Gate all suites in CI, add _on_tick integration test, finish deferred refactors

[joryirving]

Closes #234. Follow-ups from #282.

Summary

• CI (both jobs) and a new just test-all loop over every tests/test_*.gd instead of four hand-listed suites — the other 15 suites are now actually gated. The TestCase base fails any run where zero assertions executed, so a suite whose preload fails to compile can't silently exit 0.
• New tests/test_tick_integration.gd (Add full-cycle integration test for _on_tick #234): drives ColonySim.process_tick() with no scene/UI/autoloads — worker movement, gather/haul cycles, food upkeep cadence, goal rotation, dirty-flag contract, starvation build stalls, idle behavior, build completion + milestone advance. Deviation from the issue text: it's a standalone suite gated by the CI loop rather than folded into test_runner.gd, matching the post-Refactor test suite onto a shared TestCase base with a canonical assertion API #281 one-suite-per-file convention.
• Sim is now fully autoload-free: the reservation rebuild is a static ColonySim function GameState delegates to (this was required for the integration test to compile in --script mode). do_haul flattened into _deliver_carried/_pick_up_for_build under that new coverage.
• Popup visibility is a single MenuMode state machine; apply_anchor_layout's mirrored reparenting branches collapse onto an ordered node/parent/index table; structure/tile icons + short labels move to constants.gd; next_unlock_text walks BUILD_UNLOCKS; LayoutMath exports ALL_ANCHOR_FAMILIES and BOTTOM_DOCK_MIN_HEIGHT (save validation and main.gd stop mirroring them); main.gd uses Constants.X uniformly.
• *.log ignored; smoke.log/smoke_ci.log untracked.

Verification

• All 19 suites headless: exit 0, 724 passing assertions, zero script errors.
• Full-game headless smoke boot (--fixed-fps 60 --quit-after 300): no script errors.

Notes

• Minor intended behavior changes: closing the popup now always reapplies dock sizing (previously skipped on some close paths), and a failed load closes the popup instead of leaving it half-open.

[its-saffron]

AI Automated Review

Full PR review.

Analysis engine: MiniMax-M2.7@https://litellm.jory.dev/v1 (anthropic) — escalated (fast_low_confidence)

Recommendation: Approve

This PR is well-executed and addresses its stated goals comprehensively. The new tick integration test (issue PR 234), CI test discovery loop, autoload-free refactor, and popup state machine are all high-quality additions that align with repository conventions.

Change-by-Change Findings

CI / test discovery (.github/workflows/test.yml, .justfile)

• Replaces 4 hand-listed test steps with a for suite in tests/test_*.gd loop that automatically gates all 19 suites (excluding test_case base). This is strictly better than the previous approach — suites added later are automatically caught.
• .justfile gains a test-all command that mirrors the CI matrix exactly, improving local validation fidelity.
• Minor behavior change (acknowledged in PR notes): the test_runner.gd suite is now run alongside 18 others rather than as a wrapper. This is fine — the test logic is unchanged.

New integration test (tests/test_tick_integration.gd, 202 lines)

• Instantiates ColonySim directly (no scene tree, no autoloads) and drives process_tick() programmatically.
• Covers all required acceptance criteria: worker movement, gather/haul cycle, food upkeep cadence, goal completion/rotation, dirty flag contract, starvation stalls, idle behavior, and build completion with milestone advance.
• Acceptance criteria deviation: The issue explicitly requires inclusion in test_runner.gd. The PR runs it as a standalone suite via the CI loop instead, citing the post-PR 281 one-suite-per-file convention. Functionally the test runs in CI (all test_*.gd except test_case are auto-discovered), so the spirit of the criterion is met. This is acceptable.

Autoload-free ColonySim (scripts/colony_sim.gd, scripts/game_state.gd)

• rebuild_reservations_from_workers() is now a static ColonySim method. GameState delegates to it. This enables ColonySim to be instantiated in --script mode without triggering autoload initialization — a prerequisite for the integration test.
• do_haul() is flattened into _deliver_carried() and _pick_up_for_build(), with improved inline documentation. The logic is unchanged.

