A hand-only digest used to move the cursor past all status rows.
It now stops at the last by_user change, so the sync digest still
gets its status detail lines.
Digest notes now keep the lines behind their counts in
inbox_note_detail (V48). GET /api/inbox/{id}/details shows them
while the reader still follows the subject and it is public.
Add GET /api/library/dashboard-games for the top and playing tiles
in one call. It has no count query.
GET /api/library/games/{appid}?refresh=false answers from stored
data and sets achievementsStale. It does not wait for Steam.
entryPlatforms reads the view once in a materialized CTE:
planning ~33 ms -> 15 ms.
Test CORS Access-Control-Max-Age 3600 on preflight.
Refs #47
speedrun.com returns 429 or 420 when we ask too fast. These were
counted as errors: 3 in a row dropped both queues. Now the thread
pauses and retries the same game. Rate: 1 request per 2 s. Refs #48
Loose name matches gave wrong links. Now a link needs one sure hit:
same name and within 1 year of IGDB, or one hit when no year is known.
V47 marks old matches (not corrected ones) for one re-check. Refs #8
A sync that unlocks the last achievement of a game not COMPLETED
adds an ALL_UNLOCKED change. The overview hides it once the game is
COMPLETED, hidden or removed. Refs #21
Steam sync added a second row for a game the user added by hand (#9).
The manual row moves to the Steam appid and keeps its data, such as
status, notes, sets and PSN links. Manual rows on other platforms stay
and join the Steam game's set.
Refs #44. Check the last achievement unlock, not the first status change.
Old 100% games no longer pass. New reason COMPLETION_TOO_OLD.
Add earlySince to the games and wallet GET responses.
A paid FINISHED_MANUAL on a set edition blocked COMPLETED for the
Steam game forever. It now opens COMPLETED, but only when the Steam
game's first finish is on or after app.rewards.early-since. Refs #35
Played, never started, under one hour and oldest unplayed used the
counted edition's own minutes, so a set played only on PS5 showed as
never started. They and their drill-down lists now use the set total
(user_game_entry_playtime). Refs #38. Nightly snapshot (#43) unchanged.
RewardControllerTest leaves user_game_playtime_day rows for user 1,
affecting paceWeeks calculation. Delete them in seed() to ensure
clean test isolation.
A linked PlayStation title no longer blocks. Its playtime never
proves a finish. An unproven Steam finish pays 10 early points,
the rest comes with proof. Back pay only from app.rewards.early-since.
Sort, total playtime, most played, genres, moods and the public
profile use the set total of an entry, as the library card does.
New view user_game_entry_playtime (V45) is the one SQL rule.
Store PSN play per title. A PS title on a Steam row keeps its minutes.
Steam rtime_last_played now fills last played.
Dropped? questions read the newest PSN last played too.
Gitea #34 slice 2 of 3. Lists and stats count a set once.
V43 views map an entry to its editions and platforms.
Remove the hideDuplicates param; GameDto gets platforms[].
One entry per game across platforms (Gitea #34, slice 1 of 3).
Same IGDB game is linked; a user can split one key off its set.
V42 keeps status, rating and notes per set; rewards pay once per set.
Add open questions per game: Start it?, Finished?, Dropped?, Rate it?.
A "No" is saved in user_game_question (V41) and snoozes the question.
Account cleanup clears the new table too. Refs Gitea #32.
Sets map by exact title id, else by name and platform.
Trophy gains count as achievements; platinum or all earned is perfect.
Max 30 PSN calls per run; trophy lists load only when changed.
Hidden trophies stay hidden (name and detail) until earned.
Exact IGDB matches are added; unsure titles wait in a review import.
Playtime and last played come from PSN: summed per game, not editable.
PSN games give no Pile Points. Users with only a PSN link sync too
(V39 drops steam_id NOT NULL). Linking starts a sync, or schedules one
for the end of the cooldown. Inbox notes 2 days before the PSN link
ends and when it has ended.
