File size: 28,219 Bytes

4244f27

import os

import cython
from cython.cimports import libav as lib
from cython.cimports.av.bitstream import BitStreamFilterContext
from cython.cimports.av.codec.codec import Codec
from cython.cimports.av.codec.context import CodecContext, wrap_codec_context
from cython.cimports.av.codec.hwaccel import HWAccel
from cython.cimports.av.container.streams import StreamContainer
from cython.cimports.av.dictionary import Dictionary
from cython.cimports.av.error import err_check
from cython.cimports.av.packet import Packet
from cython.cimports.av.stream import Stream, wrap_stream
from cython.cimports.av.utils import dict_to_avdict, to_avrational
from cython.cimports.libc.stdint import uint8_t
from cython.cimports.libc.string import memcpy, memset


@cython.cfunc
def _set_codecpar_extradata(
    stream: cython.pointer[lib.AVStream],
    data: cython.pointer[uint8_t],
    size: cython.int,
):
    buf: cython.p_uchar = cython.cast(
        cython.p_uchar, lib.av_malloc(size + lib.AV_INPUT_BUFFER_PADDING_SIZE)
    )
    if buf == cython.NULL:
        raise MemoryError("Could not allocate extradata")

    memcpy(buf, data, size)
    memset(buf + size, 0, lib.AV_INPUT_BUFFER_PADDING_SIZE)

    lib.av_freep(cython.address(stream.codecpar.extradata))
    stream.codecpar.extradata = buf
    stream.codecpar.extradata_size = size


@cython.cfunc
def close_output(self: OutputContainer):
    if self.packet_ptr != cython.NULL and self._buffered_packets:
        buffered: list = self._buffered_packets
        self._buffered_packets = []
        packet: Packet
        for packet in buffered:
            self._mux_one(packet)

    self.streams = StreamContainer()
    if self._myflag & 12 == 4:  # enum.started and not enum.done
        # If the underlying Python IO file was already closed (e.g. during GC
        # finalization where cycle ordering is undefined), skip the trailer.
        if self.file is not None and getattr(self.file.file, "closed", False):
            self._myflag |= 8  # enum.done = True
            return
        # We must only ever call av_write_trailer *once*, otherwise we get a
        # segmentation fault. Therefore no matter whether it succeeds or not
        # we must absolutely set enum.done.
        try:
            self.err_check(lib.av_write_trailer(self.ptr))
        finally:
            if self.file is None and not (self.ptr.oformat.flags & lib.AVFMT_NOFILE):
                lib.avio_closep(cython.address(self.ptr.pb))
            self._myflag |= 8  # enum.done = True


@cython.final
@cython.cclass
class OutputContainer(Container):
    def __cinit__(self, *args, **kwargs):
        self.streams = StreamContainer()
        self._extradata_bsfs = {}
        self._buffered_packets = []
        with cython.nogil:
            self.packet_ptr = lib.av_packet_alloc()

    def __del__(self):
        close_output(self)

    def __dealloc__(self):
        with cython.nogil:
            lib.av_packet_free(cython.address(self.packet_ptr))

    def add_stream(
        self,
        codec_name,
        rate=None,
        options: dict | None = None,
        hwaccel: HWAccel | None = None,
        **kwargs,
    ):
        """add_stream(codec_name, rate=None, *, hwaccel=None)

        Creates a new stream from a codec name and returns it.
        Supports video, audio, and subtitle streams.

        :param codec_name: The name of a codec.
        :type codec_name: str
        :param dict options: Stream options.
        :param HWAccel hwaccel: Optional settings for hardware-accelerated encoding.
            Only applies to video streams (e.g. ``h264_vaapi``); software frames
            passed to :meth:`~av.codec.context.CodecContext.encode` are uploaded to
            the device automatically.
        :param \\**kwargs: Set attributes for the stream.
        :rtype: The new :class:`~av.stream.Stream`.

        .. warning::

            Configure every output stream before muxing the first packet. Writing
            the file header opens all stream codec contexts, after which structural
            properties such as format, dimensions, layout, and rate cannot change.

