Skip to content

roadmap.md

Latest commit

 

History

History
181 lines (133 loc) · 9.25 KB

roadmap.md

File metadata and controls

181 lines (133 loc) · 9.25 KB
Raw

ReaBeat Roadmap

Tracking deferred fixes and feature requests by user/source. Anything actionable lives here so it doesn't fall out of memory between sessions.

Updated: 2026-05-20

v2.0.3 candidates - UX polish (forum feedback)

Daodan #4 - Middle-mouse pan + working scrollbar

Source: Daodan post #2 (forum t=308368) Effort: ~2h Risk: LOW Status: Deferred from v2.0.2

Pan via middle-mouse drag (standard DAW behaviour) and make the existing scroll thumb at the bottom of the waveform clickable/draggable.

  • WaveformView.cpp - add mouseDown / mouseDrag handling for e.mods.isMiddleButtonDown()
  • Detect click on scrollbar Y region (bottom 4-6 px), allow drag to scrub view position
  • Watch for trackpad emulation differences between macOS and Windows

Daodan #5 - Item persistence after deselect

Source: Daodan post #2 Effort: ~30 min + careful pointer-safety testing Risk: MEDIUM (cached MediaItem* can be deleted by user) Status: Deferred from v2.0.2

Keep the last selected item visible in updateSelectedItem when count drops to 0, so users don't lose their detection state by clicking empty timeline.

  • MainComponent.cpp:869-882 - on count<=0, set sourceLabel to " (deselected)" instead of full reset
  • Before any Apply path, call ValidatePtr2(nullptr, currentItem_.item, "MediaItem*") and bail out if invalid
  • Mark detected_ as "stale" (different colour) when deselected, refresh on next select

Daodan #9 + notabot - Manual time signature dropdown

Source: Daodan post #2, notabot post #43 (6/8 misdetected as 4/4) Effort: ~2-3h Risk: MEDIUM (algorithm-touching) Status: Deferred from v2.0.2

Dropdown "Time signature: Auto / 2/4 / 3/4 / 4/4 / 6/8 / 9/8 / 12/8" that overrides detection_.timeSigNum and timeSigDenom. When user changes it after detection, recompute downbeats from the new meter.

  • MainComponent - add juce::ComboBox timeSigCombo
  • On change: re-run DownbeatCleaner::clean with new num, repaint waveform
  • For compound meters (6/8, 9/8, 12/8) treat each "main" beat as group of 3 eighth notes
  • Persist user override per-item alongside cache

v2.0.3 - Detection robustness

Sliding-window BPM for variable-tempo material

Source: Alex (90-min symphony) Effort: Large (1-2 days) Risk: HIGH

Current TempoEstimator returns a single BPM for the whole detection. For Beethoven-length classical or any score with tempo changes this is meaningless. Compute BPM in sliding windows (e.g. 30 s) and let the user pick "constant" vs "variable" output.

  • New TempoEstimator::computeVariable(beats, windowSec) returning vector<{time, bpm}>
  • "Insert Tempo Map" - Variable mode already exists per-beat; add windowed mid-grain mode
  • UI: small line plot under BPM label showing tempo curve

OnsetRefinement memory fix instead of skip

Source: Alex Effort: Medium Risk: MEDIUM

Currently we skip refinement past 10 minutes (saves the 3.8 GB magnitudes matrix). Better: stream the spectral flux frame-by-frame so only the running window is in RAM.

  • OnsetRefinement.cpp - rewrite detectOnsets to keep only kAdaptiveWindow (7) frames + previous magnitudes for diff
  • Removes the matrix allocation entirely
  • Should restore sample-level precision on long files

Open user issues (no fix yet)

bobo198504 - Detect Beats stays grayed out on Windows 11 (Unicode home dir suspected)

Source: GitHub issue #1 Status: Awaiting v2.0.2 retest by reporter

User reports Detect Beats button disabled after selecting an audio item, no progress bar visible. Two likely causes:

  1. Model download silently failing in v2.0.1 - no visible progress feedback, user can't tell. v2.0.2 fixes the symptom (visual progress bar, "Downloading model..." button label, portable model fallback).

  2. Unicode home directory path bug in BeatDetector::loadModel (src/BeatDetector.cpp:37):

    std::wstring widePath(modelPath.begin(), modelPath.end());

    This copies bytes 1:1 from std::string (UTF-8 from JUCE) to std::wstring (UTF-16). Works only for ASCII. A Windows username with Chinese/Cyrillic/Polish characters produces a corrupted wide string, ORT throws, modelLoaded_ stays false, button stays grey.

Fix (hotfix candidate for v2.0.3):

#ifdef _WIN32
    int wlen = MultiByteToWideChar(CP_UTF8, 0, modelPath.c_str(), -1, nullptr, 0);
    std::wstring widePath(wlen > 0 ? wlen - 1 : 0, L'\0');
    if (wlen > 1)
        MultiByteToWideChar(CP_UTF8, 0, modelPath.c_str(), -1, widePath.data(), wlen);
    session_ = std::make_unique<Ort::Session>(*env_, widePath.c_str(), opts);
#endif

Also audit ModelManager paths - JUCE's String::toStdString() returns UTF-8 but the assumption should be explicit.

