Pull request overview

This PR improves SpectrumWidget’s spectrum rendering and dBm scale interaction by reducing per-frame allocations/copies in the FFT display trace path and clamping UI-generated dBm ranges before they’re applied/emitted, preventing invalid scale requests during drag interactions.

Changes:

• Reworks buildFftDisplayTrace() to return a const QVector<float>& backed by reusable scratch buffers, avoiding per-frame QVector copies in the display-only smoothing/resampling path.
• Adds centralized dBm-range clamping helpers and applies them during dBm scale press/drag/release and immediate range application to prevent out-of-bounds/invalid ranges.
• Skips GPU FFT-line VBO updates (and line drawing) when the FFT line width is disabled.

Reviewed changes

Copilot reviewed 2 out of 2 changed files in this pull request and generated no comments.

Thanks @rfoust — this is a tidy, well-scoped change to SpectrumWidget. I traced both halves against the surrounding code and it holds up. CI is green across all six checks (build, macOS, Windows, CodeQL, accessibility).

What I verified

Scratch-buffer reuse (buildFftDisplayTrace) — Correct. The two scratch buffers don't alias: m_fftDisplaySmoothScratch (smoothing) is read and m_fftDisplayTraceScratch (resample) is written in the trace path, and the input bins (displaySpectrumBins() → m_smoothed/m_kiwiSdrFftTrace) is never one of the scratch vectors. Both call sites (renderGpuFrame, drawSpectrum) are compile-gated by AETHER_GPU_SPECTRUM so only one is live, and each invokes the function once per frame and consumes the result before anything else touches the buffers — so the reference return is safe today.

Line-VBO skip — Safe. The new if (m_fftLineWidth > 0.0f) guard on updateDynamicBuffer(m_fftLineVbo, …) mirrors the existing draw-side guard at the line pass (if (m_fftLineWidth > 0.0f) before cb->draw(...) on m_fftLineVbo), so the buffer is never drawn stale when the update is skipped.

dBm clamp helpers — The clamp math is robust: clampDbmRange handles non-finite inputs, inverted min/max ordering, and degenerate equal-value cases, always converging to a valid [10, 180] dB span within [-180, 80] dBm. The arrow-click / drag / release paths now route through these consistently, and the release path correctly bails (resets drag flags, no emit) on non-finite state rather than emitting a garbage range. This matches the stated goal of not sending invalid operator scale requests.

One minor note (non-blocking)

buildFftDisplayTrace changed from returning by value to returning const QVector<float>& into shared mutable scratch. It's safe for all current callers, but it's now a footgun if a future caller ever holds two results live (const auto& a = build(...); const auto& b = build(...); — a would be clobbered). A one-line comment on the declaration noting "result is valid only until the next call" would save a future maintainer the surprise. Your call — not a blocker.

Nice cleanup overall. 👍

_{🤖 aethersdr-agent · cost: $2.9860 · model: claude-opus-4-8}

Approved. Verified both halves independently against the diff and call sites:

• Scratch-buffer reuse — the two scratch buffers don't alias each other or the input (displaySpectrumBins()), and both call sites (renderGpuFrame, drawSpectrum) consume the result in place with no two-results-live, so the transient-reference return is safe. (Documented that contract on the declaration in a follow-up commit.)
• dBm clamp — robust across the edge cases (inverted, equal, span >180 / <10, non-finite); all interaction paths (arrow, control-drag, pan-drag, move, release) route through the helpers consistently, and the release path correctly bails without emitting a garbage range.

Display-only; Principle II honored (clamps UI-generated requests before sending, never overrides radio-authoritative state). CI green. Thanks @rfoust — clean, well-scoped work.