When rekordbox updates
For developers
This page is about checking and fixing platendoos for a new rekordbox version. If you just updated rekordbox, run the patcher again: see Installation.
What survives a rekordbox update by itself, how to check a new version before anyone runs it, and how to fix what broke. The macOS side is described here; the Windows side (a table per rekordbox build) is in Windows internals, under Finding things in rekordbox.exe.
What the mods depend on
Nothing in the mods is a fixed address. Three kinds of things are looked up in rekordbox:
| Kind | Written as | Found by | When it isn't found |
|---|---|---|---|
| Functions, vtables, globals | Addr A_x{"<mangled symbol>"} | the symbol's name, when rekordbox starts | the mod that uses it is switched off and logged |
| Fields inside rekordbox's objects | Offset O_x{0x64, "<function>", "ldr x20", 9} | "the number in the 9th ldr based on x20 inside that function": the field's offset as rekordbox's own code uses it | the mod is switched off and logged |
| Call sites the mods take over | HOOK(name, caller, nth, count, target, kind) in hooks.def | "the nth of the count calls to target inside caller", when the app is patched | that hook is left out, and what depends on it doesn't run |
So an update that only moves code changes nothing. A field that moved is picked up at its new offset (7.2.20 moved five effect fields by 8 bytes; betterfx followed without a change). What breaks is a function that was rewritten: a symbol that no longer exists, or a caller that no longer makes the same calls.
The log's first line says how it went:
rbmods: rekordbox build 4AF5D59D5EC937DCBD893FBC84D970AD (checked): 420 addresses, 167 fields
- "tested": the build the expected values were written for (
kTestedBuildincore/image_mac.mm, rekordbox 7.2.19). There a recipe that gives another value than expected is only logged ("field recipe … out of date"). - "checked": a later build every mod was run on (
checkedBuild(), 7.2.20). Same rules as an untested build, but the chip and the patcher don't warn about it. Add a build there once the tests below have passed on it. - "untested": any other build. A mod with anything unresolved is switched off; the PLATENDOOS chip says "untested rekordbox build: some mods may be off" and its window lists every mod and whether it is on.
Checking a new version without installing it
rekordbox's installer can be unpacked instead of run, so the installed rekordbox and the library stay as they are until the new version is known to work.
Unpack the installer:
pkgutil --expand-full <installer>.pkg <folder>gives<folder>/rekordbox.pkg/Payload/rekordbox.app.Check the field recipes against it:
python3 tools/dev/recipes.py check <rekordbox.app>/Contents/MacOS/rekordboxA recipe that gives another value is a moved field (fine: the mods read the new value) or a function that changed (look at it:
recipes.py find <binary> <function regex> <offset>lists the instructions that use an offset, each with its recipe).Check the hooks:
python3 patch_binary.py <rekordbox.app>/Contents/MacOS/rekordbox /tmp/rekordbox.patchedIt prints
patched 147 of 147 hooksand aleft out:line for each hook it couldn't place. 7.2.20: 128 of 135 at first, the other 7 in three functions rekordbox had rewritten.The patcher app does the same check with its own (C++) patcher:
PLATENDOOS_SOURCE=<rekordbox.app> "platendoos patcher.app/Contents/MacOS/platendoos patcher" --check
Fixing a hook whose call site moved
When the call a hook takes over still exists but the code around it changed (another caller, another count), give the hook a second site under the first:
HOOK(stem_sep4, "<caller in 7.2.19>", 0, 1, "<target>", -1)
HOOK_OR(stem_sep4, "<caller in 7.2.20>", 0, 1, "<target>", -1)
HOOK_OR is another site for the HOOK above it, tried when that one isn't found. It is no hook of its own: the mod's code stays the same, and both rekordbox versions keep working. Both patchers read it (patch_binary.py and patcher/macho.h; they must give the same result, change both).
When the call itself is gone, the mod needs new code for the new rekordbox: keep the old path for the old build and let the mod say which it found.
Running the new version next to the old one
./build.sh <unpacked rekordbox.app> ~/Applications/platendoos-<version>.app
builds a patched copy of the unpacked rekordbox under its own name; ~/Applications/platendoos.app and the installed rekordbox aren't touched. Before the first start, copy rekordbox's database folder (~/Library/Pioneer/rekordbox) somewhere: a newer rekordbox may update the library, and going back to the old version afterwards isn't tested. Never run two copies at once.
./install.sh <that app> then swaps a rebuilt library into it, as for the normal copy.
Test at least: the log's first lines (every mod loaded, no "switched off"), one analysis of a track whose result is known (same BPM and key as before), one deck load of a Spotify track (playable, waveforms, grid, stems), the track menu, the windows. To try the patcher app's install without touching the real copy: PLATENDOOS_SOURCE=<rekordbox.app> PLATENDOOS_DEST=<a test folder> with --install.
rekordbox 7.2.20
Checked and run on 2026-11-19, next to an installed 7.2.19.
- macOS: every address and field resolved. Seven hooks had lost their site:
Service::getPlaylistTracksandgetLookupTrackswere rewritten (spotifyapi's six hooks), andAudioPartSeparator::startwas split, with the call stemcache takes over now incheckAndPrepare. WithHOOK_ORlines for those, 7.2.19 and 7.2.20 both patch every hook (135 that day, 147 now). All 33 mods load; a re-analysis gives the same result as on 7.2.19; a deck load is playable in 0.1 s; Spotify tracks get stems; the patcher app's check and install pass. Not timed: that several Spotify lists really load at once. - Windows: see Status (rekordbox 7.2.20) in Windows internals. All 27 mods load with the table made for it.