Evaluation cache: every feature evaluation is kept by a key of its own inputs, the document context it reads and a digest of the model it starts from (bodies identified by the evaluation that made them, kept while passed through), shared with previews and scratch copies; setting a value back, undo, redo and repeated evaluations reuse results (a 20 × 20 patterned plate builds in 1.7 s instead of 7.5 s; undo/redo of a parameter edit 0 ms); cached and from-scratch evaluation compared in a test; numbers in docs/perf.md (solvecraft-cli bench-edit). Designs read now always get feature input names and occurrences filled in

Co-Authored-By: Claude Opus 5.5 <[email protected]>
Claude-Session: https://claude.ai/code/session_01AS5Tn3xieewUSg3uxrMVdL

crates/kernel/src/body.rs

@@ -250,6 +250,17 @@ fn polyline_mid(pts: &[Vec3]) -> (Vec3, f64) {
 }
 
 impl Body {
+    /// The very same body (shared data), not just an equal-looking one.
+    pub fn same(&self, other: &Body) -> bool {
+        std::sync::Arc::ptr_eq(&self.solid, &other.solid)
+            && match (&self.mesh, &other.mesh) {
+                (Some(a), Some(b)) => std::sync::Arc::ptr_eq(a, b),
+                (None, None) => true,
+                _ => false,
+            }
+            && self.color == other.color
+    }
+
     pub(crate) fn new(solid: Solid) -> Result<Body> {
         if solid.boundaries().is_empty() {
             return Err(KernelError::Failed("empty result".into()));