This ensures that the metadata items are always written
in the same order and that first comes the preface,
then the identification linked from the preface, ...
Some demuxers can't handle files where the metadata
items are in random order.
This will later be used to keep the structural metadata when
remuxing an MXF file and can also be used in Pitivi for example
to know the file structure and to select what should be used
and played.
For the tracks we still resolve the descriptors but the
source package only contains the descriptor it really references
instead of all subdescriptors of multiple descriptors.
This makes it easier later to serialize the structural metadata
again.
By searching for the following position we will add the complete
generic container (i.e. all tracks) at the same position to the
index which should make finding positions for more than one track
a bit faster.
When seeking to a position where no new offset can be
found or after the end of a track don't set the EOS flag
to TRUE as we don't (and can't) push EOS downstream then.
Instead set the current essence track position to then
end of the essence track and let the loop or chain function
set everything to EOS and push EOS downstream.
Also restart the pad's task if resolving metadata failed
(playback will continue as before) and return FALSE from
the seek if no new offset could be found.
When the material package track and source package track
edit rate are different the source package track edit
rate applies to the stored essence and the material track
edit rate only applies to the values in the track's sequence
and components.
Avid uses a custom essence container UL and custom essence element keys
that are fortunately compatible with the generic container essence
elements. Partially fixes bug #561922.
Some files contain multiple tracks with the same track number but different
track id inside the same source package. This is invalid and we simply ignore
the second and following of such traacks now instead of overwriting the
information of the first one.
Avid usually uses a custom essence container label for the essence
descriptors and stores the actual codec that is used inside the
picture essence coding field (and for sound probably in the sound
essence coding field but I have no sample files with sound).
Partially fixes bug #561922.
This is useful for seeking as we usually want to seek to the previous
keyframe. The keyframe detection is done by parsing the MPEG2
elementary stream and if a GOP or I-frame packet is found we
assume a keyframe in this edit unit.
Generate an index table for essence streams during playback
and make sure that only the correct essence elements are
used for played tracks.
Make it possible to have one essence stream used in multiple
playback tracks.
Fix some minor bugs.
When parsing an index table segment without a valid
primer pack in this partition only parse the static
local tags and ignore all dynamic ones.
This allows us to use index table segments in some
broken files.
Add an abstraction to represent essence tracks and
use this everywhere. This will later be used to keep
track of positions and to generate/handle seek tables.
Some random cleanup and renaming.
Allocate memory for the pos tables and slice offsets
of the index entries separately to a) fix alignment in
weird cases and b) to actually have something to free()
in the index table segment reset function.
Header metadata is finished after partition.header_byte_count
bytes after the first byte of the primer pack are handled.
After this there can only be index table segments, filler packets,
essence or the start of the next partition.
This fixes playback of some files that have non-standard metadata
packets in the header metadata.
Store all partitions & primer packs in memory for faster access.
This is later needed for fast seeking.
Pre-fill the list of partitions with the content of the random
index pack.
Don't parse metadata of an partition twice.
Store the MXF metadata in a hash table, keyed by the
instance UID. This simplifies resolval of the metadata
and makes looping over all metadata sets unnecessary
in most cases.
Additionally parse metadata always. If we already have
a metadata set with the same UID replace it only if
the new metadata set is from a later offset. This
fixes metadata parsing of files where following partitions
don't have a complete copy of the previous metadata.
Original commit message from CVS:
* gst/mxf/Makefile.am:
* gst/mxf/mxfdemux.c:
* gst/mxf/mxfmetadata.c: (mxf_metadata_handle_tag),
(mxf_metadata_preface_handle_tag),
(mxf_metadata_identification_handle_tag),
(mxf_metadata_content_storage_handle_tag),
(mxf_metadata_essence_container_data_handle_tag),
(mxf_metadata_generic_package_handle_tag),
(mxf_metadata_track_handle_tag),
(mxf_metadata_sequence_handle_tag),
(mxf_metadata_structural_component_handle_tag),
(mxf_metadata_generic_descriptor_handle_tag),
(mxf_metadata_locator_handle_tag),
(mxf_metadata_locator_class_init):
* gst/mxf/mxfmetadata.h:
"Generation UID" and "This Generation UID" are different so handle
them different. Also both are not part of every metadata type.
Original commit message from CVS:
* gst/mxf/mxfdemux.c:
(gst_mxf_demux_handle_header_metadata_update_streams),
(gst_mxf_demux_pad_next_component):
Don't allow negative or zero edit rates to prevent division
by zero and we support negative edit rates anyway.
