divorce
b735900f2e
ddat interop with dat js cable interop with cabal js ddat's frontend = dat-thang cable fe = cable-thang dillo support based on cel's dillo-dat bootstrap node xplatform support Signed-off-by: divo <418@cryspace.space>
3.1 KiB
Cable live media
Cable live media travels beside Cable room chat and Dat archives. A live session is not an archive, does not append to a user's permanent chat feed, and does not advance a Dat version. It is sent as volatile Cable records over authenticated Cesspool/uTP peer sessions. The current audio path fans packets out from the broadcaster to authenticated room peers; multi-hop relay is not yet wired.
Audio, first slice
cable-thang's BROADCAST button opens the default SDL microphone as 48 kHz,
mono, signed 16-bit PCM. FFmpeg's AAC-LC encoder converts 1,024-sample frames
to 96 kbps AAC and wraps each access unit in an ADTS header. The cable-live
packet header carries a random session id, sequence number, GUI timestamp,
media kind, and bounded payload. The C viewer validates packets, decodes AAC,
places PCM in a half-second maximum SDL playback queue, and plots those decoded
samples in the waveform panel. Ctrl+Shift+A toggles the local broadcast.
Received room audio begins playing as soon as its start packet is seen.
Cable audio and video are experimental and disabled by default. A normal SDL
build has no FFmpeg dependency. Enable both together with
make ENABLE_SDL2=1 ENABLE_CABLE_MEDIA=1; the makefile requires compatible
FFmpeg 5.0+ development files (libavcodec, libavformat, libavutil,
libswresample, and libswscale). To build the SDL clients without compiling
either media feature, use make ENABLE_SDL2=1 ENABLE_CABLE_MEDIA=0.
On Debian or Ubuntu, install the development headers and enable the feature:
sudo apt install libavcodec-dev libavformat-dev libavutil-dev \
libswresample-dev libswscale-dev
make ENABLE_SDL2=1 ENABLE_CABLE_MEDIA=1
With media enabled, the Cable video pane can listen to a local OBS MPEG-TS
output. Click OBS INPUT or press Ctrl+Shift+V; Cable listens on
udp://127.0.0.1:5000, decodes the video on a
worker thread, scales the local preview to at most 640x360, and displays the
latest frame. New frames are reduced to 160x90 JPEG and sent in 4,000-byte
fragments over authenticated Cesspool/uTP sessions; receivers reassemble and
decode them in the video pane. Transmission is capped at five frames per
second. In OBS, add a Custom FFmpeg Output with container mpegts, video
encoder libx264, and output URL udp://127.0.0.1:5000?pkt_size=1316. OBS
input provides local preview and direct broadcast to authenticated peers.
Camera capture without OBS and multi-hop forwarding are not implemented.
Storage boundary
No media packets are written to .dat, test-only Cabal JSON chat feeds, or the
versioned archive metadata/content feeds. Closing the room discards the
decoder and playback queue. This avoids turning a live broadcast into
permanent archive history. Hypervision's
historic approach chunked WebM media and appended the chunks to a dedicated
Hypercore feed. That is a useful precedent for chunking and peer redistribution,
but an append-only archive is not the right retention policy for this live
session path.
The packet kinds cover audio and video start/frame/stop. Direct camera capture, multi-hop relay, remote stream selection, and adaptive bitrate are follow-on work.