# `BB.Camera.AVF`
[🔗](https://github.com/beam-bots/bb_camera_avf/blob/main/lib/bb/camera/avf.ex#L5)

A camera on macOS, captured through ffmpeg's AVFoundation input.

Declared like any other sensor:

    topology do
      link :head do
        sensor :eye, {BB.Camera.AVF, device: "FaceTime HD Camera", width: 1280, height: 720}
      end
    end

Consumers never talk to this module. They go through `BB.Camera`, which
refcounts them and asks the driver to start capturing when the first one
arrives and to stop when the last one leaves. Two browser tabs on the same
camera produce one ffmpeg, not two.

## What it captures

JPEG, encoded by ffmpeg. AVFoundation hands ffmpeg decoded buffers even from
a USB webcam that sends MJPEG, so there is no passthrough path: every frame
costs an encode. `:input_format` chooses which pixel format AVFoundation is
asked for, and `:quality` the encoder's trade between size and fidelity.

## Camera access

macOS grants camera access to an *app*, and a process started from a
terminal inherits the terminal's grant. The first capture from a new
terminal app raises the system prompt. A VM started by something with no
user interface - a launchd job, a daemon, an agent runner - may get neither
a prompt nor frames: ffmpeg opens the camera and then waits forever.

So capture gives up when no frame arrives within `:start_timeout_ms`, kills
ffmpeg, and reports `{:no_frames, ms}` as the reason in
`BB.Error.Hardware.Camera.CaptureFailed`. Nothing can tell that case apart
from a camera that simply sends nothing, so the error names the timeout,
not a cause.

## What it reports

Capture is confirmed to the hub on the **first whole image**, not when ffmpeg
starts. `Port.open/2` returning says only that ffmpeg exists — a missing
camera would otherwise announce `BeginCapture` a moment before failing.

Dimensions come from the `SOF` marker of each frame, which is what was
actually encoded. They are not always the numbers in the DSL: when
AVFoundation refuses ffmpeg's configuration, ffmpeg warns and captures in
the camera's default format instead, and `BB.Camera.stream_info/2` reports
the real size.

The frame rate in `stream_info` is the one that was **requested**, not the
negotiated one. This driver does not query the device's rate back and does
not parse it out of ffmpeg's console output, so claiming it as actual would
be a guess dressed up as a measurement. What the frames actually did is
measured by `BB.Camera.Hub`, which sees every one of them, and reported
separately in `BB.Message.Camera.CaptureState`.

Nothing here enumerates the camera's capabilities. Ask for a size or rate
the camera does not list and ffmpeg exits, printing every mode the camera
does list on the console - which is also the quickest way to find them out.

## Frames do not go on pubsub

A robot's pubsub carries capture lifecycle only:
`BB.Message.Camera.BeginCapture` and `EndCapture`, published by the hub.
Frame bytes travel over `BB.Camera`'s bounded media interface, and never
appear in an event or an error.

## Options

See `options_schema/0`. The ones most worth knowing about are `:device`,
`:width`, `:height` and `:frame_rate`; the rest are bounds with defaults that
suit a desktop and may want lowering on a small target.

Any of them can be filled from a parameter with `param/1`, and changing that
parameter while the robot runs reconfigures the camera - see
`handle_options/2`.

## Limitations

- **macOS only.** It drives AVFoundation through ffmpeg. The ffmpeg-tagged
  tests run anywhere, against a synthetic source.
- **Cameras are chosen by name or index**, not by unique ID; see
  `BB.Camera.AVF.Command`.
- **Timestamps are arrival times.** The raw MJPEG stream carries no
  timestamps of its own, so a frame is stamped when the read completing it
  reached the BEAM. See `BB.Camera.FFmpeg.Capture` for what that does and
  does not mean.
- **A failed capture is not retried.** The failure is reported and capture
  ends; reopening the device is the consumer's decision, taken by subscribing
  again. A driver that quietly reopened hardware on its own would be claiming
  to capture during a window where it is not.
- **Changing an option that reaches ffmpeg restarts it**, because ffmpeg
  configures the camera once, when it opens it. Capture ends and
  begins again around the change, with the events to match; it is not a
  seamless transition and does not pretend to be.
- **ffmpeg's diagnostics go to the VM's console**, not into the error. They
  cannot be merged into the frame stream without corrupting it.

# `handle_options`

Take up options that changed while the robot was running.

Reached when a `param/1` reference in the camera's DSL options resolves to a
new value. Only a change that alters the ffmpeg invocation restarts capture:
ffmpeg configures the camera once, when it opens it, so a new size or rate
means a new ffmpeg. Capture stops and starts again around the change; it
is not a seamless transition and does not pretend to be. An option that
doesn't reach the command line - a framer bound, say - takes effect without
disturbing the stream.

**New options this driver cannot use stop the camera.** `handle_options/2`
has no way to refuse a change and stay running, so a value that satisfies
the parameter's declared type but not this driver - a width of `-1` against
a `:pos_integer` - takes the sensor down, and down again on each restart,
because the parameter was stored before anyone was asked about it. Declare
parameters with bounds that match the options they feed.

# `stream_info`

The format this camera is declared to produce.

Pure: it is called during discovery against robots that are not running, so
it reads the options and nothing else. Options that still hold unresolved
parameter references cannot imply a format, and produce an error rather than
a guess.

---

*Consult [api-reference.md](api-reference.md) for complete listing*
