Badge and TechText reached into $shared/ui/Label/config, and SearchBar into
$shared/ui/Input/types — bypassing the siblings public surface.
- Label/config.ts was explicitly shared text-styling config, not Labels
internals; relocate it to shared/ui/labelConfig.ts (a neutral peer module)
and point Label/Badge/TechText at it relatively.
- inputIconSize is a consumer-facing map; export it from Input/index.ts so
SearchBar imports it through the barrel alongside Input.
Thread the new createPersistentStore.destroy() through its owners so the
save effect.root is actually torn down: LayoutManager gains destroy();
ThemeManager and typographySettings dispose their #store in their existing
destroy().
comparisonStore had its own leak — a constructor $effect.root whose disposer
was discarded, and a module-level persistent storage created at import. Move
storage into the instance (#storage, created lazily per instance), capture
the effect.root disposer, add destroy(), and adopt createSingleton so reset
runs resetAll() + destroy(). Re-export createSingleton from the $shared/lib
barrel; give the test storage mock a destroy().
The store created an $effect.root for the save-on-change sync but returned no
disposer, so the effect leaked for the life of the process — contradicting
the rule that $effect.root owners must expose destroy(). Capture and expose
the disposer.
- add destroy() to the returned store; covered by tests (flushSync proves the
save effect runs before destroy and stops after)
- trim the bloated header (two near-duplicate @example blocks) to one concise
JSDoc — no fluff
- update typographySettings test mocks to satisfy the now-required destroy()
Consumers (LayoutManager, ThemeManager, typographySettings, comparisonStore)
do not yet call it — threading + the createSingleton migration follow.
Standardizes the getX() / __resetX() pattern hand-rolled identically across
every store: lazy construction on first get(), memoized thereafter, and a
reset() that runs an optional teardown (e.g. destroy()) and clears so the
next get() rebuilds. Lazy by construction, so owning modules stay inert at
import. Covered by unit tests (laziness, memoization, rebuild-after-reset,
teardown-once-with-live-instance, reset-before-get no-op, falsy-value
caching).
Not yet adopted by the stores — that migration is a separate step.
Strip box-drawing (──) section dividers and ===/banner headers — visual
noise with no information. Where a divider label carried a non-obvious
why (VirtualList owns scrolling; mobile footer is md:hidden because header
stats take over) it is kept as a plain one-line comment; pure restatements
of the markup (Header bar, Red hover line, Bottom: fixed controls) are
dropped. Single comment style, no fluff.
queryClient.ts constructed the TanStack client at module eval but was not
in the sideEffects allowlist, so Rollup treated the module as pure — safe
only while a value-importer keeps it alive (D-3).
- replace the eager `queryClient` singleton with a memoized getQueryClient()
factory; construction is deferred to first call
- the module is now genuinely side-effect-free, so no sideEffects exception
is needed and construction can never be legally tree-shaken away
- route all consumers through getQueryClient() (QueryProvider as first
caller; stores via class fields/observers; tests via a local alias; the
fontCatalogStore spec mock now overrides getQueryClient)
import/no-cycle (now active) flagged 17 cycles across 12 files:
- shared/ui self-barrel cycles (Logo/Stat/StatGroup/ComboControl/
FilterGroup/SectionHeader): import siblings relatively instead of
through the $shared/ui barrel that re-exports them
- shared/lib/utils: roundToStepPrecision imports getDecimalPlaces
relatively instead of via the utils barrel
- routes: lazy-load Redirect in the router so it no longer statically
imports a component that imports navigate back from it
- splitArray: replace the comma-operator reduce body with an explicit
block + return (no-sequences); behaviour unchanged
- BreadcrumbHeaderSeeded: declare the bind:this ref with $state() so it
is not flagged as never-assigned (oxlint cannot see template bindings),
matching the rest of the codebase; guard the onMount use
Cleanup surfaced once the oxlint config actually loads (no-unused-vars).
- drop dead locals/imports/params (cachedOffsetTop, elasticOut, key,
unused type imports, unused test imports; _-prefix unused mock params)
- createVirtualizer: keep the _version read (reactive subscription inside
$derived.by) but bind it to _v so it is not flagged
- scrollBreadcrumbsStore.test: keep the removeEventListener mock side
effect, drop the unread spy binding
The native Popover always renders its content (the vertical slider), so the
slider's value label is in the DOM even when closed, and opening is driven by
the browser's declarative popovertarget invoker (not simulated by jsdom on
click). Update the tests to scope value assertions to the trigger and drive
open via showPopover(), matching Popover.svelte.test.ts.
