Implements Phase 7 of CARD_PLAN.md — the entry point that takes a
user-supplied theme zip archive, validates it end-to-end, and
atomically unpacks it into the per-platform user themes directory.
Public API:
import_theme(zip_path) -> Result<ThemeId, ImportError>
Resolves user_theme_dir() and unpacks into <user>/<id>/.
import_theme_into(zip_path, target_root) -> Result<ThemeId, ImportError>
Test-friendly variant that takes the destination explicitly so
unit tests never touch the global OnceLock override.
Safety guarantees enforced:
- 20 MB hard cap on archive size (read from the central directory
before any extraction).
- Zip-slip path traversal rejected via ZipFile::enclosed_name plus a
Component::Normal-only belt-and-braces check.
- Manifest parsed via ron::de and validated via the existing
ThemeManifest::validate (Phase 2) — surfaces named diagnostics for
missing-of-52, unknown keys, duplicate keys, and meta errors.
- Every referenced face + back rasterised through rasterize_svg as a
structural validity check before any bytes hit the destination.
- Atomic install: writes to <root>/.<id>.tmp/ then std::fs::rename
into place, with a recursive copy + remove fallback for cross-
device renames. Failed extraction wipes the staging dir; the user
themes root is never touched on error.
- Id collision with an existing theme dir rejected up front.
7 new tests covering the happy path plus six failure modes (missing
manifest, missing face, oversized archive, zip-slip, missing-file,
id collision). Tests build zips in tempfile::TempDir so they never
touch the real user themes directory.
Workspace deps: zip 8.6 (default-features off + deflate only),
tempfile 3.27 (dev only).
cargo check --workspace --all-targets / clippy --workspace
--all-targets -- -D warnings clean. cargo test could not be run in
this turn because cc disappeared from the sandbox; tests compile
under cargo check --tests and will run on a normal toolchain.
Implements Phase 3 of CARD_PLAN.md — the embedded:// + themes:// asset
sources the card-theme system loads from. The bundled default-theme
manifest ships in the binary via Bevy's EmbeddedAssetRegistry; user
themes load from user_theme_dir() through a FileAssetReader-backed
source registered as `themes://`.
Registration is split across:
register_theme_asset_sources(&mut App)
Called BEFORE DefaultPlugins. Registers `themes://` while
AssetSourceBuilders is still mutable.
AssetSourcesPlugin
Added AFTER DefaultPlugins. Populates the EmbeddedAssetRegistry
that AssetPlugin's build step would otherwise overwrite.
Constants exposed for downstream consumers:
USER_THEMES = "themes" (asset-source name)
DEFAULT_THEME_MANIFEST_URL = "embedded://solitaire_engine/assets/themes/default/theme.ron"
Includes a stub default theme.ron (52 face slots + back) so
`ThemeManifest::validate()` accepts it today; PROVENANCE.md documents
the plan to drop in real SVG art (hayeah/playing-cards-assets) in a
follow-up.
4 new tests covering source registration, embedded-registry
population, manifest validation against the embedded stub, and the
manifest-URL constant matching the embedded asset path.
cargo check --workspace --all-targets / clippy --workspace
--all-targets -- -D warnings clean. cargo test could not be run in
this turn because the C linker (cc) is unexpectedly absent from the
sandbox; the test bodies compile cleanly under cargo check --tests
and will run on a normal toolchain.
Implements Phase 5 of CARD_PLAN.md. Phase 3 (asset sources) and
Phase 7 (zip importer) both depend on this so it goes first.
solitaire_engine/src/assets/user_dir.rs
user_theme_dir() -> PathBuf
Desktop (Linux/macOS/Windows): joins dirs::data_dir() with
"solitaire_quest/themes" — same parent as the rest of the
project's per-user files (settings.json, stats.json, etc.)
Mobile (Android/iOS): reads a process-wide OnceLock populated
by set_user_theme_dir() at entry-point bootstrap. Panics with a
targeted message if the override is missing — there is no
platform default we can guess that won't be wrong inside iOS
sandboxing or the Android storage model.
set_user_theme_dir(PathBuf) -> Result<(), PathBuf>
First-write-wins. Mobile entry points call this before App::run().
The plan suggested the `directories` crate; reused the existing `dirs`
workspace dep instead to keep the dependency surface minimal — both
crates share an author and the platform behaviour we need is identical.
3 new tests covering pure path composition (desktop nesting + empty
root) and a desktop-target-gated check that the detected data dir is
absolute. The OnceLock override is intentionally not unit-tested
because asserting its semantics would pollute global state for any
sibling test that calls `user_theme_dir()`.
Implements Phase 2 of CARD_PLAN.md — the data types and `.theme.ron`
asset loader that build on Phase 1's SVG rasteriser.
solitaire_engine/src/theme/
mod.rs — CardKey { suit, rank } as the HashMap lookup key
(distinct from solitaire_core::Card which carries
per-deal id + face_up state); CardKey::all() yields
the 52 keys in suit-major / rank-ascending order;
manifest_name() and parse_manifest_name() round-trip
via the canonical "{suit}_{rank}" form.
ThemeMeta with structural validation (id non-empty,
no path separators, non-zero aspect components).
CardTheme #[derive(Asset, TypePath)] storing the
53 image handles + meta.
manifest.rs — ThemeManifest { meta, back, faces } with serde for
RON round-trip. validate() returns a strongly-typed
HashMap<CardKey, PathBuf>, surfacing precise errors
for unknown face keys, missing-of-52 entries, and
duplicate keys (RON silently keeps the last; brittle
for a release).
loader.rs — AssetLoader for .theme.ron. Validates manifest, then
composes sibling SVG paths via AssetPath::resolve so
the same loader works for both embedded:// and
themes:// asset sources (Phase 3 territory).
