Edges shorter than the model tolerance are dropped from the flat cells, and the refold sews touching corners, so a cut-out across a bend line no longer fails with a degenerate edge.
* Refuse fillets and chamfers that run past their faces or cross each other (#1080)
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* Blend fit: chamfers keep their old acceptance of crossed loops; skip the volume bound along tangent chains
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* Blend fit: one selection per check, a coarse volume pass first; record the refused oversize blends in the action matrix baseline
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
A tapered extrude now offsets spline and conic curves as well as lines and arcs, so sketch text and any other profile can take a taper, and the join of tapered letters onto a face finds its boolean result.
Fillet gets a Curvature (G2) radius type (type: "curvature", alias "g2"): a quintic blend whose size is the equivalent circular setback, for convex and concave straight edges between planar faces with a perpendicular planar end face at each end. Unsupported cases are refused rather than falling back to a circular fillet, and a blend that would cross a loop in its end face is refused. Persisted as a fillet style under format 19; labels in en/de/ja/ko/zh.
Co-authored-by: midasdf <midasdf@users.noreply.github.com>
* Fillet: refuse a result with openings in its surface (#871)
A fillet on #871's part (radius 1 on a window's edges and on the sides of a
block 1.26 mm away) came back closed as topology with every face meshing, but
the rounds overlapped and the end faces' boundary loops crossed, so the model
showed holes. The constant-radius fillet now checks its result before taking it:
the body's mesh, welded within a millionth of its size, must have no edge
without as many triangles running the other way along it, except for cracks
within the mesh tolerance where fitted curves meet. A result that fails is
refused, so the corner build gets its try and the radius search treats it as
too large.
Body::open_mesh_edges is the check. Kernel tests cover the #871 part (refused
at radius 1, closed with a round on every edge at 0.5), and an engine test runs
the fillet from the report and leaves the part as it was. All 96 robustness
seeds build the same with and without it.
Fixes#871
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* Check fillets for crossed face boundaries, not free mesh edges
Free edges in the mesh also come from cracks where fitted curves meet (corner
patches, rounds ending on curved edges), so the mesh test refused good fillets.
The check is now on the cause of #871: a planar face whose boundary loops cross
each other or themselves has no inside. Body::crossed_face_loops samples each
planar face's loops within 1/1000 of the size and counts proper crossings
(touching within a millionth doesn't count).
The pyramid apex blend and the loft fillets build again. Ten action-matrix fillets
stay refused, recorded in the baseline: r = 3 on the 2 mm rim of oracle 09's
shelled box (the round runs past the inner wall; the result was 240 mm3 lighter
than any round can make it), and the six seam-edge fillets on oracle 60's bolt
holes, whose boundary jumps across the hole.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
Rolling-ball blends now handle knife edges, where the two faces have opposite normals; an edge that runs out along an end face tangent to one side; and rounds that end in an earlier round on curved faces, picked together or one after another. The three edges at a vertex of smooth faces, rounded to one radius, end in a piece of sphere. When the given order of up to four edges is refused, or leaves a face boundary off its surface, other orders are tried.
Closes#851.
Co-authored-by: lasselian <22126420+lasselian@users.noreply.github.com>
Offsets picked faces along their normals into a new surface body: planes and cylinders move exactly, other faces get a fitted B-spline within tolerance, and offsets that would fold over are refused. Adds the SurfaceOffset feature (noted under format 18), the surface.offset engine command and an Offset Surface dialog.
Closes#758.
Co-authored-by: jyviko <2501648+jyviko@users.noreply.github.com>
Surface Loft lofts open or closed curve chains, one section per sketch in pick order, into surface bodies. Adds the SurfaceLoft feature (format 18 note), the surface.loft command, a dialog with Edit Feature support, and kernel and engine tests.
Part of #757.
Surface Sweep carries open curves and closed profile boundaries along a sketch path into surface bodies, with smooth walls along curved point paths. Adds the SurfaceSweep feature (format 18 note), the surface.sweep command, a dialog with Edit Feature support, and kernel and engine tests.
Part of #757.
