Edge division: evaluate each point once (#1560)

Thanks @rlartot.

crates/kernel/src/edge_polyline_tests.rs

@@ -0,0 +1,159 @@
+//! The vendored edge division evaluates each parameter once (the ends of a piece are carried
+//! along, its midpoint is the probe and an end of both halves). Regressions against the division
+//! as it was, which evaluated both ends of every piece again: the polylines must be the same, bit
+//! for bit, including the fallbacks for samples that do not evaluate.
+#![allow(clippy::unwrap_used, clippy::panic)]
+use std::sync::Arc;
+use std::sync::atomic::{AtomicUsize, Ordering};
+
+use solidcraft_geom::Vec3;
+use truck_meshalgo::prelude::PolylineCurve;
+use truck_meshalgo::tessellation::{Deflection, TrySubs, bounded_polyline};
+use truck_modeling::{BoundedCurve, Curve, EuclideanSpace, MetricSpace, ParameterDivision1D, ParameterRange, ParametricCurve, Point3, Vector3};
+
+use crate::*;
+
+/// A curve that counts its evaluations.
+#[derive(Clone)]
+struct Counted {
+    curve: Curve,
+    calls: Arc<AtomicUsize>,
+}
+
+impl Counted {
+    fn new(curve: &Curve) -> Self {
+        Counted { curve: curve.clone(), calls: Arc::default() }
+    }
+
+    fn calls(&self) -> usize {
+        self.calls.load(Ordering::Relaxed)
+    }
+}
+
+impl ParametricCurve for Counted {
+    type Point = Point3;
+    type Vector = Vector3;
+    fn subs(&self, t: f64) -> Point3 {
+        self.calls.fetch_add(1, Ordering::Relaxed);
+        self.curve.subs(t)
+    }
+    fn der(&self, t: f64) -> Vector3 {
+        self.curve.der(t)
+    }
+    fn der2(&self, t: f64) -> Vector3 {
+        self.curve.der2(t)
+    }
+    fn parameter_range(&self) -> ParameterRange {
+        self.curve.parameter_range()
+    }
+    fn period(&self) -> Option<f64> {
+        self.curve.period()
+    }
+}
+
+impl BoundedCurve for Counted {}
+
+/// The division samples through `TrySubs`: count those evaluations.
+impl TrySubs for Counted {
+    fn try_subs(&self, t: f64) -> Option<Point3> {
+        self.calls.fetch_add(1, Ordering::Relaxed);
+        TrySubs::try_subs(&self.curve, t)
+    }
+    fn approx_subs(&self, t: f64) -> Point3 {
+        self.calls.fetch_add(1, Ordering::Relaxed);
+        TrySubs::approx_subs(&self.curve, t)
+    }
+}
+
+impl ParameterDivision1D for Counted {
+    type Point = Point3;
+    fn parameter_division(&self, range: (f64, f64), tol: f64) -> (Vec<f64>, Vec<Point3>) {
+        self.curve.parameter_division(range, tol)
+    }
+}
+
+/// `bounded_polyline` before it carried the end points: 4 evaluations per piece (both ends, the
+/// probe and the midpoint), with the same fallbacks for samples that do not evaluate.
+fn reference(curve: &Counted, range: (f64, f64), d: Deflection) -> PolylineCurve<Point3> {
+    const MAX_EDGE_POINTS: usize = 20_000;
+    let (tol, turn2) = (d.chord, d.turn2());
+    let end = |t: f64| curve.try_subs(t).unwrap_or_else(|| curve.approx_subs(t));
+    let probe = |t: f64| curve.try_subs(t).unwrap_or(Point3::new(f64::NAN, f64::NAN, f64::NAN));
+    let mut out = vec![end(range.0)];
+    let mut stack: Vec<(f64, f64, usize)> = vec![(range.0, range.1, 0)];
+    while let Some((a, b, depth)) = stack.pop() {
+        let (pa, pb) = (end(a), end(b));
+        let t = a + (b - a) * 0.4729;
+        let chord = pa + (pb - pa) * 0.4729;