Schedules every face + back load through SvgLoader
with target_size derived from meta.card_aspect.
24 new tests covering: 52-key enumeration uniqueness, manifest-name
round trip, garbage-name rejection, complete/missing/unknown/duplicate
manifest validation, RON round-trip integrity, target-size aspect
math (2:3 → 512x768; non-standard; degenerate 1:10000 clamps to 1px).
Workspace deps added: ron 0.12.
cargo build / clippy --workspace --all-targets -- -D warnings / test
all green (937 passed total — +24 from Phase 2 vs the +7 from
Phase 1's b8fb3fb baseline).
Implements the runtime SVG rasterisation pipeline that the card-theme
system (CARD_PLAN.md) is built on. Bevy 0.18 has no native SVG support;
this loader bridges usvg (parser) + resvg (renderer) + tiny-skia (CPU
pixmap) so the rest of the engine consumes themes as plain
Handle<Image>. Rasterisation happens once per (asset, settings) pair at
load time — Bevy's asset cache absorbs the cost.
solitaire_engine/src/assets/
mod.rs — module entrypoint
svg_loader.rs — SvgLoader (AssetLoader for .svg → Image)
SvgLoaderSettings { target_size: UVec2 } default 512×768
SvgLoaderError (Io / Parse / PixmapAlloc) via thiserror
rasterize_svg() helper exposed for non-asset-graph
callers (the future zip-importer validation step)
The rasteriser scales-to-fit while preserving aspect ratio, centring
the SVG inside the target box so a non-2:3 source doesn't pin to the
top-left corner.
7 new unit tests — default + custom target size, zero-dimension reject,
malformed-input reject, RGBA byte-count, extension advertisement, and
a compile-time guard that SvgLoaderSettings still satisfies the
AssetLoader::Settings trait bounds.
Workspace deps added: usvg 0.47, resvg 0.47, tiny-skia 0.12 (latest
minor versions; CARD_PLAN.md called out the placeholder numbers
needed verification).
cargo build / cargo clippy --workspace --all-targets -- -D warnings
/ cargo test --workspace all green (913 passed, 0 failed, 9 ignored —
+7 from the new loader tests).
Phase 4 ended at 5d57b67 with the doc reflecting "16 commits ahead,
working tree dirty, splash always shows". The next session opened the
binary, found three real bugs (resize lag, off-screen cards,
mis-aligned hit-test), fixed them, and layered four bonus polish
items on top: persisted window geometry, achievement tooltips, a
clippy::pedantic sweep, client-side sync round-trip tests, and the
splash skip on subsequent launches.
Rewrites the doc to reflect the post-push state:
- HEAD: 902560c, branch up to date with origin
- 906 tests pass, working tree clean
- Phase 5 table summarising the nine new commits
- Stale "in-flight" / "deferred" items removed where they landed
- Punch list shrunk to xCards URL, v0.1.0 tag, desktop packaging
Length drops 142 → 121 lines (-21) by trimming the prior session's
"original Track G prompt" callbacks and recovery options that no
longer apply.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Smoke-test report: the user could only initiate a drag from the bottom
strip of a tableau card, not its visible face. Root cause was a fan-
step mismatch between rendering and hit-testing.
card_plugin::card_positions steps face-down cards by
TABLEAU_FACEDOWN_FAN_FRAC (0.12) and face-up cards by TABLEAU_FAN_FRAC
(0.25), so a column with 6 face-down + 1 face-up at the bottom
renders the face-up card at base.y - 0.72 * card_h. input_plugin's
card_position used a uniform 0.25 step for every position, computing
the same card's hit-test centre as base.y - 1.5 * card_h — almost a
full card height below the visible sprite. The hit-test AABB and the
sprite AABB overlapped only over the bottom 0.61 * card_h, which
matches the user's observation that only the bottom of the card
responds to clicks.
card_position now mirrors card_plugin's exact logic: walk the pile's
preceding cards and step by TABLEAU_FACEDOWN_FAN_FRAC for face-down,
TABLEAU_FAN_FRAC for face-up. TABLEAU_FACEDOWN_FAN_FRAC is now public
for the same reason TABLEAU_FAN_FRAC already was — the renderer and
the hit-tester have to agree by construction or this regression
returns.
Updates the existing find_draggable_skips_face_down_cards test that
relied on the old uniform-fan geometry, and adds
find_draggable_hits_face_up_card_with_face_down_cards_above_it as a
regression test that fails without this fix.
The during-drag rendering and pile_drop_rect still use the uniform
TABLEAU_FAN_FRAC because the cards being dragged are guaranteed
face-up, and a slightly oversized drop target reads as forgiving
rather than wrong. Those call sites are intentionally untouched.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
The 1.6 s brand beat is delightful on first launch and tedious on
every subsequent one. spawn_splash now reads SettingsResource and
returns early when first_run_complete is true — the player has
already seen the splash at least once and the onboarding flow that
follows it, so dropping straight into gameplay is the right move.
Reuses the existing first_run_complete signal rather than introducing
a separate splash_seen field; the two states ("I've been here") line
up naturally and avoid carrying a one-shot flag forever.
The first run, a save reset (settings.json deleted), or a headless
test fixture that doesn't register SettingsResource all still see the
full splash — Option<Res<SettingsResource>> defaults to "show" when
absent, so the existing test fixture observes the same spawn it
always did.