Walls swept along a spline have flat B-spline tops and bottoms, so booleans between them went to the general intersection, which fails on coincident spline faces. Such faces are now put on true planes, and two prisms of the same height and direction are decided in 2D with their spline outlines fitted by arcs within a micron. The result is kept only when its volume is plausible; otherwise the general boolean takes over. The free-form flag is cleared by a drop guard so an unwinding boolean can't leave it set.
Fixes#844.
Surface Revolve turns open curves and closed profile boundaries about a sketch axis into surface bodies, with cone apexes and sphere poles where a chain ends on the axis, and a closed shell made solid. Adds the SurfaceRevolve feature (format 18 note), the surface.revolve command, a dialog with Edit Feature support and analytic kernel and engine tests.
Part of #757.
Delete Face takes heal (default true, as before). With heal: false the picked faces are removed and the gap stays open, leaving the remaining faces as a surface body (kernel remove_faces); deleting every face is refused. Saved only when false, noted under format 18.
Display meshes after opening, the face names read when a feature finds a face by name, and inspect.measure over all bodies are now worked out per body on plain scoped threads (solidcraft_kernel::for_each_on_threads / map_on_threads), in body order and with the same results; the browser build stays serial.
When an extrude cuts with profiles whose outlines stay more than 1 µm apart (the letters of a text), their tools are apart too, so they are joined and cut from the body in one boolean instead of one per profile. A ten-letter text cut drops from about 13 s to 1.5 s with the same result. Sheet metal targets and touching profiles keep cutting tool by tool. Tapered extrudes are unaffected.
A coincident face is pushed apart only when the space it moves through holds nothing that changes the result; otherwise the boolean falls through to the exact strategies.
Co-authored-by: NOWN-DEV <257771410+NOWN-DEV@users.noreply.github.com>
surface.extrude sweeps open curve chains and profile boundaries along the sketch normal into a surface body (walls only, no caps). Saved under format 17.
Refs #757
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Adds `fit: taper` to Fillet: where a constant radius doesn't fit along a straight edge, it follows the room there as a smooth spline, down to zero where a face closes to a point. Taper to Fit is in the Fillet dialog, and the feature's warning says how far it tapered.
Sheet-metal cut-outs and base flange outlines drawn with ellipses, splines or text stay exact curves in the flat pattern and the unfolded body; the folded body follows them with tangent arcs within 0.001 mm (#659).
A cut built slab by slab (the exact prism boolean) sewed the result from horizontal
slabs and only joins merged the coplanar pieces back. So a floor vent cut through the
Raspberry Pi case split the long wall into strips at the ports' tops and bottoms
(87 faces instead of 79 at 32 mm), and at 24 mm the end opening split the end wall.
- Prism booleans merge coplanar neighbours for every operation (faces only: edges,
curves and surfaces stay exact), keeping each face's construction source where all
its pieces share it, Split Face lines and regularization protection. Painted or
protected operands keep their authored pieces as before.
- The prism history now tells apart a wall's several edges with its caps (a wall
with openings meets each opening's floor), so cuts beside existing openings keep
exact history and faces keep their names.
- Tests: the Pi case's faces pinned after every feature at 24, at 24 changed to 32,
and built at 32 (each outer side one face); random openings and bosses on a
shelled case never split another side; every corpus design rebuilds a parameter
edit like a fresh build and reverts exactly.
- Action matrix baseline: demo/pi-case re-blessed (fewer, merged faces; no
regressions), demo/pi-case-0.8 recorded, and 88 new passes on
regressions/lid_ribs_flush_with_window.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Two L's extruded up and down from one 2 x 2 grid now join into one solid (the planar rebuild gives each fan of faces around a pinched vertex its own vertex copy), and a pad whose end lies half past the edge of the face it stands on fuses exactly: contact faces are imprinted with each other's outline and the common faces glued away, still held to the retry guard. Face splitting handles a line crossing an edge twice; glue matches differently parametrised curves and keeps several pieces apart. Robustness seed 16 now fuses with the exact volume.
