecpar - > sample_rate } ) ;
if ( discard_padding > 0 ) {
AV_WL32 ( side_data + 4 , discard_padding ) ;
} else {
AV_WL32 ( side_data , - discard_padding ) ;
}
2013-09-10 21:12:21 +03:00
}
2012-09-17 02:28:13 +03:00
if ( track - > ms_compat )
pkt - > dts = timecode ;
else
pkt - > pts = timecode ;
pkt - > pos = pos ;
2015-09-29 15:14:59 +02:00
pkt - > duration = lace_duration ;
2020-08-17 17:03:50 +02:00
res = avpriv_packet_list_put ( & matroska - > queue , & matroska - > queue_end , pkt , NULL , 0 ) ;
2018-03-26 20:02:37 +02:00
if ( res < 0 ) {
2018-04-04 15:54:14 +02:00
av_packet_unref ( pkt ) ;
2018-03-26 20:02:37 +02:00
return AVERROR ( ENOMEM ) ;
}
2012-09-17 02:28:13 +03:00
return 0 ;
2014-03-01 19:28:15 +03:00
2020-03-25 07:00:53 +02:00
no_output :
2013-05-27 10:44:27 +03:00
fail :
2019-12-06 10:53:34 +02:00
if ( ! buf )
avformat/matroskadec: Free right buffer on error
Since commit 979b5b89594c7628bd846c63198cb64ef9d81d16, reverting the
Matroska ContentCompression is no longer done inside
matroska_parse_frame() (the function that creates AVPackets out of the
parsed data (unless we are dealing with certain codecs that need special
handling)), but instead in matroska_parse_block(). As a consequence,
the data that matroska_parse_frame() receives is no longer always owned
by an AVBuffer; it is owned by an AVBuffer iff no ContentCompression needed
to be reversed; otherwise the data is independently allocated and needs
to be freed on error.
Whether the data is owned by an AVBuffer or not is indicated by a variable
buf of type AVBufferRef *: If it is NULL, the data is independently
allocated, if not it is owned by the underlying AVBuffer (and is used to
avoid copying the data when creating the AVPackets).
Because the allocation of the buffer holding the uncompressed data happens
outside of matroska_parse_frame() (if a ContentCompression needs to be
reversed), the data is passed as uint8_t ** in order to not leave any
dangling pointers behind in matroska_parse_block() should the data need to
be freed: In case of errors, said uint8_t ** would be av_freep()'ed in
case buf indicated the data to be independently allocated.
Yet there is a problem with this: Some codecs (namely WavPack and
ProRes) need special handling: Their packets are only stored in
Matroska in a stripped form to save space and the demuxer reconstructs
full packets. This involved allocating a new, enlarged buffer. And if
an error happens when trying to wrap this new buffer into an AVBuffer,
this buffer needs to be freed; yet instead the given uint8_t ** (holding
the uncompressed, yet still stripped form of the data) would be freed
(av_freep()'ed) which certainly leads to a memleak of the new buffer;
even worse, in case the track does not use ContentCompression the given
uint8_t ** must not be freed as the actual data is owned by an AVBuffer
and the data given to matroska_parse_frame() is not the start of the
actual allocated buffer at all.
Both of these issues are fixed by always freeing the current data in
case it is independently allocated. Furthermore, while it would be
possible to track whether the pointer from matroska_parse_block() needs
to be reset or not, there is no gain in doing so, as the pointer is not
used at all afterwards and the sematics are clear: If the data passed
to matroska_parse_frame() is independently allocated, then ownership
of the data passes to matroska_parse_frame(). So don't pass the data
via uint8_t **.
Fixes Coverity ID 1462661 (the issue as described by Coverity is btw
a false positive: It thinks that this error can be triggered by ProRes
with a size of zero after reconstructing the original packets, but the
reconstructed packets can't have a size of zero).
Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@gmail.com>
2020-05-03 08:35:25 +02:00
av_free ( pkt_data ) ;
2013-05-27 10:44:27 +03:00
return res ;
2012-09-17 02:28:13 +03:00
}
2018-04-04 23:12:53 +02:00
static int matroska_parse_block ( MatroskaDemuxContext * matroska , AVBufferRef * buf , uint8_t * data ,
2008-08-05 03:42:05 +03:00
int size , int64_t pos , uint64_t cluster_time ,
2012-09-17 03:48:02 +03:00
uint64_t block_duration , int is_keyframe ,
2013-02-14 00:51:48 +03:00
uint8_t * additional , uint64_t additional_id , int additional_size ,
2013-11-14 14:58:28 +03:00
int64_t cluster_pos , int64_t discard_padding )
2007-06-05 01:10:54 +03:00
{
2008-09-09 14:54:35 +03:00
uint64_t timecode = AV_NOPTS_VALUE ;
2008-08-05 03:40:12 +03:00
MatroskaTrack * track ;
2019-12-03 19:09:10 +02:00
AVIOContext pb ;
2008-09-07 02:44:29 +03:00
int res = 0 ;
2007-06-05 01:10:54 +03:00
AVStream * st ;
int16_t block_time ;
2019-12-03 19:09:06 +02:00
uint32_t lace_size [ 256 ] ;
2007-06-05 01:10:54 +03:00
int n , flags , laces = 0 ;
uint64_t num ;
2020-11-15 11:00:35 +02:00
int trust_default_duration ;
2007-06-05 01:10:54 +03:00
2019-12-03 19:09:10 +02:00
ffio_init_context ( & pb , data , size , 0 , NULL , NULL , NULL , NULL ) ;
if ( ( n = ebml_read_num ( matroska , & pb , 8 , & num , 1 ) ) < 0 )
2012-04-30 03:39:31 +03:00
return n ;
2007-06-05 01:10:54 +03:00
data + = n ;
size - = n ;
track = matroska_find_track_by_num ( matroska , num ) ;
2020-03-25 07:52:09 +02:00
if ( ! track | | size < 3 )
2012-02-14 01:06:19 +03:00
return AVERROR_INVALIDDATA ;
2020-03-25 07:52:09 +02:00
if ( ! ( st = track - > stream ) ) {
av_log ( matroska - > ctx , AV_LOG_VERBOSE ,
" No stream associated to TrackNumber % " PRIu64 " . "
" Ignoring Block with this TrackNumber. \n " , num ) ;
return 0 ;
}
2008-08-05 03:41:10 +03:00
if ( st - > discard > = AVDISCARD_ALL )
2007-06-05 01:10:54 +03:00
return res ;
2019-09-04 00:50:11 +02:00
if ( block_duration > INT64_MAX )
block_duration = INT64_MAX ;
2007-06-05 01:10:54 +03:00
2013-05-20 00:38:01 +03:00
block_time = sign_extend ( AV_RB16 ( data ) , 16 ) ;
2014-03-01 19:28:15 +03:00
data + = 2 ;
flags = * data + + ;
size - = 3 ;
2007-06-05 01:10:54 +03:00
if ( is_keyframe = = - 1 )
2010-03-31 15:29:58 +03:00
is_keyframe = flags & 0x80 ? AV_PKT_FLAG_KEY : 0 ;
2007-06-05 01:10:54 +03:00
2014-03-01 19:28:15 +03:00
if ( cluster_time ! = ( uint64_t ) - 1 & &
( block_time > = 0 | | cluster_time > = - block_time ) ) {
2020-11-15 10:59:48 +02:00
uint64_t timecode_cluster_in_track_tb = ( double ) cluster_time / track - > time_scale ;
timecode = timecode_cluster_in_track_tb + block_time - track - > codec_delay_in_track_tb ;
2014-03-01 19:28:15 +03:00
if ( track - > type = = MATROSKA_TRACK_TYPE_SUBTITLE & &
timecode < track - > end_timecode )
2008-09-09 15:07:10 +03:00
is_keyframe = 0 ; /* overlapping subtitles are not key frame */
2016-12-26 04:16:01 +02:00
if ( is_keyframe ) {
ff_reduce_index ( matroska - > ctx , st - > index ) ;
2014-03-01 19:28:15 +03:00
av_add_index_entry ( st , cluster_pos , timecode , 0 , 0 ,
AVINDEX_KEYFRAME ) ;
2016-12-26 04:16:01 +02:00
}
2008-09-09 14:54:35 +03:00
}
2014-03-01 19:28:15 +03:00
if ( matroska - > skip_to_keyframe & &
track - > type ! = MATROSKA_TRACK_TYPE_SUBTITLE ) {
2016-07-21 21:01:45 +02:00
// Compare signed timecodes. Timecode may be negative due to codec delay
// offset. We don't support timestamps greater than int64_t anyway - see
// AVPacket's pts.
