The exit push spawned a detached task on AppExit, which process
teardown almost always killed before the network round-trip completed —
the final session's sync silently never happened (2026-07-06 review,
finding M3). Local persistence meant no data loss, but stats stayed
unsynced until the next launch.
push_on_exit now blocks the closing app's final frame for at most
EXIT_PUSH_TIMEOUT (2s) via tokio::time::timeout: long enough for one
healthy round-trip, short enough that quitting never feels hung when
the server is unreachable. Timeout and errors are logged and skipped —
the next launch's pull/push converges as before.
Closes review finding M3.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Review feedback on #135: floating 'stable' + -D warnings lets every new
clippy release redden master with new lints — pin 1.95.0 like the
web-wasm-rebuild workflow. And ubuntu-latest lacks the ALSA/udev/X11/
Wayland dev packages the Bevy crates link against; install the standard
Bevy CI set (the project's own builder images have no desktop-Bevy
precedent — android-builder targets the NDK and web-e2e only builds the
server).
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Until now no CI workflow ran the test suite or clippy at all — the
android-release, docker-build, web-e2e, web-wasm-rebuild, and
builder-image workflows cover packaging and e2e, but a direct push to
master (including the web-wasm-rebuild bot commit) never executed
cargo test or clippy. The gate discipline in CLAUDE.md §6 existed only
on developer machines.
test.yml runs the exact §6 commands (clippy --workspace --all-targets
-D warnings; cargo test --workspace) on master pushes and PRs, with
SQLX_OFFLINE against the checked-in .sqlx cache, path-filtered so
docs-only merges don't burn CI. Toolchain/cache mirror web-e2e.yml.
Found during the 2026-07-06 large-scale review (finding H1).
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
canvas_bg.wasm shipped at 36.2 MB — plain release (opt-level 3, thin
LTO) piped through wasm-opt -O2. Download size, not throughput, is the
binding constraint for the browser canvas, so solitaire_web now builds
with a dedicated wasm-release profile: opt-level "s", fat LTO, one
codegen unit. Local result: 23.2 MB after the unchanged wasm-opt -O2
pass — 35.9% smaller.
The binaryen pass stays at -O2 (the -Oz grey-screen miscompile note in
build_wasm.sh still applies). profile.strip is deliberately NOT set: on
wasm it also removes the target_features custom section, which makes
wasm-opt reject the module ('all used features should be allowed' on
trunc_sat).
Verified with the new artifact: all 5 play_canvas e2e specs pass (debug
bridge, draw3 param, apply/undo, replay diagnostics, 40-seed autoplay
invariants). Headless pixel verification is inconclusive in this
environment — wgpu cannot create a usable adapter locally and panics
identically on the OLD artifact too — so visual parity should be
confirmed against production after the next deploy.
pkg/ artifacts are intentionally not committed: the web-wasm-rebuild
workflow is the single source of truth and will regenerate them with
this profile on merge.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
ARCHITECTURE.md predated the card_game migration (PR #88, 2026-06-22)
and still documented the pre-migration core: local Card with face_up,
PileType, stored score/undo/recycle, and a snapshot undo stack. Rewrites
the Core Game Models section around the upstream card_game/klondike
types and the derived-stats/replay-undo model, updates the
solitaire_core crate section (deps + ownership), adds solitaire_web and
solitaire_assetgen to the workspace tree, and corrects the
solitaire_app and Settings entries. Version bumped to 1.4.
Also adds the two missing crates to the CLAUDE.md crate map (AGENTS.md,
its gitignored local twin, was regenerated in place) and records the
full device checklist passing on the Fold 7 in SESSION_HANDOFF.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
CHANGELOG gains the missing [0.41.1] section (pile-marker child-resize
fix + Android relayout diagnostics). SESSION_HANDOFF is brought forward
from the stale v0.40.0 state to today's: v0.41.x releases, the July 6
review arc (PRs #121-#131), Fold 7 on-device verification results, the
remaining device-checklist items, and new architectural notes (plugin
submodule pattern, marker child-resize rule, Fold 7 inset quirks).
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Found during Fold 7 on-device verification of v0.41.0: on_window_resized
resized the marker fill sprite but never its children, so after any
resize (fold/unfold, rotation) the outline frame and the A/K watermark
kept their spawn-time size — rendering as oversized grey slabs over the
empty foundation slots.
The resize handler now re-derives both children from the new layout
(outline = card_size + 2*PILE_MARKER_OUTLINE_WIDTH, watermark font =
card_size.x * 0.28 — same formulas as spawn). Adds a regression test
and an Android-gated relayout log line (width/height/insets) as the
evidence channel for the remaining foldable layout bug (#130).
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Second phase of #118 for settings_plugin: the 2,857-line mod.rs becomes
four focused submodules along existing system boundaries —
mod.rs 561 types, markers, SettingsButton, plugin build, persistence
input.rs 632 button/hotkey handlers, focus attachment, scrolling
updates.rs 495 per-frame value-text updater systems + label helpers
ui.rs 1,231 spawn_settings_panel + row builders + thumbnails
Moved items are pub(super); mod.rs glob-imports the submodules so
system registration and tests keep their bare names. No behaviour
change; all 29 settings tests pass unchanged.
Refs #118
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Renames Unreleased to 0.41.0 (2026-07-06), noting it consolidates the
v0.40.x patch tags that were cut without sectioning. Adds entries for
today's safe-area resume re-poll fix (#116) and the plugin module
test-split phase (#118).