Fixes#679, fixes#674.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Joining ribs flush with a lid's window left a zero-thickness slit far from the rib (the planar fallback glued back only one of the faces its pieces shared) and displaced the tool on the other joins. Gluing now joins every shared face pair, coincident-face separation checks there is material behind a pushed face for every join, and any join or cut that needed a retry must be closed, keep volume monotonic and leave the part unchanged outside the tool's box, or it is refused with an error. Feature command replies carry the feature's warning.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Fillet fit: reduce lowers the radius only where it doesn't fit: picked edges are grouped where they can limit each other (sharing a face or vertex), and each group gets the requested radius where it fits, otherwise the largest that does. Refs #293.
Co-authored-by: Iraklis Kourtis <jyviko@gmail.com>
Refs #303. file.save / file.save_as can store a warm native geometry cache (cache: true, off by default) keyed to the document hash, app version and adapter revision. Opening shows a bounded preview from it until the rebuilt session replaces it. Any cache failure counts as a miss.
Display meshes also bound the angle one facet turns, so bends and small holes on a large body are drawn round instead of as chamfers or squares. Bounded per face, and only the display mesh changes.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
A SolidCraft-owned portable B-rep schema (geometry schema 2 with analytic curves and surfaces, explicit frames and ranges, fitted intersection curves) and a versioned body envelope (SCBREP03) with its own metadata records, a bounded binary codec, a validator with resource budgets, and an adapter that refuses what it can't convert exactly. Pinned by a box-and-fillet golden fixture. Not yet written by application saves. Refs #303.
* Add a one-surface fill of closed 3D loops to geom
A closed loop of curves that does not lie in a plane has no exact surface;
the usual answer is one smooth face fitted over it. fill_loop does that:
- the loop's best-fit plane (Newell normal), with a frame chosen from the
loop itself so the result does not depend on where the loop starts, its
direction, or where the part sits
- the loop seen along that normal must be simple and must not fold or run
along the normal; otherwise it is refused ("not supported yet"), never
filled with wrong geometry
- one non-rational bicubic B-spline height field over the loop's box,
fitted by least squares to the loop's samples (The NURBS Book 9.4.1,
9.4.3, eq. 9.62/9.65) with a thin-plate smoothness term
lambda*A*int(h_uu^2 + 2 h_uv^2 + h_vv^2), A one grid cell's area, so the
fit is the same at any scale; only planes are reproduced exactly
- span counts per direction (long narrow loops fit), refined until the
loop is within tolerance, else refused with the deviation reached
bspline: bisected span search, basis functions on a stack array, one
clamping helper shared by curves and surfaces, one binomial helper.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01J53sR6gEaVt56soFDpTck8
* Patch fills non-planar loops of edges with one smooth face (#250)
surface.patch with edges refused any loop that is not planar ("not
supported yet: patching a loop that is not planar"). It now fills it:
- planar loops still get one exact planar face, but planarity is certified
conservatively: lines by their ends, B-spline/NURBS edges by their control
points, intersection curves by their samples lying on a support plane
within tolerance; a loop that can't be certified is filled instead
- other loops get one face from geom's fill_loop, bounded by the body's own
chained edges, whole (no split): its history is one generated face whose
edges are the body's, and it stitches back into a closed solid
- intersection-curve edges are sampled by a seeded, guarded double
projection shared by the planarity test and the fill (plane_fit)
- the face tolerance is 10x the planarity tolerance (documented: later
operations take the face to contain its edges within it)
- edge picking runs inside the guard; the geom->truck conversion is shared
with the T-spline sheets (spline_conv); the Newell plane fit is shared with
STEP import
Tests: 3- to 8-sided, concave (dart, arrow, plus), flat-corner, two-edge and
imported rims fill as one face and stitch closed (volumes against an
independent integral); finite mesh normals, same result for any pick order
or placement; non-height-field loops and too-fine detail are refused.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01J53sR6gEaVt56soFDpTck8
* Test the Patch dialog filling a non-planar loop of picked edges
The Patch dialog's only test used a sketch profile. Picking the four
curved rim edges of a box cut by a sideways cylinder now makes one
surface.patch command whose result is a single smooth face over the rim.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01J53sR6gEaVt56soFDpTck8
---------
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
Every panic (in a command, in the kernel, anywhere else) and every unclean
exit (abort, stack overflow, kill) leaves a report in a local crash folder:
message, file:line, a trimmed backtrace, version, commit, system and renderer,
the command and recent commands, and when possible the design and replay.json
that reproduce it (solvecraft-cli crashes replay <id>).
