Repository navigation
v0.5.0 #51
Replies: 1 comment 5 replies
|
Hey just wanted to say. This is really awesome work! Super excited to see where this project goes and might toss in some contributions soon :) You might find this interesting, it's a rekordbox link export conformance suite. I've been working with Chris a bit on https://rbxport.com and got it passing most of these tests. You might find it useful to get your link export implementation closer to rekordbox! https://github.com/evanpurkhiser/link-spec |
|
Thank you that means a lot! And yes please, contributions are very welcome whenever you feel like it. The conformance suite is exactly what this project needs. The whole link/dbserver side here is reverse-engineered from captures, so today I mostly find out where our responses diverge from rekordbox the hard way, byte-diffing a pcap against the real thing one request at a time. A replayable golden suite against a real rekordbox oracle is the automated backstop I don't have, so I'll stand Vynull up as a backend and see where we land, and ideally wire it into testing. The docs alone look like they'll settle a few things I've had open for a while, the 0x4101 item types and the row layout especially. It's a small world too. prolink-connect and prolink-tools are already woven through this, the route for reading a track off a player's own USB came straight from that, so it's nice to close the loop. And I had a proper look at rbxport, we're clearly after the same thing, right down to the same deliberately-small scope. We've just built it differently: I went for a headless Linux daemon you drive over an API and a browser, where rbxport is a desktop app. Plenty of room for both, and a shared conformance target against real rekordbox only helps everyone working in this space. I'll report back once I've run our output against it. Thanks again for the pointer and the kind words ❤️ |
|
Wow ! I'm also helping on https://github.com/dvucinozd/Pajoniiir and this will be a god send to test DJlink. thx for this work ! |
|
@evanpurkhiser Reporting back as promised! I stood Vynull up as a conformance backend and started closing the gaps. We've gone from 18 to 279 of 423 cases exact across nine device models (not just a CDJ-2000NXS2 ;), and almost every fix was something I had no reason to suspect was wrong, because the decks were perfectly happy with it. A sense of what it caught: category rows served a hash of the name instead of rekordbox's real lookup IDs, track rows had wrong flags and a missing key index, the root menu was in the wrong order, and the genre drill, remixer and original artist categories, and playlists were all off. The docs did a lot of the work on their own, ROW_LAYOUT.md and the 0x4101 item-type reference especially. It also surfaced a couple of real bugs regardless of conformance, a decoder panic on malformed framing and the extended-setup handshake arity. I'll credit link-spec in the Vynull acknowledgements on the upcoming fixes branch. Still plenty left to close, the scalar lists and a few special views, and I'll keep going. Thank you for building this, it has moved the needle more than anything else I could have done. |
|
This is awesome! I think there’s still some cases it didn’t fully cover, I need to go back and see what it still thought was left, but super super glad to hear it caught so many little things. I’m sure it can be a bit pedantic to try and match rekordbox exactly, but I figure it’s good to have a baseline conformance suite :) I’m really excited to check out vynull soon. A headless daemon is actually exactly what I want on my home server. I’m working on a second little project called “cueback” for my RX3, it relies on some firmware modifications that both stream the REC out over TCP and streams the event handlers for device actions (button presses, knob turns), Cueback (another headless daemon) capture all that automatically when you start mixing and produces “sessions” that include full recordings of your mix session, and a full event log of everything you did on the device. I plan to have it run through some model to organize sessions, let you pick out and save your favorite mix ideas (hotcue here, filter there, etc) and probably have some presentable web ui for all of it! vynull would pair perfect with it for providing track metadata etc |
|
Reporting back once more: the full suite is at 423 of 423 now, exact, across all nine device models! The run is up as PR #62 if you want to see what it flagged. cueback sounds like a neat project, and a good fit with vynull. I've an OpenAPI spec (#58) in the works for the whole surface, so pulling track metadata for whatever is loaded should be a clean fetch rather than anything bespoke. I've also started building virtual CDJ emulation into it, standing up player devices that broadcast Pro DJ Link status rather than only acting as a source, validated against prolink-tools, which was a nice full-circle moment :) Thanks again for all your work, on the link-spec and in general! Ping me whenever you want to try it, and I would love to see cueback when it is ready 🙌 |
Uh oh!
There was an error while loading. Please reload this page.
Vynull v0.5.0
This release is shaped by the first round of community hardware reports. Shoutouts to @kayrozen, @przemeklo82, and @hdelplan!
Thank you to everyone who answered the call for testers!
Changes
Plugging a rekordbox USB into Vynull now serves everything rekordbox wrote to it (#36 (comment)). Point --music-dir at a stick and your playlists appear on the decks and in the web UI, tracks carry rekordbox's own beat grids, waveforms and hot/memory cues, and the web UI and the players finally agree on which track is which - the web, the deck and the stick share one track-ID space. Startup is faster too: a stick's own database replaces the tag scan entirely.
A crash reported on the XDJ-AZ is fixed (Issue #37 PR #38): Vynull's colour waveforms could carry values real rekordbox never emits, and the AZ's firmware hard-restarts on them. Credit to the reporter, who isolated the exact bytes against a live deck - the fix is byte-for-byte what they validated, confirmed against two thousand rekordbox-analyzed tracks.
The stream overlay's scrolling waveform no longer drifts and jumps. The playhead is now anchored to the deck's beat packets - the ones that fire exactly on the beat - and small corrections glide instead of snapping, so the waveform tracks the music at any BPM and pitch.
Beat grids get a calibration fix that triples sub-10ms agreement with rekordbox's grids on our reference library, and - important for anyone who fixes a grid by hand - manual beat-grid edits now survive analysis cache upgrades instead of being silently reverted.
The web UI can now create memory cues, not just hot cues (Issue #46 PR #48) - a
+ MEMbutton next to+ CUE, numbered the way rekordbox numbers them, so they flow to decks and USB exports like imported ones.Upgrading re-analyzes your library once (lazily, as tracks are touched). The DIAG tab now captures the full startup log, so if
something goes wrong you can copy it straight from the browser - no more photographing terminals.
Hardware-verified on a CDJ-2000NXS2; the XDJ-AZ fix verified by its reporter on the affected hardware.
Commit log
Full Changelog: v0.4.0...v0.5.0
This discussion was created from the release v0.5.0.
All reactions