Two new tests pin the split: splash_skipped_when_first_run_complete
asserts no SplashRoot spawns when settings say so, and
splash_still_shows_when_first_run_incomplete asserts the first-run
path is unchanged.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Server-side endpoint tests already exist in solitaire_server. This
adds the client-side counterpart: five integration tests in
solitaire_data/tests/sync_round_trip.rs that drive
SolitaireServerClient against an in-process axum::serve harness with
an in-memory SQLite database, covering:
- register_login_push_pull_round_trip — happy path: register, push
non-default stats, pull from a fresh client, assert the merged
payload reflects the pushed values
- pull_after_concurrent_pushes_merges_correctly — two clients on one
user push different games_played values, verify the server-side
merge returns the max
- unauthenticated_pull_returns_authentication_error — pull without
tokens surfaces SyncError::Auth as expected
- jwt_refresh_on_401_succeeds — replace the access token with one
whose exp is two hours stale (same signing key), pull triggers
401 → /api/auth/refresh → retry, asserts the call ultimately
succeeds
- pull_after_account_deletion_returns_default_or_error — register,
push, delete via the trait, confirm the next push surfaces a
result rather than panicking
keyring_core's mock store is installed once per process via Once;
each test uses a unique username so the shared store doesn't
cross-contaminate. Production code in sync_client.rs needed no
changes — the Box<dyn SyncProvider> design plus the mock keyring
were sufficient to drive every flow from outside.
solitaire_server is added as a path dev-dependency along with the
direct crates the harness needs (axum, sqlx, jsonwebtoken, uuid,
chrono, solitaire_sync); no runtime deps changed.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Conservative cleanup pass — applied only the high-signal pedantic
lints whose fixes either remove genuine waste or read more naturally,
skipping anything stylistic that would bloat the diff.
- map_unwrap_or: 29 .map(...).unwrap_or(...) sites collapsed to
.map_or / .is_some_and / .map_or_else equivalents
- uninlined_format_args: 7 production format!/write!/println! sites
rewritten to the inline-argument style; assert! sites in test code
intentionally untouched
- match_same_arms: 2 redundant arms collapsed where the bodies were
identical and the merger didn't obscure intent
Public API is unchanged. No dependencies added or removed. The
pedantic warning count dropped from 840 to 807 (-33). Out-of-scope
findings — needless_pass_by_value on Bevy Res params, false-positive
explicit_iter_loop on Bevy Query iterators, items_after_statements
inside test mods, and the "ask before changing" merge logic in
solitaire_sync — were intentionally deferred.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Each achievement row now carries a Tooltip whose text is derived from
the row's unlock state and the AchievementDef's reward, surfacing
information the row layout doesn't already show.
Four-state policy:
- Unlocked + reward → "Reward: <reward>." (e.g. "Reward: Card Back #1.")
- Unlocked + no reward → "Earned!"
- Locked, non-secret → "How to unlock: <description>." plus
" Reward: <reward>." when one exists
- Locked, secret → no tooltip; the existing row-spawn skip preserves
the achievement's discovery surprise
The row spawn loop tags each row with a new AchievementRow marker so
tests can locate them; the helper tooltip_for_row keeps the policy in
one place.
Six tests pin the policy: one full-flow test for unlocked + reward
mention, one secret-row negative test that asserts no tooltip
contains the verbatim secret condition or the secret reward, plus
four direct unit tests on tooltip_for_row covering all four states.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Settings gains an optional window_geometry field (size + position)
serialized via #[serde(default)] so legacy settings.json files without
the field deserialize cleanly to None. On launch the app restores
the persisted dimensions and position; first run and pre-upgrade
saves keep the existing 1280x800 centered default.
settings_plugin records changes from WindowResized and WindowMoved
into a PendingWindowGeometry resource and writes them to disk through
the existing atomic .tmp+rename path once the events have stayed
quiet for WINDOW_GEOMETRY_DEBOUNCE_SECS (0.5s). A merge_geometry
helper preserves whichever component (size or position) the latest
event burst didn't carry, so a position-only WindowMoved never wipes
the recorded size.
Pure should_persist_geometry and merge_geometry helpers are unit
tested for the boundary cases. Headless integration tests cover the
full flow: a single resize event then a quiet window persists, a
move event after a resize updates only position, a rapid storm
collapses to the final size, and a quiet frame with no events
leaves the geometry untouched.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Smoke-test report: window resize was still laggy after the card-side
throttle landed. Diagnosis pointed at the wgpu / OS layer rather than
ECS work — Bevy's default PresentMode is AutoVsync (Fifo), which gates
every frame on the monitor's vblank. On X11 / Wayland the compositor
sends WindowResized events at high frequency during a drag and the
vsync gate stalls each one, producing visible lag even when the
downstream systems do almost no work.
AutoNoVsync prefers Mailbox (triple-buffered, no blocking) and falls
back to Immediate when the backend can't honour Mailbox. Either is
fine for solitaire — the frame budget is tiny and the occasional
dropped frame from disabling vsync is imperceptible compared to the
stall this fixes.
Layered with the prior in-place resize updates and the 50ms
ResizeThrottle, this should bring the window-drag feel from "really
laggy" to native-feeling.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Smoke-test report: dragging the window edge to resize was sluggish.
Profiling showed each WindowResized event triggered ~170 entity
mutations across all 52 cards: full Sprite regeneration via
card_sprite plus despawn_related on each card's CardLabel children
followed by a fresh with_children spawn — and WindowResized fires per
pixel of drag, multiplying the cost.
