4.7 KiB

Raw Blame History

test-esp32-cam — emulating the OV2640 over QEMU, faithful

Phase plan for emulating the ESP32-CAM camera (OV2640 + DVP + I²S + DMA) as real QEMU peripherals — no library shim, no fakery. The upstream espressif/esp32-camera driver runs unmodified.

Reading order

autosearch/00_overview.md — problem statement and the three candidate paths (we picked Path B: real peripherals).
autosearch/01_state_of_the_art.md — what other projects do.
autosearch/02_qemu_lcgamboa_audit.md — confirms our QEMU fork has no camera/DVP/I²S today; lists every ESP32 device it does ship.
autosearch/03_browser_webcam_capture.md — getUserMedia → canvas → JPEG → WebSocket plumbing.
autosearch/04_proposed_architecture.md — end-to-end pipeline.
autosearch/05_open_questions.md — design decisions still open.
autosearch/06_existing_test_patterns.md — how the DHT22 / HC-SR04 live tests work; we mirror them.
autosearch/07_ov2640_sccb_spec.md — minimum register set the QEMU OV2640 device must implement.
autosearch/08_dvp_i2s_spec.md — exact I²S0 register sequence the esp32-camera driver issues, and the lldesc_t DMA descriptor format.
autosearch/09_qemu_build_blueprint.md — how a Phase-2 patch in third-party/qemu-lcgamboa/ reaches a running container.

Reference sources cloned into the tree

third-party/esp32-camera/ (Apache 2.0, cloned for offline reference). Used by autosearch and the C-side QEMU device once we ship Phase 2. Treat as read-only — never modify.

Phase plan

Phase	What lands	Status	Validating sketch / test
0	Research + tests skeleton	done	static metadata + WS-mock
1	`hw/i2c/esp32_ov2640.c` — SCCB chip-id	PASS	`test_sccb_probe_live.py`
2	`hw/misc/esp32_i2s_cam.c` — DMA + EOF	PASS	`test_dma_smoke_live.py`
3a	Host frame injection (webcam → backend → ctypes → QEMU → buf)	PASS	`test_frame_roundtrip_live.py`
3b	Upstream `esp_camera_init` + `fb_get` round-trip	xfail	`test_camera_live.py` (see autosearch/10)
4	CI build matrix + GH release upload	TODO	(build only)
5	Frontend: webcam hook + Camera button + missing pins	TODO	manual smoke

Layout

test-esp32-cam/
├── README.md                ← you are here
├── autosearch/              ← public-internet research
├── prototypes/              ← standalone validation runners
│   ├── echo_server.py       ← tiny WS echo for the HTML below
│   └── webcam_capture.html  ← getUserMedia → JPEG → WS prototype
├── sketches/
│   ├── camera_init/         ← Phase-3 reproducer (full upstream API)
│   ├── sccb_probe/          ← Phase-1 reproducer (I²C only, no I²S)
│   └── dma_smoke/           ← Phase-2 reproducer (raw I²S+DMA poke)
└── tests/
    ├── test_camera_metadata.py        ← static + frontend
    ├── test_camera_websocket.py       ← in-process WS mock
    ├── test_camera_live.py            ← upstream esp_camera API (xfail, see autosearch/10)
    ├── test_sccb_probe_live.py        ← Phase 1 PASS
    ├── test_dma_smoke_live.py         ← Phase 2 PASS
    ├── test_frame_roundtrip_live.py   ← Phase 3 e2e webcam → fb buffer PASS
    └── webcam_helper.py               ← OpenCV / PIL / synthetic JPEG source

Running

# Static + WS-mock layers (always green, no backend needed)
python -m pytest test/test-esp32-cam/tests -v

# Full live suite — needs a running backend with the QEMU library
# rebuilt to include the new peripherals.
cd backend && uvicorn app.main:app --port 8001 &
VELXIO_BACKEND_URL=http://localhost:8001 \
    python -m pytest test/test-esp32-cam/tests -v

The three *_live.py files all start as @expectedFailure. As each phase lands, the matching live test gets its decorator flipped off in the same diff that lands the QEMU change. That's how the test suite tracks emulation progress phase-by-phase.

Webcam → backend prototype (no QEMU)

pip install websockets
python test/test-esp32-cam/prototypes/echo_server.py &
# then open test/test-esp32-cam/prototypes/webcam_capture.html in a browser

Validates the browser-side path described in autosearch/03 without needing the backend or QEMU. If frame counter climbs and the preview panel updates, the transport is good.

4.7 KiB Raw Blame History