uopi/docs/WORK_PLAN.md

Work Plan — uopi

Phases Overview

Phase	Name	Focus	Milestone
0	Scaffold	Repo structure, build toolchain, CI	Empty binary serves embedded index page
1	Core Backend	Broker, WebSocket protocol, REST skeleton	Live signal fan-out working end-to-end
2	EPICS Data Source	CA connect, monitor, metadata, write	Real EPICS PVs visible in browser console
3	Synthetic Data Source	DSP engine, Lua sandbox, signal composition	Synthetic signals computed and streamed
4	Frontend Foundation	Svelte skeleton, WS client, signal stores, view mode shell	Subscribed values rendered in browser
5	View Mode Widgets	All widget types rendered in view mode	Full interactive HMI panel
6	Edit Mode	Konva canvas, drag-and-drop, resize, properties pane	Can build and save a panel
7	Edit Mode — Advanced	Multi-select, align/distribute, undo/redo	Complete editor UX
8	Historical Data	Archive integration, time navigation UI	Replay past data in widgets
9	Signal Discovery	Signal tree, CSV import, EPICS Channel Finder	Usable without manual PV entry
10	Hardening	Docs, packaging, integration tests, performance	Production-ready binary

---

Phase 0 — Scaffold ✅

Goal: working build pipeline, empty binary that serves a placeholder page.

• Initialise Go module (go mod init github.com/uopi/uopi)
• Create directory layout: cmd/uopi/, internal/, web/
• Svelte + Vite + TypeScript project in web/
• //go:embed web/dist in backend via web/embed.go; Makefile builds frontend then backend
• HTTP server on configurable port; serves embedded frontend
• TOML config loading (BurntSushi/toml) with UOPI_* env var overrides
• GitHub Actions CI: runs go test ./... and npm run check on push
• README.md with quick-start instructions

Done when: ./dist/uopi starts and serves an "under construction" page. ✅

Notes:

• Embed lives in web/embed.go (not cmd/uopi/) because //go:embed paths cannot use ..
• CI runs commands directly (not via make test) but is functionally equivalent

---

Phase 1 — Core Backend ✅

Goal: signal broker and WebSocket protocol working with a stub data source.

• Define DataSource interface (internal/datasource/iface.go)
• Implement in-memory stub data source: sine_1hz, sine_01hz, ramp_10s, toggle_1hz, noise, writable setpoint — all at 10 Hz
• Implement Broker (internal/broker/broker.go):
  • Subscribe / unsubscribe with reference counting
  • Fan-out goroutine per signal
  • Clean teardown when last subscriber leaves
• WebSocket handler (internal/server/ws.go):
  • subscribe / unsubscribe / write / history messages
  • Pushes update and meta messages to client
  • Graceful close on disconnect (all subscriptions cancelled)
• REST API skeleton (internal/api/api.go):
  • GET /api/v1/datasources — lists registered sources
  • GET /api/v1/signals?ds= — lists signals for a source
  • Interface endpoints return 501 stubs (storage in Phase 6)
• Unit tests: fan-out, teardown on last-unsub, unknown DS/signal, multi-signal (5 tests, all passing)

Done when: a wscat client can subscribe to the stub source and receive timed updates. ✅

Notes:

• WebSocket signal refs use {"ds": "stub", "name": "sine_1hz"} objects (not "stub:sine_1hz" strings) to avoid ambiguity with EPICS PV names containing colons
• broker.ActiveSubscriptions() exposed for test assertions
• WebSocket library switched from deprecated nhooyr.io/websocket to maintained github.com/coder/websocket (identical API)

---

Phase 2 — EPICS Data Source ✅

Goal: real EPICS PVs readable and writable through the broker.

