Trait ImageDecoder

Source

pub trait ImageDecoder {
    type Adapter: ImageAdapter;
    type Buffer: AsRef<[u8]>;
    type Error;

    // Required method
    fn decode(
        &mut self,
        packet: &AttachmentPacket<<Self::Adapter as ImageAdapter>::PacketExtra, Self::Buffer>,
    ) -> Result<ImageFrame<<Self::Adapter as ImageAdapter>::PixelFormat, <Self::Adapter as ImageAdapter>::FrameExtra, Self::Buffer>, Self::Error>;
}

Expand description

One-shot still-image decoder — cover art, an embedded thumbnail, a poster frame.

No Stream in the name, and no rhythm to go with it. An attachment track delivers exactly one packet (see Demuxer’s attachment contract), that packet is a whole file, and decoding it produces exactly one picture. There is nothing to queue, nothing to drain, and no end-of-stream to signal — so decode takes the packet and returns the frame, instead of the send_packet / receive_frame split the two *StreamDecoder traits need. SubtitleDecoder keeps that split only because FFmpeg’s own subtitle API is push-shaped; nothing forces it here.

The frame has no timestamps, because ImageFrame has no seats for them.

§Where the input comes from

A container’s still images arrive as TrackKind::Attachment tracks: the track row says what codec the picture is in, and the track’s one AttachmentPacket carries its bytes. So the row opens the decoder — off-trait, per the module docs — and the packet is what decode is handed.

§`&mut self`

Decoding one image needs no state across calls; the exclusive borrow is here because a backend’s decoder handle is a mutable resource (FFmpeg’s AVCodecContext is), and because it lets a backend reuse one open decoder across several attachments of the same codec rather than reopening per picture.