Three fixes layered together:
1. resize_cards_in_place is a new function the resize handler calls
instead of sync_cards. It mutates Sprite.custom_size, the card's
Transform.translation, and existing CardLabel TextFont.font_size
directly — no Sprite replacement, no despawn_related, no child
rebuild. update_card_entity stays unchanged for non-resize callers
(deals, moves, flips, settings changes) so the full-repaint path
they need is preserved.
2. collect_resize_events reads events.read().last() and stashes only
the latest size into a ResizeThrottle resource each frame, so
multiple WindowResized events in one frame collapse to one apply.
3. snap_cards_on_window_resize is gated by a 50ms throttle
(RESIZE_THROTTLE_SECS): work runs at ~20 Hz during a sustained
drag instead of ~120 Hz. When the user stops resizing the next
frame flushes the final pending size, so the steady state always
matches the released window dimensions. should_apply_resize is a
pure helper unit-tested for the threshold-and-baseline contract.
apply_stock_empty_indicator gained a QueryFilter generic so the new
resize handler can pass a Without<CardEntity> filter — the resize
query already takes &mut Sprite on cards, so the indicator query had
to disjoin to avoid aliasing.
Five new tests pin the contract: should_apply_resize at three
threshold boundaries, plus integration tests that fire WindowResized
and assert no CardLabel entities were despawned and that
TextFont.font_size shrinks in place.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
The previous formula card_width = window.x / 9 with card_height = 1.4 *
card_width ignored the window height entirely. On a 1920×1080 window a
13-card face-up tableau column extended ~377 px below the viewport
bottom — visible reproduction in the smoke test.
compute_layout now derives two card_width candidates: one from the
horizontal grid budget (window.x / 9, unchanged) and one from the
vertical budget needed to seat 13 fanned cards plus the foundation
row, vertical_gap, and h_gap bottom margin. The smaller of the two
wins, so width remains the limiter on standard landscape windows and
height takes over on tall or short-wide aspect ratios. The math is
solved algebraically in a single substitution to avoid iteration.
When height is the limiter the original layout would have squished the
grid against the left edge; col_x now folds in a horizontal centring
offset that collapses to the existing geometry whenever width is the
limiter, so no other module needed an update.
Adds MAX_TABLEAU_CARDS = 13.0 (King-down-to-Ace worst case) and a
locally mirrored TABLEAU_FAN_FRAC = 0.25 — the original lives in
card_plugin and importing it would have created a circular dep with
layout. The duplication is doc-flagged so future drift gets noticed.
Four new tests pin both regimes: the height-limiter activates on a
1920×1080 window, stays inactive on a 900×1600 portrait window, and
the worst-case 13-card column fits on both 1280×800 and 1920×1080
within the bottom margin.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
The handoff document was written mid-overhaul during Phase 3 / early
Phase 4 and accumulated stale "in-flight" sections, recovery
instructions for a long-completed background agent, and a 7-track
speculative list of future directions that no longer matches reality.
This rewrite halves the document (284 → 142 lines) and reorients it
toward the actual current state: Phase 3 and Phase 4 shipped, every
substantial release-readiness thread landed, working tree clean.
Replaced the pause-state recovery section, original prompt blocks,
and Phase 3 smoke-test checklist with a compact table summarizing
what each Phase 4 commit landed, a chronological commit list as an
audit trail, and a five-item punch list scoping what's left for v1
(xCards URL, smoke test, push, tag v0.1.0, optional polish).
The resume prompt at the end is rewritten to orient a future agent
toward release prep — push, tag, package — rather than ongoing UX
work.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
The earlier HUD tooltip pass deliberately skipped the popover row
content because the spawn helpers were inline and the popovers
ephemeral. Coming back to them now: every row in the Modes popover
(Classic / Daily Challenge / Zen / Challenge / Time Attack) and
every row in the Menu popover (Stats / Achievements / Profile /
Settings / Leaderboard) gets a one-sentence tooltip explaining what
opening that mode or screen does.
The row tuple in each popover spawn helper grew from
(Marker, label) to (Marker, label, tooltip), with Tooltip::new(...)
attached at the spawn site. No public helper signatures changed.
popover_rows_carry_tooltip_strings asserts every row's exact
canonical text by querying (With<ModeOption>, &Tooltip) and
(With<MenuOption>, &Tooltip), spawning the popovers directly via
world.commands() to keep the test independent of headless click
simulation.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Eleven Settings controls — volume up/down for SFX and music, the four
toggle pills (draw mode, animation speed, theme, color-blind), the
two picker rows (card backs, backgrounds), and Sync Now — each gain a
one-sentence tooltip in the established Balatro voice. Static labels,
section headers, and live value readouts are intentionally skipped:
they are not interactive and the action button beside each describes
the action.
icon_button, volume_row, toggle_row, and picker_row gain
&'static str tooltip parameters so the tooltip is required at the
spawn site rather than retrofittable later. The Done button stays
tooltip-free (its label and Esc-equivalent affordance speak for
themselves at a modal-action position).
settings_buttons_carry_tooltip locks down the contract: every
SettingsButton outside the modal Done button has a Tooltip, and
SyncNow's tooltip text is asserted exactly to pin the canonical
microcopy.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
The window previously snapped straight to a card deal, which read more
like a prototype than a finished game. SplashPlugin lays a fullscreen
overlay (BG_BASE backdrop, ACCENT_PRIMARY title, version subtitle) on
top of the gameplay layer for MOTION_SPLASH_TOTAL_SECS — the board
deals behind it so the splash dissolve hands off naturally to the
deal animation.