The NPSSO is used once to get a refresh token; only that token is
stored, sealed with AES-GCM (SecretBox, key APP_PSN_TOKEN_KEY; no key
= feature off). One-time import of played PS4/PS5 titles through the
import pipeline as source PSN. Unlink keeps the games. One PSN call
per user at a time. IGDB search now sends UTF-8, so non-ASCII names
like "Yōtei" match again.
Backlog games a followed user finished or played >= 5 h. One game per
friend; strongest signal first (finished, then playtime). Requires an
active follow of a public profile that shows SUMMARY or STATUS. Skips
DROPPED games, blocked and deleted users, hidden and removed games.
New V37 tables game_credit and game_credit_check store IGDB
collections and developer/publisher companies per game.
A scheduled job backfills existing games (250 per 30 s);
sync, import, manual add, merge and detail page refresh them.
SERIES row: next part of a collection the user played.
STUDIO row: backlog games from developers of top played games.
Picker returns categories in fixed order: NEXT_UP, SIMILAR, LONGEST,
RANDOM, each with up to 3 games. Empty rows are left out.
Param `count` is removed; new optional `category` loads one row (reroll).
Budget and exclude apply to all rows; mood only to RANDOM.
SIMILAR takes the 3 most played games, one similar game each.
Remove PickMix.
Picker slots rotate three sources: backlog games that share moods with
one of the top 5 played games, oldest first seen, and random. Each game
carries a reason (source, anchor game, first seen). Empty slots fall
back to random. Issue #28.
V32 stores the last role change per user, sent by Auth. A token
issued before that change, or of a blocked user, gets 401 within
about 30 s instead of working until it expires.
A visit starts a background sync when the last successful sync is
older than 20 h; one try per hour, one sync at a time. Migration V31
adds last_success_at and login_sync_tried_at. The nightly sync skips
only users with a successful sync today after 23:00. games-count
returns syncStarted.
Allows dashboard to show what the last sync changed. Migration V30
creates sync_change and sync_result tables. Sync records additions,
removals, playtime, achievements, and status changes per game.
The number and list share the same SQL query, so they cannot drift.
StatKey enum maps each statistic to its query, validation rules, and visibility.
StatGamesService reads games per key with 40-game pages; repositories share the
same set logic as the stat counts. Three endpoints open games: own library stats,
public profile stats (with ticket checks), and public landing-page stats (cached
10 min). Rewards track and bonus rows support drill-down too.
Five progress tracks: games finished, completed, weekly finish
streaks, pile-down months and play-more months. Settle books UNLOCK
rows once, also retroactively, when a track reaches an item's target.
12 unlock-only items (incl. the Steam skin) and 3 skins to buy. The
wallet returns progress; buying an unlock item answers NOT_FOR_SALE.
V29 adds game_merge_undo and game_merge_undo_user: every new merge saves
a copy of what it changed. POST /api/library/merges/{mergeId}/undo
(auditors, newest merge first) puts it back; users keep later status
and rating, nothing is paid twice. A nightly job drops copies after 30
days. Old keys follow game_alias to the kept game. Manual games can now
be hidden and shown like Steam games.
GET /api/library/catalog finds any game a user has, by name, for
auditors; GET /api/library/catalog/{appid} loads it for the correction
dialog. New correction field TIME_TO_BEAT (normal + completionist,
1..60000 min, completionist >= normal) that survives the IGDB refresh.
Every correction and reset writes GAME_CORRECT / GAME_CORRECTION_RESET
to admin_audit.
GET /api/library/audit-log for auditors and admins, paged, filter by
action. The account state push stores displayName in account_name so
the log shows names. Contract test updated.
All game counts read the counted_user_game view, which keeps only the
most advanced edition of a linked set (older edition wins a tie). Add
the hideDuplicates filter to the library query.
Tables game_edition (IGDB "same game" id per library key), game_link
(global sets set by auditors), game_link_pair (queue of IGDB and
requested pairs), game_link_request and account_name. Fetch IGDB
editions, link and unlink with audit rows, serve the paged auditor queue
with request counts. WireMock recordings for the editions query.
