Commit graph

2 commits

Author SHA1 Message Date
Erik
5f9df4d620 sky(phase-3e): drive WeatherSystem from DayGroup name — no more rogue rain
User reported rain in acdream while retail showed a clear sunny sky
after Phase 3d landed. Root cause: two independent weather systems
running in parallel.

  1. Retail DayGroup picker (FUN_00501990 port, Phase 3c/3c.1) —
     selected DayGroup[6] "Sunny" correctly.
  2. WeatherSystem.Tick (legacy stub from pre-decompile era) —
     kept rolling its own hardcoded PDF every day (60% Clear, 20%
     Overcast, 12% Rain, 5% Snow, 3% Storm), independent of the
     DayGroup picker. Its output drove the rain/snow particle
     emitters via UpdateWeatherParticles. If its hash happened to
     land on Rain for today's dayIndex, rain rendered even on a
     Sunny DayGroup day.

Retail has ONE source of truth for weather: the DayGroup roll. There
is no separate weather state machine — rain/snow/storm are implied by
the DayGroup name and its per-keyframe SkyObjectReplace settings.

Fix (Phase 3e):
- WeatherSystem.SetKindFromDayGroupName(string?) — loose substring
  match on the retail DayGroup name: "storm" → Storm, "snow" → Snow,
  "rain" → Rain, "cloud"/"overcast"/"dark"/"fog" → Overcast, else
  Clear. Case-insensitive. Covers the names observed in the live
  Dereth dat dump (Sunny, Clear, Cloudy, Rainy + inferred variants).
- WeatherSystem._externallyDriven flag disables the internal
  RollKind auto-roll once SetKindFromDayGroupName has been called at
  least once. Tests that drive Tick() directly keep the legacy
  hash-roll behavior (offline fallback). ForceWeather still works
  for debug overrides.
- GameWindow.RefreshSkyForCurrentDay calls
  Weather.SetKindFromDayGroupName(grp.Name) right after it installs
  the new SkyStateProvider. Logs the resulting WeatherKind on the
  same line as the DayGroup pick for correlation.
- New WeatherSystemTests.SetKindFromDayGroupName_MapsRetailNames
  (theory, 14 cases) + SetKindFromDayGroupName_DisablesInternalRoll.

Expected effect: Sunny/Clear DayGroups → no rain emitter. Rainy/Stormy
DayGroups → rain emitter active. The user's specific scenario
(DayGroup[6] "Sunny") now correctly maps to WeatherKind.Clear and no
particles spawn.

Build + 733 tests green (+16 new).

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
2026-04-24 09:32:27 +02:00
Erik
0df1c5b4a6 feat(world): Phase G.1 data model — dat-accurate SkyKeyframe + WeatherSystem
Expand the SkyKeyframe record with retail-exact fog fields (FogStart,
FogEnd, FogMode) per r12 §5. The existing FogDensity field is retained
for backwards compat with tests that pin it; new shipping code reads
FogStart / FogEnd / FogMode directly.

Add WeatherSystem (WeatherKind + EnvironOverride enum + 10s transition
ease + deterministic per-day-index roll) matching r12 §6.1. Roll weights
are ~60% Clear / 20% Overcast / 12% Rain / 5% Snow / 3% Storm — tuned
against retail observations. Storm mode triggers lightning flashes
every 8–30 s via an exponential-decay (200ms τ) flash level that the
shader consumes as an additive scene bump.

Add SkyDescLoader that parses the Region dat (0x13000000) into
LoadedSkyDesc — DayGroupData with SkyObjectData (visibility window +
arc sweep), per-keyframe SkyObjectReplaceData, and a shader-ready
SkyStateProvider builder. Sun/ambient colors are pre-multiplied by
DirBright/AmbBright so the shader never needs to know about retail's
scalar brightness field.

19 new tests (weather determinism, transition ease, environ override
tint, flash decay, dat-load conversion with fog + pre-mult colors).

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
2026-04-19 10:29:33 +02:00