BaseParse#

class BaseParse(**properties: Any)#

Superclasses: Element, Object, InitiallyUnowned, Object

This base class is for parser elements that process data and splits it into separate audio/video/whatever frames.

It provides for:

provides one sink pad and one source pad
handles state changes
can operate in pull mode or push mode
handles seeking in both modes
handles events (SEGMENT/EOS/FLUSH)
handles queries (POSITION/DURATION/SEEKING/FORMAT/CONVERT)
handles flushing

The purpose of this base class is to provide the basic functionality of a parser and share a lot of rather complex code.

Description of the parsing mechanism:#

Set-up phase#

BaseParse calls GstBaseParseClass::start to inform subclass
that data processing is about to start now.
BaseParse class calls GstBaseParseClass::set_sink_caps to
inform the subclass about incoming sinkpad caps. Subclass could already set the srcpad caps accordingly, but this might be delayed until calling finish_frame() with a non-queued frame.
At least at this point subclass needs to tell the BaseParse class
how big data chunks it wants to receive (minimum frame size ). It can do this with set_min_frame_size().
BaseParse class sets up appropriate data passing mode (pull/push)
and starts to process the data.

Parsing phase#

BaseParse gathers at least min_frame_size bytes of data either
by pulling it from upstream or collecting buffers in an internal Adapter.
A buffer of (at least) min_frame_size bytes is passed to subclass
with GstBaseParseClass::handle_frame. Subclass checks the contents and can optionally return GST_FLOW_OK along with an amount of data to be skipped to find a valid frame (which will result in a subsequent DISCONT). If, otherwise, the buffer does not hold a complete frame, GstBaseParseClass::handle_frame can merely return and will be called again when additional data is available. In push mode this amounts to an additional input buffer (thus minimal additional latency), in pull mode this amounts to some arbitrary reasonable buffer size increase.

Of course, set_min_frame_size() could also be used if: a very specific known amount of additional data is required. If, however, the buffer holds a complete valid frame, it can pass the size of this frame to finish_frame().
If acting as a converter, it can also merely indicate consumed input: data while simultaneously providing custom output data. Note that baseclass performs some processing (such as tracking overall consumed data rate versus duration) for each finished frame, but other state is only updated upon each call to GstBaseParseClass::handle_frame (such as tracking upstream input timestamp).
Subclass is also responsible for setting the buffer metadata: (e.g. buffer timestamp and duration, or keyframe if applicable). (although the latter can also be done by BaseParse if it is appropriately configured, see below). Frame is provided with timestamp derived from upstream (as much as generally possible), duration obtained from configuration (see below), and offset if meaningful (in pull mode).
Note that GstBaseParseClass::handle_frame might receive any small: amount of input data when leftover data is being drained (e.g. at EOS).

As part of finish frame processing, just prior to actually pushing
the buffer in question, it is passed to GstBaseParseClass::pre_push_frame which gives subclass yet one last chance to examine buffer metadata, or to send some custom (tag) events, or to perform custom (segment) filtering.
During the parsing process GstBaseParseClass will handle both srcpad
and sinkpad events. They will be passed to subclass if GstBaseParseClass::sink_event or GstBaseParseClass::src_event implementations have been provided.

Shutdown phase#

BaseParse class calls GstBaseParseClass::stop to inform the subclass that data parsing will be stopped.

Subclass is responsible for providing pad template caps for source and sink pads. The pads need to be named “sink” and “src”. It also needs to set the fixed caps on srcpad, when the format is ensured (e.g. when base class calls subclass’ GstBaseParseClass::set_sink_caps function).

This base class uses %GST_FORMAT_DEFAULT as a meaning of frames. So, subclass conversion routine needs to know that conversion from %GST_FORMAT_TIME to %GST_FORMAT_DEFAULT must return the frame number that can be found from the given byte position.

BaseParse uses subclasses conversion methods also for seeking (or otherwise uses its own default one, see also below).

Subclass start and stop functions will be called to inform the beginning and end of data processing.