Original commit message from CVS:
* gst/mxf/mxfaes-bwf.c: (mxf_bwf_handle_essence_element),
(mxf_aes3_handle_essence_element):
* gst/mxf/mxfalaw.c: (mxf_alaw_handle_essence_element):
* gst/mxf/mxfd10.c: (mxf_d10_picture_handle_essence_element),
(mxf_d10_sound_handle_essence_element):
* gst/mxf/mxfdemux.c: (gst_mxf_demux_pad_init),
(gst_mxf_demux_choose_package),
(gst_mxf_demux_handle_header_metadata_update_streams),
(gst_mxf_demux_pad_next_component),
(gst_mxf_demux_handle_generic_container_essence_element),
(gst_mxf_demux_parse_footer_metadata),
(gst_mxf_demux_handle_klv_packet), (gst_mxf_demux_src_query):
* gst/mxf/mxfdv-dif.c: (mxf_dv_dif_handle_essence_element):
* gst/mxf/mxfjpeg2000.c: (mxf_jpeg2000_handle_essence_element):
* gst/mxf/mxfmetadata.c: (mxf_metadata_sequence_init),
(mxf_metadata_structural_component_init),
(mxf_metadata_generic_picture_essence_descriptor_init):
* gst/mxf/mxfmpeg.c: (mxf_mpeg_video_handle_essence_element),
(mxf_mpeg_audio_handle_essence_element):
* gst/mxf/mxfparse.h:
* gst/mxf/mxfup.c: (mxf_up_handle_essence_element):
* gst/mxf/mxfvc3.c: (mxf_vc3_handle_essence_element):
* tests/check/elements/mxfdemux.c: (_sink_chain):
Implement support for OP2a/b/c and OP3a/b/c, i.e. tracks with
more than a single component. This currently only works for
the case where the components are stored in playback order
in the file.
Set some more default/distinguished values for the structural
metadata.
Make some types more strict by choosing the correct subclasses.
Set DISCONT flag on buffers after a component switch.
Take the last partition from the random index pack for the footer
partition of the header partition doesn't reference the footer
partition. This gives us the final structural metadata for
some more files in the beginning.
Original commit message from CVS:
* gst/mxf/mxfmpeg.c: (mxf_mpeg_es_create_caps),
(mxf_mpeg_create_caps):
Add initial support for h.264 video (SMPTE RP2008).
Original commit message from CVS:
* gst/mxf/mxfmetadata.c: (mxf_metadata_base_resolve):
* gst/mxf/mxfmetadata.h:
If resolving a metadata object failed the first time return
FALSE for a second mxf_metadata_base_resolve() call instead
of TRUE.
Original commit message from CVS:
* gst/mxf/mxfaes-bwf.c: (mxf_bwf_create_caps),
(mxf_aes3_create_caps):
* gst/mxf/mxfalaw.c: (mxf_alaw_create_caps):
* gst/mxf/mxfd10.c: (mxf_d10_create_caps):
* gst/mxf/mxfdemux.c:
* gst/mxf/mxfmetadata.c: (mxf_metadata_source_package_resolve),
(mxf_metadata_generic_picture_essence_descriptor_set_caps),
(mxf_metadata_generic_sound_essence_descriptor_set_caps):
* gst/mxf/mxfmetadata.h:
* gst/mxf/mxfmpeg.c: (mxf_mpeg_es_create_caps):
Add mxf_metadata_generic_sound_essence_descriptor_set_caps() to
set rate and channels and use this for all sound essence.
Give some debug output when setting picture essence caps with
invalid descriptor values.
Fix height calculation from the frame layout a bit more and
add a TODO to check if it's really correct now or if it needs
more fixing (especially, does the framerate need adjustments?).
Original commit message from CVS:
* gst/mxf/Makefile.am:
* gst/mxf/mxf.c: (plugin_init):
* gst/mxf/mxfvc3.c: (mxf_is_vc3_essence_track),
(mxf_vc3_handle_essence_element), (mxf_vc3_create_caps),
(mxf_vc3_init):
* gst/mxf/mxfvc3.h:
Add support for VC-3 (aka DNxHD essence, see SMPTE 2019-4).
This is untested because of lack of sample files but should
work fine.
Original commit message from CVS:
* gst/mxf/mxfdemux.c: (gst_mxf_demux_chain):
In push mode skip the run in again if we get buffers with
an offset before the run in. This can happen on seeks for example.