+        let d2 = probe(t).distance2(chord);
+        let tm = 0.5 * (a + b);
+        let dm = probe(tm).distance2(pa.midpoint(pb));
+        let flat = !(d2 >= tol * tol || dm >= tol * tol);
+        let flat = flat && !d.turns(turn2, d2.max(dm), pa.distance2(pb));
+        if flat || depth >= 24 || out.len() + stack.len() >= MAX_EDGE_POINTS {
+            out.push(pb);
+        } else {
+            stack.push((tm, b, depth + 1));
+            stack.push((a, tm, depth + 1));
+        }
+    }
+    PolylineCurve(out)
+}
+
+/// Divides `curve` both ways and checks the points, their order and the evaluations; returns
+/// the polyline.
+fn same_division(curve: &Curve, d: Deflection) -> Vec<Point3> {
+    let range = curve.range_tuple();
+    let (before, after) = (Counted::new(curve), Counted::new(curve));
+    let old = reference(&before, range, d).0;
+    let new = bounded_polyline(&after, range, d).0;
+    let bits = |p: &Point3| [p.x.to_bits(), p.y.to_bits(), p.z.to_bits()];
+    assert_eq!(old.len(), new.len(), "{d:?}");
+    for (i, (p, q)) in old.iter().zip(&new).enumerate() {
+        assert_eq!(bits(p), bits(q), "point {i} of {}: {p:?} before, {q:?} now ({d:?})", new.len());
+    }
+    // Every piece is kept (one point) or halved: n points come from 2n - 3 pieces.
+    let pieces = 2 * new.len() - 3;
+    assert_eq!(before.calls(), 1 + 4 * pieces);
+    assert_eq!(after.calls(), 2 + 2 * pieces);
+    new
+}
+
+/// The chord tolerance alone (booleans), then with the display's facet angle.
+fn deflections(tol: f64) -> [Deflection; 2] {
+    [Deflection::chord(tol), Deflection { chord: tol, angle: DISPLAY_ANGLE, floor: tol * 0.05, rulings: true }]
+}
+
+#[test]
+fn intersection_curves_of_a_drilled_cylinder_divide_as_before() {
+    // The radial hole of a clamping hub: both cylinders are NURBS, so each point of the edges
+    // is two grid searches.
+    let hub = cylinder(Vec3::new(0.0, 0.0, -10.0), Vec3::Z, 8.0, 20.0).unwrap();
+    let drill = cylinder(Vec3::new(-12.0, 0.0, 1.5), Vec3::X, 1.25, 24.0).unwrap();
+    let drilled = boolean(&hub, &drill, BoolOp::Cut).unwrap().unwrap();
+    let curves: Vec<Curve> = drilled.solid.edge_iter().map(|e| e.curve()).filter(|c| matches!(c, Curve::IntersectionCurve(_))).collect();
+    assert!(curves.len() >= 2, "{} intersection curves", curves.len());
+    // One curve with each deflection (each division takes about half a second here).
+    for (curve, d) in curves.iter().zip(deflections(0.01)) {
+        let points = same_division(curve, d);
+        assert!(points.len() > 8, "{} points ({d:?})", points.len());
+        // On the drill's wall.
+        for p in &points {
+            let r = (p.y * p.y + (p.z - 1.5) * (p.z - 1.5)).sqrt();
+            assert!((r - 1.25).abs() < 1e-3, "{p:?} at {r} from the drill's axis");
+        }
+    }
+}
+
+#[test]
+fn circle_divides_as_before() {
+    let body = cylinder(Vec3::ZERO, Vec3::Z, 5.0, 10.0).unwrap();
+    let circle = body.solid.edge_iter().map(|e| e.curve()).find(|c| !matches!(c, Curve::Line(_))).unwrap();
+    for d in deflections(0.001) {
+        let points = same_division(&circle, d);
+        assert!(points.len() > 18, "{} points ({d:?})", points.len());
+        for p in &points {
+            assert!(((p.x * p.x + p.y * p.y).sqrt() - 5.0).abs() < 1e-9, "{p:?}");
+        }
+    }
+}