Things that subclass need to take care of:

Provide pad templates
Fixate the source pad caps when appropriate
Inform base class how big data chunks should be retrieved. This is done with set_min_frame_size() function.
Examine data chunks passed to subclass with GstBaseParseClass::handle_frame and pass proper frame(s) to finish_frame(), and setting src pad caps and timestamps on frame.
Provide conversion functions
Update the duration information with set_duration()
Optionally passthrough using set_passthrough()
Configure various baseparse parameters using set_average_bitrate(), set_syncable() and set_frame_rate().
In particular, if subclass is unable to determine a duration, but parsing (or specs) yields a frames per seconds rate, then this can be provided to BaseParse to enable it to cater for buffer time metadata (which will be taken from upstream as much as possible). Internally keeping track of frame durations and respective sizes that have been pushed provides BaseParse with an estimated bitrate. A default GstBaseParseClass::convert (used if not overridden) will then use these rates to perform obvious conversions. These rates are also used to update (estimated) duration at regular frame intervals.

Methods#

class BaseParse

add_index_entry(offset: int, ts: int, key: bool, force: bool) → bool#

Adds an entry to the index associating offset to ts. It is recommended to only add keyframe entries. force allows to bypass checks, such as whether the stream is (upstream) seekable, another entry is already “close” to the new entry, etc.

Parameters:

offset – offset of entry
ts – timestamp associated with offset
key – whether entry refers to keyframe
force – add entry disregarding sanity checks

convert_default(src_format: Format, src_value: int, dest_format: Format) → Tuple[bool, int]#

Default implementation of GstBaseParseClass::convert.

Parameters:

src_format – Format describing the source format.
src_value – Source value to be converted.
dest_format – Format defining the converted format.

do_convert(self, src_format: Format, src_value: int, dest_format: Format, dest_value: int) → bool#

Parameters:

src_format
src_value
dest_format
dest_value

do_detect(self, buffer: Buffer) → FlowReturn#

Parameters:: buffer

do_get_sink_caps(self, filter: Caps) → Caps#

Parameters:: filter

do_handle_frame(self, frame: BaseParseFrame) → Tuple[FlowReturn, int]#

Parameters:: frame

do_pre_push_frame(self, frame: BaseParseFrame) → FlowReturn#

Parameters:: frame

do_set_sink_caps(self, caps: Caps) → bool#

Parameters:: caps

do_sink_event(self, event: Event) → bool#

Parameters:: event

do_sink_query(self, query: Query) → bool#

Parameters:: query

do_src_event(self, event: Event) → bool#

Parameters:: event

do_src_query(self, query: Query) → bool#

Parameters:: query

do_start(self) → bool#

do_stop(self) → bool#

drain() → None#: Drains the adapter until it is empty. It decreases the min_frame_size to match the current adapter size and calls chain method until the adapter is emptied or chain returns with error.

Added in version 1.12.

finish_frame(frame: BaseParseFrame, size: int) → FlowReturn#

Collects parsed data and pushes it downstream. Source pad caps must be set when this is called.

If frame's out_buffer is set, that will be used as subsequent frame data, and size amount will be flushed from the input data. The output_buffer size can differ from the consumed size indicated by size.

Otherwise, size samples will be taken from the input and used for output, and the output’s metadata (timestamps etc) will be taken as (optionally) set by the subclass on frame's (input) buffer (which is otherwise ignored for any but the above purpose/information).

Note that the latter buffer is invalidated by this call, whereas the caller retains ownership of frame.

Parameters:

frame – a BaseParseFrame
size – consumed input data represented by frame

merge_tags(tags: TagList | None, mode: TagMergeMode) → None#

Sets the parser subclass’s tags and how they should be merged with any upstream stream tags. This will override any tags previously-set with merge_tags().

Note that this is provided for convenience, and the subclass is not required to use this and can still do tag handling on its own.

Added in version 1.6.

Parameters:

tags – a TagList to merge, or NULL to unset previously-set tags
mode – the TagMergeMode to use, usually GST_TAG_MERGE_REPLACE

push_frame(frame: BaseParseFrame) → FlowReturn#

Pushes the frame’s buffer downstream, sends any pending events and does some timestamp and segment handling. Takes ownership of frame’s buffer, though caller retains ownership of frame.

This must be called with sinkpad STREAM_LOCK held.