Add endpoints to buy items, equip looks by slot and unequip. The shop has
8 whitelisted looks: avatar borders, badges, accents and banner frames.
Ownership is a BUY ledger row; PLATFORM_ADMIN owns every catalog item
without buying. The public profile endpoint shares only the equipped looks
filtered by ownership.
New module rewards tracks Pile Points in an append-only ledger. A settle
derives rows from the status log, playtime days and achievements, once per
user, kind and game. The balance is the sum of all rows. Settle runs on
every wallet read and nightly at 02:30. The unique key (user_id, kind, ref)
makes every run idempotent. V26 also pays a one-time starter grant for
finishes from before the status log.
Add PLATFORMS stat group to user stats: games and playtime per platform.
Add platforms, genres, and finished-per-year statistics to public stats
for the landing page.
C14 C16 C17 C18 C20 A11 A15 X10 X12 X15: dead Steam count code, GameNames, steamIdText, one query per game for the finished estimate, no API versioning config, open-in-view off, CORS from APP_FRONTEND_URL, non-root image, prune only up to the highest known id.
A blocked user gets no sync and leaves the landing stats. A new or removed Steam ID stops the nightly sync of the old one. No token check in Core: the token lives until it expires.
Every @Async names a pool: with the scheduler there are two TaskExecutor beans, and a bare @Async ran on an unbounded executor. Page views queue only stale games; the long runs after a sync get their own pool.
Profile "local" points Steam, Twitch/IGDB and speedrun.com to a WireMock
container from compose.yaml (recordings in local/wiremock) and checks
logins against the local Auth.
The IGDB lookups now keep only the ids they asked for: an answer with
another game made tags and release years drop the whole batch.
An auditor merges a game added by hand into its Steam game or another
game added by hand (V21). Every user's row moves; a user with both rows
keeps one, with the status, rating and notes set by hand. game_alias
sends later adds to the kept key, and admin_audit logs who merged what.
The game search shows games already in ToBePlayed first.
An AUDITOR corrects name, header image, speedrun link, moods, endless and genres of a game for every user (V20 game_correction, PUT/DELETE /api/library/games/{appid}/corrections/{field}). The value is also written into the cached row, so reads stay unchanged; the Steam sync, the image repair, the IGDB add and the mood and speedrun upserts skip corrected fields. SteamGame gets @DynamicUpdate so a sync no longer writes back columns it did not change.
Every service method a controller calls carries @RequiresRole, @Public or @Internal; a Spring Security method interceptor checks it while a request is served. Core now grants the active role from the token and every lower one. ArchUnit fails the build for an unmarked method, two markers, a controller calling an @Async method, or a service reaching a method that needs more. The background refreshes moved from LibraryController into the services, because an @Async method is never checked.
Auth marks a deleted user in deleted_user (no sync, not in landing stats or pile snapshots) and later purges the rows through /internal/accounts with a service token. Prune removes rows of user ids Auth does not know. Dry run by default.
A game the sync saw for the first time put its whole playtime on that
day (40 h "longest gaming day" after skip_unvetted_apps=0). Growth now
counts at most Steam's last two weeks, and one game gets at most 24 h
per day. V18 drops the bad rows and caps old ones.
Log every status change with a trigger, so no code path can miss it.
Store IGDB genres with the moods. Write a pile snapshot per user each
night. The sync moves an untouched pile game with new playtime to
PLAYING in one guarded update, so a status the user picks wins.
Old finished games get their year from the ending achievement.
GET /api/library/stats serves the groups; the public profile shows
only the groups in the profile ticket.
After an unlink in Auth, or when a verified link took the ID away, the
nightly job kept syncing the old Steam ID for up to 14 days. A
games-count request without a steamId claim now clears user_seen_at.
steam_id stays, so a later link with another ID still soft-deletes the
old games.