Visibility curves through fade-in (300ms), hold (~1s), fade-out
(300ms) using a pure splash_alpha helper that gets pinned by a unit
test rather than wired to the Bevy clock — Time<Virtual>'s 250ms
per-tick clamp makes float-tight alpha assertions around the fade
boundary brittle.
Any keystroke or mouse-button press jumps the age forward to the
fade-out window so the splash dissolves immediately. The dismiss
handler is read-only on ButtonInput / Touches, so the same press is
still visible to gameplay handlers downstream — pressing Space on the
splash both dismisses it and triggers the next-tick stock draw, as
verified by dismissal_keypress_is_visible_to_other_systems.
Z_SPLASH sits above every other UI rung (Z_TOAST + 100) so the splash
owns the viewport for its brief lifetime. The hierarchy test was
extended to enforce the new rung's monotonic position.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Applies the tooltip infrastructure to the HUD: ten readouts (Score,
Moves, Time, Mode, daily-challenge target, draw cycle, undo count,
recycle count, auto-complete badge, keyboard selection chip) and the
six action-bar buttons (Menu, Undo, Pause, Help, Modes, New Game)
each gain a one-sentence tooltip in the established Balatro voice.
The strings earn their keep by surfacing information that isn't
visible: the link between the undo counter and the No Undo
achievement, the recycle counter and Comeback, the dual count-up /
countdown semantics of the timer in Time Attack, and the keyboard
shortcuts plus side-effects on action buttons.
spawn_action_button now requires a tooltip parameter so every action
bar entry gets one — there is no opt-out, by design. The popover Mode
and Menu rows are intentionally skipped: they're inside ephemeral
overlays whose hover surfaces are brief and already labeled.
Adds hud_elements_carry_expected_tooltip_strings, asserting the exact
text on each of the 16 instrumented elements.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
A new ui_tooltip module owns a Tooltip(Cow<'static, str>) component
that turns any UI node into a hover-revealing help target. Bevy 0.18's
required-components attribute auto-inserts an Interaction so callers
just attach Tooltip and the rest is wired.
A single overlay entity is reparented above the focus ring (new
Z_TOOLTIP token = Z_FOCUS_RING + 10) and tracked from the hovered
target's GlobalTransform + ComputedNode. The chained Update systems
start a hover timer on Interaction::Hovered, show the overlay once
MOTION_TOOLTIP_DELAY_SECS (0.5s) has elapsed, hide it the moment hover
ends, and refresh the text when the hover target switches without an
intervening unhover.
Tested headless under MinimalPlugins with a 200ms ManualDuration
ticker — Bevy clamps Time<Virtual>'s max_delta to 250ms by default, so
a one-shot 1s step doesn't actually advance the clock past the
threshold; the tests step five times to exercise both pre- and
post-delay invariants.
This commit ships the infrastructure only — no entity in the engine
has Tooltip attached yet. A follow-up applies tooltips to the HUD
readouts and action bar.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Both confirm modals previously used a "Yes, <verb>" pattern that read
like a question-and-answer dialog ("Are you sure? Yes, forfeit"); the
canonical UX pattern for a destructive confirm is just the bare verb.
The Confirm New Game modal's primary button is now "New game" instead
of "Yes, abandon" — matching the verb the user originally clicked
and framing the action positively rather than as a loss.
The Forfeit Confirm modal's primary button is now "Forfeit" instead
of "Yes, forfeit" — same pattern, less ceremony.
The Pause menu's own Resume / Forfeit buttons are unchanged: it's an
action menu, not a destructive confirm, and bare verbs are already
correct there.
Two doc comments and the ui_modal.rs spawn_modal_button example
docstring are updated to reflect the new copy. Marker symbol names
(ConfirmYesButton, ForfeitConfirmButton) are kept to avoid
unnecessary churn — the rename would ripple into mouse-input handlers
without a matching user-visible benefit.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Closes the audit gap: the two achievements that previously had only
unit-level condition tests now also have full-flow tests that fire a
GameWonEvent and assert the unlock state through the same plugin
ordering production uses (update_stats_on_win runs before
evaluate_on_win, so the freshly bumped stat is visible to the
condition closure).
Four tests, headless under MinimalPlugins:
- draw_three_master_fires_on_tenth_draw_three_win — pre-seed 9 wins,
fire a Draw3 win, assert unlock
- draw_three_master_does_not_fire_at_nine_wins — pre-seed 8, fire a
Draw3 win bumping to 9, assert still locked
- zen_winner_fires_on_zen_mode_win — Zen-mode win unlocks the badge
- zen_winner_does_not_fire_for_classic_win — Classic win in same
fixture leaves it locked
After this commit every advertised achievement has an integration
test that exercises the production unlock path.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
§3, §5, and §13 all referenced bevy_kira_audio as the audio
dependency, but the workspace Cargo.toml has used kira 0.12 directly
since the kira-direct migration. Four mentions updated so the
architecture document matches the actual dependency graph.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
The textual attribution to Huub de Beer / xCards / LGPL-3.0 already
satisfies the LGPL's notice requirement on its own; the unfilled
URL placeholder was the only TODO left in the file. Removed rather
than guessed — a confirmed upstream URL can be added in a follow-up
when the project owner decides which xCards mirror or fork to point
at.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Settings was the last mouse-only surface in the engine. Phase 3 closes
that gap and finishes the keyboard-focus rollout.
