Plan A 5K60 and multi-device display test plan

[oosuhada]

Goal

Build iBridge into a practical MacBook -> iMac software external display path. Current target is BetterDisplay Virtual 16:9 over wired LAN, with receiver fullscreen UX, bounded capture/encode latency, and input relay moving toward real external-display behavior.

Current status

• Branch is feat/plan-a-5k60-benchmark; latest handoff commit is 8bbdba0.
• The 2017 iMac is immediately connectable. scripts/start_2017_imac_receiver_macos.sh builds an Intel receiver locally, uploads it to ~/ibridge-remote/latest, starts it fullscreen with --hide-status, and verifies TCP 48320.
• Receiver UI/input MVP is implemented: standard macOS fullscreen/resizable window, Command-F fullscreen toggle, Esc fullscreen exit, pointer/key capture on the receiver surface, and input relay back to the source Mac over the existing TCP connection.
• Receiver input capture is now hardened with view-level handling, AppKit local/global monitors, and a listen-only CGEventTap for pointer events.
• Primary sender listens for receiver input lines and injects pointer/key events into the captured display with CGEvent.
• Source-side ibridge-primary Accessibility permission is now confirmed: latest sender printed input_accessibility_trusted=true.

Latest measured / verified result

• Current capture target is correct: display_index=1 display_id=12 width=3840 height=2160 frame=(-3840.0, -157.0, 3840.0, 2160.0) and input_target_display_id=12.
• Receiver diagnostics are active on the 2017 iMac: receiver_local_event_monitor_installed and receiver_event_tap_installed both logged.
• Synthesized iMac-local input self-test passed end to end: receiver logged receiver_input_event and input_sent; source logged input_event_received for both key and pointer events.
• Input MVP deployment smoke before diagnostics: latest receiver restarted on the 2017 iMac, DURATION=3 scripts/start_mbp_to_2017_imac_4k60.sh connected with input_relay=on, encoded 23/180 changed frames, had 0 send failures, and receiver logged 23 frames.
• Earlier 4K60 virtual-display motion comparison over direct 1GbE, HEVC 80Mbps: backpressure 1 reduced p95 encode from 52.175 ms to 19.090 ms with 0 send failures.

Changed areas

• apps/receiver-macos/Sources/iBridgeReceiverMac/main.swift
• apps/primary-macos/Sources/iBridgePrimary/main.swift
• scripts/package_macos_alpha.sh
• docs/18_ALPHA_RELEASE.md
• docs/current-work.md

Validation commands run

swift build --package-path apps/receiver-macos -c release --arch x86_64
swift build --package-path apps/receiver-macos -c release
swift build --package-path apps/primary-macos -c release
python3 apps/shared-protocol/test_protocol_v0.py
bash -n scripts/package_macos_alpha.sh scripts/start_ibridge_virtual_capture.sh scripts/start_2017_imac_receiver_macos.sh scripts/start_mbp_to_2017_imac_4k60.sh
scripts/package_macos_alpha.sh
scripts/start_2017_imac_receiver_macos.sh
DURATION=600 scripts/start_mbp_to_2017_imac_4k60.sh
ssh gabrieljang@100.89.104.119 'tail -80 ~/ibridge-remote/receiver-macos-48320.log'
ssh gabrieljang@100.89.104.119 'grep -n "receiver_input_event\|input_sent" ~/ibridge-remote/receiver-macos-runtime.log | tail -40'
grep -n 'input_event_received' benchmarks/runs/2026-05-17_1121_mbp_to_2017_imac_4k60/primary_console.txt | tail -80

Remaining blockers / next tests

• Manual operator test still needs the physical cursor to move onto the actual 2017 iMac receiver screen via Universal Control / Link Keyboard and Mouse. Moving into the MacBook source's left-side BetterDisplay Virtual 16:9 is not the same as moving onto the iMac receiver.
• Once manual physical input is confirmed, move MBP -> 2015 iMac Ethernet tests through 2560x1440, 3840x2160, then 4096x2304.
• Dual-iMac testing needs either a switch or two Ethernet adapters, plus multi-receiver session/profile management.
• MacBook Air repeats should use conservative profiles first; current evidence still treats 2560x1440@60 as the realistic Air default.

Notes for another machine

git fetch origin
git checkout feat/plan-a-5k60-benchmark
git pull --rebase
git submodule update --init --recursive reference/BetterDisplay
scripts/package_macos_alpha.sh
scripts/start_mbp_to_2017_imac_4k60.sh

[oosuhada]

Update: added encoder fallback recovery probe (b7c3eff).

Key result:

• Restarting user VTEncoderXPCService did not recover post-tiled 4096x2304/3200x1800 fallback performance.
• 4096x2304 stayed around p95 46.5ms even after service restart and 60s wait.
• 3200x1800 stayed slow around p95 41.8ms.
• 2560x1440 remained safe at p95 10.8ms.

Decision added to docs/logs: after tiled 5K60, high-detail fallback switching is unsafe until a stronger reset/session strategy is proven; current conservative emergency fallback is 2560x1440@60.