EnsureLoadedAround skipped Get() for any active chunk already in
_inflight. That worked for the MonoGame TacticalRenderer, which calls
Get() during its own draw loop and incidentally drains pre-warm tasks.
But subscribers to OnChunkLoaded (e.g. the Godot port) saw no event
when a previously-pre-warmed chunk transitioned into the active set on
a later frame — the chunk stayed in _inflight forever, presenting as
permanently-uncached gaps in the rendered world.
Fix: drop the !_inflight.ContainsKey(cc) guard. Get() already handles
all three paths (cache hit, inflight drain, fresh generate), so passing
every active chunk through Get() guarantees OnChunkLoaded fires once
per chunk regardless of how it was scheduled.
Same flavour of bug as M1's MoistureGen FastNoiseLite race —
cross-process / event-driven consumers exercise paths the in-process
pull-based test fixtures never hit. 708/708 tests still pass.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
Formalises Content/ access from the Godot host. Content lives at the
repo root (sibling of Theriapolis.Godot/), not duplicated under res://,
so the MonoGame branch and headless Tools keep reading from the same
single source of truth.
ContentPaths.cs:
Static ContentRoot/DataDir/GfxDir resolved once via res:// walk-up
and cached. Replaces two inline ResolveDataDir copies in SmokeTest
and WorldMapView.
ContentLoader.cs:
LoadGfx(relativePath) -> ImageTexture, cached by relative path.
Bypasses the res:// import pipeline because Content/ lives outside
the project — fine for static pixel-art assets at native size, and
the project default texture filter is already Nearest. Cache is
per-process, never evicted (full atlas <1 MB).
AssetTest.cs + Main.cs --asset-test flag:
Smoke-tests the pipeline. Walks Content/Data and Content/Gfx,
reports counts, attempts to load every PNG, prints per-subdir
breakdown. Quits with non-zero on any failure.
Verified post-refactor (--asset-test):
53 JSON files in Data, 50 PNG files in Gfx (8 tactical/deco +
42 tactical/surface), 50/50 loaded, 0 failures.
Verified no regressions:
--smoke-test (M1) still produces canonical FNV hashes.
--world-map 12345 (M2) still produces 3 rivers / 91 roads /
226 settlements / 0 rails / 0 bridges.
Scope note: plan mentioned "tile/NPC/CodexUI atlases — three
separate themes". Only tactical/ exists in Content/Gfx today; NPC
and CodexUI atlases never landed during MonoGame development. M3
ships what's actually present. ContentLoader.LoadGfx works for any
future sub-directories without changes.
Closes M3 of theriapolis-rpg-implementation-plan-godot-port.md.
Next: M4 (tactical render).
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
Renders the full worldgen output as a Godot scene at visual parity with
worldgen-dump's PNG output: biome tiles, rivers/roads/rails as Line2D
polylines, settlements as filled circles. Pan + zoom via Camera2D.
WorldMapView.cs:
- Loads Content/Data via res:// walk-up, runs WorldGenerator.RunAll
- Tile palette built from BiomeDef.ParsedColor() — same source as the
PNG dump, so colours are identical
- Tiles rendered as a 256x256 Image scaled by WORLD_TILE_PIXELS to
cover world-pixel space (matches polyline coord system)
- Polyline draw order mirrors LineFeatureRenderer.cs: roads (smaller
first) -> rivers -> rail tie underlay -> rail line. Bridges as
short Line2Ds; settlements as SettlementDot (Node2D + _Draw circle)
- Line widths in world-pixel space, tuned for visibility at world-map
zoom; M4 will add zoom-aware width scaling for tactical view
- Camera fits the whole world (95% of viewport) on first frame
PanZoomCamera.cs:
- Mouse-wheel zoom centered on cursor (cursor world-point stays fixed)
- Middle/right click + drag to pan
- MinZoom/MaxZoom configurable per-instance
Main.cs:
- --world-map [seed] flag launches the view (default seed 12345)
- Arg parser now reads both GetCmdlineArgs and GetCmdlineUserArgs so
callers don't need to remember the "--" separator
- --smoke-test path and M0 hello-world fallback unchanged
Visual diff against world_seed12345.png (generated by
worldgen-dump --seed 12345) confirmed manually: same biome palette, same
rivers/roads topology, same settlement placement and tier colours.