crates/kernel/src/lib.rs

@@ -30,6 +30,8 @@ mod clearance;
 mod coplanar;
 mod curveblend;
 mod delete_face;
+#[cfg(test)]
+mod edge_polyline_tests;
 mod emboss;
 mod face_params;
 #[cfg(test)]

docs/perf.md

@@ -1,5 +1,54 @@
 # Design loading
 
+## Edges made by booleans: each point evaluated once (2026-10-16)
+
+A boolean meshes both operands with truck-meshalgo, which first divides every edge into a
+polyline (`bounded_polyline`, vendored). An edge that an earlier boolean made is an
+intersection curve. Each of its points is found by a grid search on each of the two surfaces,
+51 × 51 evaluations of a NURBS surface each, before Newton. The division evaluated both ends of
+every piece again, so 2 of its 4 evaluations per piece repeated a point it already had.
+Profiles of `main` (01b1b705) on the machine below put the division at 42 % of the CPU time
+of opening the rover, 58 % of `bench-agent` and 82-85 % of a drilled hub or a union of tubes.
+
+Each piece now carries the points of its ends, and its midpoint is both the probe and an end of
+both halves: 2 evaluations per piece. The fallbacks for samples that do not evaluate (#1364) are
+kept: a midpoint that fails is NaN as a probe and the curve's approximation as an end, exactly as
+before. The polyline is the same, bit for bit: `edge_polyline_tests` compares it with the
+previous division (points, order and number of evaluations) on the intersection curves of a
+drilled cylinder and on a circle.
+
+Release `solidcraft-cli` from `main` (73493b94) and the same with the change, 3 alternating
+pairs (before/after, after/before, before/after), each run a new process timed with
+`/usr/bin/time -v`, started once the 1-minute load average was below 2.5. Medians, with the
+range of the 3 runs. 2018 laptop: Intel Core i7-8565U (4 cores / 8 threads, 15 W), 7.4 GB,
+Ubuntu (Linux 7.0), on mains power:
+
+| Run | Evaluation | Meshes | Wall time | CPU time | Peak RSS |
+|---|---:|---:|---:|---:|---:|
+| `bench-open`, `examples/acceptance/sensor_enclosure.json` | 3,203 → 3,089 ms | 6 → 6 ms | 3.2 → 3.1 s | 8.7 → 8.5 s | 91 → 91 MB |
+| `bench-open`, a hub barrel with one radial hole cut through it | 966 → 817 ms | 878 → 481 ms | 1.9 → 1.4 s (-24 %) | 4.4 → 2.8 s (-37 %) | 27 → 30 MB |
+| `bench-open`, the frame of the `bench-agent` bicycle (one union of tilted tubes) | 8,204 → 6,277 ms | 1,711 → 893 ms | 9.9 → 7.2 s (-28 %) | 33.3 → 20.6 s (-38 %) | 44 → 46 MB |
+| `bench-open`, a rover (3,324 features, 1,395 bodies) | 26,783 → 22,792 ms | 7,960 → 6,046 ms | 35.7 → 29.8 s (-17 %) | 129.0 → 101.9 s (-21 %) | 2,076 → 2,079 MB |
+| `bench-agent` (build 33.5 → 30.2 s, -10 %) | 10,553 → 8,565 ms | 2,778 → 2,081 ms | 47.6 → 41.1 s (-14 %) | 124.0 → 91.0 s (-27 %) | 1,428 → 1,432 MB |
+
+`bench-agent`'s evaluation and meshes are those of reopening the saved design. CPU time is the
+steadier measure on this 15 W processor: its before and after ranges are apart for the four
+cases with boolean edges (the hub 3.0-4.4 s before, 2.5-2.9 s after; the rover 113.6-133.5 s,
+101.0-102.5 s), while the first pair's "before" run, which started on a cool processor, overlaps
+the "after" wall times of the two short cases. The sensor enclosure does not change: its time goes
+to grid searches in face meshing, which this does not touch. An earlier series on `main`
+(8535d229), on a quieter machine, gave -38 %, -35 %, -18 % and -16 % of wall time for the same
+four cases. The CPU time saved is close to half the division's share in the profile. The mesh
+phase gains as well, so these display meshes also call the reference mesher somewhere (where is
+not traced here; `tess::edges`, the display plans' own edge division, already evaluated each
+point once). In the rover, the second shell (which names every body's faces, so meshes them all)
+goes from 7.8 s to 5.7 s.
+
+No model changed. `solidcraft-cli eval` of the five designs (the bicycle as `bench-agent` saved
+it) is the same with both binaries apart from the per-feature milliseconds: bodies, volumes,
+areas, centres, bounding boxes, topology counts and warnings. `bench-open` meshes the same
+triangle counts, and the bicycles both binaries save differ only in the document's name.
+
 ## STEP assemblies: a repeated part is meshed once (2026-10-14)
 