- Cost: nothing until the app installs the hook (tests, CLI and web never
do); then one small journal line per top-level command of the design on
screen, taken back with a truncate when the command changed nothing.
- Unclean exits: each run holds a lock on its journal; a journal nobody holds
at the next start belongs to an app that died. Autosaves restart the
journal from the autosaved design.
- Bounded: a crash's signature (place + message, numbers ignored) is counted,
not rewritten; at most 5 new reports per run, 20 reports, 50 MB, 90 days;
recovered kernel panics are pruned first.
- Agents: diagnostics.crash_reports / crash_report / crash_reports.clear
commands, the MCP crash_reports tool, solvecraft-cli crashes, solvecraft
--crash-log; diagnostics.crash_test (with SOLVECRAFT_CRASH_TEST) checks it.
- People: the Crash Reports window opens after a crash (Help > Crash
Reports…) with a reviewed issue draft per report; nothing is sent.
- Release builds keep line tables for file:line in backtraces.
Tests: child processes that really panic in a command, in the kernel, on a
thread, and abort; dedup counts, limits, hostile folders, journal parsing,
the UI window and draft, the MCP tool. docs/crash-reports.md.
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
Co-authored-by: bherbruck <brennen.herbruck@gmail.com>
* feat(kernel): typed TooLarge refusal and fillet_max search
Blend refusals caused only by the size (larger than the neighbouring faces,
taller than the walls, longer than the next edge, larger than a corner
radius) now return KernelError::TooLarge, displayed as before. fillet_max
finds the largest radius below a request at which the fillet of the edges
works, to three significant digits, by building candidates; it is an explicit
call (up to 16 fillets), not run on every refusal.
A corner edge that doesn't run square to the edge is now reported as such
instead of as a blend larger than the neighbouring faces.
Refs #293
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01TWc1smHYwTSVrVc8SNvS44
* feat: fillet fit option reduces a radius too large for its edges
solid.fillet takes fit: exact (the default, as before: a radius too large
for the geometry around the edges is refused) or reduce: the radius is
reduced to the largest that works for all the picked edges (kernel
fillet_max, three significant digits, every candidate built). The feature
stores the requested radius, so it refits on every evaluation; when it
reduces, the feature's warning says so and the command returns
radius_applied. Pattern copies of the fillet use the same rule. The file
format writes fit only when it is reduce.
The Fillet dialog gets a Radius Fit row (Exact / Reduce to Fit, constant
radius) and, when the preview reduced the radius, an Applied Radius line;
editing the feature keeps the choice.
One radius for all edges for now; per-edge radii and taper follow.
Refs #293
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01TWc1smHYwTSVrVc8SNvS44
---------
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
Co-authored-by: bherbruck <brennen.herbruck@gmail.com>
* Add B-spline basis, knot insertion and NURBS surfaces to geom
Freeform faces need local refinement and gap-free joins between patches
with different knot vectors, which tensor B-spline grids cannot give;
T-splines can, and the first user is an N-sided Boundary Fill of
non-planar loops. This lays the univariate groundwork they rest on, in
the L0 geometry crate so it stays independent of the B-rep kernel:
- Cox-de Boor basis of one function from its local knot vector, on
half-open spans with a left-limit side for the domain end, with
derivatives (NURBS Book eq. 2.5, 2.9)
- knot insertion into a single basis function (Sederberg et al. 2004
section 3) and Bezier extraction of one function by repeated insertion
- Boehm knot insertion for curves (NURBS Book A5.1), de Casteljau
evaluation and sub-segments
- a rational tensor NURBS surface with evaluation, knot insertion and
clamping, and a rational Bezier patch type
All input is checked (finite, monotone knots, positive weights, size
caps) and returns SplineError; nothing indexes unchecked.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01J53sR6gEaVt56soFDpTck8
* Add T-spline surfaces: T-mesh, evaluation, AS check, refinement, extraction
Tensor B-spline faces refine only by whole rows and columns and cannot
join patches with different knot vectors without gaps. T-splines allow
T-junctions in the control net, which fixes both; the first user is an
N-sided Boundary Fill. This adds the maths only (no command, no change
in behaviour):
- T-mesh in index space with knot values per line, validated (outer
rectangle, rectangular faces, no empty lines, size caps); local knot
vectors read off rays from each vertex, rays leaving the mesh repeat
the boundary (Sederberg et al. 2003 section 3)
- rational T-spline evaluation and first derivatives for odd bi-degree,
with a complete (p-1)/2-line frame and the domain inside it
- T-junction extensions and the analysis-suitability check (Scott et al.