        """

        codec_obj: Codec = Codec(codec_name, "w")
        codec: cython.pointer[cython.const[lib.AVCodec]] = codec_obj.ptr

        # Assert that this format supports the requested codec.
        if not lib.avformat_query_codec(
            self.ptr.oformat, codec.id, lib.FF_COMPLIANCE_NORMAL
        ):
            raise ValueError(
                f"{self.format.name!r} format does not support {codec_obj.name!r} codec"
            )

        # Create new stream in the AVFormatContext, set AVCodecContext values.
        stream: cython.pointer[lib.AVStream] = lib.avformat_new_stream(self.ptr, codec)
        ctx: cython.pointer[lib.AVCodecContext] = lib.avcodec_alloc_context3(codec)

        # Now lets set some more sane video defaults
        if codec.type == lib.AVMEDIA_TYPE_VIDEO:
            ctx.pix_fmt = lib.AV_PIX_FMT_YUV420P
            ctx.width = kwargs.pop("width", 640)
            ctx.height = kwargs.pop("height", 480)
            ctx.bit_rate = kwargs.pop("bit_rate", 0)
            ctx.bit_rate_tolerance = kwargs.pop("bit_rate_tolerance", 128000)
            try:
                to_avrational(kwargs.pop("time_base"), cython.address(ctx.time_base))
            except KeyError:
                pass
            to_avrational(rate or 24, cython.address(ctx.framerate))

            stream.avg_frame_rate = ctx.framerate
            stream.time_base = ctx.time_base

        # Some sane audio defaults
        elif codec.type == lib.AVMEDIA_TYPE_AUDIO:
            out: cython.pointer[cython.const[cython.void]] = cython.NULL
            lib.avcodec_get_supported_config(
                cython.NULL,
                codec,
                lib.AV_CODEC_CONFIG_SAMPLE_FORMAT,
                0,
                cython.address(out),
                cython.NULL,
            )
            if out:
                ctx.sample_fmt = cython.cast(cython.pointer[lib.AVSampleFormat], out)[0]
            ctx.bit_rate = kwargs.pop("bit_rate", 0)
            ctx.bit_rate_tolerance = kwargs.pop("bit_rate_tolerance", 32000)
            try:
                to_avrational(kwargs.pop("time_base"), cython.address(ctx.time_base))
            except KeyError:
                pass

            if rate is None:
                ctx.sample_rate = 48000
            elif type(rate) is int:
                ctx.sample_rate = rate
            else:
                raise TypeError("audio stream `rate` must be: int | None")
            stream.time_base = ctx.time_base
            lib.av_channel_layout_default(cython.address(ctx.ch_layout), 2)

        # Some formats want stream headers to be separate
        if self.ptr.oformat.flags & lib.AVFMT_GLOBALHEADER:
            ctx.flags |= lib.AV_CODEC_FLAG_GLOBAL_HEADER

        # Initialise stream codec parameters to populate the codec type.
        #
        # Subsequent changes to the codec context will be applied just before
        # encoding starts in `start_encoding()`.
        err_check(lib.avcodec_parameters_from_context(stream.codecpar, ctx))

        # Construct the user-land stream
        py_codec_context: CodecContext = wrap_codec_context(ctx, codec, hwaccel)
        py_stream: Stream = wrap_stream(self, stream, py_codec_context)
        self.streams.add_stream(py_stream)

        if options:
            py_stream.options.update(options)

        for k, v in kwargs.items():
            setattr(py_stream, k, v)

        return py_stream

    def add_mux_stream(self, codec_name: str, rate=None, **kwargs) -> Stream:
        """add_mux_stream(codec_name, rate=None)

        Creates a new stream for muxing pre-encoded data without creating a
        :class:`.CodecContext`. Use this when you want to mux packets that were
        already encoded externally and no encoding/decoding is needed.

        :param codec_name: The name of a codec.
        :type codec_name: str
        :param \\**kwargs: Set attributes for the stream (e.g. ``width``, ``height``,
            ``time_base``).
        :rtype: The new :class:`~av.stream.Stream`.

        """
        # Find the codec to get its id and type (try encoder first, then decoder).
        codec_name_bytes: bytes = codec_name.encode()
        codec: cython.pointer[cython.const[lib.AVCodec]] = (
            lib.avcodec_find_encoder_by_name(codec_name_bytes)
        )
        codec_descriptor: cython.pointer[cython.const[lib.AVCodecDescriptor]] = (
            cython.NULL
        )
        if codec == cython.NULL:
            codec = lib.avcodec_find_decoder_by_name(codec_name_bytes)
        if codec == cython.NULL:
            codec_descriptor = lib.avcodec_descriptor_get_by_name(codec_name_bytes)
            if codec_descriptor == cython.NULL:
                raise ValueError(f"Unknown codec: {codec_name!r}")

        codec_id: lib.AVCodecID
        codec_type: lib.AVMediaType
        if codec != cython.NULL:
            codec_id = codec.id
            codec_type = codec.type
        else:
            codec_id = codec_descriptor.id
            codec_type = codec_descriptor.type