Original commit message from CVS:
* gst/mxf/mxfdemux.c:
(gst_mxf_demux_handle_header_metadata_resolve_references):
* gst/mxf/mxfmetadata.c: (mxf_metadata_base_resolve_default),
(mxf_metadata_base_class_init), (mxf_metadata_base_parse),
(mxf_metadata_base_resolve), (mxf_metadata_new),
(mxf_metadata_preface_resolve),
(mxf_metadata_content_storage_resolve),
(mxf_metadata_essence_container_data_resolve),
(mxf_metadata_generic_package_resolve),
(mxf_metadata_source_package_resolve),
(mxf_metadata_track_resolve), (mxf_metadata_sequence_resolve),
(mxf_metadata_dm_segment_resolve),
(mxf_metadata_generic_descriptor_resolve),
(mxf_metadata_multiple_descriptor_resolve):
* gst/mxf/mxfmetadata.h:
Rename MXFMetadataBase methods to reflect that they belong to
MXFMetadataBase and not MXFMetadata.
Original commit message from CVS:
* gst/mxf/mxfaes-bwf.c:
(mxf_metadata_wave_audio_essence_descriptor_handle_tag),
(mxf_metadata_aes3_audio_essence_descriptor_handle_tag):
* gst/mxf/mxfmetadata.c:
(mxf_metadata_multiple_descriptor_handle_tag):
Return FALSE when parsing a metadata tag failed.
Remove some useless return value assignments.
Original commit message from CVS:
* gst/mxf/mxfdemux.c: (gst_mxf_demux_handle_descriptive_metadata),
(gst_mxf_demux_handle_klv_packet):
Add a generic handler for descriptive metadata so we can get some
debug output and let users file bugs for unsupport descriptive
metadata schemes.
Original commit message from CVS:
* gst/mxf/mxfaes-bwf.c:
* gst/mxf/mxfalaw.c: (mxf_alaw_create_caps):
* gst/mxf/mxfdemux.c:
* gst/mxf/mxfdv-dif.c: (mxf_dv_dif_create_caps):
* gst/mxf/mxfjpeg2000.c:
* gst/mxf/mxfmpeg.c:
* gst/mxf/mxfup.c:
Major update of the TODO lists. There's still much to do.
Original commit message from CVS:
* gst/mxf/mxfdemux.c: (gst_mxf_demux_handle_metadata_track),
(gst_mxf_demux_handle_header_metadata_resolve_references),
(gst_mxf_demux_handle_metadata):
* gst/mxf/mxfparse.c: (mxf_metadata_track_parse),
(mxf_metadata_structural_component_parse),
(mxf_metadata_structural_component_reset):
* gst/mxf/mxfparse.h:
* gst/mxf/mxftypes.h:
Implement parsing of Event Tracks, Static Tracks, DM Segments
and DM Source Clips as a preparation for descriptive metadata
support. Next step is to implement SMPTE S380M, "Descriptive
Metadata Scheme-1".
Original commit message from CVS:
* gst/mxf/mxfjpeg2000.c: (mxf_jpeg2000_create_caps):
Implement detection of the colorspace used by the
JPEG2000 codestream. XYZ colorspace isn't supported by
our decoders yet unfortunately.
Original commit message from CVS:
* gst/mxf/mxfparse.c:
(mxf_metadata_generic_picture_essence_descriptor_set_caps):
All frame layout values except 0 are for interlaced video,
not only 1 and 2.
Original commit message from CVS:
* gst/mxf/mxfdemux.c: (gst_mxf_demux_reset_metadata),
(gst_mxf_demux_handle_primer_pack),
(gst_mxf_demux_handle_metadata_preface),
(gst_mxf_demux_handle_header_metadata_resolve_references),
(gst_mxf_demux_choose_package),
(gst_mxf_demux_handle_header_metadata_update_streams),
(gst_mxf_demux_parse_footer_metadata),
(gst_mxf_demux_handle_klv_packet):
* gst/mxf/mxfdemux.h:
* gst/mxf/mxftypes.h:
Remove the concept of "final" metadata, broken files have updated
metadata in following partitions even if the current partition
is closed and complete.
If there's a second primer pack inside a partition ignore the second
one instead of throwing an error.
Store the type of a track (audio, video, timestamp, ...) inside the
track struct.
Post no-more-pads only once for a stream and not a second time
when switching the currently playing package.
When updating all streams make sure that we could select and find
a package instead of dereferencing NULL.
Allow descriptive metadata inside metadata parts of a file when
pulling the footer metadata too.
Only switch the currently playing package if a new one was chosen
and the metadata is in a resolved state. If the latter isn't true
postpone the switch until the metadata is resolved.
Pull the footer metadata directly after parsing the header partition
pack if it's not closed or not complete instead of waiting until
the next KLV packet.
Original commit message from CVS:
* gst/mxf/mxfup.c: (mxf_up_handle_essence_element):
Don't create a subbuffer, instead just change the buffer metadata
of the input buffer.