• CGo wrapper for EPICS libca (internal/datasource/epics/):
  • Context init with ca_enable_preemptive_callback (no polling loop)
  • ca_create_channel with connection callback shim
  • ca_add_event monitor with value callback (DBR_TIME_* types)
  • ca_put for writes (float64, int64, string, bool)
  • ca_get for DBR_CTRL metadata (units, limits, enum strings)
• Map CA DBR types to internal DataType enum; CA severity → Quality
• Implement DataSource interface for EPICS (epics.go)
• Config: ca_addr_list (overrides EPICS_CA_ADDR_LIST); archive URL
• Handle disconnected channels gracefully (quality = Bad via connection callback)
• Integration test with SoftIOC (gated on EPICS_BASE + TEST_PV env vars)
• noop.go (//go:build !epics) — package compiles without EPICS; epics.Available() returns false
• archive.go — Archive Appliance JSON HTTP client for History()
• make backend-epics and make test-epics targets added to Makefile
• CI comment explains EPICS build requirements; no EPICS step in CI runners

Done when: a PV subscription round-trips from IOC → broker → WebSocket client with correct timestamp and units. ✅ (verified by design; integration test requires live IOC)

Notes:

• Uses //go:build epics on all CGo files; clean build without the tag: go build ./... passes
• Handle table (Go map, mutex-protected) maps C-side uintptr_t back to Go channels; callbacks are minimal (no alloc)
• epics.Available() checked in main.go before registration so EPICS-less builds silently skip

---

Phase 3 — Synthetic Data Source

Goal: users can define computed signals from existing signals.

• Define synthetic signal definition schema (JSON)
• Implement DAG evaluator: nodes re-evaluated on upstream change
• Built-in node types (internal/dsp/):
  • Arithmetic: gain, offset, add, subtract, multiply, divide
  • Statistics: moving average (by count, by time), RMS
  • Filters: IIR lowpass/highpass/bandpass (via gonum or biquad)
  • Calculus: finite-difference derivative, cumulative integral
  • FFT / inverse FFT (gonum/dft)
  • Threshold / clamp
  • Custom expression (simple formula parser)
• Lua node: sandboxed gopher-lua state per signal; inputs as globals
• Load synthetic definitions from synthetic.json at startup
• REST endpoint to CRUD synthetic signal definitions (persists to JSON file)
• Unit tests for each DSP node type

Done when: a synthetic moving-average of an EPICS PV is visible in the WebSocket stream.

---

Phase 4 — Frontend Foundation ✅

Goal: Svelte app connects to WebSocket and reactively displays values.

• App shell (App.svelte): mode state (view | edit), WS connect on init, disconnect banner, --dpr CSS variable
• WebSocket client (ws.ts): singleton WsClient, exponential back-off reconnect (500 ms→30 s), re-subscribe all signals on reconnect, status Svelte readable store
• Reference-counted subscriptions in WsClient: one WS message per unique signal regardless of number of callers; unsubscribe sent when ref-count hits zero
• Signal store factory (stores.ts): getSignalStore(ref) and getMetaStore(ref) — lazily created Svelte readable stores, WS subscription started on first use
• types.ts: SignalRef, SignalValue, SignalMeta, Widget, Interface
• xml.ts: parseInterface(xml) — DOMParser-based XML→Interface converter
• View mode shell (ViewMode.svelte): collapsible InterfaceList pane (260 px) + Canvas, toolbar with WS status chip
• InterfaceList.svelte: fetches /api/v1/interfaces, Import XML file picker (calls onLoad prop), "New" stub button, collapse toggle
• Canvas.svelte: absolute-position widget rendering, --dpr CSS variable, placeholder for null interface
• TextView.svelte: live {label}: {value} {unit} with quality dot (green/amber/red/grey), positioned absolutely
• EditMode.svelte: stub ("Edit mode — coming in Phase 6")
• vite.config.ts: /ws WebSocket proxy added for dev server
• npm run build and npm run check: 0 errors, 0 warnings (95 files checked)

Done when: loading a manually crafted XML interface displays live PV values. ✅

Notes:

• Uses Svelte 5 runes ($state, $props, $derived) throughout, matching existing code style
• Routing is a simple mode state variable in App.svelte (no SvelteKit, no router library)
• Unknown widget types in Canvas render as grey dashed placeholder boxes (forward-compatible)

---

Phase 5 — View Mode Widgets

Goal: all widget types rendered and interactive in view mode.

• Text view widget
• Gauge widget (SVG arc, configurable range/thresholds)
• Vertical / horizontal bar widget
• LED widget (condition evaluator, configurable colours)
• Multi-LED widget (bitset, per-bit labels)
• Set-value widget (input + Set button; sends write over WS)
• Button widget (sends fixed value on click)
• Plot widget:
  • Time-series (uPlot, streaming buffer of N points)
  • FFT, waterfall, histogram, bar chart, logic analyser (ECharts)
  • Multi-signal support; legend
• Text label (static)
• Image widget (base64 or server URL)
• Link widget (navigates to another interface)
• Right-click context menu: signal info dialog, copy name, export CSV
• Svelte component tests for each widget type (mock store)

Done when: a complete HMI panel with multiple widget types runs smoothly at 60 fps.