        # Assert that this format supports the requested codec.
        if not lib.avformat_query_codec(
            self.ptr.oformat, codec_id, lib.FF_COMPLIANCE_NORMAL
        ):
            raise ValueError(
                f"{self.format.name!r} format does not support {codec_name!r} codec"
            )

        # Create stream with no codec context.
        stream: cython.pointer[lib.AVStream] = lib.avformat_new_stream(
            self.ptr, cython.NULL
        )
        if stream == cython.NULL:
            raise MemoryError("Could not allocate stream")

        stream.codecpar.codec_id = codec_id
        stream.codecpar.codec_type = codec_type

        if codec_type == lib.AVMEDIA_TYPE_VIDEO:
            stream.codecpar.width = kwargs.pop("width", 0)
            stream.codecpar.height = kwargs.pop("height", 0)
            if rate is not None:
                to_avrational(rate, cython.address(stream.avg_frame_rate))
        elif codec_type == lib.AVMEDIA_TYPE_AUDIO:
            if rate is not None:
                if type(rate) is int:
                    stream.codecpar.sample_rate = rate
                else:
                    raise TypeError("audio stream `rate` must be: int | None")

        # Construct the user-land stream (no codec context).
        py_stream: Stream = wrap_stream(self, stream, None)
        self.streams.add_stream(py_stream)

        for k, v in kwargs.items():
            setattr(py_stream, k, v)

        return py_stream

    def add_stream_from_template(
        self, template: Stream, opaque: bool | None = None, **kwargs
    ):
        """
        Creates a new stream from a template. Supports video, audio, subtitle, data and attachment streams.

        :param template: Copy codec from another :class:`~av.stream.Stream` instance.
        :param opaque: If True, copy opaque data from the template's codec context.
        :param \\**kwargs: Set attributes for the stream.
        :rtype: The new :class:`~av.stream.Stream`.
        """
        if opaque is None:
            opaque = template.type != "video"

        if template.codec_context is None:
            return self._add_stream_without_codec_from_template(template, **kwargs)

        codec_obj: Codec
        if opaque:  # Copy ctx from template.
            codec_obj = template.codec_context.codec
        else:  # Construct new codec object.
            codec_obj = Codec(template.codec_context.codec.name, "w")

        codec: cython.pointer[cython.const[lib.AVCodec]] = codec_obj.ptr

        # Assert that this format supports the requested codec.
        if not lib.avformat_query_codec(
            self.ptr.oformat, codec.id, lib.FF_COMPLIANCE_NORMAL
        ):
            raise ValueError(
                f"{self.format.name!r} format does not support {codec_obj.name!r} codec"
            )

        # Create new stream in the AVFormatContext, set AVCodecContext values.
        stream: cython.pointer[lib.AVStream] = lib.avformat_new_stream(self.ptr, codec)
        ctx: cython.pointer[lib.AVCodecContext] = lib.avcodec_alloc_context3(codec)

        err_check(lib.avcodec_parameters_to_context(ctx, template.ptr.codecpar))
        # Reset the codec tag assuming we are remuxing.
        ctx.codec_tag = 0

        # Copy the template's stream time_base
        stream.time_base = template.ptr.time_base
        ctx.time_base = template.ptr.time_base

        # Some formats want stream headers to be separate
        if self.ptr.oformat.flags & lib.AVFMT_GLOBALHEADER:
            ctx.flags |= lib.AV_CODEC_FLAG_GLOBAL_HEADER

        # Copy flags If we're creating a new codec object. This fixes some muxing issues.
        # Overwriting `ctx.flags |= lib.AV_CODEC_FLAG_GLOBAL_HEADER` is intentional.
        if not opaque:
            ctx.flags = template.codec_context.flags

        # Initialize stream codec parameters to populate the codec type. Subsequent changes to
        # the codec context will be applied just before encoding starts in `start_encoding()`.
        err_check(lib.avcodec_parameters_from_context(stream.codecpar, ctx))

