megaproxy/rimlike
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions
97 commits 1 branch 0 tags 11 MiB
Commit graph

3 commits

Author SHA1 Message Date
megaproxy
16e04e46f0 add reachability pre-checks to plant/sleep/chop/mine 
Provider audit found 6 WorkProviders missing reachability gates before
returning jobs. Without them, pawns can be offered doomed walk-jobs
(target boxed in), JobRunner cancels each tick, Decision re-offers
same job → 20Hz busy-spin starves lower-priority work.

Fixed 4 here (mechanical pattern):
- PlantProvider._find_harvest: walkable-target check (mirrors _find_sow)
- SleepProvider: walkable bed-tile check
- ChopProvider: adjacent-walkable for impassable tree
- MineProvider: adjacent-walkable for impassable rock

Cooking/Crafting reachability changes (in the same audit's
recommendation) were attempted but caused intermittent null returns
that regressed cooking rate. Reverted those — they need more careful
work that doesn't break the existing flow. Filed separately.

Future cleanup: _find_adjacent_walkable duplicated across
ConstructionProvider, ChopProvider — extract to a base/util.

MCP-verified after revert: 2 meals + 1 bread + 2 grain in cabin crate
within 3700 ticks at ULTRA. Cooking fires, hauling fires, all
production paths operational.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
 2026-05-17 20:20:35 +01:00
megaproxy
2f76ae1639 sleep_provider deconflicts beds against other pawns' targets 
L: SleepProvider.find_best_for filters bed candidates via the existing
Job.is_target_taken_by_other mechanism that ConstructionProvider
already uses. Sets j.target_node = best_bed on the proposed job so
other pawns see the claim.

Fixes the 2/3-pawns-floor-sleep symptom (memory.md 2026-05-11) caused
by greedy nearest-neighbor convergence. The bed.claim() mechanism was
already race-free; this just prevents simultaneous proposals on the
same bed.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
 2026-05-16 18:29:19 +01:00
megaproxy
43e52ffe75 Phase 8 — Beds, sleep need, thoughts, mood, Sulking soft-break 
Three gdscript-refactor agents in parallel; Opus integrated and verified
the sleep+wake cycle via MCP runtime.

Bed entity (Agent A, scenes/entities/bed.{gd,tscn} + world.gd, ~280 lines):
- class Bed extends Node2D — bottom-anchored 3/4 perspective like Wall/Workbench
- BuildJob interface (is_buildable / on_build_tick / _complete) — same pattern
  as Wall / Crate / Workbench. blocks_pathing_when_complete=false (walkable).
- Quality-tinted sheet colours by Item.Quality tier (drab grey → blue →
  gold-brown → regal pink); white pillow + dark frame constant across tiers.
- claim(pawn) / release() / is_available() — atomic occupancy; claim re-checks
  is_available() inside to avoid race conditions during pawn walk-to-bed.
- World.beds registry + register_bed / unregister_bed (mirrors workbench pattern)

Sleep need + SleepProvider + KIND_SLEEP toil (Agent B, ~220 lines):
- Pawn.sleep: float 0..100. SLEEP_DECAY_PER_TICK=0.015 (~6667 ticks / 5.5 min
  at 1× / 1 min at Ultra to fully tire). Slower than hunger.
- is_tired() at <30; is_exhausted() at <5 (Phase 9 status interrupt hook)
- SleepProvider priority=8 (highest — sleep beats eat=7 when both urgent)
- Toil.KIND_SLEEP + Toil.sleep_in_bed(NodePath) factory
- JobRunner._tick_sleep: first-tick bed claim (with race-loss → floor fallback),
  per-tick recovery (bed=0.5/tick, floor=0.25/tick), wake-when-full at ≥99,
  emergency ceiling SLEEP_TICKS_MAX=2000 prevents stuck-asleep loops

Thoughts + mood + Sulking (Agent C, ~290 lines):
- scenes/ai/thought.gd: class Thought (RefCounted) with id, modifier, lifetime
  (PERSISTENT/EVENT), stacks, ticks_remaining; MAX_STACKS_PER_THOUGHT=5 locked
- scenes/ai/thought_catalog.gd: ThoughtCatalog with 5 Phase 8 thoughts —
  hungry(-6, PERSISTENT) / tired(-4, PERSISTENT) / well_rested(+5, EVENT 1200t)
  / slept_on_floor(-5, EVENT 1200t) / ate_meal(+3, EVENT 800t, stacks up to 3)
- Pawn extended: thoughts: Array, mood: float (base 50), sulking: bool,
  _sulk_low_ticks. add_thought (stack-merge by id), remove_thought_by_id,
  has_thought, is_sulking. _process_thoughts in sim_tick decays EVENT thoughts,
  syncs PERSISTENT thoughts to state (hungry/tired), recomputes mood, checks
  sulking transition: mood < 25 for MOOD_SULK_SUSTAIN_TICKS=600 ticks → SULKING;
  mood >= 35 → recover.
- Decision Layer 1 extended: pawn.is_sulking() → return null (sulking pawns
  refuse all work; Phase 17 may add Wandering variant)
- EventBus.pawn_mood_changed signal
- JobRunner._tick_eat: fires ate_meal thought when consuming MEAL/BREAD
- JobRunner._tick_sleep: fires well_rested or slept_on_floor on wake

Opus integration:
- world.tscn: SleepProvider node added (9 providers total)
- world.gd registers in priority order:
  sleep=8 > eat=7 > construction=6 > chop=5 ≈ plant=5 > mine=4 ≈ crafting=4 > haul=3 > rest=0
- Demo seed: 3 beds along cabin's north row at (45/47/49, 24), pre-built
  so pawns can sleep immediately when tired

Acceptance — MCP-verified end-to-end:
- Pre-tired Bram at sleep=25 → SleepProvider issued 'Sleep at (45, 24)' job
- Bram walked to bed, claimed, slept 200 ticks, woke at sleep≥99
- Bed released back to available; well_rested thought fired (+5 mood)
- After ~12000 ticks total: all 3 pawns slept (sleep recovered to 67/86/51),
  thoughts active (1-2 per pawn — well_rested + ate_meal from Phase 7 cooked
  bread consumption), beds all back to available, no claim leaks
- Mood compute working (base 50 + thought modifiers); sulking transition
  ready but didn't fire — would need misery accumulation (Phase 9 Cold +
  Bleeding statuses) to drive mood < 25 sustained

Phase 8 followups for later phases:
- Sulking returns null (stand still); Phase 17 may add Wandering soft-break
  that issues a random-walk job
- Bed ownership (_owner_pawn) reserved but not used in Phase 8 — Phase 17
  may add 'bedrooms' where each pawn claims a specific bed
- _tick_sleep's using_bed local-var reset pattern is correct but fragile;
  cleanup pass when status interrupts (Phase 9) wire into the eat/sleep
  cancellation path

Delegation report this phase:
- Agent A: Bed entity (buildable, quality-tinted, claim/release)
- Agent B: Pawn.sleep + SleepProvider + KIND_SLEEP toil + JobRunner._tick_sleep
- Agent C: Thought + ThoughtCatalog + Pawn mood/sulking + Decision Layer 1
  + JobRunner thought hooks in _tick_eat / _tick_sleep
- Opus: scene wiring + 3 beds in demo seed + MCP runtime verification

~75% of Phase 8 GDScript was subagent-authored.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
 2026-05-11 13:21:15 +01:00