[joseph-x-li]

Fixes #15.

When the weights' sum is zero or not finite, fall back to the nearest corner (weight 1
there, 0 elsewhere). That is still a valid interpolation (non-negative weights summing
to one), it's exact for values that are constant over the triangle, and for a point
that close to all three edges the triangle is effectively degenerate around it anyway.
Every other path is unchanged.

Adds a_point_near_all_three_edges_of_a_thin_triangle_gets_finite_weights to
src/values.rs, which returns NaN weights on main and passes with this change.

Diff (src/values.rs):

-        uvw /= uvw[0] + uvw[1] + uvw[2];
-        uvw
+        // Within `tol` of all three edge lines (a thin triangle at a coarse tolerance)
+        // every weight is zero; dividing would make every value NaN. Take the nearest
+        // corner instead, still a valid interpolation.
+        let sum = uvw[0] + uvw[1] + uvw[2];
+        if sum.abs() > Real::MIN_POSITIVE && sum.is_finite() {
+            uvw /= sum;
+            return uvw;
+        }
+        let d = [0, 1, 2].map(|i| (*v - pos[i]).length_squared());
+        let near = if d[0] <= d[1] && d[0] <= d[2] { 0 } else if d[1] <= d[2] { 1 } else { 2 };
+        uvw = Vec3::ZERO;
+        uvw[near] = 1.;
+        uvw

🤖 Generated with Claude Code

[komietty]

LGTM.