The either/or format meant a SubProgramCall with both a non-zero
Offset and non-zero Rotation would only show the Offset, hiding the
rotation metadata. The data model supports both independently, so the
display should too.
Also fixes a zero-field leak where the old fallback emitted
`G65 P_ R0` for calls with no rotation. Now each field is only shown
when non-zero, and `G65 P_` with no arguments is emitted when
neither is set.
Note: SubProgramCall.ToString is purely a debug/display aid. The
Cincinnati post emits sub-calls via the G52 + M98 bracket, not via
G65, so this format doesn't correspond to real machine output.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
CincinnatiSheetWriter.WriteHoleSubprogramCall emitted
`M98 P<num> X<x> Y<y>`, but per manual §3.98 ("M98 SUB-PROGRAM CALL
WITH NO ARGUMENTS") M98 takes only P and L — the X/Y had no defined
meaning to the control. The intent was to position the sub-program at
the hole center, which is what G52 is for per §1.52 ("local work
coordinate system") and which explicitly does not move the nozzle.
Emit the documented G52 bracket instead:
G52 X<hole.x> Y<hole.y>
M98 P<holeSubNum>
G52 X0 Y0
The hole sub-program is authored in hole-local coordinates, so its
first rapid (the lead-in to the pierce point) resolves to the absolute
pierce under the G52 shift and moves the tool directly there from the
previous feature's end — no phantom rapid to the hole center.
Also add docs/cincinnati-post-output.md as the reference for the full
post output format, with every emitted G/M code cross-referenced to
the Cincinnati programming manual. Un-ignore docs/ (docs/superpowers/
stays ignored) and track the PDF manual alongside the reference.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
The SubProgramCall branch in DrawRapids used to draw a rapid from the
previous feature's end to the hole center, then rely on the sub-program's
own first rapid to draw from center to the lead-in pierce. That rendered
a phantom center-hop segment that doesn't exist physically — a
SubProgramCall is a coordinate-frame shift (emitted as a G52 bracket on
Cincinnati), not a move to the hole center.
Look ahead through the sub-program for its first pierce point in
absolute coordinates and draw a single direct rapid from pos to that
pierce. Recurse into the sub with skipFirstRapid: true so the sub's
first rapid isn't drawn again on top.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
ConvertMode.ToIncremental skipped SubProgramCall codes entirely when
computing deltas, so parent motions after a sub-call were encoded as if
the tool never moved. Several traversal sites (ConvertProgram,
GraphicsHelper, PlateRenderer, CutDirectionArrows, Program.BoundingBox)
worked around this with save/restore hacks that treated sub-calls as
transparent — but DrawRapids legitimately tracks actual tool position,
so after the last hole the first perimeter rapid was applied to the
wrong base, drifting the rendered perimeter past the plate edge by
roughly the distance to the last hole.
Fix the root cause: ToIncremental and ToAbsolute now walk sub-programs
to compute where they leave the tool, and advance pos accordingly. The
other traversals capture a frameOrigin at entry and compute sub-call
placement as frameOrigin + Offset, letting pos advance naturally
through the sub recursion. All the save/restore workarounds are
removed.
Program.BoundingBox also picks up the same frame-origin treatment,
which corrects a latent bug where absolute-mode endpoints and nested
sub-calls dropped the parent's frame origin.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Don't restore pos after SubProgramCall expansion in DrawRapids — the
machine moves from hole to hole sequentially, so rapids should connect
from the previous hole's end to the next hole's center.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
The rapid from the previous feature to the hole center is implied by
the SubProgramCall offset but wasn't being drawn. Now DrawRapids
renders this traverse before recursing into the sub-program.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
ConvertMode.ToIncremental skips SubProgramCalls when computing deltas,
so all code paths that expand SubProgramCalls must: (1) set curpos to
savedPos + Offset before expanding, and (2) restore curpos afterward
so subsequent incremental codes get correct deltas.
Fixed in ConvertProgram, GraphicsHelper (AddProgram, AddProgramSplit),
PlateRenderer (DrawRapids, DrawProgramPiercePoints, GetFirstPiercePoint),
and CutDirectionArrows.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
SubProgramCalls are now treated as standalone features in the Cincinnati
post-processor. SplitByRapids emits them as single-element features
instead of splitting on rapids within sub-programs. A nest-level hole
sub-program registry deduplicates by content and assigns post numbers.
Sheet writers emit M98 calls with X/Y offsets for hole features, and
hole sub-program definitions are written after part sub-programs.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Inline sub-program geometry into the parent geometry list using Offset
as the starting curpos, replacing the Shape-wrapping approach.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Snaps lead-in angles on ArcCircle contours to a configurable
increment (default 5°), reducing unique hole variations from
infinite to 72 max. Rounding happens upstream in EmitContour
so the PlateView and post output stay in sync.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
PairEvaluator was cloning the full CNC program (including all internal
cutouts) for every candidate. For parts with many holes (e.g. 952),
this caused O(n²) overlap checks and thousands of unnecessary polygon
tessellations per candidate.
Now extracts the perimeter shape once, builds a lightweight drawing
from it, and uses that for all Part.CreateAtOrigin calls. Cutouts are
irrelevant for best fit — only the outer boundary matters for pairing.
75x speedup on a 952-hole rectangle (30s → 0.4s).
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
System.Timers.Timer fires on thread pool threads, causing GraphicsPath
objects to be accessed concurrently by hover detection and OnPaint,
triggering "Object is currently in use elsewhere" in DrawParts.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
External lead-ins now sit on the line between the last internal cutout
and the next part's first pierce point, minimizing rapid travel. Cutout
sequencing starts from the bounding box corner opposite the origin and
iterates 3 times to converge the perimeter lead-in and internal sequence.