---

Phase 6 — Edit Mode (Core)

Goal: users can build and save an interface from scratch.

• Konva stage setup in edit mode; devicePixelRatio scaling
• Widget renderer adapter: same widget components rendered on Konva layer
• Signal tree pane: fetch signal list, tree display, filter/search
• Drag signal from tree → drop on canvas → widget type picker
• Place widget at drop coordinates with default size
• Single selection: bounding box + resize handles via Konva Transformer
• Move widget by dragging body
• Delete widget (× button or Del key)
• Properties pane: common options (label, position, size)
• Per-widget property editors (range, colour, plot type, etc.)
• Save interface to server (POST/PUT XML)
• Load interface from server (GET XML, populate canvas)
• Export / import local XML file

Done when: an engineer can create a panel with 5+ widgets, save it, reload it, and see live data.

---

Phase 7 — Edit Mode (Advanced)

Goal: complete editing UX matching the functional spec.

• Multi-select: Ctrl+click toggle; rubber-band area select
• Group move: drag any selected widget to move all
• Group delete: Del key on selection
• Align toolbar: left / center-H / right / top / center-V / bottom
• Distribute toolbar: evenly by center / by gap (H and V)
• Undo / redo: command pattern, Ctrl+Z / Ctrl+Shift+Z
• Snap-to-grid (optional, toggle in toolbar)
• Add text label tool
• Add image tool (upload to server or embed base64)
• Add link tool
• Collapsible signal tree pane and properties pane (toggle buttons)
• Right-click on interface in list: Edit / Clone / Delete

Done when: the editor feels complete and the undo stack works reliably.

---

Phase 8 — Historical Data

Goal: users can navigate to past timestamps using archive data.

• EPICS Archive Appliance HTTP API client (internal/datasource/epics/archive.go)
• Broker History() dispatch: route to correct data source's History() impl
• WebSocket history request / response handling (implemented in Phase 1)
• Frontend: time range picker in view mode toolbar
• "Live" button: flushes history state, re-subscribes to live updates
• Plot widget: switch between streaming and historical range mode
• Point-value widgets: show value at selected timestamp

Done when: a plot can display 24 hours of archived data with a time slider.

---

Phase 9 — Signal Discovery

Goal: users can find signals without knowing their names in advance.

• EPICS: attempt Channel Finder or cainfo enumeration; fall back to manual entry
• Signal tree: lazy-load children on expand
• Manual add custom PV (EPICS): text input in signal tree
• New synthetic signal wizard: name, inputs, node graph UI
• CSV import: parse NAME, DataSource, DS_PARAMETERS; add to tree

Done when: a new user can find and subscribe to a PV without prior knowledge.

---

Phase 10 — Hardening

Goal: production-quality binary, docs, performance validation.

• End-to-end integration tests (real SoftIOC, headless browser via Playwright)
• Benchmark: 20 clients × 100 signals; verify < 5 ms fan-out latency
• Benchmark: frontend at 10 Hz update rate; verify 60 fps
• Static binary validation: run on RHEL 7 (Docker image centos:7)
• Security: Lua sandbox audit; XML XXE protection; write-permission guard
• Graceful shutdown: drain WS connections, flush pending writes
• Structured logging (log/slog) — done from Phase 0
• /metrics endpoint (Prometheus format) for server monitoring
• Complete README.md and operator docs
• Release: goreleaser config for Linux amd64 + arm64 binaries

Done when: binary passes integration tests on CentOS 7 container with a SoftIOC.

---

Estimated Effort

These are rough single-developer estimates. Parallel work across backend and frontend where possible will reduce calendar time.

Phase	Effort	Status
0	1–2 days	✅ Complete
1	3–5 days	✅ Complete
2	1–2 weeks	✅ Complete
3	1–2 weeks	—
4	3–5 days	✅ Complete
5	2–3 weeks	—
6	2–3 weeks	—
7	1–2 weeks	—
8	1 week	—
9	1 week	—
10	1–2 weeks	—
Total	~14–20 weeks