ADR 0007 — A C-section is an OPEN ANNULUS, not a single closed loop

Status: accepted (2026-08-21) · Phase 2

Context

The original design sketch described the split tube as returning “ONE closed C-shaped loop — the outer arc from 30 to 330, then the inner arc back from 330 to 30 — with inner=undef”. docs/usage/profiles.md then predicted the consequence: a C-section is not star-shaped about its centroid, so ut_rod’s centroid-fan end cap would produce faces crossing the void, and the section would need “its own cap strategy” — realistically a general polygon triangulator, the one piece of work ADR 0001 deliberately deferred.

Decision

Keep the split section as two loops, outer and inner, exactly like a closed tube — but mark the profile open in its meta. The emitter then:

• runs the outer and inner skins over i = 0 .. N-2 instead of wrapping,
• runs the end caps as ring strips over the same range,
• and adds two seam walls, the cut faces that close the split along the whole run: a quad strip at i = 0 and another at i = N-1.

That is one extra term in _ut_faces_hollow and one conditional index bound.

Consequences

No triangulator is needed, and none of the deferred work came due. The prediction in docs/usage/profiles.md was pessimistic because it accepted the plan’s single-loop framing; changing the representation dissolved the problem rather than solving it.

PROF-2 correspondence is preserved for free — both loops are sampled at the same angles across the material span, exactly as in the closed case.

The seam walls are not optional. Without them the surface is not closed and CGAL rejects the mesh outright, which is at least a loud failure rather than a plausible-looking wrong solid.

A split run still has no enclosed lumen, so ut_bore_rgn() and ut_shell_rgn() both return empty and the run cannot participate in a junction. That limit is unchanged and is what C-sections are for: conduit, cable channel, and parts that print open-side-up without bridging.

ut_rod’s centroid fan remains, and remains restricted to sections that are star-shaped about their centroid. A genuinely non-star-shaped closed profile would still need a triangulator — but no shipped constructor produces one.