STEP assemblies: mesh a repeated part once and place it for its instances (#1152)

The reader converts each item once, but every instance made its own display mesh
(tessellation, seam and tangent flags), the most expensive step of opening a large
assembly. A later instance of an item now shows its first instance's mesh, placed by
the difference of the two placements. Only rigid placements share; split lines and
any operation drop the link. Measurements and export read each instance's own
geometry, so eval output is unchanged.

Generated assemblies: mesh phase 2,215 -> 139 ms (901 bodies) and 1,028 -> 85 ms
(5,401 bodies), with the same triangle counts.

Co-authored-by: Claude Opus 5.5 <[email protected]>

crates/kernel/src/body.rs

@@ -45,6 +45,50 @@ pub struct Body {
     /// Exact history of the operation that made this body, relative to that operation's
     /// operands; `None` when the operation does not report one.
     pub(crate) history: Option<std::sync::Arc<crate::history::History>>,
+    /// A placed copy of a solid that another body shows already (an instance of a part a STEP
+    /// assembly lists many times): its display mesh is that body's, placed by the matrix, so the
+    /// part is tessellated once rather than once per instance. Dropped by anything that changes
+    /// what the body shows (its geometry is never changed in place).
+    pub(crate) display_of: Option<std::sync::Arc<(std::sync::Arc<SharedDisplay>, [[f64; 4]; 4])>>,
+}
+
+/// The display mesh of a body that several placed copies show (see `Body::display_of`): the
+/// body, and its mesh once made.
+#[derive(Debug)]
+pub(crate) struct SharedDisplay {
+    pub(crate) original: Body,
+    mesh: std::sync::OnceLock<Option<Mesh>>,
+}
+
+impl SharedDisplay {
+    pub(crate) fn new(original: Body) -> Self {
+        SharedDisplay { original: Body { display_of: None, ..original }, mesh: std::sync::OnceLock::new() }
+    }
+}
+
+/// `m` placed by the rigid matrix `t` (by columns, translation in `t[3]`, as the kernel's
+/// transforms take it): positions, normals and edge polylines move, everything indexed stays.
+/// `None` for anything but a rotation and translation (a reflection would turn the triangles
+/// inside out).
+fn placed_mesh(m: &Mesh, t: &[[f64; 4]; 4]) -> Option<Mesh> {
+    let r = |v: Vec3, w: f64| {
+        Vec3::new(
+            t[0][0] * v.x + t[1][0] * v.y + t[2][0] * v.z + w * t[3][0],
+            t[0][1] * v.x + t[1][1] * v.y + t[2][1] * v.z + w * t[3][1],
+            t[0][2] * v.x + t[1][2] * v.y + t[2][2] * v.z + w * t[3][2],
+        )
+    };
+    let det = t[0][0] * (t[1][1] * t[2][2] - t[1][2] * t[2][1]) - t[0][1] * (t[1][0] * t[2][2] - t[1][2] * t[2][0])
+        + t[0][2] * (t[1][0] * t[2][1] - t[1][1] * t[2][0]);
+    if !(det > 0.0 && (det - 1.0).abs() < 1e-6) || t.iter().flatten().any(|x| !x.is_finite()) {
+        return None;
+    }
+    Some(Mesh {
+        positions: m.positions.iter().map(|p| r(*p, 1.0)).collect(),
+        normals: m.normals.iter().map(|n| r(*n, 0.0)).collect(),
+        edges: m.edges.iter().map(|e| e.iter().map(|p| r(*p, 1.0)).collect()).collect(),
+        ..m.clone()
+    })
 }
 
 #[derive(Clone, Debug, PartialEq, serde::Serialize, serde::Deserialize)]
@@ -538,6 +582,7 @@ impl Body {
             face_sources: None,
             regularization_protected: false,
             history: None,
+            display_of: None,
         })
     }
 
@@ -567,6 +612,7 @@ impl Body {
             face_sources: None,
             regularization_protected: false,
             history: None,
+            display_of: None,
         })
     }
 