Every interactive button in the settings panel — icon buttons (32px
volume, draw mode, color blind, sync now), swatch pickers (5 card
backs, 5 backgrounds), and toggle pills — now opts into Focusable via
a single ancestry-walking system that mirrors the Phase 1/2 pattern.
The Done button continues to be auto-tagged through the modal path.
The two picker rows gain a new FocusRow marker. Inside a FocusRow,
Left/Right arrow keys cycle the swatches (skipping Disabled, wrapping
at endpoints) while Tab/Shift-Tab still escape to the next section's
focusable. Outside a FocusRow, arrow keys are explicit no-ops.
scroll_focus_into_view runs after the focus overlay updates and
adjusts the SettingsPanelScrollable container's ScrollPosition when
the focused button sits outside the visible viewport, with a
SPACE_2 padding so the focus ring never gets clipped at the
viewport edge. The system is a no-op when layout hasn't computed yet,
so headless tests are unaffected.
After Phase 3 every interactive UI element in the engine is
keyboard-navigable: modals (Phase 1), HUD action bar and Home mode
cards (Phase 2), Settings bespoke controls and picker rows (Phase 3).
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
The HUD action bar (Menu / Undo / Pause / Help / Modes / New Game) and
the five Home mode-launcher cards now participate in keyboard focus,
extending Phase 1's modal-only coverage.
The HUD focus group activates only when no modal is open and the
mouse is hovering an action-bar button — the design decision avoids
stealing Tab from selection_plugin's card-selection nav for the
common "playing on the board" case. Once engaged, Tab/Shift-Tab cycles
the bar in spawn order and Enter activates. Moving the mouse off the
bar clears focus so the ring doesn't linger.
Home mode cards opt into FocusGroup::Modal(home_scrim) via an
ancestry-walking system that mirrors the Phase 1 attach helper, so
spawn_mode_card's signature is unchanged. Locked cards (Zen,
Challenge, Time Attack at level <5) get the Disabled marker so Tab
skips them and Enter is a no-op — mirroring the existing visual
locked state with real keyboard semantics.
handle_focus_keys gains a Hud-on-hover branch in its active-group
resolver and a clear_hud_focus_on_unhover system. Together they
implement the agreed UX: focus follows hover when the bar is active,
Tab cycles within the hovered group, and the ring disappears the
instant the mouse leaves.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Every button spawned via spawn_modal_button is now keyboard-navigable.
Tab/Shift-Tab cycles focus within the active modal, Enter activates
the focused button via the same Interaction::Pressed signal mouse
clicks use, and the primary action auto-focuses on modal open. Mouse
clicks transfer focus so the two input modes stay in sync.
The visual indicator is a single overlay entity that's reparented
above the topmost modal scrim and tracks the focused button's
GlobalTransform + ComputedNode each frame. Sitting outside the
modal-card subtree means the ring isn't affected by the open
animation's 0.96→1.0 scale, and sitting outside any scroll container
means it can't be clipped by Settings' Overflow::scroll_y. Z-order
sits one rung above Z_MODAL_TOP via the new Z_FOCUS_RING token.
Existing 11 modals (Help, Stats, Achievements, Settings, Profile,
Leaderboard, Pause, Forfeit confirm, GameOver, Confirm new game,
Onboarding, Home) get focus support without any call-site changes —
attach_focusable_to_modal_buttons walks the ancestry of any
ModalButton lacking Focusable to find its scrim and tags it
automatically. selection_plugin's Tab handler keeps working when no
modal is open; when one is, focus consumes Tab/Enter before the
selection system sees them.
Phase 1 scope only — HUD action bar, Home mode cards, and Settings
bespoke buttons (icon, swatch, toggle) come in Phase 2/3.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
The Home modal was previously a keyboard-shortcut reference card that
mostly duplicated Help. It now opens directly into a Mode Launcher:
five mode cards (Classic, Daily Challenge, Zen, Challenge, Time
Attack) stacked vertically with a Cancel button at the bottom.
Each card dispatches the canonical request event already used by the
HUD modes-popover (NewGameRequestEvent, StartDailyChallengeRequestEvent,
StartZenRequestEvent, StartChallengeRequestEvent,
StartTimeAttackRequestEvent), so level gates, daily-seed lookup, and
session setup all flow through the existing handlers — Home is just
another entry point.
The three modes that unlock at level 5 (Zen, Challenge, Time Attack)
render with reduced opacity and a "Reach level 5 to unlock" caption
when locked; clicking a locked card is a deliberate no-op so the
player can pick a different mode without dismissing the modal.
The keyboard-shortcut reference is dropped entirely — Help (F1) still
covers it. M continues to toggle the modal open and closed.
Adds 5 new headless tests covering card spawn, locked-state click,
unlocked-state click, Classic launch + close, and Cancel close.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Lists the project's third-party assets and major dependencies with
their licenses for v1 release readiness:
- xCards @2x card artwork and back_0 (LGPL-3.0)
- FiraMono-Medium font (OFL)
- Bevy 0.18, kira 0.12, axum, sqlx, tokio, and the rest of the
ten most prominent Rust deps (MIT/Apache-2.0 across the board)
Generated assets — card backs 1-4, all backgrounds, and every WAV
file — are credited as original work produced by the project's own
solitaire_assetgen pipeline; the audio synthesis stack and the
absence of any external sample sources are documented.
The README gains a brief Credits section linking to the full list.
