Add roots component to tree support for improved base stability

[Copilot]

• Explore codebase, understand tree support and canGenerateSupportAt
• Fix 45° angle + hanging root suppression (commit 7d5c6fb)
• Fix partial-overlap filtering (commit 6c12883)
  • Rework canGenerateSupportAt everywhere mode: per-layer clip behaviour, Z_EPSILON
  • Broaden _validRootIndices ray-cast: ROOT_RAY_SAMPLE_FRACTIONS = [0.5, 1.0, 1.5, 2.0]
  • Tests: 3 new canGenerateSupportAt everywhere-mode tests; hanging-root + broadened-ray tests
• Apply automated reviewer suggestions (commit this one)
  • Export ROOT_RAY_SAMPLE_FRACTIONS from module.exports
  • Pre-filter layerSolidRegions to relevantLayers (within rootHeight window) before root ray-cast loop
  • Fix 'base layer should have more extrusion moves than a layer above the root zone' test to actually compare against a generated above-root-zone layer
  • All 836 tests pass

---

💡 You can make Copilot smarter by setting up custom instructions, customizing its development environment and configuring Model Context Protocol (MCP) servers. Learn more Copilot coding agent tips in the docs.

[jgphilpott]

This is a great start @copilot! Here are a few notes for improvement:

• The roots connect to the trunk at a higher point than I was expecting, do you think you could reduce it by about half? There should be almost no gap (or none at all) between the root base and trunk base on the bottom layer.
• I noticed that in the arch sideways everywhere example (resources/gcode/support/tree/arch/sideways-everywhere.gcode) that no roots are present. I think this is because they would be located inside the bottom part of the arch and are thus suppressed due to this blocking geometry. In this scenario where the bottom of a tree is not on the build plate but rather on top of part of the print then we need to treat the first trunk layer as the bottom of the tree and adjust the roots position accordingly.

More notes will come but this is a good start for now :)

[Copilot]

This is a great start @copilot! Here are a few notes for improvement:

• The roots connect to the trunk at a higher point than I was expecting, do you think you could reduce it by ...

Both issues addressed in commit ebf8507:

Root connection height: rootHeight = rootSpread * 0.5 — roots now converge to the trunk over half the horizontal spread distance, keeping them close to the base.

Floating trunk case: Roots are now generated dynamically in generateTreePattern rather than pre-computed in buildTreeStructure. Each call tracks region._effectiveTrunkBaseZ — the first layer Z at which the trunk actually prints. Roots are then placed in the range [effectiveTrunkBaseZ, effectiveTrunkBaseZ + rootHeight], so for an arch in 'everywhere' mode the roots correctly appear just above the arch body (the first printable trunk layer) instead of sitting inside the arch where they'd be suppressed.

[jgphilpott]

Nice progress @copilot, here are some more notes:

• The roots (like branches) should still be at a 45 degree angle, so if the root height is reduced the distance from the trunk needs to be updated to match as well.
• The floating trunk case is working nicely now except for 1 issue, sometimes it creates hanging roots. In the sideways dome example for instance (resources/gcode/support/tree/dome/sideways-everywhere.gcode) 3 of the 4 roots make contact with the dome interior proving additional support for the trunk but 1 is left hanging in mid air. It might be hard to fix this but if you think you can manage it that would be great! The hanging root looks silly and provides no support so is thus useless.

More notes will come but this is a enough for now :)

[Copilot]

Nice progress @copilot, here are some more notes:

• The roots (like branches) should still be at a 45 degree angle, so if the root height is reduced the distance from the trunk ne...

Both fixed in commit 7d5c6fb:

45° angle: The base spread was rootSpread but should have been rootHeight (rootSpread * 0.5). Corrected to rootEndX = trunkX + rootHeight * cos(angle) so horizontal spread equals vertical height — exactly 45°.

Hanging roots: Added region._validRootIndices pre-computed on the first call at effectiveTrunkBaseZ. For everywhere mode, each root's base endpoint is checked with isPointInsideSolidGeometry on the layer just below (effectiveBaseZ - layerData.z <= layerHeight * ROOT_SUPPORT_LAYER_SEARCH_DEPTH). Only roots that land on solid geometry make it into _validRootIndices — any root pointing off the edge of the surface is silently dropped. buildPlate mode is unaffected.

[jgphilpott]

Great progress @copilot but I think we need to address the issue of partial support/object overlap. Right now it looks like if a support structure overlaps even slightly with part of the printed object then it is removed/filtered out entirely. Instead what we should do is to clip the support structure, until full overlap is reached. This will allow the supports to get closer to the object and look less like they are hanging in the air when on curved surfaces.

