Pull edges towards horizontal, not just out of their own way

[matt-edmondson]

Problem

The clearance splay added in #360 only fires once a curve is at risk of hiding behind the bodies at its own ends. That caps an edge at about 75.5 degrees off horizontal and does nothing at all below that — legal, but never flat. A graph settles with edges at whatever angle the springs and gravity happen to leave them, which in the node editor reads as a tangle.

Fix

CalculateLinkFlatteningForces now also levels an edge continuously, with a force proportional to the vertical offset between its two ends. The two mechanisms do different jobs:

	fires	job
levelling (new)	always, on forward edges	makes a link lie flat
clearance splay (#360)	only below the bezier bound	guarantees a link stays visible

A backward edge is exempt from levelling. It is still being reordered, and pulling it level would fight the vertical slide that #362's reorder needs to get its endpoints past one another.

Measured

On a six-body graph with four edges, shaped like the editor screenshot that prompted this:

	mean angle off horizontal	worst overlap
before	27.4°	0px
after	13.1°	0px

Per-edge after: 36.6°, 13.5°, 1.3°, 0.9° — two of the four land within about a degree of flat.

Neither mechanism can make every edge horizontal at once: a node with several inputs cannot have all of them level with it, so the edges that share a busy node keep some angle. This is a preference the overlap pass and springs push back on, not a constraint. LinkFlatteningStrength now scales both it and the clearance splay; 0 still disables both.

On the overlapping nodes in the same screenshot

I tried to reproduce those and could not. The no-overlap invariant holds at rest both for a two-body pair placed overlapping and for a crowded six-body graph with four backward edges — worst overlap 0px in every case, before and after this change. The most likely explanations are a frame caught mid-settle, or physics not enabled in that editor. Worth another look if it persists with physics running; I did not want to claim a fix I could not demonstrate.

Testing

• LinkFlattening_PullsAForwardEdgeTowardsHorizontal — a steep forward edge closes to under a quarter of its starting vertical offset.
• LinkFlattening_ZeroStrength_LeavesAnEdgeAsSteepAsItStarted — with the preference off, the spring alone leaves the edge steeper than it is wide.

ForceDirectedLayout.Tests 35/35, ImGui.NodeEditor.Tests 71/71, and ForceDirectedLayout.Native / ImGuiAppDemo build clean in Release.

Not verified visually — this container is headless and the demo UI suite cannot start here (the repo's PNGs are Git LFS pointers and git-lfs is not installed). The angles are measured, but whether LinkFlatteningStrength = 0.5 is the right amount of flattening is a judgement worth making in the real editor; the demo slider covers 0–2.

🤖 Generated with Claude Code

https://claude.ai/code/session_01D2KNUKr1xJPTEdDmeF2tQN

---

Generated by Claude Code

[sonarqubecloud]

Quality Gate passed

Issues
2 New issues
0 Accepted issues

Measures
0 Security Hotspots
100.0% Coverage on New Code
0.0% Duplication on New Code

See analysis details on SonarQube Cloud