        # Construct the user-land stream
        py_codec_context: CodecContext = wrap_codec_context(ctx, codec, None)
        py_codec_context._ctxflags |= 1  # _template_initialized = True
        py_stream: Stream = wrap_stream(self, stream, py_codec_context)
        self.streams.add_stream(py_stream)

        for k, v in kwargs.items():
            setattr(py_stream, k, v)

        return py_stream

    def _add_stream_without_codec_from_template(
        self, template: Stream, **kwargs
    ) -> Stream:
        codec_type: cython.int = template.ptr.codecpar.codec_type
        if codec_type not in {lib.AVMEDIA_TYPE_ATTACHMENT, lib.AVMEDIA_TYPE_DATA}:
            raise ValueError(
                f"template stream of type {template.type} has no codec context"
            )

        stream: cython.pointer[lib.AVStream] = lib.avformat_new_stream(
            self.ptr, cython.NULL
        )
        if stream == cython.NULL:
            raise MemoryError("Could not allocate stream")

        err_check(lib.avcodec_parameters_copy(stream.codecpar, template.ptr.codecpar))

        # Mirror basic properties that are not derived from a codec context.
        stream.time_base = template.ptr.time_base
        stream.start_time = template.ptr.start_time
        stream.duration = template.ptr.duration
        stream.disposition = template.ptr.disposition

        py_stream: Stream = wrap_stream(self, stream, None)
        self.streams.add_stream(py_stream)

        py_stream.metadata = dict(template.metadata)

        for k, v in kwargs.items():
            setattr(py_stream, k, v)

        return py_stream

    def add_attachment(self, name: str, mimetype: str, data: bytes):
        """
        Create an attachment stream and embed its payload into the container header.

        - Only supported by formats that support attachments (e.g. Matroska).
        - No per-packet muxing is required; attachments are written at header time.
        """
        # Create stream with no codec (attachments are codec-less).
        stream: cython.pointer[lib.AVStream] = lib.avformat_new_stream(
            self.ptr, cython.NULL
        )
        if stream == cython.NULL:
            raise MemoryError("Could not allocate stream")

        stream.codecpar.codec_type = lib.AVMEDIA_TYPE_ATTACHMENT
        stream.codecpar.codec_id = lib.AV_CODEC_ID_NONE

        # Allocate and copy payload into codecpar.extradata.
        payload_size: cython.size_t = len(data)
        if payload_size:
            buf = cython.cast(cython.p_uchar, lib.av_malloc(payload_size + 1))
            if buf == cython.NULL:
                raise MemoryError("Could not allocate attachment data")
            # Copy bytes.
            for i in range(payload_size):
                buf[i] = data[i]
            buf[payload_size] = 0
            stream.codecpar.extradata = cython.cast(cython.p_uchar, buf)
            stream.codecpar.extradata_size = cython.cast(cython.int, payload_size)

        # Wrap as user-land stream.
        meta_ptr = cython.address(stream.metadata)
        err_check(lib.av_dict_set(meta_ptr, b"filename", name.encode(), 0))
        mime_bytes = mimetype.encode()
        err_check(lib.av_dict_set(meta_ptr, b"mimetype", mime_bytes, 0))

        py_stream: Stream = wrap_stream(self, stream, None)
        self.streams.add_stream(py_stream)
        return py_stream

    def add_data_stream(self, codec_name=None, options: dict | None = None):
        """add_data_stream(codec_name=None)

        Creates a new data stream and returns it.

        :param codec_name: Optional name of the data codec (e.g. 'klv')
        :type codec_name: str | None
        :param dict options: Stream options.
        :rtype: The new :class:`~av.data.stream.DataStream`.
        """
        codec: cython.pointer[cython.const[lib.AVCodec]] = cython.NULL
        codec_descriptor: cython.pointer[cython.const[lib.AVCodecDescriptor]] = (
            cython.NULL
        )

        if codec_name is not None:
            codec_name_bytes: bytes = codec_name.encode()
            codec = lib.avcodec_find_encoder_by_name(codec_name_bytes)
            if codec == cython.NULL:
                codec = lib.avcodec_find_decoder_by_name(codec_name_bytes)
            if codec == cython.NULL:
                codec_descriptor = lib.avcodec_descriptor_get_by_name(codec_name_bytes)
                if codec_descriptor == cython.NULL:
                    raise ValueError(f"Unknown data codec: {codec_name}")

            # Verify format supports this codec
            codec_id = codec.id if codec != cython.NULL else codec_descriptor.id
            if not lib.avformat_query_codec(
                self.ptr.oformat, codec_id, lib.FF_COMPLIANCE_NORMAL
            ):
                raise ValueError(
                    f"{self.format.name!r} format does not support {codec_name!r} codec"
                )

