Wire IDistanceComputer into RotationSlideStrategy, replacing inline
CPU/GPU branching. BestFitFinder constructs the appropriate implementation.
Replace PushDirection enum with direction vectors in BuildOffsets.
Rename IBestFitStrategy.Type and PairCandidate.StrategyType to StrategyIndex
for clarity (JSON field name unchanged for backward compatibility).
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Extract distance computation from RotationSlideStrategy into a pluggable
IDistanceComputer interface. CpuDistanceComputer adds leading-face vertex
culling (~50% fewer rays per direction) with early exit on overlap.
GpuDistanceComputer wraps ISlideComputer with Line-to-flat-array conversion.
SlideOffset struct uses direction vectors (DirX/DirY) instead of PushDirection.
SpatialQuery.RayEdgeDistance(dirX,dirY) made public for CPU path.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Three bugs fixed in NfpSlideStrategy pipeline:
1. NoFitPolygon.Reflect() incorrectly reversed vertex order. Point
reflection (negating both axes) is a 180° rotation that preserves
winding — the Reverse() call was converting CCW to CW, producing
self-intersecting bowtie NFPs.
2. PolygonHelper inflation used OffsetSide.Left which is inward for
CCW perimeters. Changed to OffsetSide.Right for outward inflation
so NFP boundary positions give properly-spaced part placements.
3. Removed incorrect correction vector — same-drawing pairs have
identical polygon-to-part offsets that cancel out in the NFP
displacement.
Also refactored NfpSlideStrategy to be immutable (removed mutable
cache fields, single constructor with required data, added Create
factory method). BestFitFinder remains on RotationSlideStrategy
as default.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
NFP strategy has coordinate correction issues causing overlaps.
The slide-based approach is fast and accurate — keeping it as default.
NfpSlideStrategy and PolygonHelper remain in the codebase for future use.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Calling ConvexMinkowskiSum directly with manual reflection produced
wrong winding/reference-point handling, causing all pairs to overlap.
Route through Compute which handles reflection correctly. Hull inputs
keep it fast — few triangles means trivial Clipper union.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
ConvexMinkowskiSum is O(n+m) with no boolean geometry ops.
The concave Minkowski path was doing triangulation + pairwise
sums + Clipper2 Union, which hung at 100% CPU for complex parts.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Replace RotationSlideStrategy with NfpSlideStrategy in BuildStrategies,
and add integration tests covering the end-to-end FindBestFits pipeline.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Creates PolygonHelper.cs in OpenNest.Engine.BestFit with ExtractPerimeterPolygon
(returning PolygonExtractionResult with polygon + correction vector) and RotatePolygon.
AutoNester.ExtractPerimeterPolygon and RotatePolygon become thin delegates.
Adds MakeSquareDrawing/MakeLShapeDrawing to TestHelpers and 6 PolygonHelperTests.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Normalize pair bounding box to landscape (width >= height) in
PairEvaluator for consistent display and filtering. Fix
BestFitViewerForm where BoundingWidth/BoundingHeight were passed
in the wrong order to the plate Size constructor.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Move fill algorithms to OpenNest.Engine.Fill namespace:
FillLinear, FillExtents, PairFiller, ShrinkFiller, Compactor,
RemnantFiller, RemnantFinder, FillScore, Pattern, PatternTiler,
PartBoundary, RotationAnalysis, AngleCandidateBuilder, and
AccumulatingProgress.
Move strategy layer to OpenNest.Engine.Strategies namespace:
IFillStrategy, FillContext, FillStrategyRegistry, FillHelpers,
and all built-in strategy implementations.
Add using directives to all consuming files across Engine, UI,
MCP, and Tests projects.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
The Size fix (d4222db) changed Size.Width to Y axis and Size.Length to
X axis but only updated DrawPlate/LayoutViewGL. BoundingBox, WorkArea,
rotations, DXF export, and engine code still used the old Width=X
convention, causing the fill engine to get a swapped work area (60x120
instead of 120x60) and parts to fill in the wrong direction.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
The BestFitFilter's aspect ratio cap of 5.0 was rejecting valid pair
candidates needed for narrow plates (e.g. 60x6.5, aspect 9.2) and
remainder strips on normal plates. Three fixes:
- BestFitFinder: derive MaxAspectRatio from the plate's own aspect
ratio so narrow plates don't reject all elongated pairs
- SelectPairCandidates: search the full unfiltered candidate list
(not just Keep=true) in strip mode, so pairs rejected by aspect
ratio for the main plate can still be used for narrow remainder
strips
- BestFitCache.Populate: skip caching empty result lists so stale
pre-computed data from nest files doesn't prevent recomputation
Also fixes console --size parsing to use LxW format matching
Size.Parse convention, and includes prior engine refactoring
(sequential fill loops, parallel FillPattern, pre-sorted edge
arrays in RotationSlideStrategy).