Extract NonRetryableError into its own tanstack-free module and drop the
./api re-export from the Font slice barrel. Importing $entities/Font no
longer transitively loads @tanstack/query-core or constructs the QueryClient
singleton via the ./api and ./lib (errors) chains — light consumers (domain,
types, consts) and unit specs stop paying for TanStack.
Note: ./ui still pulls the stores; addressed separately.
BaseQueryStore pulls @tanstack/query-core. Re-exporting it through the
broad $shared/lib and $shared/lib/helpers barrels made every consumer of
those barrels eager-load TanStack at module-eval time (no tree-shaking in
vitest/vite-node), which is what surfaced the queryClient mock init-order
failure. Its single consumer now imports it by path.
The proxy returned `{fonts: null, total: 0}` for empty results, which
fetchProxyFonts surfaced as a generic Error. fontCatalogStore wrapped it
as FontNetworkError, and TanStack retried 3× with exponential backoff —
pinning the loading skeleton for ~7s before settling on an empty list.
Schema mismatches are deterministic; retrying only delays surfacing the
contract violation.
- shared/api/queryClient: introduce NonRetryableError marker class.
The default retry handler short-circuits when it sees this so any
store using the shared client gets fail-fast behavior for free.
- entities/Font/lib/errors: FontResponseError extends NonRetryableError.
- entities/Font/api/proxy/proxyFonts: throw FontResponseError (was a
bare Error). Document that ProxyFontsResponse.fonts is always an array.
- entities/Font/.../fontCatalogStore.fetchPage: preserve a
FontResponseError raised lower in the stack instead of re-wrapping
it as FontNetworkError.
- features/FilterAndSortFonts/api/filters: throw NonRetryableError on
invalid filters payloads and document the array-never-null contract.
Dark mode unchanged. Targets that were reported as "barely visible" in
light theme:
Surfaces / dividers
- --color-border-subtle (light) bumped from rgb(0 0 0 / 0.05) to
--neutral-300 (matches the Input underline variant's border color and
yields a visible hairline on bg-surface / bg-paper).
- New bg-subtle utility (same color as border-subtle but as
background-color) — used by Divider component and the TypographyMenu
inline column separator. Replaces ad-hoc 'bg-black/5 dark:bg-white/10'
and 'bg-black/10 dark:bg-white/10' bands.
- FontSearch + ComparisonView Search wrapper borders switched from
hand-written 'border-swiss-black/5 dark:border-white/10' to
border-subtle so they participate in the palette.
Muted text
- Button tertiary inactive text (light) bumped neutral-400 → neutral-600
(~2.7:1 → ~7.5:1 contrast). Covers the A/B toggle and the font-list
rows in the sidebar.
- Label/TechText muted variant (light) bumped neutral-400 → neutral-600.
Covers the ComboControl value text.
- Link text aligned to neutral-500 / neutral-400 (subtle but visible).
No behavior changes; pure styling.
Replace inline class clusters with the design-system utilities and
tokens established in the prior two commits. No behavior changes
intended beyond two real bug fixes.
Bug fixes:
- SampleList.svelte: 'border-border-subtle bg-background-40' was a
silent no-op (both classes mis-spelled). Now 'border-subtle
bg-background/40' applies as intended.
- FontList.svelte: 'h-[44px]' → 'h-11' (44px = 2.75rem = spacing-11,
no need for arbitrary value).
Sweeps:
- TypographyMenu: popover + floating bar now use surface-popover /
surface-floating + shadow-popover.
- FontList + FilterGroup: tertiary list buttons use the new
Button layout="block-list-row" variant; skeleton fills use
the skeleton-fill utility.
- Footer / BreadcrumbHeader: surface-floating absorbs the
bg-surface/blur/border cluster. Footer bumped to z-20 with a
comment explaining the stacking against SidebarContainer (z-40/50).
- FontSampler: surface-card + hover shadow-stamp-card token.
- SliderArea: surface-canvas, flex-center, shadow-floating-panel
tokens (light + dark variants).
- Sidebar / Header / ButtonGroup / Layout / SidebarContainer:
bg-surface dark:bg-dark-bg → surface-canvas (8 sites);
SidebarContainer mobile panel uses shadow-overlay.
- Loader / Thumb: flex items-center justify-center → flex-center;
Thumb durations → duration-fast.
- ComboControl: trigger uses surface-card-elevated when open,
popover uses surface-card-elevated, label cluster → text-label-mono,
flex-center for the trigger interior.
