[ten9876]

Summary

Closes #6249. Stacked on #6247. The base is fix/dvk-wiki-conformance, so the diff shows only this pass; retarget to main once #6247 merges.

This moves the Digital Voice Keyer panel onto the AetherSDR style guide (RFC #6226), using the vocabulary of the unified title bar (#6198). The pass covers layout, controls and states as well as colour.

Structure

• Header. Modelled on a title-bar radio tab: the title in ink, a status line led by a status dot, then a hairline. The status line moves up from the bottom of the panel. It reserves two lines, so a wrapped refusal never shifts the slots, and it reads like the title bar's subline: Idle, Previewing · slot 3 · 0.4s of 4.2s, On air · slot 3 · …, Play failed (slot 3): DVK requires an active SmartSDR+ subscription.
• Slot rows. 8 px canon.nested cards with a canon.line hairline. Hover brightens the edge, and the selected row takes the title bar's active-tab pair (titlebar.tab.active.*).
• Label boxes fixed. The old row rule (QFrame { … }) also matched the row's QLabels, because QLabel is a QFrame subclass, so every name and length was drawn as a box inside the box. The rule is now scoped by property.
• Controls. F-keys are 22 px key caps and the transport buttons are 24 px, both using the canon secondary-control recipe from the About window: control fill, lineHi edge, cyan text, aqua hover and focus ring, 4 px corners.
• Enabled only when useful. STOP is disabled while idle, PLAY and PREV are disabled on an empty slot, and the other start controls are disabled while an operation runs. They also grey out while a start awaits its reply or a WAV transfer runs. That admission rule comes from fix(dvk): import converts to the radio's 24 kHz mono and uses the wiki's two-step upload; slot names keep their spaces. Principle I. #6247's review fixes (DvkModel::canStartOperation(), signalled by admissionChanged()); this PR only shows it.
• Slot menu. It is the title bar's rounded menu. The title bar's helper and rules move into RoundedMenu.{h,cpp}, so the two menus stay identical; the title bar's own behaviour is unchanged.
• Rename field. Control fill with a cyan edge.

States (provisional, as agreed; to be adjusted after review)

• On air uses the MOX family: PLAY, the playing slot's F-key, its progress bar and the status dot take background.tx, tx.mox.border and tx.mox.text.
• Recording and preview use the canon cyan active treatment.
• Refusals use accent.danger and always say "failed".
• No red anywhere else, since the canon keeps red for transmit indicators.

Mechanism. One tracked stylesheet on the panel (DvkPanel::styleTemplate()) resolves every colour from color.canon.*, color.titlebar.*, the MOX tokens and accent.danger. Widgets carry their state as properties (selected, empty, live, mode, state, tone). All 11 colour literals are gone, and no token values change.

Constitution principle honored

Principle VIII. Each state is checked against a real panel driven by recorded firmware status lines, not just asserted. The colour ratchet moves down.

Rebased onto #6247's review fixes (b8bac9d4c). PLAY keeps its TX-keying marker. dvk_panel_admission_test now links RoundedMenu.cpp, and it passes against the themed panel too: a refused start can't even be pressed now.

Test plan

• Full local build and ctest: 668 tests pass offscreen (after the rebase). That includes the four title-bar tests that compile TitleBar.cpp directly; they now link RoundedMenu.cpp.

• New dvk_panel_theme_test (33 cases). It drives a real DvkPanel from firmware 4.2.20 status lines and checks:

  • every panel token resolves in both Default Dark and Default Light
  • the panel's stylesheet has no colour literals, and no tokens outside the canon, title-bar and MOX groups
  • rows and labels carry no per-widget stylesheet
  • each state (selected, empty, preview, on air, recording, refusal and recovery, light theme) lands on the right widgets
  • a wrapped refusal doesn't move the slots

  Set DVK_PANEL_GRAB_DIR to save a PNG of each state for review. That is how I checked them, the on-air state included, without transmitting.

• Live app, offscreen, through the bridge. I checked the panel, the slot menu (rounded, with Clear and Export greyed on an empty slot) and the rename field.

• tools/audit_colours.py --strict against the base branch: unique colours 587 → 582, references 2522 → 2487, setStyleSheet calls 905 → 893.

• check_a11y reports nothing on the touched files. check_test_registration, check_theme_seed, check_engine_boundary, check_ci_test_gate, check_command_plane and gen_touchpoint_manifest --check pass.

• Not yet seen on a real display or with a radio connected. A smoke run on the maintainer's display is the natural next step.

For review:

• Light theme, on air. tx.mox.text / tx.mox.border are pale on light canon cards. These are the same tokens the MOX button uses, and the guide says a migration PR doesn't change token values, so I left them.
• Title bar menu, disabled items. The shared rounded-menu rules give disabled items no colour of their own. The DVK menu adds canon.muted locally; the title bar's menu has the same gap and is left as is here.