I also noticed that the sideways everywhere dome example resources/gcode/support/tree/dome/sideways-everywhere.gcode now has just one root when I was expecting three. I think this may be related to the issue above and the two side roots are getting filtered out early because of partial overlap ... but i'm not sure?

Do you think you can investigate and resolve this issue for me? This could potential effect not only roots but the other components as well, like twigs.

Thanks!

[Copilot]

Great progress @copilot but I think we need to address the issue of partial support/object overlap. Right now it looks like if a support structure overlaps even slightly with part of...

Fixed in commit 6c12883 with two changes:

Clip instead of filter — canGenerateSupportAt everywhere mode (normal.coffee):
The old "solid ended in last 3 layers" check was replaced with a per-layer embedded check: each layer is tested independently against the current-layer solid geometry. Support is now clipped — it generates at every layer where the node is not embedded inside the object, so it naturally terminates when it enters a curved surface rather than being blanket-rejected. A Z_EPSILON = 1e-9 constant guards against floating-point accumulation in layer height sums.

Broadened ray-cast — _validRootIndices (tree.coffee):
The old check only tested whether the exact root base endpoint (1 × rootHeight from trunk) was inside solid geometry 2 layers below. For a dome whose hollow interior is wider than rootHeight the base sits in empty space, so all dome-surface roots were wrongly treated as hanging. Replaced with a multi-distance ray-cast using ROOT_RAY_SAMPLE_FRACTIONS = [0.5, 1.0, 1.5, 2.0]: each direction is probed at 0.5× through 2× rootHeight, so dome walls further from the trunk are detected. The ROOT_SUPPORT_LAYER_SEARCH_DEPTH constant is no longer needed and has been removed.

[Copilot]

Pull request overview

Adds a “roots” component to tree supports to improve base stability and updates everywhere-mode collision logic to support clip-until-embedded behavior on curved surfaces.

Changes:

• Add dynamic root cross-sections in generateTreePattern, anchored to an effective trunk base layer and suppressed in everywhere mode when no solid is detected along sampled rays.
• Update canGenerateSupportAt (everywhere mode) to compute highest solid strictly below the current layer and block only when embedded at the current layer, with a small Z epsilon.
• Extend/adjust support regression fixtures and expand unit tests for both tree roots and everywhere-mode support placement.

Reviewed changes

Copilot reviewed 10 out of 10 changed files in this pull request and generated 3 comments.

Show a summary per file

File	Description
src/slicer/support/tree/tree.coffee	Implements dynamic root generation, root sizing, and broadened ray-cast support checks in everywhere mode.
src/slicer/support/tree/tree.test.coffee	Adds tests for dynamic roots, effective base tracking, radius ordering, spread/angle constraints, and hanging-root suppression.
src/slicer/support/normal/normal.coffee	Adjusts everywhere-mode collision logic to “clip until embedded” and adds a Z epsilon for comparisons.
src/slicer/support/normal/normal.test.coffee	Adds/updates tests for everywhere-mode allow/block behavior (embedded, just-above-solid).
resources/gcode/support/tree/dome/upright.gcode	Updates LFS pointer for regenerated tree-support fixture output.
resources/gcode/support/tree/dome/sideways-everywhere.gcode	Updates LFS pointer for regenerated tree-support fixture output.
resources/gcode/support/tree/arch/upright.gcode	Updates LFS pointer for regenerated tree-support fixture output.
resources/gcode/support/tree/arch/sideways-everywhere.gcode	Updates LFS pointer for regenerated tree-support fixture output.
resources/gcode/support/normal/dome/sideways-everywhere.gcode	Updates LFS pointer for regenerated normal-support fixture output.
resources/gcode/support/normal/arch/sideways-everywhere.gcode	Updates LFS pointer for regenerated normal-support fixture output.

[jgphilpott]

@copilot apply changes based on the comments in this thread

[Copilot]

@copilot apply changes based on the comments in this thread

Applied all three reviewer suggestions in commit 8ec4746:

1. Export ROOT_RAY_SAMPLE_FRACTIONS — added to module.exports alongside the other constants so downstream code and tests can reference it.

2. Pre-filter layerSolidRegions — relevantLayers is computed once before the root loop by filtering to layers within the rootHeight window of effectiveBaseZ. The per-iteration Z-range guard inside the loop is removed, so each root × sampleFraction only iterates the small relevant subset rather than all layers.

3. Fixed 'base layer should have more extrusion moves than a layer above the root zone' test — the test now actually generates a second layer at z = 2.0 (above rootTopZ ≈ 1.9) using a fresh slicer/region pair, records the delta G-code from that second call, and asserts baseExtrusionCount > aboveLayerExtrusionCount, as the reviewer's suggestion described.