        # Create new stream in the AVFormatContext
        stream: cython.pointer[lib.AVStream] = lib.avformat_new_stream(self.ptr, codec)
        if stream == cython.NULL:
            raise MemoryError("Could not allocate stream")

        # Set up codec context and parameters
        ctx: cython.pointer[lib.AVCodecContext] = cython.NULL
        if codec != cython.NULL:
            ctx = lib.avcodec_alloc_context3(codec)
            if ctx == cython.NULL:
                raise MemoryError("Could not allocate codec context")

            # Some formats want stream headers to be separate
            if self.ptr.oformat.flags & lib.AVFMT_GLOBALHEADER:
                ctx.flags |= lib.AV_CODEC_FLAG_GLOBAL_HEADER

            # Initialize stream codec parameters
            err_check(lib.avcodec_parameters_from_context(stream.codecpar, ctx))
        else:
            # No codec available - set basic parameters for data stream
            stream.codecpar.codec_type = lib.AVMEDIA_TYPE_DATA
            if codec_descriptor != cython.NULL:
                stream.codecpar.codec_id = codec_descriptor.id

        # Construct the user-land stream
        py_codec_context: CodecContext | None = None
        if ctx != cython.NULL:
            py_codec_context = wrap_codec_context(ctx, codec, None)

        py_stream: Stream = wrap_stream(self, stream, py_codec_context)
        self.streams.add_stream(py_stream)

        if options:
            py_stream.options.update(options)

        return py_stream

    @cython.ccall
    def start_encoding(self):
        """Write the file header! Called automatically."""
        if self._myflag & 4:  # started
            return

        # TODO: This does NOT handle options coming from 3 sources.
        # This is only a rough approximation of what would be cool to do.
        used_options: set = set()
        stream: Stream

        # Finalize and open all streams.
        for stream in self.streams:
            ctx = stream.codec_context
            if ctx is not None and not ctx.is_open:
                for k, v in self.options.items():
                    ctx.options.setdefault(k, v)

                if not (ctx._ctxflags & 1):  # template_initialized
                    ctx.open()

                    # Track option consumption.
                    for k in self.options:
                        if k not in ctx.options:
                            used_options.add(k)

            stream._finalize_for_output()

        # Open the output file, if needed.
        name_obj: bytes = os.fsencode(self.name if self.file is None else "")
        name: cython.p_char = name_obj
        if self.ptr.pb == cython.NULL and not self.ptr.oformat.flags & lib.AVFMT_NOFILE:
            err_check(
                lib.avio_open(cython.address(self.ptr.pb), name, lib.AVIO_FLAG_WRITE)
            )

        # Copy the metadata dict.
        dict_to_avdict(
            cython.address(self.ptr.metadata),
            self.metadata,
            encoding=self.metadata_encoding,
            errors=self.metadata_errors,
        )

        all_options: Dictionary = Dictionary(self.options, self.container_options)
        options: Dictionary = all_options.copy()
        self.err_check(lib.avformat_write_header(self.ptr, cython.address(options.ptr)))

        # Track option usage...
        for k in all_options:
            if k not in options:
                used_options.add(k)

        # ... and warn if any weren't used.
        unused_options = {
            k: v for k, v in self.options.items() if k not in used_options
        }
        if unused_options:
            import logging

            log = logging.getLogger(__name__)
            log.warning(f"Some options were not used: {unused_options}")

        self._myflag |= 4

    @property
    def supported_codecs(self):
        """
        Returns a set of all codecs this format supports.
        """
        result: set = set()
        codec: cython.pointer[cython.const[lib.AVCodec]] = cython.NULL
        opaque: cython.p_void = cython.NULL

        while True:
            codec = lib.av_codec_iterate(cython.address(opaque))
            if codec == cython.NULL:
                break

            if (
                lib.avformat_query_codec(
                    self.ptr.oformat, codec.id, lib.FF_COMPLIANCE_NORMAL
                )
                == 1
            ):
                result.add(codec.name)

        return result

    @property
    def default_video_codec(self):
        """
        Returns the default video codec this container recommends.
        """
        return lib.avcodec_get_name(self.format.optr.video_codec)

    @property
    def default_audio_codec(self):
        """
        Returns the default audio codec this container recommends.
        """
        return lib.avcodec_get_name(self.format.optr.audio_codec)