Trigger hotfix if bobo confirms v2.0.2 download succeeds but plugin still doesn't load.

reaperfreaker - Debian Trixie not loading

Source: Forum post #6 Status: Awaiting diagnostics

Need from user:

  • ldd ~/.config/REAPER/UserPlugins/reaper_reabeat-x86_64.so
  • REAPER --loglevel=verbose 2>&1 | grep -i reabeat (if REAPER supports a verbose flag)
  • dmesg | tail -20 after attempting to load

Candidate causes:

  • glibc/libstdc++ ABI: we build on Ubuntu 22.04 (glibc 2.35) - Debian Trixie has 2.41, should be forward-compatible
  • Wayland-only session: REAPER under native Wayland can't open X11 socket, our XBridge fails
  • Missing system libs: libfreetype6, libcurl4, mesa-utils

Mitigations to consider:

  • Build on Ubuntu 20.04 instead of 22.04 (older glibc baseline)
  • -static-libstdc++ -static-libgcc link flags

reaperfreaker - macOS Mojave 10.14 support

Source: Forum post #6 (separate from Debian) Status: Open, not yet implemented

Plugin currently builds with CMAKE_OSX_DEPLOYMENT_TARGET=10.15 for the Intel build. ORT 1.16.3 (now bundled for darwin64) actually supports minos 10.14 per otool -l, so the ORT dep is fine. Only the plugin's own deployment target is blocking Mojave.

To enable Mojave:

  • Lower CMAKE_OSX_DEPLOYMENT_TARGET to 10.14 in .github/workflows/build.yml macOS-x86_64 matrix entry
  • Verify no C++20 features used that require macOS 10.15+ (probably fine - main culprit would be filesystem APIs)
  • Test that built plugin actually loads on Mojave (REAPER 7+ supports Mojave per their forums)

Effort: ~30 min. Risk: LOW (worst case we revert to 10.15). Trigger if reaperfreaker confirms he still uses Mojave.

flark - Font size adjustment

Source: Forum post #32 (Linux user, "getting old") Status: Open, no specific plan

User wants larger text in the plugin UI. Currently sizes are hardcoded throughout MainComponent.cpp and WaveformView.cpp.

Options:

  • DPI scaling already partially handled by JUCE - audit and fix any hardcoded font sizes
  • Add a "UI scale" combo in plugin (1.0x / 1.25x / 1.5x / 2.0x)
  • Multiply all juce::FontOptions(N.0f) calls through a single scale factor

Effort: ~2-3h. Risk: MEDIUM (visual regressions possible). Probably v2.0.4+ once we have a feel for who else wants it.

Long-term ideas

  • C++ unit tests - none exist today. At minimum: MelSpectrogram numerical regression test vs Python output, BeatInterpolator gap-fill logic, ReaperActions stretch marker construction (without actually calling REAPER API).
  • Cancel button during detection - reuse "Detect Beats" as "Cancel" while detecting. Pattern from reamix.me_native: alive_ shared atomic + threadShouldExit() between stages.
  • Free mono buffer after resample - release ~900 MB on long files. Requires changing detect() signature to take rvalue ref or moving the call site.
  • Compound meter heuristic in TimeSigDetector (BPM > 130 && bar=2 → 6/8 hint). Better than nothing, but Daodan #9 dropdown is more honest.
  • CI maintenance:
    • ubuntu-22.04 GitHub-hosted runners will be removed at some point - bump to 24.04 when needed (or stay on older for glibc compat)
    • actions/checkout@v4, actions/upload-artifact@v4 use Node 20 which is deprecated June 2026
    • windows-latest redirects to windows-2025-vs2026 by June 15, 2026 - confirm v2026 toolset works
  • Forum:
    • Post v2.0.2 announcement (draft ready in drafts/forum_post_v2.0.2.txt)
    • Email Alex Shturmak (draft ready in drafts/email_alex_shturmak_v2.0.2.txt)
    • Reply per user after v2.0.2 retest reports come in
  • Sample-perfect alignment for long files - if streaming OnsetRefinement still can't keep up, alternative: run refinement only within a small window around each model-predicted beat (sparse, O(beats x small window) instead of O(audio_length)).

Post-v2.0.2 verification still needed

Just shipped, not yet confirmed by external users:

  • 80icio retest on Catalina Intel (was ORT 1.20.1 incompatibility - now 1.16.3)
  • fightclxb / squibs retest on Windows 11 (was missing VC++ Redist - now statically linked)
  • plush2 retest of Insert Stretch Markers on Windows 11 (was silent fail - now status messages)
  • Mercado_Negro retest on macOS Tahoe + M4 (was MIDI item false-enable - now disabled)
  • gkurtenbach retest of Insert Tempo Map preserving other-item markers
  • bobo198504 retest of Detect Beats button (waiting since GitHub Issue #1 reply)
  • Daodan retest of 5 UX points (item drag, suggestion color, N hint, dock activate, download progress)
  • Alex Shturmak retest of long-file detection (waiting since email reply queued in drafts/)