Files
Ferrous-Solitaire/SESSION_HANDOFF.md
T
funman300 e63046700c docs(handoff): record win_move_index data field; B's next finite step
Captures `ab857bb` (Replay::win_move_index data field) under "Since
the v0.21.3 cut". Updates the Visual-identity follow-up entry for
B-2 to flag the data-layer prerequisite as landed and identifies
the WIN MOVE scrub-bar marker UI as the natural next finite commit.
Bumps test count to 1212.

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
2026-05-08 14:45:59 -07:00

16 KiB
Raw Blame History

Solitaire Quest — Session Handoff

Last updated: 2026-05-08 — v0.21.3 cut and tagged at 3d92a91, working tree clean, all post-tag work pushed to origin.

v0.21.3 is a patch release with one through-line: accessibility arc closure. v0.21.2 explicitly carved out "dynamic-paint sites" (HUD action buttons, modal buttons, radial menu rim) on the assumption that their existing paint cycles would race the central update_high_contrast_borders system. v0.21.3 walks the actual code, finds the carve-out was over-cautious, and closes it. Bonus: the first real consumer of ToastVariant::Warning also lands here, making the ToastVariant enum fully load-bearing (every variant has at least one driver).

Full v0.21.3 detail lives in CHANGELOG.md § [0.21.3]. This file from here on focuses on what's open post-cut and how to resume.

Status at pause

  • HEAD locally: see git rev-parse HEAD. The cut commit is 3d92a91; the post-cut Replay::win_move_index data field (ab857bb) rides on top of that.
  • HEAD on origin: matches local. v0.21.3 is fully on origin.
  • Working tree: clean. No WIP outstanding.
  • artwork/ directory: still untracked. Intentional.
  • Build: cargo clippy --workspace --all-targets -- -D warnings clean.
  • Tests: 1212 passing / 0 failing across the workspace (1207 from v0.21.3's stats + 5 from ab857bb's win_move_index coverage).
  • Tags on origin: v0.9.0 through v0.21.3. v0.21.3 is on 3d92a91; v0.21.2 stays on f23df3b; v0.21.1 stays on daa655a; v0.21.0 stays on 04f9bf9; v0.20.0 stays on 41a009a.

Since the v0.21.3 cut

  • ab857bbReplay::win_move_index data field landed. First finite step toward the B-2 replay screen-takeover redesign. Additive optional Option<usize> on Replay with #[serde(default)] so older latest_replay.json / replays.json files load unchanged (no schema bump). Populated at the live recording site via a new with_win_move_index builder; for fresh recordings the value is always Some(moves.len() - 1) because recording freezes on win, but storing it explicitly lets the playback UI read the WIN MOVE position directly without re-deriving on every render. UI consumption (the WIN MOVE marker on the scrub bar, plus the broader screen-takeover work — move-log scroller, mini-tableau preview, playback controls) lands in subsequent commits. 5 new tests (1207 → 1212): default, builder set / set-None, on-disk round-trip, legacy-JSON-loads-with-None backward-compat.

Open punch list

Phase Android (build + persistence shipped; runtime gaps remain)

  • APK launch verification on AVD / device. adb install then adb logcat against the bevy_test AVD or an x86_64 device. The build works and persistence is wired, but no end-to-end device run has been logged. Shakes out runtime bugs the build + unit tests can't catch.
  • JNI ClipboardManager bridge. Replaces the Android stub for the Stats "Copy share link" toast. arboard doesn't ship an Android backend; small custom JNI call.
  • Android Keystore for credentials. keyring is target-gated to a stub returning KeychainUnavailable; replace with Android Keystore via JNI when sync auth ships on mobile.
  • Google Play Games (gpgs) integration. Listed as a Phase-Android target since Phase 1; now unblocked by the build target.
  • Cosmetic cargo apk build --lib workaround. Post-sign panic doesn't affect the APK on disk but produces noisy stderr. Either upstream a cargo-apk fix or document --lib as canonical in the runbook.