 The reader already converts each B-rep item once and keeps it for the instances that list it

vendor/README.md

@@ -108,6 +108,11 @@ Changes:
   point that does not evaluate counts as flat instead of recursing). Surface grids refine only
   the direction that bends, and a ruling left undivided gets cells comparable to the other
   direction (long triangles from a trim loop to the grid's far ends cut through cylinders).
+- `tessellation/triangulation.rs`: an edge's division evaluates each parameter once. A piece
+  carries the points of its ends and its midpoint is an end of both halves: 2 evaluations per
+  piece instead of 4, and the same polyline (a point of an intersection curve costs two grid
+  searches of its surfaces). `bounded_polyline` and `Deflection::turn2`/`turns` are public but
+  hidden, for the kernel's regression against the old division.
 - `tessellation/mod.rs`, `tessellation/triangulation.rs`, for display: a `Deflection` is the
   chord tolerance plus an optional bound on the angle one facet spans, and
   `RoundMeshableShape::robust_triangulation_within` divides by it: an edge or a grid direction

vendor/truck-meshalgo/src/tessellation/mod.rs

@@ -25,6 +25,13 @@ pub fn preview_face<S: RobustMeshableSurface>(surface: &S, loops: &[Vec<Point3>]
     triangulation::preview_face(surface, loops, steps)
 }
 
+/// SolidCraft: an edge's polyline as the meshers divide it (at most 20 000 points). Exposed for
+/// the kernel's regressions in the workspace CI.
+#[doc(hidden)]
+pub fn bounded_polyline<C: PolylineableCurve>(curve: &C, range: (f64, f64), d: Deflection) -> PolylineCurve {
+    triangulation::bounded_polyline(curve, range, d)
+}
+
 /// SolidCraft: sampling a curve that may fail to evaluate. truck's `IntersectionCurve::subs`
 /// unwraps the projection of its leader's point onto both surfaces, which does not converge
 /// everywhere; the mesher samples through this trait so that a failed sample is data, not a panic.
@@ -145,8 +152,9 @@ impl Deflection {
 
     /// The squared sag over squared length above which a facet turns more than `angle`: a
     /// chord spanning `a` of a circle sags `tan(a / 4) / 2` of its length. Infinite without
-    /// an angle bound.
+    /// an angle bound. SolidCraft: public for the kernel's regressions.
-    pub(super) fn turn2(&self) -> f64 {
+    #[doc(hidden)]
+    pub fn turn2(&self) -> f64 {
         match self.angular() && self.angle < std::f64::consts::PI {
             true => (f64::tan(self.angle / 4.0) / 2.0).powi(2),
             false => f64::INFINITY,
@@ -154,8 +162,9 @@ impl Deflection {
     }
 