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
First increment of the module-split plan: card_plugin.rs becomes
card_plugin/{mod.rs, tests.rs} (2,536 + 1,592 lines), following the
replay_overlay/ pattern. Pure mechanical move via git mv — no runtime
code changes; all 70 tests preserved and passing.
Refs #118
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Polls the docker-build / web-wasm-rebuild workflow runs and prints a
compact status block until the newest deploy is live. Reads the API
token from ~/.config/tea/config.yml and never prints it.
Closes#119
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Resolves the Draw-1 recycle question from the June 500-game audit:
unlimited recycling with upstream score penalties is deliberate,
matching mainstream digital solitaire rather than strict 3-pass
tournament rules. The difficulty seed catalog and the winnable-deal
solver are verified under this rule, so a hard pass limit must not be
introduced casually.
- ARCHITECTURE.md: new 'Rules decisions' note in the solitaire_core
section
- GameState::draw() doc comment points at the decision record
- New lock-in test draw_one_recycling_is_unlimited_by_design asserts
10+ recycles are never rejected
Closes#117
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
refresh_insets gated its loop on insets.is_populated(), so once insets
resolved at first launch, rearm_on_resumed's poll-counter reset was a
no-op and JNI was never queried again. Insets that changed while the
app was backgrounded (fold/unfold, rotation, gesture/3-button nav
switch) stayed stale until process restart.
Gate the loop on the poll counter alone and settle a cycle by
exhausting it once a populated reading arrives. The cached resource is
rewritten only when the value actually differs, so resumes where
nothing moved trigger no change detection and no relayout — preserving
the no-flash resume behaviour. Also correct the stale on_app_resumed
doc that still described the old inset-zeroing approach.
Closes#116
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
MAX_DYNAMIC_FAN_FRAC 0.6 -> 0.9 so the dynamic tableau fill spreads
further on very tall / narrow viewports (e.g. a foldable cover screen),
which were left ~40% empty at the cap. Fills the unfolded near-square
screen to ~100% and lets the cover screen fill further (and the rest fills
as columns deepen during play). Normal phones are unaffected — their fill
fraction is already below the cap. apply_dynamic_tableau_fan still floors
at TABLEAU_FAN_FRAC and deeper columns drive the fraction down, so nothing
overflows and hit-testing stays in sync.
Vertical centring of the residual was investigated but dropped: a 21:9
phone is aspect-identical to the cover screen, so centring can't be
targeted to foldables without also disconnecting the board from the HUD on
tall phones.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
On a near-square viewport (e.g. an unfolded Galaxy Fold) a fresh deal left
the bottom ~40% of the screen empty: the dynamic fan (update_tableau_fan_frac)
measured only face-up column depth and returned early at a fresh deal (face-up
depth 1), so it never spread the tableau, and the deep face-down stacks were
ignored. The cold-start deal also never fired StateChangedEvent, and on Android
the safe-area-inset resize (frames 1-3) reset the fan to compute_layout's
sparse worst-case value, so even the post-move fill was wiped.
Move the fill into layout::apply_dynamic_tableau_fan, driven by each column's
TOTAL weighted depth (face-down cards count, scaled by the face-down/face-up
step ratio) so the deepest column fills the available height. Run it in three
places so every path stays filled: PostStartup (cold-start deal), on
StateChangedEvent (moves), and inside on_window_resized after compute_layout
(safe-area resize + fold/unfold). MAX_DYNAMIC_FAN_FRAC caps the spread so a
near-empty column keeps readable overlap; TABLEAU_FAN_FRAC floors it. Deeper
columns drive the fraction down so everything still fits — no overflow.
card_position/card_positions read the same fractions, so hit-testing stays
in sync.
Adds regression tests: cold-start deal fills the fan, and a resize re-fills it.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Every move fired StateChangedEvent -> sync_cards, which rebuilt the full
visual for all 52 cards: despawning and respawning every child entity
(drop-shadow, border frame, and on Android a Text2d corner label needing
a glyph re-layout) even for the ~50 cards that did not move. That ~250
entity despawn/spawns plus 52 text re-layouts in a single frame spiked
the StateChangedEvent frame and stuttered the slide animation on
high-resolution devices (reported on a Galaxy Fold 7).
Add a CardChildrenKey component capturing the only inputs the child
entities depend on (face_up, card_size, color_blind, high_contrast).
update_card_entity now rebuilds children only when that key changes (a
flip, resize/fold, or accessibility toggle); a position-only move just
updates the Transform. The Sprite is still refreshed every sync (a cheap
handle swap), so theme/card-back image changes need no child rebuild.
Adds a regression test asserting an appearance-neutral StateChangedEvent
no longer despawns/respawns card label children.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
The apksigner step relied on auto scheme selection, which produced an APK
carrying invalid v1 (JAR) signature files: META-INF/*.SF and *.RSA were
present but failed v1 verification (apksigner reports `v1 scheme: false`
while v2/v3 verify). Android installs such an APK fine via v2/v3, but
Obtainium parses the legacy v1 certificate at install time, gets an empty
cert list, and crashes with:
RangeError (length): Invalid value: valid value range is empty: 0
This is why the app adds fine in Obtainium (Gitea API only) but fails on
install (APK parse). minSdk is 26, so v1/JAR signing is unnecessary —
sign explicit v2+v3 only (matching modern Android tooling for minSdk >= 24)
and pass --min-sdk-version 26. Adds a post-sign guard that fails the build
if any META-INF v1 signature files remain.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Elevates the physical-device smoke test as the single remaining task for the
v0.40.0 release and clarifies the Matomo validation is an independent task,
not a release blocker.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
funman300
funman300