e99fa231dd
Root cause: build_bass_track() never received the kick_cache in main(). Second issue: kick times were WAV-relative (0-12 beats) but bass expects absolute positions (16+ beats). Added loop-duration projection to convert relative → absolute positions across clip duration. 285 CC11 events now generated in output. 302/302 tests pass.
77 lines
3.4 KiB
Markdown
77 lines
3.4 KiB
Markdown
# Proposal: Fix Sidechain CC11 — Zero Events
|
|
|
|
## Intent
|
|
|
|
`scripts/compose.py` populates `_kick_cache` via `_get_kick_cache()` → `DrumLoopAnalyzer.analyze()` but never passes it to `build_bass_track()` at line 767. Result: `kick_cache` parameter defaults to `{}`, zero CC11 events generated in every `.rpp`. Drumloop WAV files exist on disk (verified — all 5 variants present). 302/302 tests pass because tests call `build_bass_track(kick_cache={...})` directly, bypassing the broken `main()` wiring.
|
|
|
|
## Scope
|
|
|
|
### In Scope
|
|
- Pass `kick_cache=_kick_cache` to `build_bass_track()` in `main()` (1 line)
|
|
- Add diagnostic log: kick count per drumloop after cache population (2 lines)
|
|
- Add integration test exercising `_get_kick_cache()` → `build_bass_track()` full path, asserting `midi_cc` non-empty (catches regression)
|
|
|
|
### Out of Scope
|
|
- Refactoring module-level `_kick_cache` state
|
|
- Fixing double-build of `drumloop_track` (separate optimization)
|
|
- Changing CC11 duck depth, shape, or DrumLoopAnalyzer accuracy
|
|
|
|
## Capabilities
|
|
|
|
### New Capabilities
|
|
- `sidechain-cc-diagnostics`: diagnostic output logs kick count per analyzed drumloop at composition time
|
|
|
|
### Modified Capabilities
|
|
- `bass-generation`: `build_bass_track()` now receives live kick cache in production path (was always `{}`); bass clips gain CC11 Expression ducking events
|
|
|
|
## Approach
|
|
|
|
**Fix (1 line, `scripts/compose.py`):**
|
|
```python
|
|
build_bass_track(sections, offsets, key_root, key_minor, kick_cache=_kick_cache),
|
|
```
|
|
|
|
**Diagnostic log (after `_get_kick_cache()` call):**
|
|
```python
|
|
for path, kicks in _kick_cache.items():
|
|
print(f" Kick cache: {len(kicks)} kicks in {Path(path).name}")
|
|
```
|
|
|
|
**Test:** New `test_bass_track_populates_cc_from_main_path` in `test_compose_integration.py` — calls `_get_kick_cache()` with real drumloop paths, then `build_bass_track(kick_cache=_kick_cache)`, asserts at least one bass clip has `len(midi_cc) > 0`.
|
|
|
|
**Why existing tests missed it:** All 4 sidechain tests pass `kick_cache` explicitly to `build_bass_track()`, so the `main()` wiring gap is untested.
|
|
|
|
**Verified non-causes:**
|
|
- Pipeline: `CCEvent` dataclass ✅, `ClipDef.midi_cc` ✅, `RPPBuilder` emits `B0 0B` E-lines ✅, `build_bass_track()` filter+triplet logic ✅
|
|
- Disk: all 5 drumloop WAV files exist at `ABLETON_DRUMLOOP_DIR`
|
|
|
|
## Affected Areas
|
|
|
|
| Area | Impact | Description |
|
|
|------|--------|-------------|
|
|
| `scripts/compose.py` | 3 lines added/changed | Pass `kick_cache`, diagnostic log |
|
|
| `tests/test_compose_integration.py` | +1 test | Full-path kick cache → bass CC regression test |
|
|
|
|
## Risks
|
|
|
|
| Risk | Likelihood | Mitigation |
|
|
|------|------------|------------|
|
|
| DrumLoopAnalyzer fails on WAV | Med | `try/except` returns `[]` per path — no crash, just no ducking for that section |
|
|
| librosa unavailable on CI | Low | Already a project dependency; existing analyzer tests pass |
|
|
| CC events malform REAPER playback | Very Low | Valid delta-encoded E-lines interleaved with notes |
|
|
|
|
## Rollback Plan
|
|
|
|
Remove `kick_cache=_kick_cache` argument. Reverts to zero CC events (current behavior). No schema changes.
|
|
|
|
## Dependencies
|
|
|
|
None — pure wiring fix using existing code paths.
|
|
|
|
## Success Criteria
|
|
|
|
- [ ] `rg "B0 0B" output/test.rpp` returns non-empty results after `compose.py` run
|
|
- [ ] Diagnostic prints kick counts (e.g., "Kick cache: 32 kicks in 90bpm reggaeton antiguo drumloop.wav")
|
|
- [ ] All 302 existing tests pass
|
|
- [ ] New integration test fails before fix, passes after
|