2016-07-28 03:33:30 +02:00
if ( ( int64_t ) timecode < ( int64_t ) matroska - > skip_to_timecode )
2007-06-05 01:10:54 +03:00
return res ;
2013-09-11 09:00:31 +03:00
if ( is_keyframe )
matroska - > skip_to_keyframe = 0 ;
2020-10-09 09:22:36 +02:00
else if ( ! st - > internal - > skip_to_keyframe ) {
2012-04-14 01:17:30 +03:00
av_log ( matroska - > ctx , AV_LOG_ERROR , " File is broken, keyframes not correctly marked! \n " ) ;
matroska - > skip_to_keyframe = 0 ;
}
2007-06-05 01:10:54 +03:00
}
2019-12-03 19:09:08 +02:00
res = matroska_parse_laces ( matroska , & data , size , ( flags & 0x06 ) > > 1 ,
2019-12-03 19:09:10 +02:00
& pb , lace_size , & laces ) ;
2019-12-03 19:09:09 +02:00
if ( res < 0 ) {
av_log ( matroska - > ctx , AV_LOG_ERROR , " Error parsing frame sizes. \n " ) ;
2019-12-03 19:09:06 +02:00
return res ;
2019-12-03 19:09:09 +02:00
}
2007-06-05 01:10:54 +03:00
2020-11-15 11:00:35 +02:00
trust_default_duration = track - > default_duration ! = 0 ;
if ( track - > audio . samplerate = = 8000 & & trust_default_duration ) {
2013-07-15 18:13:45 +03:00
// If this is needed for more codecs, then add them here
2016-04-10 21:58:15 +02:00
if ( st - > codecpar - > codec_id = = AV_CODEC_ID_AC3 ) {
if ( track - > audio . samplerate ! = st - > codecpar - > sample_rate | | ! st - > codecpar - > frame_size )
2013-07-15 18:13:45 +03:00
trust_default_duration = 0 ;
}
}
if ( ! block_duration & & trust_default_duration )
2012-09-20 22:46:35 +03:00
block_duration = track - > default_duration * laces / matroska - > time_scale ;
2007-06-05 01:10:54 +03:00
2012-09-20 21:37:26 +03:00
if ( cluster_time ! = ( uint64_t ) - 1 & & ( block_time > = 0 | | cluster_time > = - block_time ) )
2012-09-17 03:48:02 +03:00
track - > end_timecode =
FFMAX ( track - > end_timecode , timecode + block_duration ) ;
2012-09-14 19:51:49 +03:00
2012-09-17 02:28:13 +03:00
for ( n = 0 ; n < laces ; n + + ) {
2012-09-20 22:46:35 +03:00
int64_t lace_duration = block_duration * ( n + 1 ) / laces - block_duration * n / laces ;
2019-12-06 10:53:34 +02:00
uint8_t * out_data = data ;
int out_size = lace_size [ n ] ;
if ( track - > needs_decoding ) {
res = matroska_decode_buffer ( & out_data , & out_size , track ) ;
if ( res < 0 )
return res ;
/* Given that we are here means that out_data is no longer
* owned by buf , so set it to NULL . This depends upon
* zero - length header removal compression being ignored . */
av_assert1 ( out_data ! = data ) ;
buf = NULL ;
}
2012-09-15 01:59:05 +03:00
2019-12-02 12:04:55 +02:00
if ( track - > audio . buf ) {
2019-12-06 10:53:34 +02:00
res = matroska_parse_rm_audio ( matroska , track , st ,
out_data , out_size ,
2012-09-21 00:43:20 +03:00
timecode , pos ) ;
2019-12-06 10:53:34 +02:00
if ( ! buf )
av_free ( out_data ) ;
2012-09-17 02:28:13 +03:00
if ( res )
2019-12-03 19:09:06 +02:00
return res ;
2016-04-10 21:58:15 +02:00
} else if ( st - > codecpar - > codec_id = = AV_CODEC_ID_WEBVTT ) {
2013-08-09 01:40:03 +03:00
res = matroska_parse_webvtt ( matroska , track , st ,
2019-12-06 10:53:34 +02:00
out_data , out_size ,
2013-08-09 01:40:03 +03:00
timecode , lace_duration ,
pos ) ;
2019-12-06 10:53:34 +02:00
if ( ! buf )
av_free ( out_data ) ;
2013-08-09 01:40:03 +03:00
if ( res )
2019-12-03 19:09:06 +02:00
return res ;
2012-09-17 02:28:13 +03:00
} else {
avformat/matroskadec: Free right buffer on error
Since commit 979b5b89594c7628bd846c63198cb64ef9d81d16, reverting the
Matroska ContentCompression is no longer done inside
matroska_parse_frame() (the function that creates AVPackets out of the
parsed data (unless we are dealing with certain codecs that need special
handling)), but instead in matroska_parse_block(). As a consequence,
the data that matroska_parse_frame() receives is no longer always owned
by an AVBuffer; it is owned by an AVBuffer iff no ContentCompression needed
to be reversed; otherwise the data is independently allocated and needs
to be freed on error.
Whether the data is owned by an AVBuffer or not is indicated by a variable
buf of type AVBufferRef *: If it is NULL, the data is independently
allocated, if not it is owned by the underlying AVBuffer (and is used to
avoid copying the data when creating the AVPackets).
Because the allocation of the buffer holding the uncompressed data happens
outside of matroska_parse_frame() (if a ContentCompression needs to be
reversed), the data is passed as uint8_t ** in order to not leave any
dangling pointers behind in matroska_parse_block() should the data need to
be freed: In case of errors, said uint8_t ** would be av_freep()'ed in
case buf indicated the data to be independently allocated.
