Following lists are public; followers are visible to the user and to mutual followers.
Adds GET /users/{id}/followers and /following, exposes isFollowing and canViewFollowers
on the profile, and returns isFollowedByViewer from user search so the web player no
longer calls /is-following per user.
Discover used a hardcoded list of genre tiles that searched track titles,
and the upload form kept its own copy of the genre list and compatibility
rules. Both now come from the backend.
Backend:
- GET /genres returns genres with at least one track and their counts.
- GET /genres/available returns every assignable genre with the genres
it cannot be combined with, so the rules live only in
GenreCompatibility.
- GET /tracks accepts ?genre= (case-insensitive, 400 on unknown values).
web-player:
- Discover tiles are built from /genres; each opens /explore?genre=X, a
song list filtered by that tag. Unknown genres fall back to a label
derived from their name.
- The upload form loads /genres/available, disables incompatible genres
as you pick, and the server validates before reserving upload URLs.
- lib/data/trackGenres.ts is removed; its helpers merge into
lib/data/genres.ts.
- Regenerate the typed API client.
The counted-listen tests lived in one catch-all UncountedListensTests
file. Each now sits next to the handler it covers, and the cleanup
handler gets its own test file.
"Listening time this week" summed the full duration of every track whose
stream ticket was issued, so pressing play and skipping counted the whole
song. Playback is now measured on the client and reported to the API.
Backend:
- ListeningHistory gains PlayedSeconds, LastProgressAt and IsCounted
(migration TrackPlayedSeconds marks existing rows as counted).
- /tracks/{id}/stream returns a listenId; new
PUT /tracks/listens/{listenId}/progress stores the accumulated seconds,
capped by track duration and elapsed wall-clock time, monotonic and
owner-only.
- A listen counts once it reaches min(30s, 50% of the track). Weekly
stats, streak, recent history/albums, mixes and listensCount only use
counted listens; seconds sum PlayedSeconds.
- ListeningHistoryCleanupJob (daily) deletes uncounted listens older
than 8 days.
web-player:
- ListenTracker accumulates played time (seeks are ignored) and flushes
on pause, track change, ended, every 30s, and on pagehide.
- Proxy route /api/listens/[id]/progress; repeat starts a new listen per
loop.
Remove deploy/nginx and the nginx service from compose.prod.yml: the
shared edge in Infrastructure owns ports 80/443 and serves every
domain. Make the deploy job fail when zitadel-init or seaweedfs-init
fail, and document the runner and edge setup.
0.65 was still too dark per feedback. The original concern that made
us drop accentL from 0.74 (accent surfaces stacking too bright over
the darkened spotlight/hero cards) is already fixed architecturally by
the earlier color-mix() conversion, which makes bg-accent-tint/
bg-accent-btn a solid precomputed value that no longer stacks with
whatever's underneath. So there's no reason to keep accentL dimmed for
that reason anymore.
0.55 (dropped from the original 0.74 to fix accent surfaces looking
too bright when stacked on the darkened spotlight/hero cards) made
bg-accent-tint on the plain page background barely visible. Split the
difference at 0.65, closer to how it looked originally.
--mf-accent-soft-bg/--mf-accent-btn-bg/-hover were translucent alpha
overlays, so an accent-tinted element (e.g. the active-row highlight,
or the accent Button) rendered differently depending on what was
already underneath it: fine over the plain page background, but over
an already-tinted surface like the spotlight card, the two translucent
layers stacked and the combined result read brighter/more saturated
than intended.
Same fix as bg-btn earlier: replace the alpha overlay with a solid
color-mix() against --mf-bg, computed once, so the token always
renders identically regardless of what it's placed on top of. Verified
the active-row highlight and the accent button now compute to the
exact same color whether they sit on the plain background or on the
spotlight card.
Per feedback, --mf-hero-bg/--mf-spotlight-bg in dark mode were still
too bright: drop their tc() base lightness from 42 to 38 and alpha
from 0.2 to 0.1, so the cards read as a subtle wash close to the page
background instead of a distinct bright block.
That made bg-accent-tint (Button's accent variant, ListRow/TrackList's
active-row highlight) look comparatively too light against the now
much darker cards, since accentL was still 0.74. Drop dark's accentL
to 0.55 so every accent-derived surface (rings, accent-tint, accent-btn
hover) darkens to match. Text (titleL, still 0.87) is untouched since
it needs to stay legible regardless of how dark the surface under it
gets.
Cleaned up the now-unused ACCENT_LIGHTNESS import/L constant.
MenuItem used text-sm (14px) for its label, which read as oversized
for a dropdown; drop to text-xs (12px). This is a shared component
(AccountMenu, ContextMenu, MobileHeader's menu), so all of them get
the more compact sizing.
Add a border prop to GlassMenu (default true, unchanged for
ContextMenu) and set it false specifically on AccountMenu's dropdown,
per feedback that its border didn't fit.