LeadInAssigner and PlateProcessor both use a two-pass approach: first
pass collects pierce points, second pass refines with next-part knowledge.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Auto-assign lead-ins silently reused existing plate parameters with no
way to change them after the first assignment. Now a dialog with the
full CuttingPanel is shown every time, pre-populated with the current
settings, so the user can review and modify before confirming.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
FilterPanel.LoadItem was hardcoding all layer and line type checkboxes
to checked, ignoring actual visibility state. Now reads Layer.IsVisible
and entity IsVisible to set correct checked state.
Also guard DetermineWinding against shapes with fewer than 3 polygon
points (defaults to CCW) to prevent crash when applying lead-ins.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
LoadItem was resetting all entity visibility to true, overriding the
suppression state set by LoadDrawings. Now stores suppressed entity IDs
on FileListItem and re-applies after the reset. Also auto-unchecks
layers where all entities are suppressed, and syncs suppression state
back to the FileListItem when filters change.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Entities now have a Guid Id for stable identity across edit sessions.
Drawing stores the full source entity set (SourceEntities) and a set of
suppressed entity IDs (SuppressedEntityIds), replacing the previous
SuppressedProgram approach. Unchecked entities in the converter are
suppressed rather than permanently removed, so they can be re-enabled
when editing drawings again.
Entities are serialized as JSON in the nest file (entities/entities-N)
alongside the existing G-code programs. Backward compatible with older
nest files that lack entity data.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
EditDrawingsInConverter was replacing Drawing objects with new instances,
but Part.BaseDrawing is readonly — parts kept referencing the old drawings
with stale programs (e.g. etch lines that were removed). Now matches by
name and updates existing drawings in-place, then refreshes all parts.
Also fixes Part.Update() which applied rotation backwards and was missing
UpdateBounds() and lead-in state reset.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Extract ArcToLineClosestDistance helper to eliminate duplicate Phase 3
arc-to-line loops, remove redundant MaxValue guard in curve-to-curve
check, consolidate CollectVertices overloads, and add entity-based
PushDirection overload for API consistency.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Corner arcs from offset perimeters could slip past vertex sampling,
causing compactor push to undershoot by ~halfSpacing. Use ClosestPointTo
to find the actual nearest point on each arc to each line before firing
the directional ray.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Adds a toolbar button on the Drawings tab that opens the CAD converter
pre-populated with the current nest drawings, allowing users to revisit
layer filtering, quantities, and other settings without re-importing.
Also fixes PlateView stealing focus from text inputs on mouse enter
and FilterPanel crashing when loaded before form handle is created.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Add a Shape Library dialog (Nest > Shape Library) for creating drawings
from built-in parametric shapes. Supports configuration presets loaded
from JSON files — ships with 136 standard pipe flanges. Parameters use
TextBox inputs with architectural unit parsing (feet/inches, fractions).
- ShapeLibraryForm with split layout: shape list, preview, parameters
- ShapePreviewControl for auto-zoom rendering with info overlay
- ArchUnits utility for parsing architectural measurements
- SetPreviewDefaults() on all ShapeDefinition subclasses
- Convention-based config discovery (Configurations/{ShapeName}.json)
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Add Phase 3 curve-to-curve direct distance in CpuDistanceComputer to
catch contacts that vertex sampling misses between curved entities.
Enforce minimum copy distance in FillLinear to prevent bounding box
overlap when circumscribed polygon boundaries overshoot true arcs.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Pull duplicated vertex collection, edge conversion, sorting, and
ray-circle solving into reusable private methods. Delegate the
no-offset DirectionalDistance overload to the offset version.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
The vertex-to-entity approach in DirectionalDistance only sampled 4
cardinal points per circle, missing the true closest contact when
circles are offset diagonally from the push direction. This caused
the distance to be overestimated, pushing circles too far and
creating overlap that worsened with distance from center.
Add a curve-to-curve pass that computes exact contact distance by
treating the problem as a ray from one center to an expanded circle
(radius = r1 + r2) at the other center. Includes arc angular range
validation for arc-to-arc and arc-to-circle cases.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Circle.Rotation was lost in three places, causing reversed circles to
still offset inward instead of outward:
- ConvertGeometry.AddCircle hardcoded CCW instead of using circle.Rotation
- ConvertProgram.AddArcMove created Circle without setting Rotation from arc
- Shape.OffsetOutward/OffsetInward copied Circle without setting Rotation
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Raise ViewScaleMax from 3000 to 10000 for deeper zoom. Catch
InvalidOperationException in hoverTimer_Elapsed when GraphicsPath is
concurrently used by the paint thread.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
MicrotabLeadOut was an unimplemented stub (Generate returned empty list)
that duplicated tab functionality. Existing saved configs with "Microtab"
selected will gracefully fall back to NoLeadOut.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
- Iterate all remnants instead of only the first when packing and filling
- Improve ScoreZone with estimated part count and aspect ratio matching
- Cache bounding boxes in SortItems and remnants in TryPlaceOnExistingPlates
- Make TryConsolidateTailPlates loop until stable, trying all donor/target pairs
- Fix consolidation grouping to use BaseDrawing reference instead of name
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Move the expensive per-part hit-test out of OnMouseMove and into
the hoverTimer callback. The hit-test now only runs once after
1000ms of stillness, not on every mouse move event.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Add a hoverTimer that restarts on each mouse move over a part.
Tooltip only renders after the timer fires, hiding while the
cursor is in motion.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Update hoverPoint on every mouse move while over a part, not just
when the hovered part changes.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Path.IsVisible was consuming 52% of CPU on mouse move. Add a cheap
GetBounds().Contains() check first so only parts under the cursor
hit the expensive GDI+ path test.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Plate setter is called in the constructor before actionManager is
initialized, causing a NullReferenceException on startup.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