2012 section 2)
- local refinement that keeps the surface exactly: requested knot lines
are cut in, then every original blending function is carried as a
record and split (violation 1) or the mesh gains edges (violations 2
and 3) until each record is a blending function of the refined mesh
(Sederberg et al. 2004 section 3); analysis-suitable input stays
analysis-suitable by carrying conflicting T-junctions across faces
- NURBS <-> T-spline conversion for meshes without T-junctions, and
Bezier extraction over the rectangles no blending-function knot line
crosses
Measured: regular T-spline vs NURBS < 1e-12, round trip exact, partition
of unity with T-junctions < 1e-14, refinement deviation <= 1.4e-14 over
40 random refinements, Bezier patches vs surface < 1e-14.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01J53sR6gEaVt56soFDpTck8
* Convert T-splines to sewn NURBS sheet bodies in the kernel
A T-spline is only useful to the modeller once it is a B-rep face set.
tspline_sheet builds an open shell: a mesh without T-junctions becomes
one clamped NURBS face; otherwise each Bezier element is a face, and
sides are split at the T-points where neighbours' corners land, so all
faces share vertices and edges and the sheet is one oriented shell
(V - E + F = 1). Every truck call runs under guard. No command yet;
Boundary Fill will be the first caller.
Tests: one face for a regular surface, sewn Bezier faces with
T-junctions, tessellated points on the surface to ~1e-9 mm, and hostile
control nets (coincident, 1e300, needle-thin) never panic.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01J53sR6gEaVt56soFDpTck8
---------
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
Fillets (local, loop, rolling ball, corner, variable, chord), chamfers
(cutters, local corners, loops, two sides), shell, draft, offset faces,
split face/body, delete face, patch/copy/stitch/trim/extend and
transforms record their history over [body] while they build: faces
rebuilt from the body's, blends in place of edges (corner patches in
place of the edges at the corner), shell walls as offsets, edges as
pieces of the body's or where two faces meet. Multi-step operations
compose each step. The push-face offset fallback now has history, and
the bore union composes its narrowed tool.
A reflection reorders a body's edges, so transforms now match edges by
identity (mirrored copies reported the wrong edges).
Refs #136
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01J53sR6gEaVt56soFDpTck8
ops::boolean's result carries its history over [a, b] whichever path made
it, recorded by the code that builds the faces:
- truck path (incl. jitter retries): shapeops history mapped through the
order-preserving copies and the cut's turned tool, then through healing
- heal records which input faces each healed face merges and which input
edges each joined or exact edge replaces (HealMap); opt-in, so heal_keep
callers pay nothing
- trivial shortcuts, the void, glue, faces pushed apart (push_faces records
its own history; the slab cut back on a union maps to the operands'
faces in that plane), the nudged retry, by_lumps and join composition
- prism booleans: section segments carry tags (operand cap edge, the wall
beside it, its vertices) through the 2D boolean; walls, caps and edges of
the extruded or sewn result follow from them
- planar and faceted polygon booleans: polygons carry their operand face;
faces join the pieces across edges, edges come from the pieces either side
History gains renumber/over (and consuming forms) for composition;
record gains regions/traced/build helpers. truck-shapeops
*_with_history now return the solid with an Option history. Paths that
cannot report (bore narrowing via offset, inexpressible lineage) leave the
result without history. Boolean results, tolerances and fallback order are
unchanged.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01J53sR6gEaVt56soFDpTck8
One numbered loop type (crate::numbered) carries each segment's profile
numbers through orientation, snapping, circle re-seaming, re-sampling and
rotation; every builder orients its regions through it once. Builders that
make their own side faces pair each face with its provenance as it is made
(ruled walls by the wire edges each face holds, smooth lofts, coils and path
sweeps by segment), and cap_and_close appends the caps itself. Column
following is kept only for faces truck sweeps internally (rsweep, tsweep
along paths, cones, where an apex vertex stays put). Extrude derives face
sources and history in one topology pass. One attach entry point takes
operands with their edges already indexed.
