WireRestShape::initTopology corrupts topology for N≥3 sections

[lkarstensen]

Summary

WireRestShape::initTopology builds the visual EdgeSetTopologyContainer by iterating over section materials and placing points along the wire's curvilinear abscissa. Two accumulator variables are updated via assignment instead of accumulation:

• prev_length = length (should be prev_length += length)
• prev_edges = nbrVisuEdges (should be prev_edges += nbrVisuEdges)

For any wire with three or more sections this collapses point coordinates from section 2 onward and corrupts edge indices, producing degenerate edges and a scene-init failure (or silently malformed visual topology).

Two-section wires are unaffected because after the first iteration the assignment and accumulation forms produce the same value.

Affected component

• File: src/BeamAdapter/component/engine/WireRestShape.inl
• Method: WireRestShape<DataTypes>::initTopology
• Confirmed in: release-v24.12 and master

The bug

Real prev_length = 0.0;
int prev_edges = 0;
int startPtId = 0;
for (sofa::Size i = 0; i < l_sectionMaterials.size(); ++i)
{
    int nbrVisuEdges = l_sectionMaterials.get(i)->getNbVisualEdges();
    Real length = fabs(keyPts[i + 1] - keyPts[i]);
    Real dx = length / nbrVisuEdges;

    for (int i = startPtId; i < nbrVisuEdges + 1; i++) {
        l_topology->addPoint(prev_length + i * dx, 0, 0);
    }
    for (int i = prev_edges; i < prev_edges + nbrVisuEdges; i++) {
        l_topology->addEdge(i, i + 1);
    }

    prev_length = length;        // BUG: should be prev_length += length
    prev_edges = nbrVisuEdges;   // BUG: should be prev_edges += nbrVisuEdges
    startPtId = 1;
}

• length is the local length of the current section (fabs(keyPts[i+1] - keyPts[i])).
• prev_length is intended to be the global cumulative offset for point placement — assigning resets it to just the last section length.
• prev_edges is intended to be the global edge index offset — assigning resets it to just the last section's count, causing edges for section 2+ to overwrite indices already used by earlier sections.

Why N=2 works

After iteration 0, both prev_length = length_0 and prev_length += length_0 yield the same value (starting from 0). The discrepancy only appears from iteration 2 onward.

Worked example (3 sections of 100, 50, 25 mm, 10/5/4 edges each)

Iter	length	prev_length on entry (buggy)	Point range (buggy)	Point range (correct)
0	100	0	0 → 100 ✓	0 → 100
1	50	100	100 → 150 ✓	100 → 150
2	25	50 (bug)	50 → 75 ✗	150 → 175

Iteration 2 places its points at curvilinear abscissa 50–75 mm, overlapping section 1. Additionally, edges for section 2 are added at global indices 5–8 instead of 15–18, corrupting the topology.

keyPoints is correct

WireRestShape::initLengths uses proper accumulation:

keyPointList[i + 1] = keyPointList[i] + rodSection->getLength();

Mechanical sampling paths (getMechanicalSampling, getCollisionSampling, getRestTransformOnX, getBeamSectionAtX) all consult keyPts directly and are unaffected. Only initTopology, which builds the visual edge set, is broken.

Proposed fix

Two one-line changes:

prev_length += length;       // was: prev_length = length
prev_edges += nbrVisuEdges;  // was: prev_edges = nbrVisuEdges

Zero effect on N≤2 wires (already produces identical output).

[lkarstensen]

Closing — the reported fix does not produce the expected improvement in practice. Further investigation needed.