Checklist

• Commits are signed
• No new flat-key AppSettings calls
• Clean-room
• No meter UI touched
• Documentation: no doc describes the panel's look. CHANGELOG is untouched.
• No security-sensitive changes

🤖 Generated with Claude Code

[aethersdr-agent]

Issue fit

Mostly yes. #6249 names four defects and a direction; the diff closes the first three cleanly and the fourth almost completely.

#6249 requirement	Where	Verdict
11 hard-coded colours + pre-canon tokens	src/gui/DvkPanel.cpp:24-88, one tracked template	Closed — I extracted all 18 {{…}} tokens and all 18 resolve in both bundled themes (walked scopes.root.tokens, which mixes nested and dotted keys)
"Boxes inside boxes" — QFrame { … } also matching its QLabels	:146 property selector QFrame[dvkRole="slot"], plus border: none; background: transparent on the name/length rules	Closed — the slot rule can no longer reach a QLabel, and the test asserts the per-widget sheets are gone (tests/dvk_panel_theme_test.cpp:111-112)
Red/green checked states; PLAY should read like MOX	:63-66 QPushButton#dvkPlay:checked → background.tx / tx.mox.*; REC/PREV → cyan active	Closed
"Buttons that can't act stay enabled"	refreshTransport(), :447-459	Partially — see nit 1: STOP is enabled (and the only enabled control) when the radio reports DVK Disabled
Header modelled on a title-bar radio tab	:114-135	Closed
Slot menu uses the title bar's rounded menu	RoundedMenu.{h,cpp} extraction + DvkPanel.cpp:716-724	Closed

Coverage layer is right: tests/dvk_panel_theme_test.cpp is a socket-free CTest driving a real DvkPanel from fw 4.2.20 status text through CommandParser::parseLine → DvkModel::applyStatus — no fake peer, no radio, no TX. No socket test is added or modified by this PR (I grepped the PR's own test source for QTcpServer/QUdpSocket/QLocalServer/bind/listen/connectToHost/Fake*: zero hits).

Scope

File / group	Change	Claimed?	Verdict
src/gui/DvkPanel.{h,cpp}	Theming, property-driven states, transport enablement, status wording	Yes — the whole of #6249	In scope
src/gui/RoundedMenu.{h,cpp} (new)	Verbatim lift of kRoundedMenuProperty / kHasChecksProperty / syncMenuCheckColumn / roundMenuTree out of TitleBar.cpp's anonymous namespace, plus the rules string	Yes — the issue's "Menu" bullet; sharing requires the extraction	In scope
src/gui/TitleBar.cpp	Deletes the moved code, includes the header	Follows from the extraction	In scope
CMakeLists.txt:1446, tests/tests.cmake (4 titlebar targets)	RoundedMenu.cpp added	Mechanical consequence	In scope
tests/tests.cmake:3163-3176	New dvk_panel_theme_test	Required for a behaviour change	In scope

Everything in the diff is explained by #6249. Notably nothing is out of scope: no CHANGELOG.md entry, no settings keys, no new protocol verb or capability field. styleTemplate() / verbLabel() are public statics but they are test seams on a GUI class, not public protocol surface.

Preference check (§6): the status-line rewording ("Status: Playback 3 / 0:04" → "On air · slot 3 · 0:04"), the middle dots, and the amber/cyan assignment are preference-shaped changes to working UI — but #6249 is maintainer-requested, carries maintainer-review, specifies the wording direction ("always with the word 'failed'") and explicitly marks the colour split provisional. Ratified, not smuggled. One sentence is all this needs.

Blockers

None. Two things I expected to be blockers both held up:

• roundMenuTree moved from TitleBar.cpp's anonymous namespace into namespace AetherSDR in the new header — the unqualified calls left behind at TitleBar.cpp:974 and :1112 still resolve, because TitleBar.cpp:51 opens namespace AetherSDR { and closes it at :1749.
• All four other tests.cmake targets that compile src/gui/TitleBar.cpp (lines 7242, 7274, 7308, 7335) gained src/gui/RoundedMenu.cpp. None left behind. aetherdesktop_support does not contain ${GUI_SOURCES} (CMakeLists.txt:1665-1672), so compiling DvkPanel.cpp/RoundedMenu.cpp into the new test and also linking that library is not a duplicate-symbol risk.

Nits (non-blocking)

1. src/gui/DvkPanel.cpp:458 — STOP is the one enabled control on an unlicensed radio, which is the inverse of #6249's fourth bullet.

    m_stopBtn->setEnabled(!idle);

DvkModel::canStartOperation() is m_status == Idle || m_status == Unknown (src/models/DvkModel.h:34), so idle is false for Disabled. On dvk status=… enabled=0 the panel shows "Disabled · SmartSDR+ required" with REC/PLAY/PREV correctly greyed and STOP alone clickable — and pressing it sends dvk *_stop to a radio that will refuse it. The new test never feeds enabled=0, so it cannot catch this.

    m_stopBtn->setEnabled(!idle && s != DvkModel::Disabled);

(and a feed(model, QStringLiteral("S1|dvk status=idle enabled=0")) step asserting all four are off would pin it.)

