Pull request overview

Stabilizes VFO flag placement for diversity-receive A/H pairs by making pair side-locking and overlay updates authoritative (vs. transient screen position), adding short “echo hold” windows during fast drag to suppress stale radio echoes, and using z-order to resolve overlap with nearby slice flags.

Changes:

• Add diversity metadata to slice overlays and use it to lock left/right flag sides for attached diversity pairs (plus active-group z-order handling).
• Update tune/drag paths to push frequency changes to both member overlays together and suppress stale pan/center echoes during fast drag.
• Extend the VFO flag placement test coverage for edge/locking behavior and diversity ordering.

Reviewed changes

Copilot reviewed 10 out of 10 changed files in this pull request and generated 2 comments.

Thanks for this, @rfoust — solid, well-structured fix. The extraction of the flag-placement geometry into a pure, testable VfoWidget::placementForMarker(...) / diversityPairOrderKey(...) and the consolidation of the three near-identical direction-assignment blocks (GPU repositionVfoFlags, software paintEvent, and vfoFlagOnLeftForSlice) into the shared assign*Directions helpers is a real readability win and removes a known drift hazard between those paths. Null-handling looks careful throughout (pushSliceFrequencyToOverlays, applyActiveVfoZOrder, and overlaysAreAttachedDiversityPair all guard their pointers), and the SliceModel change to emit diversityChanged on any of child/parent/diversity/index transitioning (not just the diversity bool) is more correct. CI is green across build/macos/windows/CodeQL.

A couple of points, both echoing Copilot — I verified both against the diff and agree they're worth addressing:

1. Stale displayMhz capture in the drag-echo singleShot (MainWindow_Wiring.cpp:619-633) — valid, minor. The lambda captures displayMhz by value at schedule time, but its own guard reads the live m_sliceDragTargetMhz. If the drag advances before the singleShot(0) runs, the lambda pushes the older captured frequency — a momentary backward nudge, which is exactly the snap-back class this PR is suppressing. A following echo would correct it, so it's transient, but since you already read the member in the guard, just read it for the push too:

QTimer::singleShot(0, this, [this, sliceId]() {
    const double displayMhz = m_sliceDragTargetMhz;
    if (displayMhz <= 0.0) return;
    if (!m_sliceDragInProgress
        && QDateTime::currentMSecsSinceEpoch() >= m_sliceDragEchoHoldUntilMs) return;
    if (SliceModel* slice = m_radioModel.slice(sliceId))
        pushSliceFrequencyToOverlays(slice, displayMhz);
});

2. Duplicated diversity-pair detection — valid, maintainability. The same "are these two an attached A/H diversity pair?" check now exists in four spots with slightly different shapes: sameDiversityReceivePair() (MainWindow.cpp, file-static), the activeDiversityPartner and dragDiversityPartner lambdas (MainWindow_Wiring.cpp), and overlaysAreAttachedDiversityPair() (SpectrumWidget.cpp, over overlays). The two Wiring lambdas are byte-for-byte identical apart from the slice they test, and they replicate sameDiversityReceivePair exactly. Since pairing rules for diversity_index/roles have already changed once, this is a real drift risk. The slice-based copies could collapse to one shared bool isSameDiversityReceivePair(const SliceModel*, const SliceModel*) reused at all three call sites; the overlay variant necessarily stays separate (overlays carry no panId).

Neither is blocking. Nice work, and the focused test additions for the geometry and ordering helpers are appreciated.

_{🤖 aethersdr-agent · cost: $2.2082 · model: claude-opus-4-8}

Approved. Reviewed the current head (e9a1f74):

• Solid follow-on to #3866 — extracts flag geometry into pure, testable placementForMarker()/diversityPairOrderKey() and consolidates the three direction-assignment blocks into shared assign*Directions helpers, removing the drift hazard between the GPU/software/vfoFlagOnLeftForSlice paths.
• Diversity pair placement is now authoritative (stable L/R sides), both overlays update together, stale echoes suppressed during fast drag, and overlap resolves by active-group z-order rather than pushing flags to a lower row. SliceModel emitting diversityChanged on any child/parent/diversity/index transition is more correct.
• Both review notes addressed in e9a1f74: the drag-echo singleShot reads the live m_sliceDragTargetMhz (no stale-capture backward nudge — verified), and the duplicated pair detection is collapsed into a shared isSameDiversityReceivePair() helper used at all slice-based call sites.
• Not a keying path (Principle VI N/A); vfo_flag_placement_test extended for the geometry + ordering helpers. CI green on all 6.

Thanks @rfoust — clean fix.