3 rivers, 91 roads, 226 settlements (138 PoIs), 0 rails (ENABLE_RAIL=false),
0 bridges (this seed has no road/river crossings). All match the PNG.
Settlement dot sizes iterated twice from user feedback — final values
in tile units, scaled to world-pixel space, so they shrink at world-map
zoom and grow toward tactical zoom (the right "scale with the map"
behaviour).
Closes M2 of theriapolis-rpg-implementation-plan-godot-port.md.
Next: M3 (asset pipeline).
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
Proves Theriapolis.Core works untouched under Godot's csproj — the worldgen
pipeline produces byte-identical output whether invoked from the Tools CLI
or from inside the Godot process. This is the determinism contract surviving
the port.
Architecture test:
CoreNoDependencyTests now forbids Godot.* and GodotSharp in addition to
Microsoft.Xna and MonoGame. Both bans stay in force for the duration of
the port so neither engine can leak into Core.
Determinism oracle:
New worldgen-hash Tools command runs the full pipeline and prints FNV-1a
hashes for every channel (elevation, moisture, temperature, biomes,
settlements, polylines) plus per-stage hashes. Pairs with the Godot
smoke-test for cross-process verification.
Godot-side smoke test:
SmokeTest.cs runs WorldGenerator.RunAll inside the Godot process; Main.cs
fires it on --smoke-test <seed>. Resolves Content/Data via res:// walk-up.
M0 hello-world behaviour preserved when launched without the flag.
Verification (seed 12345):
- dotnet run -- worldgen-hash and Godot --headless --smoke-test agree on
all 6 channels and all 14 per-stage hashes (diff produces zero output)
- 10-run sweeps stable on both sides post-determinism-fix
- dotnet test: 708/708 pass
Closes M1 of theriapolis-rpg-implementation-plan-godot-port.md.
Next: M2 (world map render).
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
FastNoiseLite lazily populates its internal _perm[512] table on the first
GetNoise call via EnsurePerm(). When called concurrently from a Parallel.For
loop, threads race on this initialization and may read a partially-populated
table, producing different moisture/temperature values per row across runs.
Empirical: a 10-run worldgen-hash sweep on seed 12345 produced 4+ distinct
moisture hashes and 3+ distinct temperature hashes. All other channels
(elevation, biomes, settlements, polylines) remained stable; biomes only
because their bucket thresholds happened to absorb the upstream float noise.
The fix is the same one ElevationGenStage:125-130 and BorderDistortionGenStage:
102-104 already apply: call GetNoise once on the main thread before the
Parallel.For so _perm is fully initialized when worker threads start reading.
MoistureGenStage and TemperatureGenStage were missing this; now they have it.
WorldgenDeterminismTests didn't catch this because xUnit's WorldCache fixture
runs both pipeline variants in the same process, where consecutive runs hit
the same JIT/thread-pool state and produce the same corrupted output. The
Godot port surfaced it by invoking Core from a fresh process with different
threading.
Verified: post-fix 10-run sweep produces stable hashes on all six channels
(0xA8F99BB9795D8CF8 moisture, 0xAA05F3FB1523F6C3 temperature, seed 12345).
708/708 tests still pass.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
Captures the pre-Godot-port state of the codebase. This is the rollback
anchor for the Godot port (M0 of theriapolis-rpg-implementation-plan-godot-port.md).
All Phase 0 through Phase 6.5 work is included; Phase 7 is in flight.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
Christopher Wiebe