acdream

erik/acdream

Fork 0

Commit graph

Author	SHA1	Message	Date
Erik	56975f8919	fix(terrain): align per-cell triangle geometry with ACE's ConstructPolygons convention Our LandblockMesh, terrain.vert corner tables, and TerrainSurface.SampleZ used the OPPOSITE diagonal for each CellSplitDirection enum value from what ACE (and the decompiled retail client at FUN_00532a50) picks for the same sign bit. Same formula, same sign-bit mapping, inverted geometry. Symptom: remote players rendered at server-broadcast Z hovered or clipped by up to ~1m on sloped cells. Flat cells masked the bug because all four corner heights were equal so any triangle pair returned the same Z. Live diagnostic confirmed +0.79m hover on cell (7,5) at lb(AA,B4) — a ~20° slope — while flat neighbors agreed to floating-point noise. Three coordinated edits so CPU mesh + GPU corner lookup + CPU sampler all agree on the retail geometry: - LandblockMesh: SWtoNE branch now emits {BL,BR,TR}+{BL,TR,TL} (y=x cut), SEtoNW emits {BL,BR,TL}+{BR,TR,TL} (x+y=1 cut). - terrain.vert: corner-index tables updated to match. - TerrainSurface.SampleZ: swapped the two branches' interpolation. After the fix, 19 live DIAG samples across flat + two slope transitions all land within 0.01m of server Z. Staircase pattern during remote motion on slopes is a separate bug (no per-frame collision resolution) and will be addressed via the transition/FindValidPosition port. Cross-verified against: ACE LandblockStruct.ConstructPolygons lines 221- 244, decompiled retail FUN_00532a50 (chunk_00530000.c:2235), ClientReference IsSWtoNECut (tests/AcDream.Core.Tests/Terrain/ClientReference.cs). Updated test SplitDirection_TerrainSurface_AgreesWith_TerrainBlending with corrected expectations (Z values swap between the two branches). All 717 tests green. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>	2026-04-21 13:20:59 +02:00
Erik	05749f52e0	test: port ACME ClientReference + conformance tests Ports the decompiled AC client ground-truth oracle and exhaustive conformance test suite from WorldBuilder-ACME-Edition into acdream's test project. ClientReference.cs: faithful C# port of CLandBlockStruct.cpp with IsSWtoNECut, GetPalCode, GetVertexHeight, GetVertexPosition. ClientConformanceTests.cs verifies acdream's implementations match: - SplitDirection: 9 spot-checks + 25,600-cell full sweep (0 mismatches) - PalCode: 5 spot-checks + 256 exhaustive roads + 1M exhaustive types - Height sampling: flat terrain exact match, vertex corners match, interpolated points in-range - TerrainSurface.SampleZ agrees with TerrainBlending split direction - Constants match (CellSize=24, CellsPerBlock=8, BlockLength=192) 27 new tests. 310 total (201 core + 109 net), all green. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>	2026-04-12 22:45:20 +02:00

Author

SHA1

Message

Date

Erik

56975f8919

fix(terrain): align per-cell triangle geometry with ACE's ConstructPolygons convention

Our LandblockMesh, terrain.vert corner tables, and TerrainSurface.SampleZ
used the OPPOSITE diagonal for each CellSplitDirection enum value from
what ACE (and the decompiled retail client at FUN_00532a50) picks for the
same sign bit. Same formula, same sign-bit mapping, inverted geometry.

Symptom: remote players rendered at server-broadcast Z hovered or clipped
by up to ~1m on sloped cells. Flat cells masked the bug because all four
corner heights were equal so any triangle pair returned the same Z. Live
diagnostic confirmed +0.79m hover on cell (7,5) at lb(AA,B4) — a ~20°
slope — while flat neighbors agreed to floating-point noise.

Three coordinated edits so CPU mesh + GPU corner lookup + CPU sampler all
agree on the retail geometry:
 - LandblockMesh: SWtoNE branch now emits {BL,BR,TR}+{BL,TR,TL} (y=x cut),
   SEtoNW emits {BL,BR,TL}+{BR,TR,TL} (x+y=1 cut).
 - terrain.vert: corner-index tables updated to match.
 - TerrainSurface.SampleZ: swapped the two branches' interpolation.

After the fix, 19 live DIAG samples across flat + two slope transitions
all land within 0.01m of server Z. Staircase pattern during remote motion
on slopes is a separate bug (no per-frame collision resolution) and will
be addressed via the transition/FindValidPosition port.

Cross-verified against: ACE LandblockStruct.ConstructPolygons lines 221-
244, decompiled retail FUN_00532a50 (chunk_00530000.c:2235), ClientReference
IsSWtoNECut (tests/AcDream.Core.Tests/Terrain/ClientReference.cs).

Updated test SplitDirection_TerrainSurface_AgreesWith_TerrainBlending
with corrected expectations (Z values swap between the two branches).
All 717 tests green.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

2026-04-21 13:20:59 +02:00

Erik

05749f52e0

test: port ACME ClientReference + conformance tests

Ports the decompiled AC client ground-truth oracle and exhaustive
conformance test suite from WorldBuilder-ACME-Edition into acdream's
test project.

ClientReference.cs: faithful C# port of CLandBlockStruct.cpp with
IsSWtoNECut, GetPalCode, GetVertexHeight, GetVertexPosition.

ClientConformanceTests.cs verifies acdream's implementations match:
- SplitDirection: 9 spot-checks + 25,600-cell full sweep (0 mismatches)
- PalCode: 5 spot-checks + 256 exhaustive roads + 1M exhaustive types
- Height sampling: flat terrain exact match, vertex corners match,
  interpolated points in-range
- TerrainSurface.SampleZ agrees with TerrainBlending split direction
- Constants match (CellSize=24, CellsPerBlock=8, BlockLength=192)

27 new tests. 310 total (201 core + 109 net), all green.

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

2026-04-12 22:45:20 +02:00

2 commits