README

HSI pictures for the wiki's HSI page

Made 2026-11-07 for The HSI. Two kinds: bench_* are captures from the Beelink bench, the rest are offscreen renders.

Bench flight captures

bench_panel_approach.png and the other bench_* pictures except bench_no_data.png come from sim-bench session 8f9eb80b0daa (processes/sim_bench.md in maos-workspace), flown 2026-11-07 05:11-05:33 UTC. X-Plane flew fix-gateway's fp8-ksmx-rnav30-missed profile: the KSMX RNAV (GPS) RWY 30 via GVO from 2 nm before PAMJI, the missed approach, the 1 nm offsets on HANNU-WINCH and the WINCH hold. The bench ran pyEFIS dev 4c4f4dc (the same code as 980497e; the difference is wiki pages) and fix-gateway dev 5f1625c, with X-Plane position from the bench's standing RREF block (AER-4970).

bench_flight/docs_flight.py ran on the bench, detached. It changed no config and restarted nothing. Once the session was flying it published the approach to the flight-plan engine (the FP8 sessions' fp8_proc_route.py, in ../fp8_bench/bench_scripts/), sent ACT at PAMJI, and sent RESUME 5 s after the engine's auto-SUSP at the MAP. It grabbed the 1920 x 1080 screen with ffmpeg -f x11grab under the bench lock every 30 s and 3 s after each change of leg, phase, state, TO/FROM or nav source, then put the route block, NAVSRC and GPSSRC back as it found them. bench_flight/events.log is its log, and bench_flight/keys.jsonl holds the FIX keys it read just after each grab.

A grab can show the previous grab's state. About half the x11grab frames show the screen as it was at the previous grab, 5 to 30 s earlier, whatever the interval; the altitude readout against ALT shows it. The staleness tracks the grabs, not the clock, so it comes from the capture, not from the display. Each caption on the page is therefore taken from the readouts in the picture itself, and the frame it was cut from is listed here:

Picture	Frame	What it shows (readouts)
bench_panel_approach.png	051432	The whole panel, 1.9 nm before CIMSO (the FAF), 2,210 ft, LNAV, scaled to 1280 x 720
bench_gps_leg_off_course.png	052235	HANNU-WINCH, 0.86 nm right of course on the 1 nm offset: CRS 296, MAG 317
bench_susp_at_map.png	051829	Past RW30 (the MAP), auto-SUSP: FROM, CRS 319
bench_resume_missed.png	051840	After RESUME: the missed approach to HANNU, TO, CRS 319
bench_hold_1_crossing.png	053006	1.9 nm past WINCH, still on 300: CRS 122, the outbound course
bench_hold_2_entry_turn.png	053107	In the entry turn: MAG 069, CRS 122
bench_hold_3_outbound.png	053138	Outbound: MAG 147, CRS 122
bench_hold_4_turn_inbound.png	053239	Turning back inbound: MAG 087, CRS 080

The HSI crops are 420 x 410 px at (420, 640) of the panel.

The diamond right of the rose in these captures is the glidepath pointer. On this bench GSV for GPS comes from X-Plane's own GPS glidepath flag whatever GPSSRC is, so it shows beside the internal plan's lateral guidance. That is a defect, not intended behaviour: fix-gateway#137.

To capture again: queue the profile (POST /sessions with {"profile": "fp8-ksmx-rnav30-missed"}), copy bench_flight/docs_flight.py, fp8_proc_route.py and nfx.py into one directory on the bench, set its DEADLINE, and start it with the session id and the sim-bench service URL:

setsid nohup ~/pyefis-venv/bin/python docs_flight.py <session-id> http://<bench-host>:<port> > follow.out 2>&1 < /dev/null &

Renders

The other pictures are the real pyEFIS HSI widget (pyEFIS dev 980497e), rendered offscreen at 420 x 420 px. It is connected over netfix to a real fix-gateway (dev 5f1625c) running only netfix, compute and flightplan. The options are the Beelink bench panel's (screens/managed.yaml), applied through the screen builder's own apply_options. The face is drawn over a dark green fill standing in for the synthetic vision, because the bench panel sets bg_opacity: 30.

• GPS scenes (gps_*): hsi_docs.py publishes KSBA, DEANO, HENER, VTU (FAA CIFP coordinates) and flies a kinematic aircraft along the engine's own guidance: desired track plus a cross-track correction, 120 kt, turns of up to 3 deg/s, MAGVAR -11.449. Heading equals track, so there is no wind. The fly-by is the 61 degree turn at HENER. The hold scene adds an HM hold at HENER: right turns, inbound 073, 4 nm legs. The page shows the bench flight's leg and hold instead, so gps_leg_right_of_course.png and the three gps_hold_*.png are kept only as the script's output.
• NAV radio scenes (vor1_*, loc1_*): the script writes the radio keys an X-Plane or CAN-FIX source would publish (NAV1CRS, NAV1CDI, NAV1TF, NAV1DH, NAV1GSI, NAV1GSF, VOR1BRG).
• vor1_crs_selected.png selects CRS for the knob, as a touch on it does.
• evidence_guidance_failed.png is not on a page. It is the evidence for the defect where the HSI keeps drawing the bar and course while guidance has failed: position stopped, so the engine flags its guidance fail.

The fonts are DejaVu Sans; the bench panel asks for B612. The font_family falls back the same way wherever B612 is not installed.

Regenerate

# 1. A fix-gateway with only netfix + compute + flightplan enabled
cp -r <fix-gateway>/src/fixgw/config gwcfg
cat > gwcfg/preferences.yaml.custom <<'YAML'
includes:
  QUORUM_CONFIG: connections/quorum.yaml
enabled:
  QUORUM: false
  XPLANE: false
  DEMO: false
  ANNUNCIATE: false
  SYSTEM: false
  NETFIX: true
  COMPUTE: true
  FLIGHTPLAN: true
  STATE_PERSIST: false
YAML
(cd <fix-gateway> && PYTHONPATH=src python3 fixGw.py --config-file "$PWD/../gwcfg/default.yaml") &

# 2. The renders (scenes: gps, hold, vor, edit; default all)
cd docs/images/hsi
RENDER_FONT_DIR=/usr/share/fonts/truetype/dejavu QT_QPA_PLATFORM=offscreen \
  PYTHONPATH=../../../src python3 -u hsi_docs.py . gps hold vor edit

The script prints the FIX keys behind every picture as it saves it.

Bench, no data

bench_no_data.png is the Beelink bench (pyEFIS dev 980497e), grabbed with kill -USR1 under the bench lock on 2026-11-07 at 02:05 UTC, with no simulator running. It is cropped to the HSI from the 1920 x 1080 panel.