Visual-identity follow-ups (post-v0.21.0)

The visual-identity arc is effectively complete: token system, chrome migration, splash boot screen, replay-overlay banner, card-face artwork (both rendering paths), and the ACCENT_PRIMARY palette refresh all shipped in v0.20.0 + v0.21.0. What stays open:

  • Replay-overlay screen-takeover redesign. The full mockup (docs/ui-mockups/replay-overlay-mobile.html) calls for a mini-tableau preview, playback controls, move-log scroll, and a WIN MOVE marker on the scrub bar. Banner-local pieces all shipped in v0.21.0 (c84d9f4 + 6204db8 + 54005d5 + e080b49); the floating MOVE chip above the focused card shipped in v0.21.2 (2fb2d63). The data-layer prerequisite for the WIN MOVE marker — Replay::win_move_index: Option<usize> — landed post-v0.21.3 in ab857bb (additive serde-default field; populated at the recording site; exhaustively tested). What still needs to land for the screen-takeover: the WIN MOVE scrub-bar marker UI (consumes the new field), a move-log scroller, a mini-tableau preview, and playback controls. Multi-session.
  • Floating MOVE N/M chip above the focused card during playback — closed 2026-05-08 by 2fb2d63. World-space Text2d entity sibling to the banner overlay; uses the same LayoutResource pile coordinates so it survives window resizes without UI/camera math.
  • Toast Warning variant wiring — closed 2026-05-08 by 279e23d. Daily-challenge-expiry toast fires once per daily.date when within 30 min of UTC midnight reset and today is incomplete. ToastVariant is now fully load-bearing (every variant has at least one real driver). Future Warning drivers can either reuse the generic WarningToastEvent(String) carrier or add their own domain message + animation_plugin handler.
  • Toast Error variant wiring — closed 2026-05-08 by 68d50b5. MoveRejectedEvent now fires a 2-second pink-bordered "Invalid move" toast as the third leg of the audio + visual + text rejection-feedback stool.
  • High-contrast accessibility mode — closed 2026-05-08 by c5787c6 + 07e0357 (engine + UI) + v0.21.2's HC chrome rollout (c9af1ea + d87761d + ec804d5) + post-cut dynamic-paint rollout (c153363). Card text rendering plus 8 static-border chrome surfaces (modal scaffold, tooltip, onboarding key chips, help panel key chips, stats panel cells, home Level/XP/Score row, home mode buttons, home mode-hotkey chips, 4 settings panel surfaces) all boost borders to BORDER_SUBTLE_HC under HC via the HighContrastBorder marker. The previously-carved-out dynamic-paint sites are now also covered: HUD action buttons and modal buttons take the same marker (their paint cycles only mutate BackgroundColor, so no race); the radial menu rim folds HC into its per-frame spawn via radial_rim_outline so the focused rim boosts to BORDER_SUBTLE_HC under HC (preserving focused-vs-resting hierarchy that naive marker substitution would invert).
  • Reduced-motion mode — closed 2026-05-08 by c5787c6 + v0.21.2's ed152e2. effective_slide_secs forces 0 on card animations; pulse_splash_cursor skips the per-frame pulse multiplier; spawn_splash skips the scanline overlay entirely. Future scope: gate any future card-lift z-bump animation, warning-chip pulse (when one materialises).

Carried forward from v0.19.0

  • App icon round — closed 2026-05-08 by 3eb3a26 + 716a025. Runtime Window::icon wired (Linux/macOS/Windows); 9-size PNG hierarchy at assets/icon/icon_<size>.png covers Linux hicolor + downstream .icns/.ico packaging needs. The .ico and .icns bundle-format files themselves are not generated — both would need new crate deps (ico and icns respectively) and only matter at app-bundle time (cargo-bundle / packaging), not at cargo run. Open if the project later ships as a packaged macOS / Windows app.