@@ -869,8 +915,20 @@ impl Body {
     }
 
     /// The same body with the lines its faces were split along on purpose.
+    /// This body shows `shared`'s mesh placed by `t` (it is that body's solid, placed by `t`).
+    pub(crate) fn with_display_of(mut self, shared: std::sync::Arc<SharedDisplay>, t: [[f64; 4]; 4]) -> Body {
+        if self.splits.is_none() && !self.is_mesh() {
+            self.display_of = Some(std::sync::Arc::new((shared, t)));
+        }
+        self
+    }
+
     pub fn with_split_lines(mut self, s: Option<std::sync::Arc<Vec<Vec<Vec3>>>>) -> Body {
         self.splits = s.filter(|l| !l.is_empty());
+        // Split lines are drawn: the shared mesh does not have them.
+        if self.splits.is_some() {
+            self.display_of = None;
+        }
         self.regularization_protected |= self.splits.is_some();
         self
     }
@@ -1228,6 +1286,12 @@ impl Body {
     /// The reference mesher's angular display mesh: what [`Body::display_mesh_angular`] falls
     /// back to when no plan can be made.
     pub fn reference_display_mesh_angular(&self, tol: f64, angle: f64) -> Result<Mesh> {
+        if let Some((shared, t)) = self.display_of.as_deref()
+            && let Some(placed) =
+                shared.mesh.get_or_init(|| shared.original.reference_display_mesh_angular(tol, angle).ok()).as_ref().and_then(|m| placed_mesh(m, t))
+        {
+            return Ok(placed);
+        }
         if self.is_mesh() {
             return self.flagged(self.tessellate(tol)?);
         }

crates/kernel/src/meshbody.rs

@@ -112,6 +112,7 @@ pub fn mesh_body(positions: &[Vec3], triangles: &[[u32; 3]]) -> Result<Body> {
         splits: None,
         face_sources: None,
         history: None,
+        display_of: None,
         regularization_protected: false,
     })
 }
@@ -208,6 +209,7 @@ impl Body {
             splits: self.splits.clone(),
             face_sources: None,
             history: None,
+            display_of: None,
             regularization_protected: self.regularization_protected,
         })
     }

crates/kernel/src/native.rs

@@ -93,6 +93,7 @@ pub fn decode_native_body(bytes: &[u8]) -> Result<Body> {
             splits: None,
             face_sources: None,
             history: None,
+            display_of: None,
             regularization_protected: false,
         };
         metadata::apply(&mut body, saved.metadata)?;

crates/kernel/src/splitface.rs

@@ -558,6 +558,7 @@ pub(crate) fn split_edges_at(body: &Body, points: &[Vec3]) -> Result<Body> {
             splits: body.splits.clone(),
             face_sources: None,
             history: None,
+            display_of: None,
             regularization_protected: true,
         };
         Ok(crate::record::rebuilt_body(split, &[&body.solid], &rec))
@@ -1053,6 +1054,7 @@ pub(crate) fn split_any(body: &Body, faces: &[usize], tool: &SplitTool) -> Resul
             splits: body.splits.clone(),
             face_sources: None,
             history: None,
+            display_of: None,
             regularization_protected: true,
         };
         Ok((crate::record::rebuilt_body(split, &[&body.solid], &imp.rec), imp))

crates/kernel/src/step_in/mod.rs

@@ -248,6 +248,9 @@ struct Reader<'a> {
     /// Items the file converts exactly once (see `index`): keeping their solids would hold a
     /// second copy of the whole import for nothing.
     once: HashSet<u64>,
+    /// An item placed more than once: its first instance's display, shared by the later ones
+    /// (with the placement that first instance had), so the part is meshed once.
+    shown: HashMap<u64, (Arc<crate::body::SharedDisplay>, mt::Matrix4)>,
     used_items: HashSet<u64>,
     /// Items already instanced for the product being read.
     node_items: HashSet<u64>,
@@ -280,6 +283,7 @@ impl<'a> Reader<'a> {
             opacity: HashMap::new(),
             solids: HashMap::new(),
             once: HashSet::new(),
+            shown: HashMap::new(),
             used_items: HashSet::new(),
             node_items: HashSet::new(),
             visits: 0,
@@ -680,6 +684,16 @@ impl<'a> Reader<'a> {
                                     .map(|face| Some(crate::FaceSource::Imported { item, face: *face, payload_sha256: None, producer: None }))
                                     .collect();
                                 let body = body.with_face_sources(Some(std::sync::Arc::new(identities)));
+                                // A later instance of an item shows its first instance's mesh, placed.
+                                let body = match self.shown.get(&item) {
+                                    Some((shared, first)) => body.with_display_of(shared.clone(), to_array(&mat_mul(m, &rigid_inverse(first)))),
+                                    None => {
+                                        if !self.once.contains(&item) {
+                                            self.shown.insert(item, (Arc::new(crate::body::SharedDisplay::new(body.clone())), *m));
+                                        }
+                                        body
+                                    }
+                                };
                                 self.bodies.push(ImportedBody {
                                     name,
                                     body,

crates/kernel/src/step_in/tests.rs

@@ -523,6 +523,31 @@ fn hand_written_assembly() {
     assert!((root.children[1].transform[3][0] - 100.0).abs() < 1e-9);
 }
 