Yet there is a problem with this: Some codecs (namely WavPack and
ProRes) need special handling: Their packets are only stored in
Matroska in a stripped form to save space and the demuxer reconstructs
full packets. This involved allocating a new, enlarged buffer. And if
an error happens when trying to wrap this new buffer into an AVBuffer,
this buffer needs to be freed; yet instead the given uint8_t ** (holding
the uncompressed, yet still stripped form of the data) would be freed
(av_freep()'ed) which certainly leads to a memleak of the new buffer;
even worse, in case the track does not use ContentCompression the given
uint8_t ** must not be freed as the actual data is owned by an AVBuffer
and the data given to matroska_parse_frame() is not the start of the
actual allocated buffer at all.
Both of these issues are fixed by always freeing the current data in
case it is independently allocated. Furthermore, while it would be
possible to track whether the pointer from matroska_parse_block() needs
to be reset or not, there is no gain in doing so, as the pointer is not
used at all afterwards and the sematics are clear: If the data passed
to matroska_parse_frame() is independently allocated, then ownership
of the data passes to matroska_parse_frame(). So don't pass the data
via uint8_t **.
Fixes Coverity ID 1462661 (the issue as described by Coverity is btw
a false positive: It thinks that this error can be triggered by ProRes
with a size of zero after reconstructing the original packets, but the
reconstructed packets can't have a size of zero).
Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@gmail.com>
2020-05-03 08:35:25 +02:00
res = matroska_parse_frame ( matroska , track , st , buf , out_data ,
2019-12-06 10:53:34 +02:00
out_size , timecode , lace_duration ,
pos , ! n ? is_keyframe : 0 ,
2014-03-08 01:46:37 +03:00
additional , additional_id , additional_size ,
discard_padding ) ;
2012-09-17 02:28:13 +03:00
if ( res )
2019-12-03 19:09:06 +02:00
return res ;
2007-06-05 01:10:54 +03:00
}
2012-09-17 02:28:13 +03:00
if ( timecode ! = AV_NOPTS_VALUE )
2012-09-20 21:37:26 +03:00
timecode = lace_duration ? timecode + lace_duration : AV_NOPTS_VALUE ;
2012-09-17 02:28:13 +03:00
data + = lace_size [ n ] ;
2007-06-05 01:10:54 +03:00
}
2019-12-03 19:09:06 +02:00
return 0 ;
2007-06-05 01:10:54 +03:00
}
2019-05-17 00:29:58 +02:00
static int matroska_parse_cluster ( MatroskaDemuxContext * matroska )
2012-04-19 21:12:24 +03:00
{
2019-05-17 00:29:59 +02:00
MatroskaCluster * cluster = & matroska - > current_cluster ;
MatroskaBlock * block = & cluster - > block ;
int res ;
2019-05-17 00:30:08 +02:00
av_assert0 ( matroska - > num_levels < = 2 ) ;
if ( matroska - > num_levels = = 1 ) {
2019-05-17 00:30:18 +02:00
res = ebml_parse ( matroska , matroska_segment , NULL ) ;
2019-05-17 00:30:08 +02:00
if ( res = = 1 ) {
/* Found a cluster: subtract the size of the ID already read. */
cluster - > pos = avio_tell ( matroska - > ctx - > pb ) - 4 ;
res = ebml_parse ( matroska , matroska_cluster_enter , cluster ) ;
if ( res < 0 )
return res ;
}
2012-04-19 21:12:24 +03:00
}
2019-05-17 00:30:08 +02:00
if ( matroska - > num_levels = = 2 ) {
/* We are inside a cluster. */
res = ebml_parse ( matroska , matroska_cluster_parsing , cluster ) ;
if ( res > = 0 & & block - > bin . size > 0 ) {
2021-02-16 17:18:02 +02:00
int is_keyframe = block - > non_simple ? block - > reference . count = = 0 : - 1 ;
2019-05-17 00:29:59 +02:00
uint8_t * additional = block - > additional . size > 0 ?
block - > additional . data : NULL ;
2012-04-19 21:12:24 +03:00
2019-05-17 00:30:09 +02:00
res = matroska_parse_block ( matroska , block - > bin . buf , block - > bin . data ,
2019-05-17 00:29:59 +02:00
block - > bin . size , block - > bin . pos ,
2019-05-17 00:30:08 +02:00
cluster - > timecode , block - > duration ,
is_keyframe , additional , block - > additional_id ,
block - > additional . size , cluster - > pos ,
2019-05-17 00:29:59 +02:00
block - > discard_padding ) ;
2019-05-17 00:30:08 +02:00
}
2012-04-19 21:12:24 +03:00
2019-05-17 00:30:13 +02:00
ebml_free ( matroska_blockgroup , block ) ;
memset ( block , 0 , sizeof ( * block ) ) ;
2019-05-17 00:30:08 +02:00
} else if ( ! matroska - > num_levels ) {
2019-05-17 00:30:11 +02:00
if ( ! avio_feof ( matroska - > ctx - > pb ) ) {
avio_r8 ( matroska - > ctx - > pb ) ;
if ( ! avio_feof ( matroska - > ctx - > pb ) ) {
av_log ( matroska - > ctx , AV_LOG_WARNING , " File extends beyond "
" end of segment. \n " ) ;
return AVERROR_INVALIDDATA ;
}
}
2019-05-17 00:30:08 +02:00
matroska - > done = 1 ;
return AVERROR_EOF ;
}
2012-04-19 21:12:24 +03:00
return res ;
}
2008-08-05 03:42:05 +03:00
static int matroska_read_packet ( AVFormatContext * s , AVPacket * pkt )
2007-06-05 01:10:54 +03:00
{
MatroskaDemuxContext * matroska = s - > priv_data ;
2016-09-27 21:00:29 +02:00
int ret = 0 ;
2007-06-05 01:10:54 +03:00
avformat/matroskadec: Introduce a "last known good" position
Currently, resyncing during reading packets works as follows:
The current position is recorded, then a call to matroska_parse_cluster
is made and if said call fails, the demuxer tries to resync from the
earlier position. If the call doesn't fail, but also doesn't deliver a
packet, then this is looped.
There are two problems with this approach:
1. The Matroska file format aims to be forward-compatible; to achieve
this, a demuxer should simply ignore and skip elements it doesn't
know about. But it is not possible to reliably distinguish unknown
elements from junk. If matroska_parse_cluster encounters an unknown
element, it can therefore not simply error out; instead it returns zero
and the loop is iterated which includes an update of the position that
is intended to be used in case of errors, i.e. the element that is
skipped is not searched for level 1 element ids to resync to at all if
later calls to matroska_parse_cluster return an error.
Notice that in case that sync has been lost there can be a chain of
several unknown/possibly junk elements before an error is detected.
2. Even if a call to matroska_parse_cluster delivers a packet, this does
not mean that everything is fine. E.g. it might be that some of the
block's data is missing and that the data that was presumed to be from
the block just read actually contains the beginning of the next element.
This will only be apparent at the next call of matroska_read_packet,
which uses the (false) end of the earlier block as resync position so
that in the (not unlikely) case that the call to matroska_parse_cluster
fails, the data believed to be part of the earlier block is not searched
for a level 1 element to resync to.
To counter this, a "last known good" position is introduced. When an
element id that is known to be allowed at this position in the hierarchy
(according to the syntax currently in use for parsing) is read and some
further checks (regarding the length of the element and its containing
master element) are passed, then the beginning of the current element is
treated as a "good" position and recorded as such in the
MatroskaDemuxContext. Because of 2., only the start of the element is
treated as a "good" position, not the whole element. If an error occurs
later during parsing of clusters, the resync process starts at the last
known good position.