- Slider: shadow-sm → shadow-rest, duration-150 → duration-fast.
- text-secondary → text-subtle across Input, Slider, ComboControl
(matches the rename in the styles commit).
- Link: reverted earlier surface-floating attempt — Link's original
bg-surface/80 backdrop-blur pattern was thinner than surface-floating
(no border, smaller blur), and the Footer was overlaying its own
border-subtle on top, fighting the utility. Kept the original style.
- New layout prop with values 'inline' (default) and 'block-list-row'.
The block-list-row variant bakes in full-width, left-aligned content
with trailing icon and text-sm, replacing the ~10-class override
duplicated across FilterGroup, FontList, and similar list-row sites.
- primary variant's three hard-offset shadows now reference the
shadow-stamp-{rest,hover,pressed} tokens; the 0.0625rem translate
becomes translate-{x,y}-px.
- Base classes use text-label-mono and duration-normal utilities
instead of inline 'font-primary font-bold tracking-tight uppercase'
and 'duration-200'.
- The icon variant's background uses surface-canvas (semantic naming;
picks up dark-mode automatically via --color-surface).
- text-secondary → text-subtle (avoids collision with the @theme
--color-secondary token; see earlier styles commit).
New exported type: ButtonLayout.
Long-tail cleanup: threshold and default-value literals in shared
helpers get named module-level constants.
- CharacterComparisonEngine: CHAR_PROXIMITY_RANGE_PCT (5),
DEFAULT_RENDER_SIZE_PX (16) — kept local instead of importing
DEFAULT_FONT_SIZE from \$entities/Font because \$shared/lib cannot
legally upward-import from \$entities per FSD (also avoids an init
cycle through the mocks barrel).
- typographySettingsStore: BASE_SIZE_EPSILON (0.01) — rounding-jitter
guard for baseSize reconciliation.
- createDebouncedState: DEFAULT_DEBOUNCE_MS (300) — exported so callers
can mirror the default.
- createVirtualizer: MEASUREMENT_EPSILON_PX (0.5) — minimum height
delta before committing a re-measured row.
- createPerspectiveManager: PERSPECTIVE_TOGGLE_THRESHOLD (0.5) — the
halfway point on the 0-1 spring that flips isBack/isFront.
Skipped #19 (PerspectivePlan defaults) per review — marginal gain.
- Use existing MULTIPLIER_S/M/L from \$entities/Font in SliderArea instead
of inlining the 0.5/0.75/1 literals (constants already existed but were
duplicated at the call site).
- Centralize API base URL in \$shared/api/endpoints.ts (was duplicated
between proxyFonts and FilterAndSortFonts filters api).
- Promote every 'glyphdiff:...' localStorage key to a named module-level
STORAGE_KEY constant. Test files now import the source constant rather
than redeclaring it (eliminates silent-typo divergence risk).
createFontRowSizeResolver was reaching into TextLayoutEngine, which
internally called pretext's heavy layoutWithLines and then walked
per-grapheme cursors to build chars arrays. The resolver discarded all
that work and used only totalHeight.
Replace with direct prepare + layout from @chenglou/pretext — pretext
docs explicitly recommend layout() (not layoutWithLines) for the resize
hot path: pure arithmetic on cached segment widths, no canvas calls, no
string allocations.
Test spies on TextLayoutEngine.prototype.layout migrated to vi.mock-ed
pretext layout (pretext's ESM exports are frozen — vi.spyOn fails with
"Cannot redefine property").
TextLayoutEngine marked @deprecated since it has no remaining
consumers; slated for removal after a release cycle.
Both names were vague or overloaded:
- fontStore / FontStore -> fontCatalogStore / FontCatalogStore
Three font-related stores live in this slice; the new name names the
paginated catalog specifically.
- appliedFontsManager / AppliedFontsManager -> fontLifecycleManager /
FontLifecycleManager
"Applied" collided with the filter-side appliedFilterStore (different
meaning). The class actually orchestrates a load-use-evict lifecycle
with FontBufferCache + FontEvictionPolicy + FontLoadQueue
collaborators, so "Manager" is justified. Companion types file moved
alongside (appliedFonts.ts -> fontLifecycle.ts).
Directories, file basenames, factory (createFontStore ->
createFontCatalogStore), and the AppliedFontsManagerDeps interface all
renamed. All consumers (ComparisonView, SampleList, FontList,
FontApplicator, FontVirtualList, FilterAndSortFonts bindings,
createFontRowSizeResolver, mocks) updated.
Ilia Mashkov