73 lines
3.3 KiB
Markdown
73 lines
3.3 KiB
Markdown
|
# Design
|
||
|
|
|
||
|
|
## Context
|
||
|
|
|
||
|
|
See `proposal.md` — Why. fpagent pushes `fprint-device-status`
|
||
|
|
`{available, device, captureAgentId}` to each connection on connect and to all
|
||
|
|
connections when availability changes (2 s watcher; probes skipped while an
|
||
|
|
enrollment runs). `available` means "a supported reader is attached", not that a
|
||
|
|
capture will succeed. The wizard currently learns only `agent-hello` and the
|
||
|
|
enrollment events, and its header tag reflects only the socket connection.
|
||
|
|
|
||
|
|
## Goals / Non-Goals
|
||
|
|
|
||
|
|
**Goals:** surface scanner availability in the wizard; stop a capture from
|
||
|
|
starting when no reader is attached; keep working against agents that do not
|
||
|
|
emit the event.
|
||
|
|
|
||
|
|
**Non-Goals:** changing the agent or the event contract; blocking capture when
|
||
|
|
the agent is merely unreachable (pre-existing behavior); a proactive device
|
||
|
|
picker or wiring guidance.
|
||
|
|
|
||
|
|
## Decisions
|
||
|
|
|
||
|
|
1. **The client is a pure listener.** The agent already emits on connect and on
|
||
|
|
change, so the wizard adds one handler and no polling or request event.
|
||
|
|
|
||
|
|
2. **Availability is tri-state, not boolean.** `agent.available` starts `null`
|
||
|
|
(unknown) and becomes `true`/`false` on the first status. An agent that never
|
||
|
|
emits the event leaves it `null`, so captures are not blocked — the feature
|
||
|
|
degrades to today's behavior instead of deadlocking older field laptops.
|
||
|
|
|
||
|
|
3. **The status tag has four labels.** Offline (socket down), "Scanner not
|
||
|
|
connected" (`available: false`), "Agent ready" (`available: true`), and
|
||
|
|
"Agent connected" (unknown). The `title` carries the version and the device
|
||
|
|
name when known.
|
||
|
|
|
||
|
|
4. **Capture is gated at both layers.** The buttons are disabled when
|
||
|
|
`available === false`, and `capture()` returns early on the same condition so
|
||
|
|
a race (availability changing between click and handler) cannot start an
|
||
|
|
enrollment. The agent remains authoritative and still rejects a capture with
|
||
|
|
no reader.
|
||
|
|
|
||
|
|
5. **The gating logic gets a small vm-sandbox test.** `app.js` is a browser
|
||
|
|
IIFE, so `test/fp-wizard.test.js` loads it in a `node:vm` sandbox with
|
||
|
|
stubbed `Vue`/`io`/`document` and exercises the availability state, the tag
|
||
|
|
labels, and the capture refusal directly. No new dependency and no browser
|
||
|
|
runtime; the remaining end-to-end pass against a real agent is manual, since
|
||
|
|
`/fingerprint` requires a session.
|
||
|
|
|
||
|
|
## Risks / Trade-offs
|
||
|
|
|
||
|
|
- [Old agent never emits the event] → availability stays `null`, captures are
|
||
|
|
allowed, behavior matches today. The status tag shows "Agent connected".
|
||
|
|
- [Brief wrong tag between socket connect and first status] → the agent emits
|
||
|
|
the status in the same connection handler as `agent-hello`, so the window is
|
||
|
|
negligible; the `null` state shows "Agent connected" rather than a false
|
||
|
|
"Scanner not connected".
|
||
|
|
- [`available` read as "ready to capture"] → the agent's contract says it means
|
||
|
|
"attached"; enrollment remains authoritative and a capture can still fail.
|
||
|
|
- [Watcher is skipped during a capture] → availability is not updated mid-scan,
|
||
|
|
which is correct: the reader is in use by that capture.
|
||
|
|
|
||
|
|
## Migration Plan
|
||
|
|
|
||
|
|
Ship the wizard; it is backward compatible with agents that lack the event.
|
||
|
|
No agent change is required to deploy the wizard, and the full benefit appears
|
||
|
|
once the field laptops run the agent that emits `fprint-device-status`. Rollback
|
||
|
|
= revert the wizard.
|
||
|
|
|
||
|
|
## Open Questions
|
||
|
|
|
||
|
|
None.
|