Clean Geometry

clean-geometry

Make the currently selected Rhino curves CAD-clean. Sibling to clean-similar-curves (which only removes duplicates/overlaps). Same architecture: two scripts next to this file, invoked via a one-line loader; a non-destructive cleangeo_diagnose.py stashes a plan in sc.sticky; a destructive cleangeo_apply.py applies ONE category chosen by a KEY variable.

Checks

• A simplify — Curve.Simplify (merge colinear segs, straighten near-lines, arcs, cull knots). Lossless-ish. KEY="simplify".
• B short → join / orphan — curves shorter than SHORT_LEN; if an endpoint meets another curve → JOIN cluster; if isolated → ORPHAN. KEY="join" (safe); KEY="orphan" deletes isolated shorts (opt-in).
• C messy → tidy — curves with > CP_MAX control points fit to line/arc/circle/ellipse within FIT_TOL (conservative; freeforms protected by a residual gate). KEY="tidy".
• D near-orthogonal — lines within ORTHO_ANG of world X/Y but not exact. REPORT-ONLY in v1 (no apply).
• E snap-gap / burst corner — endpoint near-misses (tol < dist ≤ GAP_TOL) that should meet. Fixed the user's way: Fillet radius=0, trim=true, join=true (Curve.CreateFilletCurves). KEY="gap".

Procedure (do in order)

1. Resolve paths. Glob **/clean-geometry/cleangeo_diagnose.py and **/clean-geometry/cleangeo_apply.py. Do NOT hardcode.
2. Diagnose (non-destructive).

   exec(open(r"<ABS>\cleangeo_diagnose.py").read())
   

   Prints per-check counts + a length histogram. SHORT_LEN defaults to None → check B is skipped until set.
3. Set SHORT_LEN from the histogram (pick a value below the main length cluster) and re-run diagnose to populate B:

   SHORT_LEN=2.0; exec(open(r"<ABS>\cleangeo_diagnose.py").read())
   

4. Confirm scope with AskUserQuestion. Note risk: A/C use Replace (guid-stable, ≤ tol / ≤ FIT_TOL); B join is lossless topology change; orphan deletes (opt-in); E gap moves endpoints (extend/trim). D is report-only.
5. Apply, one KEY at a time, recommended order:

   KEY="simplify"; exec(open(r"<ABS>\cleangeo_apply.py").read())
   KEY="tidy";     exec(open(r"<ABS>\cleangeo_apply.py").read())
   KEY="join";     exec(open(r"<ABS>\cleangeo_apply.py").read())
   KEY="gap";      exec(open(r"<ABS>\cleangeo_apply.py").read())
   # optional: KEY="orphan"; ...   (deletes isolated short curves)
   

   Re-diagnose after the Replace stages (simplify, tidy) and before endpoint-moving stages (join, gap) for a clean plan — Replace shifts endpoints by ≤ tol; a fresh plan avoids stale clusters.
6. Report created/modified/deleted/failed. Recommend re-running clean-similar-curves after gap (fillet may create tiny overlaps). Offer memory update.

Critical gotchas (baked in — do not "fix" out)

• Simplify angle arg is RADIANS (ModelAngleToleranceRadians), despite Rhinoscript's degrees API.
• A must run before C. Simplify lowers control-point count; the C "too many CPs" gate is evaluated on the simplified curve (diagnose does this internally). Applying tidy before simplify still works (tidy is self-contained) but re-diagnosing keeps counts honest.
• C is conservative by design: CP_MAX gate + FIT_TOL + GetDistancesBetweenCurves residual gate + new-CP-count-must-drop. A wandering freeform near no primitive is correctly LEFT ALONE. No Rebuild/Fit CP-reduction fallback in v1.
• B join edges are short-incident only — two long curves sharing an endpoint are NOT auto-joined; only clusters containing a short curve are. Prevents over-joining the whole network.
• Orphan = no neighbor within the selection. A real continuation that isn't selected looks orphaned. Never auto-deleted; KEY="orphan" is opt-in.
• E excludes settled endpoints (already coincident within tol) so closing a gap never rips apart an existing join. GAP_TOL > SNAP_TOL keeps the join band (B) and gap band (E) disjoint.
• A/C use live rg.Curve objects stashed in sc.sticky (cleangeo_geo_simplify, cleangeo_geo_tidy) — general curves can't be expressed as p0/p1. Apply = sc.doc.Objects.Replace(guid, curve); keeps Id, layer, attributes, selection. Diagnose==apply.
• Created objects inherit donor attributes (Attributes.Duplicate()) and are re-selected; join/gap donor = the longest member.
• IronPython 2.7: TryGet* / GetDistancesBetweenCurves / CreateFilletCurves return tuples/arrays; use [0] for the bool. Float division. rg.CurveSimplifyOptions.All.

Parameters

Top of cleangeo_diagnose.py, # ---- params ----: SHORT_LEN (None→histogram, TUNE), SIMPLIFY_DIST/ANG, SNAP_TOL, CP_MAX (30), FIT_TOL, ORTHO_ANG (2°), GAP_TOL (20×tol; must be > SNAP_TOL). None → doc tolerance.

Not yet implemented (deferred to v2)

• D near-orthogonal APPLY (report-only in v1 — endpoint snap-to-axis breaks networks; design is stashed in plan['ortho']).
• E extend-to-intersection is via Fillet; parallel/collinear near-miss uses a midpoint-pull fallback.
• Non-curve types (surface/brep/mesh); T-junction / mid-curve endpoint snapping; aggressive Rebuild CP reduction; multi-primitive decomposition of one messy curve.

Reference

Built 2026-07-01 as a sibling to clean-similar-curves. v1 scope decided with user: A/B/C + E fix, D report-only, conservative tidy, histogram-driven SHORT_LEN, opt-in orphan delete. See memory rhino-clean-geometry-method and the dedup sibling rhino-curve-dedup-method. Verified by an in-doc synthetic test (per-check exact-geometry asserts).