Other small candidates

  • Prev/Next selector chips spawn site. v0.19.0's 9b065e5 noted Prev/Next markers exist in stats_plugin but no spawn site renders them today — the Shareable badge therefore lands on the single-replay caption. If/when Prev/Next is plumbed, the badge will need to follow.
  • Toast queue / immediate unification. The two toast paths (spawn_queued_toast for InfoToastEvent queue; spawn_toast for fire-and-forget) now share visual treatment but remain separate functions because they serve different temporal needs (sequential vs. parallel). If overlap becomes a UX issue, merge into one queue with priority lanes.

Process notes

  • The desktop-adaptation spec is the canonical reference for geometry decisions when porting any future plugin. Read docs/ui-mockups/desktop-adaptation.md first; apply the universal rules to every surface; consult the per-screen table for the priority surfaces. The 9 missing-plugin screens (splash now ported; eight remaining) inherit the universal rules without dedicated guidance.
  • Stitch generate_variants is unreliable for layout-only adaptation prompts as of 2026-05-07. The first call timed out and no variant ever landed in list_screens. If a future session wants visual desktop mockups, prefer generate_screen_from_text with a fresh narrow prompt per screen rather than generate_variants against existing mobile screens.
  • Token-port pattern. v0.20.0's chrome-migration commits set a reusable shape for "centralised design system applied across N plugins":
    1. Constants module (ui_theme.rs) is the source of truth.
    2. Const sites that can't call Alpha::with_alpha (not yet const on stable) use a literal RGB matching the token, with a unit test pinning the RGB to the token (e.g. MARKER_VALID, HINT_PILE_HIGHLIGHT_COLOUR, RIGHT_CLICK_HIGHLIGHT_COLOUR).
    3. Cross-plugin duplication (e.g. MARKER_DEFAULTPILE_MARKER_DEFAULT_COLOUR) collapses to a single promoted const re-exported from one plugin and imported by the other — replaces "kept in sync" doc comments with a compile-time invariant.
    4. Domain colours (suit pips, card faces, lerp helpers) stay as literals with a comment naming the rationale; only UI chrome routes through tokens.
  • SplashFadable scaffolding pattern (introduced in cacb19c). Any future overlay that needs to fade N >> 3 elements together should follow the same shape: one tiny marker carrying the full-alpha base colour, one global query that lerps every marker's alpha each frame, no per-element query plumbing. Cleanly outscales the Without<X>, Without<Y> query exclusion pattern that the old splash was hitting at three siblings.

Canonical remote

github.com/funman300/Rusty_Solitaire is the canonical repo. Always push there. As of v0.21.0 origin matches local; the next push happens when post-cut work accumulates and is ready to roll into a v0.21.1 / v0.22.0 cut.

Design direction (Terminal — base16-eighties)

  • Tone: retro-terminal / synthwave — flat depth (no box-shadows), monospaced-forward typography (JetBrains Mono / FiraMono), tight 16 px edge margins, 8 px card radius.
  • Palette: near-black surface ramp (#151515 / #202020 / #2a2a2a / #353535), brick-red primary CTA (#a54242 — swapped from cyan #6fc2ef in v0.21.0 commit a292a7e), lime success (#acc267), gold warning (#ddb26f), pink error / suit-red (#fb9fb1), lavender celebration (#e1a3ee), teal info (#12cfc0).
  • Two-color suits. Red = #fb9fb1, black = #d0d0d0. Outlined glyphs for diamonds & clubs are always on; the Settings "color-blind mode" toggle swaps red → lime #acc267 (was red → cyan pre-v0.21.0; lime is the next-best non-red base16-eighties accent now that the primary itself is red).
  • Card glyphs render upright in both corners — no 180° inverted-corner-indicator rotation. Single-orientation digital play doesn't benefit from the traditional flip- readback convention. design-system.md § Game Cards documents this deliberate deviation.

