After the addBox extraction, 21 procedural mesh primitives
each still open-coded the same 5-line addV vertex-emit lambda.
Add inline addVertex(WoweeModel&, vec3, vec3, vec2) to
cli_box_emitter.hpp (header so the per-vertex hot loop stays
inlineable), plus a per-float overload for the four handlers
(--gen-mesh-stairs, --gen-mesh-tube, --gen-mesh-capsule,
--gen-mesh-arch) that compute pos/normal/uv components inline
rather than building intermediate glm vectors.
Each lambda site collapses from a 5-line body to a 3-line
forwarding wrapper. Output bytes verified identical via
--info-mesh-stats: firepit surface area 2.1100 m² unchanged
across all variants tested (stairs/tube/capsule/arch/
firepit/mushroom).
Sets up the pattern so the next "shared mesh helper" is one
line of include, not another paste-in.
36 procedural mesh primitives in cli_gen_mesh.cpp each open-
coded the same ~30-line lambda for emitting a flat-shaded
axis-aligned box (per-face vertices for unique normals,
6 faces × 4 verts × 2 tris). Hoist the implementation into
cli_box_emitter.{hpp,cpp} as addFlatBox(WoweeModel&, ...)
with two overloads:
• addFlatBox(wom, cx, cy, cz, hx, hy, hz)
center + half-extents form, used by 34 primitives
• addFlatBox(wom, glm::vec3 lo, glm::vec3 hi)
lo/hi corners form, used by --gen-mesh-archway and
--gen-mesh-fence which compute corners directly
Each lambda site collapses from ~30 lines to a 3-line thin
wrapper: cli_gen_mesh.cpp drops from 7374 to 6989 lines.
Output bytes verified identical via --info-mesh-stats
(firepit surface area 2.1100 m² unchanged, vertex/triangle
counts match across firepit/canopy/dock/archway/fence).
Future box-based primitives now opt in by including one
header instead of pasting the lambda again.