    @property
    def default_subtitle_codec(self):
        """
        Returns the default subtitle codec this container recommends.
        """
        return lib.avcodec_get_name(self.format.optr.subtitle_codec)

    def close(self):
        close_output(self)

    def mux(self, packets):
        # We accept either a Packet, or a sequence of packets. This should smooth out
        # the transition to the new encode API which returns a sequence of packets.
        if isinstance(packets, Packet):
            self.mux_one(packets)
        else:
            for packet in packets:
                self.mux_one(packet)

    def mux_one(self, packet: Packet):
        if not (self._myflag & 4) and self._buffer_for_extradata(packet):
            return

        self._mux_one(packet)

    @cython.cfunc
    def _mux_one(self, packet: Packet):
        self.start_encoding()

        # Assert the packet is in stream time.
        if (
            packet.ptr.stream_index < 0
            or cython.cast(cython.uint, packet.ptr.stream_index) >= self.ptr.nb_streams
        ):
            raise ValueError("Bad Packet stream_index.")

        stream: cython.pointer[lib.AVStream] = self.ptr.streams[packet.ptr.stream_index]
        packet._rebase_time(stream.time_base)

        # Make another reference to the packet, as `av_interleaved_write_frame()`
        # takes ownership of the reference.
        self.err_check(lib.av_packet_ref(self.packet_ptr, packet.ptr))

        with cython.nogil:
            ret: cython.int = lib.av_interleaved_write_frame(self.ptr, self.packet_ptr)
        self.err_check(ret)

    @cython.cfunc
    def _buffer_for_extradata(self, packet: Packet):
        """Buffer ``packet`` until extradata is known for all mux streams that
        need it. Returns True if the packet was buffered (caller should stop)."""
        if not (self._myflag & 16):  # extradata_planned
            self._myflag |= 16
            if self.ptr.oformat.flags & lib.AVFMT_GLOBALHEADER:
                stream: Stream
                for stream in self.streams:
                    if (
                        stream.codec_context is not None
                        or stream.ptr.codecpar.extradata != cython.NULL
                    ):
                        continue
                    try:
                        bsf = BitStreamFilterContext(
                            "extract_extradata", in_stream=stream
                        )
                    except Exception:
                        continue  # Codec does not support extradata extraction.
                    self._extradata_bsfs[stream.ptr.index] = bsf

        if not self._extradata_bsfs:
            return False  # Nothing to wait for; mux normally.

        self._try_extract_extradata(packet)
        self._buffered_packets.append(packet)
        if self._extradata_bsfs:
            return True  # Still waiting on some stream's extradata.

        # All extradata is resolved: write the header and flush buffered packets.
        buffered: list = self._buffered_packets
        self._buffered_packets = []
        buffered_packet: Packet
        for buffered_packet in buffered:
            self._mux_one(buffered_packet)
        return True

    @cython.cfunc
    def _try_extract_extradata(self, packet: Packet):
        idx: cython.int = packet.ptr.stream_index
        if idx not in self._extradata_bsfs:
            return

        bsf_wrapper: BitStreamFilterContext = self._extradata_bsfs[idx]
        bsf: cython.pointer[lib.AVBSFContext] = bsf_wrapper.ptr

        tmp: cython.pointer[lib.AVPacket] = lib.av_packet_alloc()
        if tmp == cython.NULL:
            raise MemoryError("Could not allocate packet")

        size: cython.size_t = 0
        sd: cython.pointer[uint8_t]
        try:
            # Clone the packet so the filter does not consume the caller's data.
            if lib.av_packet_ref(tmp, packet.ptr) < 0:
                return

            if lib.av_bsf_send_packet(bsf, tmp) < 0:
                lib.av_packet_unref(tmp)  # send failed; we still own the ref
                return

            # The filter rejects packets that are already length-prefixed rather
            # than annex-b, returning an error here; treat that and EOF/EAGAIN
            # alike as "no in-band extradata".
            while lib.av_bsf_receive_packet(bsf, tmp) == 0:
                sd = lib.av_packet_get_side_data(
                    tmp, lib.AV_PKT_DATA_NEW_EXTRADATA, cython.address(size)
                )
                if sd != cython.NULL and size > 0:
                    _set_codecpar_extradata(
                        self.ptr.streams[idx], sd, cython.cast(cython.int, size)
                    )
                    lib.av_packet_unref(tmp)
                    break
                lib.av_packet_unref(tmp)
        finally:
            lib.av_packet_free(cython.address(tmp))
            # A stream's first packet is the only reliable place to find in-band
            # parameter sets, so stop waiting on this stream regardless of the
            # result, falling back to the muxer's default behavior.
            del self._extradata_bsfs[idx]