Popup state machine (scripts/main.gd)

• MenuMode enum (CLOSED, MENU, MANAGEMENT, SETTINGS) replaces scattered sidebar_scroll.visible / menu_actions.visible / management_panels.visible / settings_panel.visible flag juggling.
• set_menu_mode() is the single mutating entry point; all open/close paths (toggle_menu, close_menu, open_build_popup, open_settings, close_settings, save_game, _load_failed, load_saved_game, begin_build_placement, queue_structure_at, cancel_build_placement, _on_dock_side_selected) delegate to it.
• Minor behavior change (acknowledged in PR notes): closing the popup now always reapplies dock sizing (previously skipped on some paths), and _load_failed closes the popup instead of leaving it half-open. These are correct fixes.
• _apply_header_tree() consolidates the mirrored reparenting branches in apply_anchor_layout onto an ordered [node, parent, index] table. Identical style properties for both anchors are now shared after the branch.

Constants consolidation (scripts/constants.gd, scripts/main.gd)

• STRUCTURE_ICONS, TILE_ICONS, TILE_SHORT_LABELS added to Constants. tile_icon() and tile_amount_text() in main.gd now delegate to these tables instead of inline match statements.
• next_unlock_text() now walks Constants.BUILD_UNLOCKS — adding a new structure tier only requires a table entry, not editing function logic.
• main.gd preloads cleaned up: individual const extractions removed; code consistently uses Constants.X.

LayoutMath exports (scripts/layout_math.gd, scripts/game_state.gd, scripts/main.gd)

• ALL_ANCHOR_FAMILIES and BOTTOM_DOCK_MIN_HEIGHT exported as public constants. game_state.gd's _expected_grid_sizes() now iterates LayoutMath.ALL_ANCHOR_FAMILIES instead of hardcoding ["bottom", "side"]. main.gd uses LayoutMath.BOTTOM_DOCK_MIN_HEIGHT instead of the literal 460.

Test base hardening (tests/test_case.gd)

• Added zero-assertion guard: if test_pass == 0 and test_fail == 0, exits with code 1. This catches preload compile failures that would silently exit 0.

Standards Compliance

• Autoload-free sim: ColonySim is a RefCounted class with no autoload dependencies, enabling standalone --script mode testing. ✅
• Save/version migration: Reservation rebuild is a static function callable from both load path (GameState) and runtime (ColonySim.rebuild_reservations()). ✅
• Constants over scattered literals: Icon tables, BOTTOM_DOCK_MIN_HEIGHT, ALL_ANCHOR_FAMILIES are centralized. ✅
• Popup as overlay: The state machine keeps the sidebar transient; dock sizing is reapplied on every mode change. ✅
• No tile_is_square stub: Removed the no-op tile_is_square function (always returned true by comment; body was "by construction"). This was dead code. ✅
• test_case exclusion: Both CI and .justfile skip test_case (the base class), matching the expected pattern. ✅

Linked Issue Fit

Issue PR 234 — "Add full-cycle integration test for _on_tick"

Criterion	Status
tests/test_tick_integration.gd exists and runs headless	✅ Added; gated by CI loop
Covers worker movement, resource gathering	✅ test_tick_advances_and_worker_moves, test_gather_and_haul_cycle
Covers food upkeep	✅ test_food_upkeep_interval
Covers goal tick	✅ test_goal_completion_and_rotation
Runs without full UI scene	✅ Instantiates ColonySim directly; no Control/autoload
Included in test_runner.gd	⚠️ Deviates — runs as standalone suite via CI auto-discovery loop instead of being folded into test_runner.gd. Functionally included (CI runs all test_*.gd except test_case), and matches post-PR 281 convention. Acceptable.

Tool Harness Findings

No tool harness output present in corpus.

CI Check Results

All CI checks passed (6/6): Export validation (Linux/macOS/Windows), Script test suite, macOS validation, Headless smoke test. Test suite exited 0 with 724 passing assertions. CI is green.

Unknowns / Needs Verification

None. All evidence is complete and consistent.