Given that when the header is damaged the subsequent resync never skips over
data and is therefore unaffected by both issues, the "last known good"
concept is not used there.
Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@gmail.com>
2019-05-17 00:30:05 +02:00
if ( matroska - > resync_pos = = - 1 ) {
// This can only happen if generic seeking has been used.
matroska - > resync_pos = avio_tell ( s - > pb ) ;
}
2011-07-06 20:57:11 +03:00
while ( matroska_deliver_packet ( matroska , pkt ) ) {
2007-06-05 01:10:54 +03:00
if ( matroska - > done )
2016-09-27 21:00:29 +02:00
return ( ret < 0 ) ? ret : AVERROR_EOF ;
2019-05-17 00:30:11 +02:00
if ( matroska_parse_cluster ( matroska ) < 0 & & ! matroska - > done )
avformat/matroskadec: Introduce a "last known good" position
Currently, resyncing during reading packets works as follows:
The current position is recorded, then a call to matroska_parse_cluster
is made and if said call fails, the demuxer tries to resync from the
earlier position. If the call doesn't fail, but also doesn't deliver a
packet, then this is looped.
There are two problems with this approach:
1. The Matroska file format aims to be forward-compatible; to achieve
this, a demuxer should simply ignore and skip elements it doesn't
know about. But it is not possible to reliably distinguish unknown
elements from junk. If matroska_parse_cluster encounters an unknown
element, it can therefore not simply error out; instead it returns zero
and the loop is iterated which includes an update of the position that
is intended to be used in case of errors, i.e. the element that is
skipped is not searched for level 1 element ids to resync to at all if
later calls to matroska_parse_cluster return an error.
Notice that in case that sync has been lost there can be a chain of
several unknown/possibly junk elements before an error is detected.
2. Even if a call to matroska_parse_cluster delivers a packet, this does
not mean that everything is fine. E.g. it might be that some of the
block's data is missing and that the data that was presumed to be from
the block just read actually contains the beginning of the next element.
This will only be apparent at the next call of matroska_read_packet,
which uses the (false) end of the earlier block as resync position so
that in the (not unlikely) case that the call to matroska_parse_cluster
fails, the data believed to be part of the earlier block is not searched
for a level 1 element to resync to.
To counter this, a "last known good" position is introduced. When an
element id that is known to be allowed at this position in the hierarchy
(according to the syntax currently in use for parsing) is read and some
further checks (regarding the length of the element and its containing
master element) are passed, then the beginning of the current element is
treated as a "good" position and recorded as such in the
MatroskaDemuxContext. Because of 2., only the start of the element is
treated as a "good" position, not the whole element. If an error occurs
later during parsing of clusters, the resync process starts at the last
known good position.
Given that when the header is damaged the subsequent resync never skips over
data and is therefore unaffected by both issues, the "last known good"
concept is not used there.
Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@gmail.com>
2019-05-17 00:30:05 +02:00
ret = matroska_resync ( matroska , matroska - > resync_pos ) ;
2007-06-05 01:10:54 +03:00
}
2019-02-06 16:29:38 +02:00
return 0 ;
2007-06-05 01:10:54 +03:00
}
2008-08-05 03:42:05 +03:00
static int matroska_read_seek ( AVFormatContext * s , int stream_index ,
int64_t timestamp , int flags )
2007-06-05 01:10:54 +03:00
{
MatroskaDemuxContext * matroska = s - > priv_data ;
2014-11-06 22:59:54 +02:00
MatroskaTrack * tracks = NULL ;
2007-06-05 01:10:54 +03:00
AVStream * st = s - > streams [ stream_index ] ;
2019-05-17 00:29:45 +02:00
int i , index ;
2007-06-05 01:10:54 +03:00
2011-07-09 08:48:43 +03:00
/* Parse the CUES now since we need the index data to seek. */
2012-02-12 16:09:03 +03:00
if ( matroska - > cues_parsing_deferred > 0 ) {
2011-07-09 08:48:43 +03:00
matroska - > cues_parsing_deferred = 0 ;
2012-02-12 16:09:03 +03:00
matroska_parse_cues ( matroska ) ;
2011-07-09 08:48:43 +03:00
}
2020-10-09 09:22:36 +02:00
if ( ! st - > internal - > nb_index_entries )
2012-02-12 16:09:03 +03:00
goto err ;
2020-10-09 09:22:36 +02:00
timestamp = FFMAX ( timestamp , st - > internal - > index_entries [ 0 ] . timestamp ) ;
2008-08-25 02:57:29 +03:00
2020-10-09 09:22:36 +02:00
if ( ( index = av_index_search_timestamp ( st , timestamp , flags ) ) < 0 | | index = = st - > internal - > nb_index_entries - 1 ) {
matroska_reset_status ( matroska , 0 , st - > internal - > index_entries [ st - > internal - > nb_index_entries - 1 ] . pos ) ;
while ( ( index = av_index_search_timestamp ( st , timestamp , flags ) ) < 0 | | index = = st - > internal - > nb_index_entries - 1 ) {
2008-08-27 22:58:55 +03:00
matroska_clear_queue ( matroska ) ;
if ( matroska_parse_cluster ( matroska ) < 0 )
break ;
}
2008-08-27 22:57:42 +03:00
}
2007-06-05 01:10:54 +03:00
2007-12-29 20:35:38 +02:00
matroska_clear_queue ( matroska ) ;
2020-10-09 09:22:36 +02:00
if ( index < 0 | | ( matroska - > cues_parsing_deferred < 0 & & index = = st - > internal - > nb_index_entries - 1 ) )
2012-02-12 16:09:03 +03:00
goto err ;
2007-12-29 20:35:38 +02:00
2014-11-06 22:59:54 +02:00
tracks = matroska - > tracks . elem ;
2014-03-01 19:28:15 +03:00
for ( i = 0 ; i < matroska - > tracks . nb_elem ; i + + ) {
tracks [ i ] . audio . pkt_cnt = 0 ;
2011-02-26 13:52:01 +02:00
tracks [ i ] . audio . sub_packet_cnt = 0 ;
2014-03-01 19:28:15 +03:00
tracks [ i ] . audio . buf_timecode = AV_NOPTS_VALUE ;
tracks [ i ] . end_timecode = 0 ;
2008-09-09 15:10:25 +03:00
}
2019-05-17 00:30:03 +02:00
/* We seek to a level 1 element, so set the appropriate status. */
2020-10-09 09:22:36 +02:00
matroska_reset_status ( matroska , 0 , st - > internal - > index_entries [ index ] . pos ) ;
2013-02-07 23:34:35 +03:00
if ( flags & AVSEEK_FLAG_ANY ) {
2020-10-09 09:22:36 +02:00
st - > internal - > skip_to_keyframe = 0 ;
2013-02-07 23:34:35 +03:00
matroska - > skip_to_timecode = timestamp ;
} else {
2020-10-09 09:22:36 +02:00
st - > internal - > skip_to_keyframe = 1 ;
2020-10-09 09:22:36 +02:00
matroska - > skip_to_timecode = st - > internal - > index_entries [ index ] . timestamp ;
2013-02-07 23:34:35 +03:00
}
matroska - > skip_to_keyframe = 1 ;
2014-03-01 19:28:15 +03:00
matroska - > done = 0 ;
2020-10-09 09:22:36 +02:00
ff_update_cur_dts ( s , st , st - > internal - > index_entries [ index ] . timestamp ) ;
2007-06-05 01:10:54 +03:00
return 0 ;
2012-02-12 16:09:03 +03:00
err :
// slightly hackish but allows proper fallback to
// the generic seeking code.
