[tridge]

RTSP connections previously entered publisher admission, preventing clients such as Mission Planner from viewing proxied streams by RTSP URL. Reserve /view/vN for authenticated playback of the existing MPEG-TS ring as RTP, preserving publisher splices and disconnecting viewers that fall behind.

Bound unauthenticated admission, support RTCP keepalives and TCP transport negotiation, and provide a copyable Mission Planner pipeline in the video UI.

[AP-Review]

Deprecated — see below for the updated review.

Previous review (2026-10-20)

Automated review note — AI-generated (Claude+Codex), validated against the live diff. Please sanity-check before acting.
Verdict: COMMENT

Reviewed at head 4d3b5c8606.
Full report: https://firmware.ardupilot.org/Tools/APReview/DevCallReviews/PRReviews/ardupilot/supportproxy/47/1.html#prSupportProxy-47

This adds RTSP-over-TCP playback for viewers. It builds cleanly and its new tests pass, but two viewer-side issues should be fixed.

Should fix

• videortspview.cpp:277 Closing viewer busy-spins the video child: when closing_ it returns without recv, but EPOLLIN stays level-triggered, so leftover input or a half-close after TEARDOWN spins the shared child at full CPU until the stall timeout (~10s). Drain and discard input (return false on EOF) while closing, or drop EPOLLIN.

Low

• videortspview.cpp:255 PAUSE then PLAY reloads read_pos_ from scanner.join_offset(), rewinding to the latest keyframe and resending already-delivered TS data (PTS goes backwards in the same RTP session). Keep read_pos_ on resume while it is still resident, and re-anchor only if it was lapped.

[AP-Review]

Automated review note — AI-generated (Claude+Codex), validated against the live diff. Please sanity-check before acting.
Verdict: ACCEPT

Reviewed at head 7e45231576.
Full report: https://firmware.ardupilot.org/Tools/APReview/DevCallReviews/PRReviews/ardupilot/supportproxy/47/2.html#prSupportProxy-47

This RTSP-over-TCP viewer change now looks good to merge: it builds cleanly at 7e45231 and the RTSP viewer tests pass under the review harness.

Since the last review (at 4d3b5c8), commit 7e45231 fixed both earlier issues. A closing viewer now drains its input instead of busy-spinning the video child, and PAUSE/PLAY resumes from the buffered position instead of replaying from the last keyframe. New regression tests fail on the old code and pass now.

[rmackay9]

This was tested successfully yesterday on my aerial photography copter, thanks!