Resume prompt

You are a senior Rust + Bevy developer working on Solitaire Quest.
Working directory: <Rusty_Solitaire clone path on this machine>.
Branch: master. v0.21.3 is tagged at 3d92a91 (cut 2026-05-08, a
patch release rolling up the accessibility-arc closure: HC reaches
the previously-carved-out dynamic-paint sites, and the first real
consumer of `ToastVariant::Warning` lands as the daily-challenge
expiry toast). v0.21.2 stays at f23df3b, v0.21.1 at daa655a,
v0.21.0 at 04f9bf9. Working tree clean. See CHANGELOG.md §
[0.21.3] for full detail.

State: HEAD locally — see `git rev-parse HEAD`. All workspace tests
pass (1207+; check with `cargo test --workspace`), clippy clean.

READ FIRST (in order, before doing anything):
  1. SESSION_HANDOFF.md  — this file
  2. CHANGELOG.md        — [0.21.3] section is the most recent cut
  3. CLAUDE.md           — unified-3.0 rule set
  4. CLAUDE_SPEC.md      — formal architecture spec
  5. ARCHITECTURE.md     — crate responsibilities + data flow
  6. docs/ui-mockups/    — design system + 24-mockup library +
                           desktop-adaptation.md (the rules-based
                           companion to the mockups; read this
                           before any plugin port)
  7. docs/android/*      — Android setup + build runbook
  8. ~/.claude/projects/<this-project>/memory/MEMORY.md
                         — saved feedback / project context
                           (machine-local; may be missing on a
                           fresh machine)

DECISION TO ASK THE PLAYER FIRST:
  A. APK launch verification on AVD / device — `adb install` +
     `adb logcat` to shake out runtime bugs the build / unit
     tests can't catch. Likely surfaces JNI ClipboardManager
     and Android Keystore stubs that need real bridges. Larger
     scope; needs an Android device or emulator running.
  B. Replay-overlay screen-takeover redesign — multi-session
     work: WIN MOVE marker on the scrub bar (data field
     `Replay::win_move_index` already landed in `ab857bb`, just
     needs UI consumption), move-log scroller, mini-tableau
     preview, playback controls. The smaller floating-MOVE-chip
     piece of B already shipped in v0.21.2 (`2fb2d63`); the
     data-layer prerequisite shipped post-v0.21.3 in `ab857bb`.
     The natural next commit on B is the WIN MOVE marker UI on
     the scrub bar — finite, single-system, consumes the new
     field directly.
  C. Phase 8 (sync) — local storage scaffolding, self-hosted
     Axum server, `SolitaireServerClient` impl, GPGS stub
     wired into Settings. The biggest open arc by scope; rolls
     up several Phase Android dependencies (Keystore,
     ClipboardManager).

WORKFLOW NOTES:
  - Use the system git config (already correct).
  - When attributing playtester feedback in commits/docs, use
    "Quat" not "Rhys" (saved feedback memory).
  - Sub-agents stage + verify only; orchestrator commits.
  - Every commit must pass build / clippy / test before pushing.
  - Push to GitHub (origin) — gh auth setup-git wired on
    primary dev box; verify on laptop before first push.
  - Token-port pattern: when migrating tokens, walk every
    concrete artifact downstream of the token (PNG textures,
    embedded SVGs, hardcoded literals, comment color names),
    not just the token name. v0.21.0 surfaced three "the
    migration walked past this" follow-ups that all matched
    this shape — codified here so future similar work can
    pattern-match instead of rediscovering.
  - Doc-vs-implementation drift pattern: v0.21.1's pile-marker
    visibility fix (`4d48cad`) implemented an invariant that
    had been declared in a module doc comment but was never
    enforced in code. When future work touches a module with
    a "this does X" doc comment, verify the code actually does
    X and add a test if not. Two layers, two checks.

OPEN AT THE START: ask which of AC. Don't pick unilaterally.
Note: every remaining option is multi-session by nature (A is
gated on Android tooling, B and C are explicitly multi-session
arcs). A fresh session is a better fit for any of them than the
tail of a long working stretch.