Dark mode's accent chroma came from the ACCENT_CHROMA constant (0.145,
originally meant as a generic default), used unconditionally for every
accent-derived token regardless of the actual song's extracted vividness
- extractAccent() already clamps a song's real chroma to 0.045-0.11, but
buildThemeTokens() never used it, only the hue. That made every accent
button, ring and highlight uniformly vivid. Drop it to 0.09.
Also soften --mf-accent-title (the accent-colored track title text,
e.g. active rows) from 0.07 to 0.05 chroma, since that was called out
specifically as too strong.
Cleaned up the now-unused C/ACCENT_CHROMA import.
#e6e6e8 still read as too white per feedback. Use #c4c4c8, a clearly
gray tone instead of an off-white, while text-ink still contrasts
comfortably against it.
--mf-cta-strong-bg was #ffffff in dark mode, reading as too bright/
glaring against the dark surfaces around it. Use #e6e6e8 instead, a
still-white but slightly dimmed tone. Light mode's own value (#17171a,
already dark) is untouched.
secondary used text-fg-2 (--mf-text-2), which in dark mode is a fairly
dim mid-gray (#8a8a8e) that read poorly against the dark bg-btn fill,
even though the same class looks great in light mode where --mf-text-2
is a dark, high-contrast gray. Use text-fg instead, which is tuned
bright in dark mode and dark in light mode, so both modes get good
contrast from the same class. Drop the now-redundant hover:text-fg
since resting state already uses it.
--mf-hero-bg/--mf-spotlight-bg were multi-stop, directional gradients
in dark mode (fading from a vivid corner to transparent), unlike light
mode which uses the same color at both gradient stops so it reads as
flat. Match that: both now use a single tc() tone repeated at both
stops, so the hero and spotlight cards render as one solid tint in
dark mode too.
The two radial glows behind app-backdrop were fairly small (52%/38%
and 46%/34% ellipses) and faded out by half their radius, so they only
covered the top portion of the viewport at a fairly strong 0.26/0.2
alpha. Enlarge both ellipses and push the fade-out point out to 88%,
so the tint reaches much further across the screen like light mode's
flat wash does, while lowering the alpha so it reads as subtle instead
of a bright patch in the corners.
Sidebar and Queue painted their own opaque bg-bg, which sat in front of
the fixed, full-viewport app-backdrop div (z-index: -10) that carries
--mf-ambient's radial-gradient glow in dark mode. Since the main content
area has no background of its own, the glow only ever showed through
there — sidebar and queue blocked it with a flat, gradient-less fill of
the same base color.
Drop bg-bg from both so the same app-backdrop (color + ambient gradient)
shows through underneath them too. In light mode this is a no-op
(--mf-ambient is 'none' and app-backdrop's flat color already matches
what bg-bg painted), so only dark mode is visually affected.
Per feedback, the solid color-mix bar was wrong — sidebar, queue and
the mobile header should stay on the plain --mf-bg (the same subtly
hue-shifting black used everywhere else), exactly like light mode
already does, not a separately-boosted colored panel. Drop --mf-chrome-bg/
--color-chrome and switch those three back to bg-bg.
The general dark-mode richness bump (tc(), --mf-bg/--mf-elevated/etc.)
and the glass-bar background-color fix from the previous commit stay;
only the separate chrome treatment is reverted.
Mirror the earlier light-mode pastel/colorful pass for dark mode:
boost tc()'s lightness and chroma, and richen --mf-bg/--mf-elevated/
--mf-panel-bg/--mf-bar-bg/--mf-hairline the same way, so dark mode
reads as colorful instead of near-grayscale.
Sidebar, Queue, and MobileHeader used bg-bg directly, so in dark mode
they stayed almost black regardless of hue. Add --mf-chrome-bg (an
alias to --mf-bg in light mode, a solid color-mix of --mf-hero-tint
into --mf-bg in dark mode) and switch those three to bg-chrome, so
they read as a full, solid colored bar in dark mode instead of a
fading gradient (per feedback: a whole bar, not a gradient).
Also fixes glass-bar (PlayerBar/TabsBar): it set
`background-image: var(--mf-bar-bg)`, but --mf-bar-bg is a plain
color, not a valid <image>, so the declaration was silently dropped
and these bars never actually showed their tint in either mode. Changed
to background-color.
ApiTrackLike/QueueItem had no duration field, so toQueueItem(s) always
dropped it and toTrack() hardcoded a new track's duration to 0, relying
entirely on the <audio> element's durationchange/loadedmetadata events
to fill it in once that track actually became current. Upcoming queue
rows never reach that point, so "A continuación" always showed a blank
duration even though the backend already returns it.
Add duration to both types, thread it through toQueueItems/toQueueItem/
trackFromQueueItem, and seed toTrack() from it. The real-time audio
sync still overwrites it once a track actually plays, so this only
fills the gap for tracks that haven't played yet.
Also map MixItemApplicationResponse's durationSeconds through
mixItemTrack, since mixes use a differently-named field.
Admin
NotS1mp