2019-05-17 00:30:03 +02:00
matroska_reset_status ( matroska , 0 , - 1 ) ;
avformat/matroskadec: Introduce a "last known good" position
Currently, resyncing during reading packets works as follows:
The current position is recorded, then a call to matroska_parse_cluster
is made and if said call fails, the demuxer tries to resync from the
earlier position. If the call doesn't fail, but also doesn't deliver a
packet, then this is looped.
There are two problems with this approach:
1. The Matroska file format aims to be forward-compatible; to achieve
this, a demuxer should simply ignore and skip elements it doesn't
know about. But it is not possible to reliably distinguish unknown
elements from junk. If matroska_parse_cluster encounters an unknown
element, it can therefore not simply error out; instead it returns zero
and the loop is iterated which includes an update of the position that
is intended to be used in case of errors, i.e. the element that is
skipped is not searched for level 1 element ids to resync to at all if
later calls to matroska_parse_cluster return an error.
Notice that in case that sync has been lost there can be a chain of
several unknown/possibly junk elements before an error is detected.
2. Even if a call to matroska_parse_cluster delivers a packet, this does
not mean that everything is fine. E.g. it might be that some of the
block's data is missing and that the data that was presumed to be from
the block just read actually contains the beginning of the next element.
This will only be apparent at the next call of matroska_read_packet,
which uses the (false) end of the earlier block as resync position so
that in the (not unlikely) case that the call to matroska_parse_cluster
fails, the data believed to be part of the earlier block is not searched
for a level 1 element to resync to.
To counter this, a "last known good" position is introduced. When an
element id that is known to be allowed at this position in the hierarchy
(according to the syntax currently in use for parsing) is read and some
further checks (regarding the length of the element and its containing
master element) are passed, then the beginning of the current element is
treated as a "good" position and recorded as such in the
MatroskaDemuxContext. Because of 2., only the start of the element is
treated as a "good" position, not the whole element. If an error occurs
later during parsing of clusters, the resync process starts at the last
known good position.
Given that when the header is damaged the subsequent resync never skips over
data and is therefore unaffected by both issues, the "last known good"
concept is not used there.
Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@gmail.com>
2019-05-17 00:30:05 +02:00
matroska - > resync_pos = - 1 ;
2012-02-12 16:09:03 +03:00
matroska_clear_queue ( matroska ) ;
2020-10-09 09:22:36 +02:00
st - > internal - > skip_to_keyframe =
2012-02-12 16:09:03 +03:00
matroska - > skip_to_keyframe = 0 ;
matroska - > done = 0 ;
return - 1 ;
2007-06-05 01:10:54 +03:00
}
2008-08-05 03:42:05 +03:00
static int matroska_read_close ( AVFormatContext * s )
2007-06-05 01:10:54 +03:00
{
MatroskaDemuxContext * matroska = s - > priv_data ;
2008-08-05 03:40:31 +03:00
MatroskaTrack * tracks = matroska - > tracks . elem ;
2008-08-05 03:41:13 +03:00
int n ;
2007-06-05 01:10:54 +03:00
2007-12-29 20:32:47 +02:00
matroska_clear_queue ( matroska ) ;
2007-06-05 01:10:54 +03:00
2014-03-01 19:28:15 +03:00
for ( n = 0 ; n < matroska - > tracks . nb_elem ; n + + )
2008-08-05 03:40:31 +03:00
if ( tracks [ n ] . type = = MATROSKA_TRACK_TYPE_AUDIO )
2015-01-06 13:48:38 +02:00
av_freep ( & tracks [ n ] . audio . buf ) ;
2008-08-05 03:40:58 +03:00
ebml_free ( matroska_segment , matroska ) ;
2007-06-05 01:10:54 +03:00
return 0 ;
}
2014-07-07 22:52:37 +03:00
typedef struct {
int64_t start_time_ns ;
int64_t end_time_ns ;
int64_t start_offset ;
int64_t end_offset ;
} CueDesc ;
2017-04-17 19:59:31 +02:00
/* This function searches all the Cues and returns the CueDesc corresponding to
2014-07-07 22:52:37 +03:00
* the timestamp ts . Returned CueDesc will be such that start_time_ns < = ts <
* end_time_ns . All 4 fields will be set to - 1 if ts > = file ' s duration .
*/
static CueDesc get_cue_desc ( AVFormatContext * s , int64_t ts , int64_t cues_start ) {
MatroskaDemuxContext * matroska = s - > priv_data ;
CueDesc cue_desc ;
int i ;
2020-10-09 09:22:36 +02:00
int nb_index_entries = s - > streams [ 0 ] - > internal - > nb_index_entries ;
AVIndexEntry * index_entries = s - > streams [ 0 ] - > internal - > index_entries ;
2014-07-07 22:52:37 +03:00
if ( ts > = matroska - > duration * matroska - > time_scale ) return ( CueDesc ) { - 1 , - 1 , - 1 , - 1 } ;
for ( i = 1 ; i < nb_index_entries ; i + + ) {
if ( index_entries [ i - 1 ] . timestamp * matroska - > time_scale < = ts & &
index_entries [ i ] . timestamp * matroska - > time_scale > ts ) {
break ;
}
}
- - i ;
cue_desc . start_time_ns = index_entries [ i ] . timestamp * matroska - > time_scale ;
cue_desc . start_offset = index_entries [ i ] . pos - matroska - > segment_start ;
if ( i ! = nb_index_entries - 1 ) {
cue_desc . end_time_ns = index_entries [ i + 1 ] . timestamp * matroska - > time_scale ;
cue_desc . end_offset = index_entries [ i + 1 ] . pos - matroska - > segment_start ;
} else {
cue_desc . end_time_ns = matroska - > duration * matroska - > time_scale ;
// FIXME: this needs special handling for files where Cues appear
// before Clusters. the current logic assumes Cues appear after
// Clusters.
cue_desc . end_offset = cues_start - matroska - > segment_start ;
}
return cue_desc ;
}
static int webm_clusters_start_with_keyframe ( AVFormatContext * s )
{
MatroskaDemuxContext * matroska = s - > priv_data ;
2019-05-17 00:30:03 +02:00
uint32_t id = matroska - > current_id ;
2014-07-07 22:52:37 +03:00
int64_t cluster_pos , before_pos ;
int index , rv = 1 ;
2020-10-09 09:22:36 +02:00
if ( s - > streams [ 0 ] - > internal - > nb_index_entries < = 0 ) return 0 ;
2014-07-07 22:52:37 +03:00
// seek to the first cluster using cues.