2. src/gui/DvkPanel.cpp:124 — the two-line reservation silently clips the longest refusals, which is the #3377 symptom the comment above it preserves.

    m_statusLabel->setFixedHeight(2 * m_statusLabel->fontMetrics().lineSpacing() + 2);

setWordWrap(true) plus a hard setFixedHeight means a third line is clipped with no ellipsis and no tooltip. The longest refusal is 68 characters — "Play failed (slot 3): DVK requires an active SmartSDR+ subscription" (DvkModel::dvkErrorString, src/models/DvkModel.cpp:107) — which is right at the edge of two lines in a ~230 px label at 11 px, and the WAV path can be longer still: DvkWavTransfer.cpp:346 emits "Cannot open file: " + f.errorString(), so announceStatus renders "Transfer failed: Cannot open file: <Qt error string>". Reasoned from the strings, not measured. Keeping the fixed height (the slot-stability property the test asserts) and adding the full text as a tooltip costs one line in announceStatus:

    m_statusLabel->setText(text);
    m_statusLabel->setToolTip(text);

3. src/gui/DvkPanel.cpp:124 — the reservation is computed before the sheet is applied. applyStyleSheet(this, styleTemplate()) lands at :269, so fontMetrics() here is the app default font, not the font-size: 11px the rule sets. Harmless direction (over-reserves), but it does mean "a wrapped refusal keeps the slots where they were" (tests/dvk_panel_theme_test.cpp:181-183) asserts that setFixedHeight was called, not that a wrap is actually absorbed.

4. src/gui/DvkPanel.cpp:546 — dropping the "Status: " prefix changes what a screen reader hears. m_statusLabel has no accessibleName (the a11y pass in e9584dcb6 named the panel, slots and controls but not this label), so the QAccessible::NameChanged event at :409 now announces a bare "Idle" / "On air · slot 3 · 0:04". The U+00B7 separators are also announced verbatim by some readers. docs/a11y.md has no sentence covering text-only status labels, so this is a convention note, not a rule.

5. CI has told you nothing about compilation. Only "Static checks" and "Sanitizer option configures" have reported on 582d681 — no build or test lane. The new target and the TitleBar.cpp extraction are compile-unverified in CI as of this review; dvk_panel_theme_test is not in any ctest -R filter in ci.yml, so it will only ever run in the full-suite lane.

What I tried to break (and could not)

• Theme tokens. I expected a false-resolving token (a dotted color.canon.* grep returns nothing even when defined). Flattened scopes.root.tokens in both resources/themes/default-{dark,light}.json handling the mixed nested/dotted keys: all 18 panel tokens and all 4 RoundedMenu.h tokens are present in both. I also checked the test's own API shapes — ThemeManager::color(const QString&) takes the full dotted form extractReferencedTokens returns (ThemeManager.h:86, .cpp:1571) and setActiveTheme returns bool (ThemeManager.h:198) — so the "resolves every panel token" assertion is real, not vacuous.
• The colour-literal assertion. The rendered diff makes tests/dvk_panel_theme_test.cpp:89 look like it has bare \b and \( escapes, which would make the regex invalid and the check pass unconditionally. In the file they are \\b and \\(. The assertion bites.
• The refusal test. DvkModel::handleCommandResponse (src/models/DvkModel.cpp:79-100) does emit commandFailed for 0x50004001 and does not also emit statusChanged(Disabled), so nothing re-enters onStatusChanged to clobber the failure text after the diff's deliberate ordering. The #3377 clobber the comment at :316-321 warns about does not reopen.
• Target wiring. Every tests.cmake and CMakeLists.txt consumer of TitleBar.cpp was checked for the matching RoundedMenu.cpp (all four patched), the new test's symbols (DvkModel, DvkWavTransfer, DvkWavConverter, ThemeManager) all come from aethercore/aetherdesktop_support, ${THEME_TEST_RESOURCES} is present so the offscreen run has themes, and QFrame reaches the test transitively through <QLabel>. TIMEOUT 60 is an explicit per-test opt-out of the 300 s default, which is the intended shape.
• Base ancestry. e9584dcb6 (the a11y pass) and e8a2affb7 (the import/upload fix) that the hunks sit on are both already on main, so this is not stacked on an unmerged PR and the diff is answering the real base.
• QString::arg folding in the new status strings (:587, :594) — the two-argument arg(QString, QString) overload fills %2/%3 in one pass after the arg(int) for %1, which is what the format strings expect.
• Could not check: anything requiring a build, a run, or pixels. I have no compiler and no display here, so every finding above is reasoned from the code in the head checkout rather than reproduced at runtime — in particular nits 2 and 3 are text-metric arguments, not measurements, and the DVK_PANEL_GRAB_DIR screenshots are the right way to settle them.