Parameters:: frame – a BaseParseFrame

set_average_bitrate(bitrate: int) → None#

Optionally sets the average bitrate detected in media (if non-zero), e.g. based on metadata, as it will be posted to the application.

By default, announced average bitrate is estimated. The average bitrate is used to estimate the total duration of the stream and to estimate a seek position, if there’s no index and the format is syncable (see set_syncable()).

Parameters:: bitrate – average bitrate in bits/second

set_duration(fmt: Format, duration: int, interval: int) → None#

Sets the duration of the currently playing media. Subclass can use this when it is able to determine duration and/or notices a change in the media duration. Alternatively, if interval is non-zero (default), then stream duration is determined based on estimated bitrate, and updated every interval frames.

Parameters:

fmt – Format.
duration – duration value.
interval – how often to update the duration estimate based on bitrate, or 0.

set_frame_rate(fps_num: int, fps_den: int, lead_in: int, lead_out: int) → None#

If frames per second is configured, parser can take care of buffer duration and timestamping. When performing segment clipping, or seeking to a specific location, a corresponding decoder might need an initial lead_in and a following lead_out number of frames to ensure the desired segment is entirely filled upon decoding.

Parameters:

fps_num – frames per second (numerator).
fps_den – frames per second (denominator).
lead_in – frames needed before a segment for subsequent decode
lead_out – frames needed after a segment

set_has_timing_info(has_timing: bool) → None#

Set if frames carry timing information which the subclass can (generally) parse and provide. In particular, intrinsic (rather than estimated) time can be obtained following a seek.

Parameters:: has_timing – whether frames carry timing information

set_infer_ts(infer_ts: bool) → None#

By default, the base class might try to infer PTS from DTS and vice versa. While this is generally correct for audio data, it may not be otherwise. Sub-classes implementing such formats should disable timestamp inferring.

Parameters:: infer_ts – True if parser should infer DTS/PTS from each other

set_latency(min_latency: int, max_latency: int) → None#

Sets the minimum and maximum (which may likely be equal) latency introduced by the parsing process. If there is such a latency, which depends on the particular parsing of the format, it typically corresponds to 1 frame duration.

If the provided values changed from previously provided ones, this will also post a LATENCY message on the bus so the pipeline can reconfigure its global latency.

Parameters:

min_latency – minimum parse latency
max_latency – maximum parse latency

set_min_frame_size(min_size: int) → None#

Subclass can use this function to tell the base class that it needs to be given buffers of at least min_size bytes.

Parameters:: min_size – Minimum size in bytes of the data that this base class should give to subclass.

set_passthrough(passthrough: bool) → None#

Set if the nature of the format or configuration does not allow (much) parsing, and the parser should operate in passthrough mode (which only applies when operating in push mode). That is, incoming buffers are pushed through unmodified, i.e. no GstBaseParseClass::handle_frame will be invoked, but GstBaseParseClass::pre_push_frame will still be invoked, so subclass can perform as much or as little is appropriate for passthrough semantics in GstBaseParseClass::pre_push_frame.

Parameters:: passthrough – True if parser should run in passthrough mode

set_pts_interpolation(pts_interpolate: bool) → None#

By default, the base class will guess PTS timestamps using a simple interpolation (previous timestamp + duration), which is incorrect for data streams with reordering, where PTS can go backward. Sub-classes implementing such formats should disable PTS interpolation.

Parameters:: pts_interpolate – True if parser should interpolate PTS timestamps

set_syncable(syncable: bool) → None#

Set if frame starts can be identified. This is set by default and determines whether seeking based on bitrate averages is possible for a format/stream.

Parameters:: syncable – set if frame starts can be identified

set_ts_at_offset(offset: int) → None#

This function should only be called from a handle_frame implementation.

BaseParse creates initial timestamps for frames by using the last timestamp seen in the stream before the frame starts. In certain cases, the correct timestamps will occur in the stream after the start of the frame, but before the start of the actual picture data. This function can be used to set the timestamps based on the offset into the frame data that the picture starts.

Added in version 1.2.

Parameters:: offset – offset into current buffer

Properties#

class BaseParse

props.disable_passthrough: bool#: The type of the None singleton.

Virtual Methods#

class BaseParse

do_convert(src_format: Format, src_value: int, dest_format: Format, dest_value: int) → bool#