index = av_index_search_timestamp ( s - > streams [ 0 ] , 0 , 0 ) ;
if ( index < 0 ) return 0 ;
2020-10-09 09:22:36 +02:00
cluster_pos = s - > streams [ 0 ] - > internal - > index_entries [ index ] . pos ;
2014-07-07 22:52:37 +03:00
before_pos = avio_tell ( s - > pb ) ;
while ( 1 ) {
2019-05-17 00:29:49 +02:00
uint64_t cluster_id , cluster_length ;
int read ;
2014-07-07 22:52:37 +03:00
AVPacket * pkt ;
avio_seek ( s - > pb , cluster_pos , SEEK_SET ) ;
// read cluster id and length
2019-06-25 03:08:56 +02:00
read = ebml_read_num ( matroska , matroska - > ctx - > pb , 4 , & cluster_id , 1 ) ;
2019-05-17 00:29:49 +02:00
if ( read < 0 | | cluster_id ! = 0xF43B675 ) // done with all clusters
break ;
read = ebml_read_length ( matroska , matroska - > ctx - > pb , & cluster_length ) ;
if ( read < 0 )
2014-07-07 22:52:37 +03:00
break ;
2019-05-17 00:30:03 +02:00
matroska_reset_status ( matroska , 0 , cluster_pos ) ;
2014-07-07 22:52:37 +03:00
matroska_clear_queue ( matroska ) ;
if ( matroska_parse_cluster ( matroska ) < 0 | |
2018-03-26 20:02:37 +02:00
! matroska - > queue ) {
2014-07-07 22:52:37 +03:00
break ;
}
2018-03-26 20:02:37 +02:00
pkt = & matroska - > queue - > pkt ;
2019-05-17 00:29:49 +02:00
// 4 + read is the length of the cluster id and the cluster length field.
cluster_pos + = 4 + read + cluster_length ;
2014-07-07 22:52:37 +03:00
if ( ! ( pkt - > flags & AV_PKT_FLAG_KEY ) ) {
rv = 0 ;
break ;
}
}
2019-05-17 00:30:03 +02:00
/* Restore the status after matroska_read_header: */
matroska_reset_status ( matroska , id , before_pos ) ;
2014-07-07 22:52:37 +03:00
return rv ;
}
static int buffer_size_after_time_downloaded ( int64_t time_ns , double search_sec , int64_t bps ,
double min_buffer , double * buffer ,
double * sec_to_download , AVFormatContext * s ,
int64_t cues_start )
{
double nano_seconds_per_second = 1000000000.0 ;
double time_sec = time_ns / nano_seconds_per_second ;
int rv = 0 ;
int64_t time_to_search_ns = ( int64_t ) ( search_sec * nano_seconds_per_second ) ;
int64_t end_time_ns = time_ns + time_to_search_ns ;
double sec_downloaded = 0.0 ;
CueDesc desc_curr = get_cue_desc ( s , time_ns , cues_start ) ;
if ( desc_curr . start_time_ns = = - 1 )
return - 1 ;
* sec_to_download = 0.0 ;
// Check for non cue start time.
if ( time_ns > desc_curr . start_time_ns ) {
int64_t cue_nano = desc_curr . end_time_ns - time_ns ;
double percent = ( double ) ( cue_nano ) / ( desc_curr . end_time_ns - desc_curr . start_time_ns ) ;
double cueBytes = ( desc_curr . end_offset - desc_curr . start_offset ) * percent ;
double timeToDownload = ( cueBytes * 8.0 ) / bps ;
sec_downloaded + = ( cue_nano / nano_seconds_per_second ) - timeToDownload ;
* sec_to_download + = timeToDownload ;
// Check if the search ends within the first cue.
if ( desc_curr . end_time_ns > = end_time_ns ) {
double desc_end_time_sec = desc_curr . end_time_ns / nano_seconds_per_second ;
double percent_to_sub = search_sec / ( desc_end_time_sec - time_sec ) ;
sec_downloaded = percent_to_sub * sec_downloaded ;
* sec_to_download = percent_to_sub * * sec_to_download ;
}
if ( ( sec_downloaded + * buffer ) < = min_buffer ) {
return 1 ;
}
// Get the next Cue.
desc_curr = get_cue_desc ( s , desc_curr . end_time_ns , cues_start ) ;
}
while ( desc_curr . start_time_ns ! = - 1 ) {
int64_t desc_bytes = desc_curr . end_offset - desc_curr . start_offset ;
int64_t desc_ns = desc_curr . end_time_ns - desc_curr . start_time_ns ;
double desc_sec = desc_ns / nano_seconds_per_second ;
double bits = ( desc_bytes * 8.0 ) ;
double time_to_download = bits / bps ;
sec_downloaded + = desc_sec - time_to_download ;
* sec_to_download + = time_to_download ;
if ( desc_curr . end_time_ns > = end_time_ns ) {
double desc_end_time_sec = desc_curr . end_time_ns / nano_seconds_per_second ;
double percent_to_sub = search_sec / ( desc_end_time_sec - time_sec ) ;
sec_downloaded = percent_to_sub * sec_downloaded ;
* sec_to_download = percent_to_sub * * sec_to_download ;
if ( ( sec_downloaded + * buffer ) < = min_buffer )
rv = 1 ;
break ;
}
if ( ( sec_downloaded + * buffer ) < = min_buffer ) {
rv = 1 ;
break ;
}
desc_curr = get_cue_desc ( s , desc_curr . end_time_ns , cues_start ) ;
}
* buffer = * buffer + sec_downloaded ;
return rv ;
}
/* This function computes the bandwidth of the WebM file with the help of
* buffer_size_after_time_downloaded ( ) function . Both of these functions are
* adapted from WebM Tools project and are adapted to work with FFmpeg ' s
* Matroska parsing mechanism .
*
* Returns the bandwidth of the file on success ; - 1 on error .
* */
static int64_t webm_dash_manifest_compute_bandwidth ( AVFormatContext * s , int64_t cues_start )
{
MatroskaDemuxContext * matroska = s - > priv_data ;
AVStream * st = s - > streams [ 0 ] ;
double bandwidth = 0.0 ;
2014-07-16 01:06:15 +03:00
int i ;
2020-10-09 09:22:36 +02:00
for ( i = 0 ; i < st - > internal - > nb_index_entries ; i + + ) {
2014-07-07 22:52:37 +03:00
int64_t prebuffer_ns = 1000000000 ;
2020-10-09 09:22:36 +02:00
int64_t time_ns = st - > internal - > index_entries [ i ] . timestamp * matroska - > time_scale ;
2014-07-07 22:52:37 +03:00
double nano_seconds_per_second = 1000000000.0 ;
int64_t prebuffered_ns = time_ns + prebuffer_ns ;
double prebuffer_bytes = 0.0 ;
int64_t temp_prebuffer_ns = prebuffer_ns ;
int64_t pre_bytes , pre_ns ;
double pre_sec , prebuffer , bits_per_second ;
CueDesc desc_beg = get_cue_desc ( s , time_ns , cues_start ) ;
// Start with the first Cue.
CueDesc desc_end = desc_beg ;
// Figure out how much data we have downloaded for the prebuffer. This will
// be used later to adjust the bits per sample to try.
while ( desc_end . start_time_ns ! = - 1 & & desc_end . end_time_ns < prebuffered_ns ) {
// Prebuffered the entire Cue.
prebuffer_bytes + = desc_end . end_offset - desc_end . start_offset ;
temp_prebuffer_ns - = desc_end . end_time_ns - desc_end . start_time_ns ;
desc_end = get_cue_desc ( s , desc_end . end_time_ns , cues_start ) ;
}
if ( desc_end . start_time_ns = = - 1 ) {
// The prebuffer is larger than the duration.