Note: the upstream xCards source URL is left as a placeholder pending
confirmation; downstream license obligations are unaffected.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
LeaderboardResource was a tuple struct of Option<Vec<Entry>>: None for
pre-fetch and empty Vec for both "actually empty" and "fetch failed"
— the user couldn't tell a network error from a legitimately quiet
leaderboard. The resource is now a four-state enum (Idle / Error /
Loaded), with Loaded covering both populated and empty rows. A
transient error no longer wipes a previously populated list, and the
panel renders "Couldn't reach the leaderboard. Try again later."
when the most recent fetch failed.
The Opt In / Opt Out buttons used to render unconditionally and
silently no-op under LocalOnlyProvider. The panel now reads the
SyncProviderResource backend name and, when no remote is configured,
replaces the buttons with a single line directing the player to
configure cloud sync in Settings.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
On first launch the Stats grid previously mixed "0" cells (Games
Played / Won / Lost) with "—" cells (Best Score / Win Rate / Avg
Time), reading as inconsistent. Now every cell renders an em-dash
when games_played == 0, and a "Play a game to start tracking stats."
caption sits above the grid using the existing TYPE_CAPTION /
TEXT_SECONDARY tokens. Once a game has been played the original
formatters resume.
The Profile screen gains a one-line welcome ("Welcome! Play games to
earn XP and unlock achievements.") that renders only when both
total_xp and the daily streak are zero, breaking up the wall of
zero-valued readouts that greeted users on first launch.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
The Help modal previously used "Close" while the other five overlay
modals (Home, Stats, Achievements, Settings, Profile, Leaderboard)
used "Done"; standardising on "Done" removes the outlier.
The final onboarding slide changes from "Start playing" to
"Let's play". The microcopy audit suggested matching the win modal's
"Play Again", but that verb is semantically wrong on first launch —
the player has not yet played. "Let's play" reads warmer and matches
the project's Balatro-tone direction without overloading "Play Again"
across two contexts that mean different things.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
ICON_BUTTON_PX moves from 28 to 32 to clear the desktop hit-target
threshold. The change is self-contained: icon buttons are centered in
flex rows whose neighbours retain their alignment, and the swatch
buttons (40px) still dominate the visual hierarchy.
The settings sync status fallback string changes from "Status: not
configured" to "Status: local only" so users running without a remote
backend read it as a deliberate choice rather than incomplete setup.
The other status strings (Idle / Syncing / LastSynced / Error) flow
from sync_status_label and are unaffected.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Reflects that Track B (window polish) and Track G (modal + score
animations) are now landed, and brings the resume prompt and
release-readiness scope in line with the post-commit state.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Score readouts now react to mutations: ScorePulse drives a triangular
1.0 → 1.1 → 1.0 scale on the HUD score over MOTION_SCORE_PULSE_SECS,
and jumps of at least SCORE_FLOATER_THRESHOLD points spawn a floating
"+N" that drifts up 40px and fades over 2× the pulse duration before
despawning. Detection runs after GameMutation so the visuals trail the
state update by exactly one frame.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Modals now animate in via the new ModalEntering component: scrim alpha
ramps from 0 to its full value while the card scales from 0.96 to 1.0
over MOTION_MODAL_SECS using an ease-out curve. AnimSpeed::Instant
collapses the duration to zero so reduced-motion users see the modal
snap into place on the first frame.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Sets Window::name so X11/Wayland taskbars group the game correctly,
centers the window on the primary monitor on startup, and installs a
panic hook that appends a timestamped crash record to crash.log under
the platform data dir (gracefully no-ops when the directory is
unavailable).
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Five test-only lints surfaced by --all-targets were blocking CI under
-D warnings: a useless vec! in a leaderboard sort test, a
field_reassign_with_default in tuning tests, and three
assertions_on_constants in card_plugin sanity tests. The constant
assertions are now wrapped in const blocks so they run at compile time;
the runtime-formatted values were dropped from their messages because
const-block assert messages must be string literals.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Replaces the 3-line resume prompt with a senior-Rust-developer framing
plus seven concrete next-direction tracks (A–G) covering the Home modal
decision, window/release polish, sound, sync, achievements, release
backlog, and UI/UX professional polish. Each track names specific
files/tokens so the next session can act on a chosen direction without
re-discovery.
Migrates the last remaining colour, spacing, font-size, and z-index
literals in animation_plugin (toasts), hud_plugin (action bar +
Modes/Menu popovers), and win_summary_plugin (full win modal restyle)
onto the ui_theme token system established in step 1. Win summary now
uses SCRIM/BG_ELEVATED/ACCENT_PRIMARY/STATE_* with a yellow Play Again
button. Sprite-tinted gameplay art (cards, felt, drop-zone hints,
pile markers) and sub-rung pixel sizes (1px borders, fixed cell
widths) are intentionally left untouched.
cargo build / clippy --workspace -- -D warnings / test --workspace
all green (819 passed, 0 failed, 8 ignored).
Slide animations now interpolate through MotionCurve::SmoothSnap via
sample_curve() at the call site (no struct field added). Slide and
cascade durations route through ui_theme::scaled_duration with
MOTION_SLIDE_SECS / MOTION_CASCADE_STAGGER_SECS / MOTION_CASCADE_SLIDE_SECS.
Settle bounce in feedback_anim_plugin scoped to MoveRequestEvent and
DrawRequestEvent receivers — only the top `count` cards of the
destination pile (or top of waste) bounce; undo and other state
changes no longer trigger a global all-tops settle.
Deal stagger gains a deterministic ±10% jitter via DefaultHasher on
card_id (no rand dep). Per-card stagger = base * (1.0 + jitter).
