ADR 0001 — uniTube takes no dependencies

Status: accepted (2026-08-21) · supersedes an earlier proposal to depend on BOSL2

Context

A tube swept along a path needs a mesh backend: stamp a cross-section at each station frame, stitch the sides, cap the ends. BOSL2’s sweep() does this and accepts a region (outer loop plus inner loop), which makes a hollow tube one polyhedron with annular end caps and no CSG along a run. That capability is what made the whole redesign affordable and it was the original recommendation.

The counter-pressure: the project’s own rules prefer no dependency where one can be avoided, and the owner’s position was explicit — “I really don’t want this lib having deps if I can avoid it, I’d rather do this math and geometry from ground up.”

Decision

No dependencies. The backend is written from the geometry in src/ut_mesh.scad.

The decisive argument is that ADR 0002 (the two-level IR) already means uniTube computes its own frames. That reduced the dependency’s contribution to a mesh stitcher — and a prototype of exactly that was written and validated before this decision was taken:

probe	result
Annular sweep, 90° elbow, ~40 lines, no libraries	`Simple: yes`, 2178 facets, 0.4 s forced CGAL
Closed loop (torus, caps suppressed, index wrap)	`Simple: yes`
Non-circular bore (square in round shell)	`Simple: yes`, once correspondence-aligned
`ut_volume()` — divergence theorem, ~6 lines	6109.76 vs analytic 6109.762 — exact to 6 s.f.

Consequences

Zero-install. A fresh clone renders immediately; just setup only checks the toolchain. For an OpenSCAD library, where users drop a folder into their library path, this is a real adoption advantage.
No upstream licence to accommodate, which simplifies ADR 0003.
Cost: roughly one extra day in Phase 1, and no join_prism for junction fillets later. That matters less than it appears — join_prism cannot join to a curved swept surface (its base must be a plane, sphere, straight cylinder or straight prism), which is binding for a library whose premise is curved paths. v1 ships creased junctions regardless.
Do not copy from BOSL2 as a shortcut. BSD-2 permits it with the notice retained, so it is legally clean, but its internals are deeply interconnected and a copied function drags helpers that drag more helpers.
One invariant surfaced from the probe that would otherwise have appeared in Phase 2 as an unexplained CGAL assertion: see PROF-2 in ADR 0006.