The picker takes mood=SPEEDRUN for backlog games with a speedrun.com
game that has at least one verified run. The game detail returns the
link as speedrunUrl.
speedrun.com has no Steam or IGDB link, so games are matched by name
and, when IGDB knows it, by release year (one year off at most). One
queue serves all users at 1 request per second. Results are kept 30
days in game_speedrun.
GetOwnedGames leaves out "unvetted" apps unless skip_unvetted_apps=0
is set. So a released game owned through a developer license was
missing from the library.
Such an app can come without a name. The sync now keeps the stored
name, or uses "App <appid>" for a new game, so the NOT NULL name
column does not fail the whole sync.
Moods come from IGDB genres, themes and game modes and are stored per game in game_mood (V9), refreshed like the time to beat. Picking a mood also refreshes the moods of the backlog in the background, so users without a Steam sync get them too.
A HIGH game is 4 times as likely as a LOW or unset one, a
MEDIUM game twice. Not HIGH first like the picker: the
dashboard shuffles by excluding only the current game, so a
strict order would swap between two games. The picker now
starts a stale Steam sync like the library does.
IGDB lists every e-Reader card as a game, and its own order
puts obscure entries first. Core now drops e-Reader-only
games, asks for 50 hits and sorts exact name first, then by
rating count. When a search finds fewer than 3 games, Core
searches again without doubled letters and with single
words, and adds those hits by Levenshtein distance after
IGDB's own hits.
Also fixes a flaky test that read the time to beat before
the background refresh wrote it.
The search listed DLC and e-Reader levels, and the limit of
10 pushed the main games out. Keep main games, standalone
expansions, seasons, remakes, remasters, expanded games and
ports, and show 20 hits.
Games the user owns outside Steam get the key minus their
IGDB id, or the Steam appid when IGDB knows the game on
Steam. The sync never removes a MANUAL row and takes it
over once the user owns the game on Steam. Users without
a Steam ID can use the library.
Core reads the login token from the tbp_access cookie or a Bearer
header, whichever verifies, and parses it once per request. Unsafe
requests need an X-TBP-Client or Authorization header, CORS allows
only the exact frontend origin, and a missing or expired token
answers 401 instead of 403.
Core finds the key by the token kid and loads the JWKS again for an
unknown kid, at most every 10 s. JWT_PUBLIC_KEY is used only until the
first JWKS load works, so a key removed from the JWKS cannot verify
later. While that key still works, requests do not wait for a slow
Auth.
Achievements were fetched only when a game detail page was opened, and a Steam error was stored as "no achievements" or "hidden". A background run after each library sync now fetches schemas and the unlocks of games with new playtime, one Steam call per second for all runs together, and never overwrites stored data on a Steam error. Adds GET /api/library/achievements/summary, achievement counts on GameDto and the public profile, and migration V7 (playtime_at_refresh).
GET /api/library/picker returns random backlog games for a time budget, HIGH priority first. PUT /games/{appid}/status changes only the status, so the picker does not need the game detail (and its Steam calls) to keep rating, notes and priority.
GET /api/public/profiles/library with X-Profile-Ticket returns totals, status
counts, playing and finished games with rating; never notes, priority or the
Steam ID. Login filters reject the ticket, and Core keeps no copy of the
privacy flag.
Each sync fetches the public wishlist and the store name, price and discount
(batches of 100) and replaces the user's wishlist_game rows, without games
the user owns. A Steam error keeps the old list, a failed store batch keeps
the last known price, and a Steam ID change drops the old account's list.
GET /api/library/wishlist returns it newest first.
Each sync records the growth of playtime per game and day (Europe/Berlin).
The first successful sync of a Steam account is only the baseline, tracked in
user_steam_sync.history_steam_id, so a new or failed account never turns its
whole playtime into one day. A nightly sync at 23:55 runs the recently used
accounts one after the other. GET /api/library/activity returns minutes per
day and the games of Steam's last two weeks.
lobo