Skip to content

README.md

Latest commit

 

History

History
127 lines (104 loc) · 6.45 KB

README.md

File metadata and controls

127 lines (104 loc) · 6.45 KB
Raw

libopenboxxx-export

Standalone C++17 library that writes an unencrypted rekordbox/CDJ USB image (export.pdb + per-track ANLZ files) from a plain intermediate model. No Qt, no Mixxx dependency — that keeps the byte-fiddly serialization testable in isolation before it's wired into Mixxx (see ../docs/export-design.md, Phase 0).

Layout

include/openboxxx/   public headers (model, byteio, writers, exporter, reader)
src/
  pdb/     export.pdb writer (LITTLE-endian DeviceSQL pages) + device_sql_string
  anlz/    ANLZ .DAT writer (BIG-endian tagged sections)
  usb/     PIONEER/ path assignment
  mapping/ rekordbox colour/rating/key encodings (invert Mixxx importer)
  verify/  round-trip check (tier 1) — stub pending Kaitai parser wiring
  diag/    diagnostic manifest for the in-app bug reporter
  adapter/ mixxxdb.sqlite -> ExportModel reader + beats-blob protobuf parser (Phase 1)
tools/     openboxxx_export_cli   — Phase 0 driver / smoke test
           openboxxx_from_mixxx   — Phase 1 reader CLI (real library -> USB)
tests/     dependency-free unit tests (+ SQLite fixture tests for the adapter)

Build & test

cmake -S . -B build
cmake --build build
ctest --test-dir build --output-on-failure
./build/openboxxx_export_cli               # smoke test on a built-in demo model
./build/openboxxx_export_cli --out /tmp/u  # also write the USB image to disk

Verification harnesses (optional, read-back oracles)

Two independent parsers prove our bytes are real rekordbox format:

  • anlz_oracle — parses our ANLZ with pyrekordbox (MIT).
  • pdb_roundtrip — compiles third_party/crate-digger/rekordbox_pdb.ksy (EPL-1.0, test-only) to a throwaway parser via kaitai-struct-compiler and reads back our export.pdb.
pip install -r tools/requirements.txt          # pyrekordbox + kaitaistruct
brew install kaitai-struct-compiler            # (or your platform's package)
cmake -S . -B build -DOPENBOXXX_ORACLE_PYTHON=$(which python3)
ctest --test-dir build   # openboxxx_export_tests + anlz_oracle + pdb_roundtrip

# or run either directly:
python3 tools/verify_anlz.py OURS.DAT --ref /path/to/real/ANLZ0000.DAT
kaitai-struct-compiler -t python --outdir /tmp/p third_party/crate-digger/rekordbox_pdb.ksy
PYTHONPATH=/tmp/p python3 tools/verify_pdb.py OURS/PIONEER/rekordbox/export.pdb

Each oracle test registers only if its tools are present, so environments without them simply skip it.

Export from a real Mixxx library (Phase 1, reader path)

If SQLite is available at configure time, the build also produces openboxxx_from_mixxx — a standalone CLI that reads a Mixxx mixxxdb.sqlite directly (no Mixxx build required) and writes a CDJ USB image. The eventual in-Mixxx export job fills the same ExportModel, so all mapping decisions are shared.

# dry run — read the library and report counts, write nothing
./build/openboxxx_from_mixxx --db "$HOME/.mixxx/mixxxdb.sqlite"

# write the PIONEER/ tree (and optionally copy the audio for a playable stick)
./build/openboxxx_from_mixxx --db /path/to/mixxxdb.sqlite --out /Volumes/USB --copy-audio

#   --limit N          export only the first N tracks (quick tests)
#   --ids A,B,C        export only these Mixxx track ids (targeted test sets)
#   --no-intro-outro   don't map Mixxx Intro/Outro cues as memory cues

The DB is opened read-only; the tool never modifies the user's library. Units decoded and validated against a real 2.x library: cue positions are fractional stereo samples, beatgrid first-beat is in frames, colours are 0x00RRGGBB. macOS default DB path is ~/Library/Containers/org.mixxx.mixxx/Data/Library/Application Support/Mixxx/mixxxdb.sqlite.

Status (Phase 0)

Piece State
byteio explicit LE/BE ✅ implemented + tested
device_sql_string encoder ✅ implemented + tested
mapping colour/rating ✅ implemented + tested
ANLZ container + PPTH + PQTZ ✅ implemented; validated via anlz_oracle
ANLZ PCOB / PCPT cue entries ✅ layout validated via anlz_oracle (cue ordering TODO)
PDB page allocator (PdbPage) ✅ framing validated via pdb_roundtrip
PDB tracks + lookups + playlists (buildExportPdb) ✅ round-trips via pdb_roundtrip
PDB full 20-table set + colours palette ✅ emitted; round-trips (colour_id resolves)
verify round-trip (native C++) ⛔ TODO — port the read-back into the lib for in-app use

The MVP export (all 20 standard tables incl. the 8-colour palette; tracks with colour_id, artists/albums/genres/keys; playlists; ANLZ beatgrid + hot/memory cues) now round-trips through independent parsers. Remaining before a hardware test: refine the observed track_row constants / index_shift across rekordbox versions, and add artwork. ANLZ Phase-2 (cosmetic): vbr, wf_preview, wf_tiny_preview.

Status (Phase 1 — reader path)

Piece State
beats blob protobuf parser (BeatGrid-2.0 / BeatMap-1.0) ✅ implemented + unit-tested
mixxxdb.sqlite reader → ExportModel ✅ implemented + SQLite-fixture tested
openboxxx_from_mixxx CLI (real library → USB) ✅ works; audio copy behind --copy-audio
Validated on a real ~2,900-track library ✅ export.pdb round-trips; ANLZ cue/beat times match the DB exactly
Tier-2 diff vs genuine rekordbox ANLZ ✅ shared sections (PPTH/PQTZ/PCOB×2) match field-for-field; only vbr+waveforms missing (Phase 2)
— found + fixed: negative-position hot cue dropped ✅ now clamped to 0 (was silently losing a hot cue)
— found + fixed: duplicate memory cue (MainCue≡Intro) ✅ now deduped by millisecond
In-Mixxx RekordboxExportJob + dialog + Library-menu button ✅ built (behind REKORDBOX_EXPORT) in a Mixxx fork; fills the same model from live objects
In-Mixxx "Cue Sheet → Tracklist" tool ✅ built (.cue → .txt)
Cross-platform build ✅ compiles + links on macOS Intel/ARM, Windows, Linux (CI green); downloadable Releases
Import test in rekordbox desktop (tier 3) ⛔ pending — not yet confirmed
Physical CDJ/XDJ hardware (tier 4) ⛔ pending — the real proof; testers welcome

Known minor deviation (not yet changed): rekordbox writes the hot-cue PCOB before the memory PCOB; we emit memory first. Readers key off each list's cue_type field, not position, so this is cosmetic — left as-is until a populated real reference confirms the convention.