- RatingService.getRatingValue(type, userId) centralizes per-type salts/special cases
(pp 6969/0-30, iq 123/0-140, mustache 888/0-100 + id overrides, gay 456, fat 999, boykisser 777)
- Pp/Iq/Mustache/Fat/Gay/Boykisser commands now delegate to it
- RatingOverviewCommand no longer reads random generateratings DB rows; iterates
members (no bots/super-admin) and computes min/max like the real commands do
- birb: some-random-api.com is dead -> api.alexflipnote.dev/birb
- AnimalApi: drop Caffeine URL cache (30s TTL made !neko return the same photo)
- RatingsRepository.getExtremeRating(guild, type, excluded, highest) with NOT IN
- !ratingoverview (super-admin only): per-category max/min, excluding bots + super-admin
- RatingsRepositoryTest: 7 new tests for getExtremeRating
Auxiliary music/radio bots don't enable GUILD_MEMBERS, so JDA checkIntents
threw 'Cannot use MemberCachePolicy.ALL without GatewayIntent.GUILD_MEMBERS'
and the whole cluster crashed at startup. Apply member cache + chunking only
for the main bot; aux bots keep MemberCachePolicy.NONE / ChunkingFilter.NONE
(also saves memory).
lavalink/plugins/youtube-plugin-1.18.2.jar never reached the repo - the
'*.jar' ignore rule silently dropped it from git add. Add an exception and
force-add the jar so Lavalink actually loads the YouTube source.
Lavalink 4.2.2 re-downloads/verifies config-declared plugins on startup and
failed writing to the git-managed plugins bind mount (Permission denied,
container uid 322 vs root-owned files after git pull). Jars in the plugins
directory are loaded directly; remove the duplicate config block.
- COPY build/libs/*.jar instead of hardcoded discord-bot-1.0.0.jar (breaks on version bump)
- lavalink: add /v4/info healthcheck; bot waits for service_healthy before start;
lavalink now depends on spotify-tokener
- lavalink host port bound to 127.0.0.1 (bot talks over docker network; no internet exposure)
- .dockerignore: exclude .github + lavalink/ from build context
- VoiceChannelManager: read trigger_channel_id/category_id live from ConfigRepository instead of caching at construction, so !voice-setup takes effect immediately without restart
- CounterManager: read channel_id/format live from config (same stale-cache fix); count members excluding bots
- CounterSetupPrefixCommand: use bot-excluding member count
- VoiceChannelManagerTest: lenient stubbing for live config reads
- TODO: remove quests, achievements, wiki from roadmap
- LavalinkManager: httpTimeout back to 30s (stagger delays are the real fix)
- checkstyle: comprehensive ruleset from LoParkour (200-line files, complexity, naming)
- PMD: ruleset with CognitiveComplexity, CloseResource, TooManyMethods etc
- Both tools run with ignoreFailures=true (warnings only, don't block build)
- Fix onMessageDelete: search audit log for entry matching message ID
instead of blindly taking l.get(0) (was showing random moderator)
- Increase audit log fetch limit from 1 to 5 for better matching
- Fix onMessageBulkDelete: same approach
- Unify all handlers to use findEntry() for target-matching
- Remove unused handle() helper, simplify send() logic
- Add final to every method/constructor parameter that is not assigned in the method body
- Handle annotations, varargs, generics, throws clauses
- Skip record components, lambdas, catch clauses, enum constructors
- Skip parameters that are reassigned in the method body
- Add spotify-tokener service (ghcr.io/topi314/spotify-tokener:master)
- Configure lavasrc to use it via customTokenEndpoint
- Enable preferPartnerApi for playlist loading via Partner API
- spotify-tokener handles the TOTP token generation properly
- lavasrc 4.8.2 uses /tracks endpoint for playlists (4.8.3 /items
returns 401 for new Spotify apps)
- Remove preferPartnerApi (anonymous token extraction broken on
current Spotify web player)
- playlistLoadLimit stays at 10 (1000 tracks)
RadioManager creates GuildMusicManager with null queueRepository (radio has no
queue persistence). Constructor called loadQueueAsync() which dereferenced null,
throwing NPE after voice connect — bot joined voice but track never loaded.
- remove 30s+30s VoiceHelper.waitForCondition that blocked JDA event thread
- remove abort on voice/player timeout (TrackScheduler retries sendTrack itself)
- remove openAudioConnection() that conflicted with Lavalink voice interceptor
- preserve stream URL across stopRadio/restart instead of losing it
- pass null queueRepository to GuildMusicManager (radio has no queue persistence)
- run startRadio in virtual thread from RadioBotListener/RadioCommand
The local copy of JDAVoiceUpdateListener.onVoiceServerUpdate() checked
if the bot's cached voice channel was null and returned early, silently
dropping the voice server update. Discord often sends the voice server
update BEFORE the voice state update is processed by JDA, so the channel
was always null on first attempt -> Lavalink never received the update
-> voice connection never established -> timeout.
Fix: pass null channelId to Lavalink's VoiceState instead of aborting.
This matches the Kotlin original library code.
The radio bot's own LavalinkClient may not be connected yet when
startRadio() is called (async connection). Without this wait, the
voice server update is forwarded to a not-yet-connected client and
lost, causing a voice connection timeout.
Also ensure the radio bot uses its own LavalinkManager (separate
LavalinkClient with radio bot's user ID) instead of sharing the
main bot's, which caused user ID mismatch.
Radio bot is a separate Discord bot account but shared the main bot's
LavalinkManager (created with main bot's user ID). When the radio bot
connected to voice, Lavalink received voice server updates mismatched
with the expected user ID, causing voice connection timeout.
Now the radio bot creates its own LavalinkManager with its own token,
matching the pattern already used by the music bot.