     /// Whether the angle bound splits a facet of squared sag `sag2` and squared length `len2`
-    /// (`turn2` from [`Deflection::turn2`]).
+    /// (`turn2` from [`Deflection::turn2`]). SolidCraft: public for the kernel's regressions.
-    pub(super) fn turns(&self, turn2: f64, sag2: f64, len2: f64) -> bool {
+    #[doc(hidden)]
+    pub fn turns(&self, turn2: f64, sag2: f64, len2: f64) -> bool {
         turn2.is_finite() && sag2 > self.floor * self.floor && sag2 > turn2 * len2
     }
 }

vendor/truck-meshalgo/src/tessellation/triangulation.rs

@@ -861,30 +861,34 @@ pub(super) fn face_cap(faces: usize) -> usize {
 /// bisection depth of 24; a midpoint that does not evaluate (NaN) counts as flat (truck's
 /// version recursed on it 100 levels deep, which never finishes). A piece within the chord
 /// tolerance is still halved while it turns more than the deflection's angle.
-fn bounded_polyline<C: PolylineableCurve>(curve: &C, range: (f64, f64), d: Deflection) -> PolylineCurve {
+pub(super) fn bounded_polyline<C: PolylineableCurve>(curve: &C, range: (f64, f64), d: Deflection) -> PolylineCurve {
     let (tol, turn2) = (d.chord, d.turn2());
     // A sample that does not evaluate (an intersection curve whose projection fails): an interval
     // end falls back to the curve's approximation so the polyline stays finite; an interior probe
-    // is NaN, which counts as flat.
+    // is NaN, which counts as flat. Each parameter is evaluated once: a piece carries the points
+    // of its ends, and its midpoint, an end of both halves, serves as probe and as end (a point of
+    // an intersection curve costs two grid searches).
     let end = |t: f64| curve.try_subs(t).unwrap_or_else(|| curve.approx_subs(t));
-    let probe = |t: f64| curve.try_subs(t).unwrap_or(Point3::new(f64::NAN, f64::NAN, f64::NAN));
+    let nan = Point3::new(f64::NAN, f64::NAN, f64::NAN);
-    let mut out = vec![end(range.0)];
+    let (p0, p1) = (end(range.0), end(range.1));
-    let mut stack: Vec<(f64, f64, usize)> = vec![(range.0, range.1, 0)];
+    let mut out = vec![p0];
-    while let Some((a, b, depth)) = stack.pop() {
+    let mut stack: Vec<(f64, f64, Point3, Point3, usize)> = vec![(range.0, range.1, p0, p1, 0)];
-        let (pa, pb) = (end(a), end(b));
+    while let Some((a, b, pa, pb, depth)) = stack.pop() {
         let t = a + (b - a) * 0.4729;
         let chord = pa + (pb - pa) * 0.4729;
-        let d2 = probe(t).distance2(chord);
+        let d2 = curve.try_subs(t).unwrap_or(nan).distance2(chord);
         let tm = 0.5 * (a + b);
-        let dm = probe(tm).distance2(pa.midpoint(pb));
+        let sm = curve.try_subs(tm);
+        let dm = sm.unwrap_or(nan).distance2(pa.midpoint(pb));
         let flat = !(d2 >= tol * tol || dm >= tol * tol);
         let flat = flat && !d.turns(turn2, d2.max(dm), pa.distance2(pb));
         if flat || depth >= 24 || out.len() + stack.len() >= MAX_EDGE_POINTS {
             out.push(pb);
         } else {
+            let pm = sm.unwrap_or_else(|| curve.approx_subs(tm));
             // Right half after the left one (stack order).
-            stack.push((tm, b, depth + 1));
+            stack.push((tm, b, pm, pb, depth + 1));
-            stack.push((a, tm, depth + 1));
+            stack.push((a, tm, pa, pm, depth + 1));
         }
     }
     PolylineCurve(out)