Win cascade switched from CardAnim to CardAnimation with
MotionCurve::Expressive and a deterministic ±15° per-card Z-rotation
via Fibonacci hash. Win screen shake routes through
MOTION_WIN_SHAKE_SECS / MOTION_WIN_SHAKE_AMPLITUDE; ScreenShakeResource
gained a `total` field so decay computes correctly under Fast / Instant.
cargo build / clippy --workspace -- -D warnings / test --workspace
all green (819 passed, 0 failed, 8 ignored).
AnimSpeed is a fieldless enum; adding Copy lets `scaled_duration`
and other helpers take it by value through `&AnimSpeed` deref
without requiring a `.clone()` at every call site. Prerequisite
for the upcoming animation token-routing work.
Replace the bespoke side-panel with the ui_modal scaffold. Layout
collapses into four sections: Audio (SFX / Music volume), Gameplay
(Draw Mode / Anim Speed), Cosmetic (Theme / Color-blind / Card Back
/ Background), and Sync (status + manual Sync Now).
Body lives in a scrollable child of the modal card with
max_height: Vh(60.0) so tall content stays reachable on short
windows. Done is a primary button outside the scroll so it's always
one click away regardless of scroll offset.
All colours, spacing, typography, and z-index from ui_theme tokens.
Two file-local sub-rung sizes (SWATCH_PX = 40, ICON_BUTTON_PX = 28)
remain as documented literals — they're smaller than SPACE_2 (8 px)
which is the smallest rung.
Existing systems (handle_settings_buttons, update_*_text,
scroll_settings_panel, persistence) untouched; the SettingsPanel /
SettingsPanelScrollable / SettingsScrollNode markers and every
button marker carry over so all existing tests and click handlers
keep working.
cargo build / cargo clippy --workspace -- -D warnings / cargo test
-p solitaire_engine all green (444 passed, 0 failed).
Replace the single-screen first-run banner with a 3-slide flow built
on the ui_modal scaffold:
1. Welcome
2. How to play (drag-and-drop / double-click / right-click hints)
3. Keyboard shortcuts (8 rows mirroring help_plugin's canonical list)
Navigation: primary Next button (advances; final slide reads
"Start playing" and writes first_run_complete), secondary Back button
(slide >0), tertiary Skip on slide 0. Arrow / Enter / Esc keep
working as accelerators.
OnboardingSlideIndex resource persists across despawn/respawn so the
rebuild system always knows which slide to show next.
All colours, spacing, typography come from ui_theme tokens; no
literals in the new code.
cargo build / cargo clippy --workspace -- -D warnings / cargo test
--workspace all green (813 passed, 0 failed, 8 ignored).
Two fixes the smoke test surfaced:
1. The forfeit-confirm modal at `Z_PAUSE_DIALOG` (225) was invisible
behind the pause card at `Z_PAUSE` (220). In Bevy 0.18, root-level
UI nodes don't reliably sort across stacking contexts via plain
`ZIndex` alone, so `spawn_modal` now adds `GlobalZIndex(z_panel)`
alongside the existing `ZIndex(z_panel)`. Every overlay built on
`ui_modal` (pause, forfeit-confirm, confirm-new-game, help, home,
leaderboard, profile, achievements, stats, game-over) inherits the
fix.
2. `handle_forfeit_request` no longer silently drops the request when
`move_count == 0` — pressing G or clicking the pause modal's
Forfeit button on a freshly-dealt game now opens the confirm modal,
and the only short-circuit is "game is already won", which now
fires an `InfoToastEvent` ("No game to forfeit") so the player
gets feedback. The `move_count > 0` half of the gate was the
reason a fresh-deal G press appeared to do nothing.
The G-key gate in `handle_keyboard_forfeit` is simplified to just
"not paused"; the rest of the forfeit-eligibility check moves into
`handle_forfeit_request` so it can surface the toast.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Pause overlay drops its bespoke full-screen layout and is rebuilt on
the standard `ui_modal` scaffold: uniform scrim, centred card, real
Resume (Primary, Esc) and Forfeit (Tertiary, G) action buttons. The
Draw Mode row stays inline in the body so the existing toggle still
fires `SettingsChangedEvent`.
The G-key double-press toast countdown is replaced with a real
modal: `G` (or clicking Forfeit on Pause) fires the new
`ForfeitRequestEvent`, which `PausePlugin` answers by spawning
`ForfeitConfirmScreen` at `Z_PAUSE_DIALOG` (above pause). The modal
exposes Cancel + "Yes, forfeit" buttons plus Y/Enter/N/Esc
accelerators; confirmation despawns both modals, clears
`PausedResource`, and fires `ForfeitEvent` for `StatsPlugin`.
`toggle_pause` now early-returns when a forfeit modal is visible (and
runs `.before(handle_forfeit_keyboard)`) so an Esc that closes the
forfeit modal doesn't also re-open pause in the same frame.
The legacy `forfeit_countdown` field, `FORFEIT_CONFIRM_WINDOW`
constant, and the six pure-function countdown tests are removed; new
tests cover the modal-spawn / confirm / cancel paths and the active-
game predicate that still gates the G hotkey.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Pause point in the UX overhaul Phase 3. Documents:
- The 11 commits landed this session (foundation + HUD + 8 overlay
conversions).
- Test status (797 / 797 pass, clippy clean).
- The 5 steps still pending (Pause + Forfeit, Settings, Onboarding,
animation upgrades, final literal sweep) and what each touches.
- A smoke-test checklist for the user to run on the current state.
- A copy-pasteable kickoff prompt for the next session.
- An open question about whether to keep, repurpose, or drop the
Home overlay (now that the action bar + Help cover its surface).
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
funman300