2014-10-01 20:13:31 +03:00
if ( matroska - > duration * matroska - > time_scale > = prebuffered_ns )
return - 1 ;
bits_per_second = 0.0 ;
} else {
// The prebuffer ends in the last Cue. Estimate how much data was
// prebuffered.
pre_bytes = desc_end . end_offset - desc_end . start_offset ;
pre_ns = desc_end . end_time_ns - desc_end . start_time_ns ;
pre_sec = pre_ns / nano_seconds_per_second ;
prebuffer_bytes + =
pre_bytes * ( ( temp_prebuffer_ns / nano_seconds_per_second ) / pre_sec ) ;
prebuffer = prebuffer_ns / nano_seconds_per_second ;
// Set this to 0.0 in case our prebuffer buffers the entire video.
bits_per_second = 0.0 ;
do {
int64_t desc_bytes = desc_end . end_offset - desc_beg . start_offset ;
int64_t desc_ns = desc_end . end_time_ns - desc_beg . start_time_ns ;
double desc_sec = desc_ns / nano_seconds_per_second ;
double calc_bits_per_second = ( desc_bytes * 8 ) / desc_sec ;
// Drop the bps by the percentage of bytes buffered.
double percent = ( desc_bytes - prebuffer_bytes ) / desc_bytes ;
double mod_bits_per_second = calc_bits_per_second * percent ;
if ( prebuffer < desc_sec ) {
double search_sec =
( double ) ( matroska - > duration * matroska - > time_scale ) / nano_seconds_per_second ;
// Add 1 so the bits per second should be a little bit greater than file
// datarate.
int64_t bps = ( int64_t ) ( mod_bits_per_second ) + 1 ;
const double min_buffer = 0.0 ;
double buffer = prebuffer ;
double sec_to_download = 0.0 ;
int rv = buffer_size_after_time_downloaded ( prebuffered_ns , search_sec , bps ,
min_buffer , & buffer , & sec_to_download ,
s , cues_start ) ;
if ( rv < 0 ) {
return - 1 ;
} else if ( rv = = 0 ) {
bits_per_second = ( double ) ( bps ) ;
break ;
}
2014-07-07 22:52:37 +03:00
}
2014-10-01 20:13:31 +03:00
desc_end = get_cue_desc ( s , desc_end . end_time_ns , cues_start ) ;
} while ( desc_end . start_time_ns ! = - 1 ) ;
}
2014-07-07 22:52:37 +03:00
if ( bandwidth < bits_per_second ) bandwidth = bits_per_second ;
}
return ( int64_t ) bandwidth ;
}
2017-04-21 17:40:35 +02:00
static int webm_dash_manifest_cues ( AVFormatContext * s , int64_t init_range )
2014-07-07 22:52:37 +03:00
{
MatroskaDemuxContext * matroska = s - > priv_data ;
EbmlList * seekhead_list = & matroska - > seekhead ;
MatroskaSeekhead * seekhead = seekhead_list - > elem ;
char * buf ;
2014-08-25 19:15:13 +03:00
int64_t cues_start = - 1 , cues_end = - 1 , before_pos , bandwidth ;
2014-07-07 22:52:37 +03:00
int i ;
2017-04-20 16:02:54 +02:00
int end = 0 ;
2014-07-07 22:52:37 +03:00
// determine cues start and end positions
for ( i = 0 ; i < seekhead_list - > nb_elem ; i + + )
if ( seekhead [ i ] . id = = MATROSKA_ID_CUES )
break ;
if ( i > = seekhead_list - > nb_elem ) return - 1 ;
before_pos = avio_tell ( matroska - > ctx - > pb ) ;
cues_start = seekhead [ i ] . pos + matroska - > segment_start ;
if ( avio_seek ( matroska - > ctx - > pb , cues_start , SEEK_SET ) = = cues_start ) {
2014-10-01 20:13:30 +03:00
// cues_end is computed as cues_start + cues_length + length of the
2019-05-17 00:29:54 +02:00
// Cues element ID (i.e. 4) + EBML length of the Cues element.
// cues_end is inclusive and the above sum is reduced by 1.
uint64_t cues_length , cues_id ;
int bytes_read ;
2019-06-25 03:08:56 +02:00
bytes_read = ebml_read_num ( matroska , matroska - > ctx - > pb , 4 , & cues_id , 1 ) ;
2019-05-17 00:29:54 +02:00
if ( bytes_read < 0 | | cues_id ! = ( MATROSKA_ID_CUES & 0xfffffff ) )
return bytes_read < 0 ? bytes_read : AVERROR_INVALIDDATA ;
bytes_read = ebml_read_length ( matroska , matroska - > ctx - > pb , & cues_length ) ;
if ( bytes_read < 0 )
return bytes_read ;
cues_end = cues_start + 4 + bytes_read + cues_length - 1 ;
2014-07-07 22:52:37 +03:00
}
avio_seek ( matroska - > ctx - > pb , before_pos , SEEK_SET ) ;
2014-08-25 19:15:13 +03:00
if ( cues_start = = - 1 | | cues_end = = - 1 ) return - 1 ;
2014-07-07 22:52:37 +03:00
// parse the cues
matroska_parse_cues ( matroska ) ;
// cues start
2014-07-29 22:10:39 +03:00
av_dict_set_int ( & s - > streams [ 0 ] - > metadata , CUES_START , cues_start , 0 ) ;
2014-07-07 22:52:37 +03:00
// cues end
2014-07-29 22:10:39 +03:00
av_dict_set_int ( & s - > streams [ 0 ] - > metadata , CUES_END , cues_end , 0 ) ;
2014-07-07 22:52:37 +03:00
2019-05-17 00:30:13 +02:00
// if the file has cues at the start, fix up the init range so that
2017-04-21 17:40:35 +02:00
// it does not include it
if ( cues_start < = init_range )
av_dict_set_int ( & s - > streams [ 0 ] - > metadata , INITIALIZATION_RANGE , cues_start - 1 , 0 ) ;
2014-07-07 22:52:37 +03:00
// bandwidth
bandwidth = webm_dash_manifest_compute_bandwidth ( s , cues_start ) ;
if ( bandwidth < 0 ) return - 1 ;
2014-07-29 22:10:39 +03:00
av_dict_set_int ( & s - > streams [ 0 ] - > metadata , BANDWIDTH , bandwidth , 0 ) ;
2014-07-07 22:52:37 +03:00
// check if all clusters start with key frames
2014-07-29 22:10:39 +03:00
av_dict_set_int ( & s - > streams [ 0 ] - > metadata , CLUSTER_KEYFRAME , webm_clusters_start_with_keyframe ( s ) , 0 ) ;
2014-07-07 22:52:37 +03:00
// store cue point timestamps as a comma separated list for checking subsegment alignment in
// the muxer. assumes that each timestamp cannot be more than 20 characters long.