Security audit

Ran as required on the harness's file list. Every finding is pre-existing on main; this PR introduces none of them — #6251's diff does not touch any of the five flagged files. Credited to security-audit (CodeGuard):

• src/core/DvkWavTransfer.cpp:190/:226 — worth routing as its own issue, not a finding against this PR. The DVK export path does m_server->listen(QHostAddress::Any, port) on a port taken from the radio's reply and then adopts nextPendingConnection() unconditionally, so during the 10 s connect window any reachable host can race the radio and have its bytes written to the operator's chosen export path as a successful "Exported slot N". Bounded (5 MB cap, user-picked destination, nothing executed) and LAN-adjacent. Verified byte-identical to main.
• Discarded as false positives: the flagged memcpy in DvkWavConverter.cpp:40 is a 4-byte float/quint32 type-pun sized from the destination, and the frame walk at :176-183 is in bounds (dataBytes clamped, blockAlign >= channels * bytesPerSample enforced at :155); upload() caps size pre-read and re-encodes rather than passing bytes through; the CommandParser.cpp and TciProtocol.cpp index accesses are all length-guarded, and the "path" triage reason for those three files matches comment prose only.
• Triage staleness worth noting to the maintainers: DvkWavTransfer::validateWavFile, named in the triage reason, does not exist anywhere in the tree — DvkWavConverter::convertForRadio took that role over.

Recommendation

Approve with nits. This is a careful, well-structured pass: one tracked template, state carried as properties with a correct unpolish/polish helper, the QFrame-cascade defect actually fixed at the selector level rather than papered over, and a test that drives real firmware status text through a real panel and checks states on the specific widgets rather than asserting the implementation back to itself. Nit 1 is the only functional gap against the issue and is a one-line change plus one test step; nit 2 is worth a tooltip before merge. Neither needs a new round of review — fold them in and this is ready for the maintainer's call on the provisional amber/cyan split. Thanks for the extraction of RoundedMenu as a shared seam rather than copy-pasting the rules; that was the right call.

---

_{🤖 aethersdr-agent · cost: $9.2865 · model: claude-opus-5}

[jensenpat]

Reviewed 11dbe6bd40d9d4ec45bf1001354ca2a68f399570 (replay of a9f12087 onto main 248bccd23 after #6247). Issue fit for #6249: the DVK panel moves onto the style-guide / title-bar vocabulary, tokens only, no token-value edits. Flex-only panel. No code blockers.

Scope

Path	What landed
src/gui/DvkPanel.cpp / .h	One styleTemplate() stylesheet (color.canon.*, color.titlebar.*, MOX, accent.danger); state via properties; header, slot cards, F-keys, transport, rename field
src/gui/RoundedMenu.cpp / .h	Verbatim lift of the title-bar rounded-menu helper and rules
src/gui/TitleBar.cpp	Calls the shared helper; call sites unchanged
CMakeLists.txt, tests/tests.cmake	RoundedMenu.cpp in the GUI target and in every test that compiles TitleBar.cpp or DvkPanel.cpp
tests/dvk_panel_theme_test.cpp	Real panel, both themes, state properties, no colour literals

Nits (non-blocking)

• DvkPanel::refreshTransport enables STOP whenever canStartOperation() is false (DvkPanel.cpp:475). That includes status Disabled and an in-flight WAV transfer. stopActiveOperation() does not stop a transfer and does not act on Disabled, so STOP is the only enabled control in those states.
• Empty-slot F-keys stay enabled and only look muted. The click still no-ops through the existing guards.

Verification

Offscreen, Qt 6.12, on this SHA: dvk_panel_theme_test and dvk_panel_admission_test passed. unified_title_bar_test failed Alt+F opening the File menu, and therefore the late-submenu rounding that depends on that popup. Rebuilding the same test against main's TitleBar.cpp (this extraction removed) failed those same two checks and passed the same earlier ones, including existing menus rounded and menu text at x=32 with or without a check column. Linux full-suite on main already passed unified_title_bar_test. The local offscreen Alt+F miss is pre-existing.

Required checks on this SHA are green: build, check-macos, check-windows, Static checks, Sanitizer option configures. The per-PR gate does not run dvk_panel_theme_test. The app bundle was not built. Socket-owning DVK transfer tests were not run.

Follow-ups (not merge prerequisites)

• STOP-while-Disabled and STOP-during-transfer are unasserted.
• Colour audit numbers in the PR body were not re-run here.