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
- Add verbose per-file and per-sheet-size console output during collection
- Skip already-processed parts at the sheet-size level instead of all-or-nothing
- Precompute best-fits once per part and reuse across all sheet sizes
- Clear best-fit cache after each part to prevent memory growth
- Save best-fits in separate bestfits/ zip entries instead of embedding in nest.json
- Filter to Keep=true results only and scope to plate sizes in the nest
- Set nest name to match filename (includes sheet size and part count)
- Add TrainingDatabase with per-run skip logic and SQLite schema
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Store hull edge angles in BestFitResult at evaluation time so they
don't need to be recomputed during the fill phase. Extract
GetHullEdgeAngles(Polygon) overload from FindHullEdgeAngles.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Thread ISlideComputer through BestFitCache → BestFitFinder →
RotationSlideStrategy. RotationSlideStrategy now collects all offsets
across 4 push directions and dispatches them in a single batch (GPU or
CPU fallback). Also improves rotation angle extraction: uses raw geometry
(line endpoints + arc cardinal extremes) instead of tessellation to avoid
flooding the hull with near-duplicate edge angles, and adds a 5-degree
deduplication threshold.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
ISlideComputer abstracts batched directional-distance computation so GPU
implementations can process all slide offsets in a single kernel launch.
GpuSlideComputer uses ILGPU with prepared edge data (precomputed inverse
deltas and min/max bounds) and caches stationary/moving buffers across
calls. GpuEvaluatorFactory exposes a singleton factory method.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
"Length" is more natural than "height" for flat plate materials.
Renames the field on OpenNest.Geometry.Size, Box.Height property,
and all references across 38 files.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Narrow remnant strips now get more parts by:
- Sweeping rotations every 5° when the strip is narrower than the part
- Including all pairs that fit the strip width (not just top 50 by area)
- Placing individual parts from incomplete pattern copies that still fit
- Using finer polygon tolerance (0.01) for hull edge angle detection
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Add PushDirection.Right and PushDirection.Up to RotationSlideStrategy so
parts can approach from all four directions. This discovers concave
interlocking arrangements (e.g. L-shaped parts nesting into each other's
cavities) that the original Left/Down-only slides could never reach.
Introduce BestFitCache so best-fit results are computed once at step size
0.25 and shared between the viewer and nesting engine. The GPU evaluator
factory is configured once at startup instead of being wired per call
site, and NestEngine.CreateEvaluator is removed.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
- Move BuildPairParts to BestFitResult.BuildParts() instance method
- Extract BinConverter (RectanglePacking) for Part/NestItem/Bin conversions
- Extract RotationAnalysis for FindBestRotation and FindHullEdgeAngles
NestEngine reduced from 484 to 287 lines — now purely orchestration,
strategy selection, and comparison logic.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
- Fill(NestItem) and Fill(List<Part>) now delegate to their Box overloads
- Add Part.CreateAtOrigin() to replace repeated 4-line build-at-origin pattern
used in NestEngine, RotationSlideStrategy, and PairEvaluator
- Remove dead code: FillArea overloads, Fill(NestItem, int), FillWithPairs(NestItem),
ConvertTileResultToParts, PackBottomLeft.FindPointHorizontal, Pattern.GetLines/GetOffsetLines,
unused count variable in FillNoRotation
- Simplify IsBetterValidFill to delegate to IsBetterFill after overlap check
NestEngine reduced from 717 to 484 lines.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
The perpendicular sweep started at perpMin (e.g. -8.75) which with
coarse step sizes never landed on offset=0, missing the perfect
side-by-side and stacked same-orientation patterns for rectangular parts.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