2020-10-09 09:22:36 +02:00
buf = av_malloc_array ( s - > streams [ 0 ] - > internal - > nb_index_entries , 20 ) ;
2014-07-07 22:52:37 +03:00
if ( ! buf ) return - 1 ;
strcpy ( buf , " " ) ;
2020-10-09 09:22:36 +02:00
for ( i = 0 ; i < s - > streams [ 0 ] - > internal - > nb_index_entries ; i + + ) {
2018-09-03 14:18:02 +02:00
int ret = snprintf ( buf + end , 20 ,
2020-10-09 09:22:36 +02:00
" % " PRId64 " %s " , s - > streams [ 0 ] - > internal - > index_entries [ i ] . timestamp ,
i ! = s - > streams [ 0 ] - > internal - > nb_index_entries - 1 ? " , " : " " ) ;
if ( ret < = 0 | | ( ret = = 20 & & i = = s - > streams [ 0 ] - > internal - > nb_index_entries - 1 ) ) {
2017-04-20 16:02:54 +02:00
av_log ( s , AV_LOG_ERROR , " timestamp too long. \n " ) ;
2017-05-07 05:29:08 +02:00
av_free ( buf ) ;
2017-04-20 16:02:54 +02:00
return AVERROR_INVALIDDATA ;
}
end + = ret ;
2014-07-07 22:52:37 +03:00
}
2019-11-10 06:07:31 +02:00
av_dict_set ( & s - > streams [ 0 ] - > metadata , CUE_TIMESTAMPS ,
buf , AV_DICT_DONT_STRDUP_VAL ) ;
2014-07-07 22:52:37 +03:00
return 0 ;
}
static int webm_dash_manifest_read_header ( AVFormatContext * s )
{
char * buf ;
int ret = matroska_read_header ( s ) ;
2017-04-21 17:40:35 +02:00
int64_t init_range ;
2014-07-07 22:52:37 +03:00
MatroskaTrack * tracks ;
MatroskaDemuxContext * matroska = s - > priv_data ;
if ( ret ) {
av_log ( s , AV_LOG_ERROR , " Failed to read file headers \n " ) ;
return - 1 ;
}
2019-08-30 15:18:29 +02:00
if ( ! matroska - > tracks . nb_elem | | ! s - > nb_streams ) {
av_log ( s , AV_LOG_ERROR , " No track found \n " ) ;
2020-06-13 23:58:32 +02:00
ret = AVERROR_INVALIDDATA ;
goto fail ;
2016-11-08 01:42:23 +02:00
}
2014-07-07 22:52:37 +03:00
2015-04-01 01:51:57 +02:00
if ( ! matroska - > is_live ) {
buf = av_asprintf ( " %g " , matroska - > duration ) ;
2020-06-13 23:58:32 +02:00
if ( ! buf ) {
ret = AVERROR ( ENOMEM ) ;
goto fail ;
}
2019-11-10 06:07:31 +02:00
av_dict_set ( & s - > streams [ 0 ] - > metadata , DURATION ,
buf , AV_DICT_DONT_STRDUP_VAL ) ;
2015-04-01 01:51:57 +02:00
// initialization range
// 5 is the offset of Cluster ID.
2017-04-21 17:40:35 +02:00
init_range = avio_tell ( s - > pb ) - 5 ;
av_dict_set_int ( & s - > streams [ 0 ] - > metadata , INITIALIZATION_RANGE , init_range , 0 ) ;
2015-04-01 01:51:57 +02:00
}
2014-07-07 22:52:37 +03:00
// basename of the file
2017-12-30 00:30:14 +02:00
buf = strrchr ( s - > url , ' / ' ) ;
av_dict_set ( & s - > streams [ 0 ] - > metadata , FILENAME , buf ? + + buf : s - > url , 0 ) ;
2014-07-07 22:52:37 +03:00
// track number
tracks = matroska - > tracks . elem ;
2014-07-29 22:10:39 +03:00
av_dict_set_int ( & s - > streams [ 0 ] - > metadata , TRACK_NUMBER , tracks [ 0 ] . num , 0 ) ;
2014-07-07 22:52:37 +03:00
// parse the cues and populate Cue related fields
2017-04-12 06:33:28 +02:00
if ( ! matroska - > is_live ) {
2017-04-21 17:40:35 +02:00
ret = webm_dash_manifest_cues ( s , init_range ) ;
2017-04-12 06:33:28 +02:00
if ( ret < 0 ) {
av_log ( s , AV_LOG_ERROR , " Error parsing Cues \n " ) ;
2020-06-13 23:58:32 +02:00
goto fail ;
2017-04-12 06:33:28 +02:00
}
}
// use the bandwidth from the command line if it was provided
if ( matroska - > bandwidth > 0 ) {
av_dict_set_int ( & s - > streams [ 0 ] - > metadata , BANDWIDTH ,
matroska - > bandwidth , 0 ) ;
}
return 0 ;
2020-06-13 23:58:32 +02:00
fail :
matroska_read_close ( s ) ;
return ret ;
2014-07-07 22:52:37 +03:00
}
static int webm_dash_manifest_read_packet ( AVFormatContext * s , AVPacket * pkt )
{
return AVERROR_EOF ;
}
2015-04-01 01:51:57 +02:00
# define OFFSET(x) offsetof(MatroskaDemuxContext, x)
static const AVOption options [ ] = {
2015-11-21 23:05:07 +02:00
{ " live " , " flag indicating that the input is a live file that only has the headers. " , OFFSET ( is_live ) , AV_OPT_TYPE_BOOL , { . i64 = 0 } , 0 , 1 , AV_OPT_FLAG_DECODING_PARAM } ,
2017-04-12 06:33:28 +02:00
{ " bandwidth " , " bandwidth of this stream to be specified in the DASH manifest. " , OFFSET ( bandwidth ) , AV_OPT_TYPE_INT , { . i64 = 0 } , 0 , INT_MAX , AV_OPT_FLAG_DECODING_PARAM } ,
2015-04-01 01:51:57 +02:00
{ NULL } ,
} ;
static const AVClass webm_dash_class = {
. class_name = " WebM DASH Manifest demuxer " ,
. item_name = av_default_item_name ,
. option = options ,
. version = LIBAVUTIL_VERSION_INT ,
} ;
2011-01-26 00:03:28 +02:00
AVInputFormat ff_matroska_demuxer = {
2011-07-16 23:18:12 +03:00
. name = " matroska,webm " ,
2012-07-25 00:51:41 +03:00
. long_name = NULL_IF_CONFIG_SMALL ( " Matroska / WebM " ) ,
2019-12-28 05:05:44 +02:00
. extensions = " mkv,mk3d,mka,mks,webm " ,
2011-07-16 23:18:12 +03:00
. priv_data_size = sizeof ( MatroskaDemuxContext ) ,
. read_probe = matroska_probe ,
. read_header = matroska_read_header ,
. read_packet = matroska_read_packet ,
. read_close = matroska_read_close ,
. read_seek = matroska_read_seek ,
2014-03-14 00:14:43 +03:00
. mime_type = " audio/webm,audio/x-matroska,video/webm,video/x-matroska "
2007-06-05 01:10:54 +03:00
} ;
2014-07-07 22:52:37 +03:00
AVInputFormat ff_webm_dash_manifest_demuxer = {
. name = " webm_dash_manifest " ,
. long_name = NULL_IF_CONFIG_SMALL ( " WebM DASH Manifest " ) ,
. priv_data_size = sizeof ( MatroskaDemuxContext ) ,
. read_header = webm_dash_manifest_read_header ,
. read_packet = webm_dash_manifest_read_packet ,
. read_close = matroska_read_close ,
2015-04-01 01:51:57 +02:00
. priv_class = & webm_dash_class ,
2014-07-07 22:52:37 +03:00
} ;