Geometry, face order, face sources and recorded origins are unchanged.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01J53sR6gEaVt56soFDpTck8
Extrude, revolve (partial and full turn), tapered extrude, loft (incl. to a
point), sweep, path sweep, coil, primitives, lip, emboss and thicken attach
the exact history of the bodies they build: each face's profile segment, cap
or operand face, and each edge's operand edge, recorded by topology identity
inside the building call. Nothing is sampled; a body whose faces do not follow
the expected construction carries no history. Face sources are unchanged.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01J53sR6gEaVt56soFDpTck8
Bodies can now carry the exact history of the operation that made them:
for each result face its origins among the operation's operands (kept,
modified, split, merged, or generated from a profile segment, cap, edge or
offset) and for each edge the faces it separates and the operand edge or
face pair it came from. History is recorded by the operation, never
sampled, and checked against the body when attached.
- history.rs: the vocabulary (moved from provenance.rs), History::check,
identity, substitute (composes an operation with the one that made an
operand, including bodies made from a sketch) and face-source propagation.
- One edge order for a body (body::edge_order), shared by tessellation and
history, so history edge indices are the mesh's edge indices.
- Transforms report an identity history; placed_topology documents the
face/edge order guarantee it relies on.
- provenance.rs keeps the geometric trace for operations that do not yet
report history.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01J53sR6gEaVt56soFDpTck8
* test: reproduce #261 (Rectangular pattern of a hole cut scales super-linearly)
* Patterns of many hole cuts scale near-linearly (#261)
Two quadratic costs:
- Meshing a face tested every point against every segment of every boundary
loop, so a plate face with hundreds of holes meshed in quadratic time. The
vendored truck-meshalgo inside test now skips loops whose parameter box is
clear of the point (they wind zero times around it).
- A pattern cut each copy in turn, rebuilding and re-checking an ever more
complex target per copy. Copies whose boxes are clear of each other are now
one tool of many pieces (kernel::union_apart, no intersection work), cut once.
512 holes now take ~7 s instead of not finishing. The test's volume tolerance is
widened from 1e-6 relative to 0.1 % of each hole: the measure meshes the bores
and reads each hole ~0.06 % small, on main too.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* Hole-pattern test: a budget CI can meet; coarser apart-box meshes (#261)
CI runs every test binary at once and is about ten times slower than a
machine running this test alone (15 tests passed 60 s in the last run), so the
512-hole pattern (about 6 s alone) hit the 60 s timeout there. The budget now
only has to tell linear from the quadratic hang: 300 s timeout, 120 s budget.
union_apart meshes each copy coarsely (2 % of its size, box grown by twice that).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* Hole-pattern test checks the scaling, not a wall-clock time (#261)
CI runs every test binary at once: the 512-hole pattern (about 6 s alone, 10 s
CPU single-threaded) took 177 s there, so no fixed time is reliable. The test
now patterns 128 and then 512 holes and requires the second to take under 8x
the first (linear is 4x, measured 3.5-4.2x here; the old quadratic cost was 16x
or more), each with the volume check and a 900 s hang guard.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
---------
Co-authored-by: bherbruck <brennen.herbruck@gmail.com>
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
Irakli Kourtis