Skip to content

Changelog — 2026-08-25

Repo hygiene: dead-reference sweep, internal archive move, contributing fix (2026-08-25 ~03:55 CEST)

Second batch of follow-ups to the external AI project-analysis review, covering the items decided with Jerome on 2026-08-24 late evening (roadmap section "8s"). Build-system and documentation hygiene only — no source code changed.

Context

The git-tracked documentation referenced files that live in git-excluded folders (historical daily changelogs under docs/changelogs/, superseded planning documents moved to docs/archive/ earlier this batch). On a fresh clone those links are dead. Decision: no tracked file may reference an excluded path; excluded material stays available locally only.

Changes

  • Dead-reference sweep (docs/implementation-roadmap.md): all 24 references to docs/changelogs/changelog-*.md rewritten as dated "dev log" mentions without file paths (batches 2026-07-17 x10, 2026-07-23 x8, 2026-08-02 x2, 2026-08-03 x4). Quoted user sentences and historical test counts preserved verbatim.
  • Historical changelogs moved: git ls-files confirmed none of the ~37 daily logs (2026-06-09 -> 2026-08-10) was tracked; the folder was moved from docs/changelogs/ to docs/internal/changelogs/ (covered by the existing .gitignore entry docs/internal/). No .gitignore change needed.
  • Contributing fix: CONTRIBUTING.md + docs/contributing.md — the "add a changelog entry" instruction now points to the tracked docs/changelog/ folder instead of the excluded docs/changelogs/.
  • Archived-plan reference fix (docs/implementation-roadmap.md, MIDI Import section): path link to the archived implementation-plan-midi-import-drag-and-drop.md replaced by an "archived locally" note.
  • Kept on purpose: the historical note in changelog-2026-08-01.md that documents the "no references to git-excluded changelogs" rule itself (it is the archive of the rule).

Earlier items of the same batch (late 2026-08-24)

Already applied on 2026-08-24 evening, logged here for completeness: - i18n sweep 2: remaining French text in CMakeLists.txt comments and the PowerShell / bash helper scripts translated to plain ASCII English. - test/ cleanup: temporary test/decode_settings.py removed (its conclusion stays documented in roadmap entry HC.11); test/dsp/click_count.txt artifact added to .gitignore. - Legacy option removal: AUTOTUNE_BUILD_TESTS CMake option deleted (leftover from the autotune_clone era); tests target builds unconditionally like every other target. - Docs archival: five superseded planning documents moved to the local git-excluded docs/archive/ folder; the two active plans stay tracked. - Roadmap updated with section "8s. Repo hygiene & docs archival follow-up (2026-08-24 -> 2026-08-25)" (entries HS.1-HS.6).

Validation

  • Repo-wide grep: zero references from tracked files to git-excluded paths (except the intentional 2026-08-01 rule note).
  • Full Release rebuild OK (SharedCode, VST3, Standalone, OpenVoxTunerTests; a sandbox restriction blocked unrelated NVIDIA ShadowPlay log writes at the very end of the msbuild run — build outputs were all produced).
  • Unit-test suite: Result: 97 OK, 0 FAILED (incl. PitchShifterOutput / PitchShifterOutputRms / PitchShifterClick suites exercising the modified click test).

Roadmap accuracy audit vs codebase (2026-08-25 ~05:04 CEST)

Verified every unchecked roadmap item against the actual codebase, and sampled checked items for removed features.

Findings and fixes

  • Backlog "Note name labels on additional piano keys (D, E, F, G, A, B)" was still unchecked although implemented (Source/ui/PianoKeyboard.cpp: white-key labels C/D/E/F/G/A/B with octave suffix on C, height-gated

    = 20 px). Marked [x] with an implementation note.

  • Section 9 "Pitch Visualizer improvements documentation": annotated (archived locally, no longer in the tracked tree) after yesterday's docs archival.
  • Stale header "Last updated: 2026-07-31" refreshed to the current date.

Confirmed correct (no change)

  • Bookmark positions, responsive small-screen layout, accessibility: no matching code found — correctly unchecked.
  • MIDI Import "Validation": no MidiImporter test exists — correctly unchecked (the drag-and-drop feature itself is implemented and checked in section 5b; only the automated test coverage is missing).
  • LP.7 MUSHRA harness: pending by design (listening panel + corpus).
  • LV2 format: under reflection, correctly open.
  • Website: no swipe support, no testimonials component, no user-facing theme toggle button (only automatic prefers-color-scheme handling in Layout.astro) — all three correctly unchecked.
  • Checked items sampled (piano-roll mode, spectral/EQ view, PresetGallery, MIDI target, key detection, Dark/Light theme, image export): all still present in the codebase. Deprecated features (FlexTune, attack-aware correction) properly struck through.

Audit correction (user feedback, ~05:15 CEST)

The initial audit wrongly reported "touch gestures: no matching code" — the grep patterns missed JUCE's mouseMagnify API. Pinch-to-zoom IS implemented on macOS trackpads (mouseMagnify overrides on PitchVisualizer + PitchCurveEditor, Ctrl/Cmd+wheel equivalent, and pinch-vs-smooth-scroll disambiguation). Roadmap item updated to [~] "largely implemented"; only raw multi-touch touchscreen finger gestures remain open (JUCE emulates single-touch mouse there).

Backlog cleanup decision (user, ~05:20 CEST)

The two stale wishlist items "Bookmark positions" (save/restore frequency-range view presets) and "Responsive layout for small screens" (adapt plugin UI at tiny host window sizes) were removed from the backlog and from section 4, on Jerome's decision.

Tests: MidiImporter validation suite (2026-08-25 ~07:26 CEST)

Soldered the last open item of roadmap section 5b (MIDI Import): the missing automated test coverage.

New file test/dsp/MidiImporterTest.cpp (14 sub-tests)

Self-contained synthetic MIDI generation (in-memory juce::MidiFile with 480 TPQN + explicit 120 BPM tempo meta written to per-test temp files, RAII-deleted). Coverage: - analyzeFile: monophonic single-channel summary; channel 10 percussion excluded from the summary and note counts; multi-channel listing sorted by channel with per-channel min/max/count/duration. - Rejection paths: garbage bytes ("Invalid MIDI file format"), nonexistent path ("Cannot open file"), syntactically valid file with no notes. - importFrom: monophonic sequence -> one point per note at exact times and frequencies; polyphonic reduction Highest/Lowest with staggered extents (documents the sweep-line transition point emitted when a selected note ends while another sustains); Loudest strategy; SpecificChannel filtering; percussion-only -> empty curve; sub-20 ms artefact notes filtered; dangling note-on (analyze counts it, import needs a paired note-off); garbage file -> empty curve.

Build integration

Source/dsp/MidiImporter.cpp added to the OpenVoxTunerTests target sources in CMakeLists.txt; test included in test/Main.cpp.

Validation

Full suite: 111 OK / 0 FAILED (was 97 + 14 new). Two initial test expectations were corrected during bring-up (percussion note counting; transition-point emission on note boundaries) — no product code changed.

Docs

Roadmap section 5b Validation checked with details; docs/testing.md tree + coverage list + sample output updated.