fix(front-ends): ARDOP, KISS and the TUI ignored [audio] device (#1311)

[dc0sk]

Closes #1311.

What was wrong

ModemEngine::set_default_device was called at exactly one construction site in the workspace — the daemon's main engine. ARDOP, KISS and the TUI built engines and passed a hardcoded None for the per-call device, so on any multi-card host they took the OS default card and [audio] device was silently ignored.

openpulse-tnc is the shipped Pat surface. An operator with a USB soundcard interface plus onboard audio — the ordinary amateur configuration — got the onboard card, with no diagnostic.

Why it happened six times

The engine resolves a device as device.or(self.default_device), so a front-end may name it per call or once at construction. The two correct surfaces used two different mechanisms:

• daemon → set_default_device
• CLI → threads --device into every call

Nothing documents which a front-end should use, and three of them used neither. There are also seven copies of the backend-selection match across the workspace, so there was no shared place where pinning the device would have been the obvious next line.

Because the per-call argument falls back to the default, one guarded line each is sufficient — no call sites needed touching.

The TUI needed one extra change: it takes its backend from --backend, has no device flag at all, and loaded its config after building the engine. The config load moved ahead of it.

Census, corrected twice

My issue filing said "four front-ends", which was a claim about a filter that had only looked at the sites #1308's review named. Reading every site:

site	mechanism	verdict
daemon main engine	set_default_device	correct
openpulse-cli	--device per call	correct — I had it wrong in the other direction too
openpulse-ardop	neither	fixed here
openpulse-kiss	neither	fixed here
openpulse-tui	neither	fixed here
openpulse-mesh	neither	moot — real-audio capability removed, no_real_audio.rs keeps it out
daemon repeater RX/TX	neither	#1308, blocked on a ruling about which device they should use at all

Why a source scan and not the shared builder

#1308's review floated a shared engine builder, and ptt_builder (#1258) is the in-repo precedent for exactly this shape — N front-ends each open-coding a backend match. It is right in principle.

It also needs cpal feature forwarding across five crates, which is too wide to carry inside a fix for a silent-wrong-card bug. Deduping the seven matches stays separable and available.

What actually prevents recurrence is the scan — the instrument this repo already uses for keying (ptt_keys_every_transmit). Its exclusions are written into its own source with a reason each, so the next front-end gets added to it rather than quietly omitted.

Validation, in four directions

A scan that matches nothing passes exactly like a scan over compliant code, and this repo has been bitten by that before — a pattern copied between crates matched nothing because the calls were multi-line chains, and its planted control still passed. So:

• planted violation → flagged at the right line
• planted fix → accepted
• set_default_device placed outside the 12-line window → rejected
• and the real one: deleting the pin from openpulse-kiss itself fails with "openpulse-kiss builds a ModemEngine at line(s) [82] without calling set_default_device"

Consumer

openpulse-tnc (ARDOP), openpulse-kisstnc, openpulse-tui — the three shipped binaries whose main builds an engine and whose call sites all pass None.

Prior art

server.rs:111 is the one correct set_default_device, the line this copies. ptt_builder (#1258) is the shared-builder precedent, deliberately not followed here for scope. ptt_keys_every_transmit is the source-scan precedent, which is followed.

Twins

The census above is the twins check, and it is complete rather than sampled: every ModemEngine::new( in the workspace was read, not grepped. Two sites are correct by different mechanisms, three are fixed, one is moot, two are #1308.

Implements: REQ-DEV-01

🤖 Generated with Claude Code

https://claude.ai/code/session_0188ATCj6DZ9aRVQ2vSirua6