+/// A part an assembly places twice is meshed once: the second instance shows the first one's
+/// display mesh, placed (the same triangles and faces, at its own place), and keeps its own
+/// placed geometry for everything else.
+#[test]
+fn later_instances_show_the_first_instances_mesh_placed() {
+    let imp = step_import(&two_instance_assembly()).unwrap();
+    let (first, second) = (&imp.bodies[0].body, &imp.bodies[1].body);
+    assert!(first.display_of.is_none() && second.display_of.is_some());
+    let a = first.reference_display_mesh_angular(0.01, crate::DISPLAY_ANGLE).unwrap();
+    let b = second.reference_display_mesh_angular(0.01, crate::DISPLAY_ANGLE).unwrap();
+    assert_eq!(a.triangles, b.triangles);
+    assert_eq!(a.tri_face, b.tri_face);
+    assert_eq!(a.edges.len(), b.edges.len());
+    // Turned 90 degrees about Z and moved 100 mm along X.
+    for (p, q) in a.positions.iter().zip(&b.positions) {
+        assert!(q.dist(Vec3::new(100.0 - p.y, p.x, p.z)) < 1e-9, "{p:?} -> {q:?}");
+    }
+    let own = measure(second).unwrap().bbox;
+    let shown = b.bounds();
+    assert!(shown.min.dist(own.min) < 1e-6 && shown.max.dist(own.max) < 1e-6, "{shown:?} vs {own:?}");
+    // Anything that changes what the body shows drops the shared mesh.
+    let split = second.clone().with_split_lines(Some(std::sync::Arc::new(vec![vec![Vec3::new(91.0, 0.0, 5.0), Vec3::new(91.0, 10.0, 5.0)]])));
+    assert!(split.display_of.is_none());
+}
+
 /// The import keeps a conversion only where the file can ask for the item again. A flat file of
 /// separate solids converts each item exactly once (no second copy of the whole import is held);
 /// an assembly that instances one item keeps it, and each instance still gets geometry of its

crates/kernel/src/surfaces.rs

@@ -41,6 +41,7 @@ fn surface_of(shell: mt::Shell) -> Body {
         splits: None,
         face_sources: None,
         history: None,
+        display_of: None,
         regularization_protected: false,
     }
 }

docs/perf.md

@@ -1,5 +1,31 @@
 # Design loading
 
+## 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
+again. Every instance still became a body of its own geometry (it has to: features cut and join
+it where it is), and each of those bodies made its own display mesh: tessellation, seam and
+tangent flags, the most expensive step of opening a large assembly. A board with 60 of each
+resistor meshed the same resistor 60 times.
+
+A later instance of an item now shows its first instance's display mesh, placed by the
+difference of the two placements: the same triangles, faces and edge flags, moved. The mesh is
+made once, by whichever instance is drawn first. Only a rotation and translation is shared
+this way (anything else meshes itself), and anything that changes what a body shows (split
+lines; any operation, which makes a new body) drops the link. Measurements, booleans and export
+read the instance's own geometry as before, so `solidcraft-cli eval` of an assembly is unchanged.
+
+Two generated assemblies (SolidCraft designs with placed copies of components, exported as STEP
+AP214 assemblies), `solidcraft-cli bench-open`, release build, shared 16-core Linux box:
+
+| file | bodies | mesh phase before | after | CPU before | after | peak RSS before | after |
+|---|---:|---:|---:|---:|---:|---:|---:|
+| 4 parts, 500 instances (bolts with a torus ring, brackets) | 901 | 2,215 ms | 139 ms | 28.4 s | 4.1 s | 918 MB | 749 MB |
+| 10 filleted parts with 8 pins, 600 instances | 5,401 | 1,028 ms | 85 ms | 16.7 s | 6.0 s | 1,095 MB | 783 MB |
+
+Both import to the same triangle counts (4,404,412 and 4,316,412). Each instance's display plan
+(the GPU viewport draws bodies from plans) is still made per instance; sharing those the same
+way is the next step.
 ## Designs built by agents: work per command that grew with the design (2026-10-14)
 
 A design an agent builds over MCP is made of many small commands: a cylinder per spoke, a box