Original commit message from CVS:
* gst/mxf/mxfdemux.c: (gst_mxf_demux_pad_finalize),
(gst_mxf_demux_pad_class_init), (gst_mxf_demux_pad_init),
(gst_mxf_demux_remove_pad), (gst_mxf_demux_reset_metadata),
(gst_mxf_demux_combine_flows),
(gst_mxf_demux_handle_header_metadata_resolve_references),
(gst_mxf_demux_handle_header_metadata_update_streams),
(gst_mxf_demux_handle_generic_container_essence_element),
(gst_mxf_demux_pull_random_index_pack), (gst_mxf_demux_loop),
(gst_mxf_demux_src_query):
Rename GstMXFPad to GstMXFDemuxPad to prevent name clashes
with the GstMXFMuxPad in the muxer (later).
Pull the random index pack only directly after skipping the run
in and don't allocate an array for it if parsing it failed.
Original commit message from CVS:
* gst/mxf/mxfd10.c: (mxf_d10_sound_handle_essence_element),
(mxf_d10_create_caps):
Use the mapping data for storing the channel count and the
width instead of parsing caps for every buffer.
Original commit message from CVS:
* gst/mxf/mxfup.c: (mxf_up_handle_essence_element),
(mxf_up_rgba_create_caps):
Handle the image start and end offsets, otherwise we output
too large image buffers.
Original commit message from CVS:
* gst/mxf/Makefile.am:
* gst/mxf/mxfdemux.c:
(gst_mxf_demux_handle_header_metadata_update_streams):
* gst/mxf/mxftypes.h:
* gst/mxf/mxfup.c: (mxf_is_up_essence_track),
(mxf_up_handle_essence_element), (mxf_up_rgba_create_caps),
(mxf_up_create_caps):
* gst/mxf/mxfup.h:
Add initial support for uncompressed video essence (SMPTE S384M).
* gst/mxf/mxfparse.c:
(mxf_metadata_rgba_picture_essence_descriptor_handle_tag),
(mxf_metadata_rgba_picture_essence_descriptor_reset):
Fix parsing of the RGBA descriptor and add support for parsing
the pixel layout.
Original commit message from CVS:
* gst/mxf/mxfdemux.c: (gst_mxf_demux_parse_footer_metadata):
The previous partition field of the partition is relative
to the current partition and not to the offset of the header
partition.
Original commit message from CVS:
* gst/mxf/mxfd10.c: (mxf_d10_sound_handle_essence_element):
Document decoding of the AES3 data and also skip the 32 bit
header in the beginning of every buffer, otherwise we get
one broken audio channel.
Original commit message from CVS:
* gst/mxf/mxfdemux.c:
(gst_mxf_demux_handle_header_metadata_resolve_references):
Generalize some lines of codes that are more or less the same
by a macro.
Original commit message from CVS:
* gst/mxf/mxfdemux.c:
(gst_mxf_demux_handle_generic_container_essence_element):
Ignore zero sized essence elements. They're often inserted
to fulfill the streaming requirements.
Original commit message from CVS:
* gst/mxf/mxfparse.c:
(mxf_metadata_generic_picture_essence_descriptor_set_caps):
Fix setting of the height/width and PAR of video streams.
Original commit message from CVS:
* gst/mxf/Makefile.am:
* gst/mxf/mxfd10.c: (mxf_is_d10_essence_track),
(mxf_d10_picture_handle_essence_element),
(mxf_d10_sound_handle_essence_element), (mxf_d10_create_caps):
* gst/mxf/mxfd10.h:
* gst/mxf/mxfdemux.c:
(gst_mxf_demux_handle_header_metadata_update_streams):
Add support for SMPTE D10 essence (SMPTE 386M).
* gst/mxf/mxfparse.c:
(mxf_metadata_generic_picture_essence_descriptor_set_caps):
Don't set width/height and PAR on the caps as those values are
wrong for most files (height is sometimes the height of a field
and aspect ratio is some random value).
* gst/mxf/mxfaes-bwf.c: (mxf_bwf_create_caps),
(mxf_aes3_create_caps):
Fix calculation of block align if it isn't set in the descriptor.
Original commit message from CVS:
* gst/mxf/mxfdemux.c:
(gst_mxf_demux_handle_header_metadata_update_streams),
(gst_mxf_demux_base_init):
Use a single "track_%u" src pad template and use the track ID
of the playback package for the pad name instead of the track ID
of the source package, as the latter could be the same for
more than one track if they're in different packages.
Original commit message from CVS:
* gst/mxf/mxfjpeg2000.c: (mxf_jpeg2000_create_caps):
Use new image/x-jpc caps for the raw JPEG2000 codestream.
For now assume sRGB as colorspace, has to be read from the descriptor
somehow later.
