P2.2 overlap-as-edge on CurveSegmentString (OV-P2 / Bar 2)

[grootstebozewolf]

Stacked on cursor/p21-curve-segment-string-7422 @ d3e0843 (draft #16). Do not merge this into #7. This is OV-P2 / Bar 2, not OV-P1. Does not merge #12 / #13 / #15 / #16 / #7. Does not push to feature/sfa-curve-rgr or master.

Sentence

The noder is the named-miss suite, one assembly rule at a time. Nodes first (P2.1), then overlap as an edge. Densify is never a noder.

What this is

P2.1 CurveSegmentString + CurveSegmentNoder emit a discrete node set, or null. MIXED was the first miss because collinear overlap is an interval, not a cheap node set.

P2.2 teaches those strings that a shared run is an edge (an interval on one circle, or a collinear segment), not a pair of endpoints pretending to be the whole story.

CircleSweepOverlay.sharedSweep is the same-circle interval the kits already sew. The noder lifts that onto the string. Collinear overlap uses LineIntersector.COLLINEAR_INTERSECTION endpoints as a segment. A different-circle tangent (intersectCircles length 1, on both sweeps) is a zero-length degenerate edge. No face walk. No CurvePolygon for a kiss.

Package-private. Reuses CurveSegmentString, CircleSweepOverlay, intersectCircles, isOnSweep. No new public API. No noder in jts-core. Name is OverlayNGCurve.

Edges vs named misses

Fixture	This rung
H-SHELL-N-MIXED	Edge. Overlapping diameter (-1 0)–(1 0). nodes() stays null (not a point set). Overlay kit still refuses — no face walk.
H-SAME-CIRCLE	Edge. Shared quarter of CIRCLE_5, length 2.5π, already sewn by CircleSweepOverlay. Locked on the string. Endpoint-only kiss is not an interval.
H-SAME-CIRCLE two-shell	Edge. HALF_DISC ∩ HALF_RIGHT lifts that quarter onto two CompoundCurve shells.
H-ANNULUS-TANGENT	Degenerate edge at (5 0). Zero-width pinch (d+r=R). Overlay does not punch an annulus. nodes() stays null.
TOUCH-ext	Degenerate edge at (1 0). Unit discs, centres (0,0) and (2,0), d=r1+r2. Oracle FF2F01212 (EXT_TANGENT). nodes() stays null (not a 2-node set). CircularDiscOverlay stays null — a point is not a CurvePolygon.
R-LL collinear overlap	Edge on the noder: (2 0)–(10 0). Public helper still misses; public overlay stays approximate (R2). Not chordsaw-as-exact.
R1.5 / R1.6-2 / H-SHELL-2 / even-n / D2-odd	Nodes, bit-identical. Crossing is not a shared run.
H-SHELL-HOLE-CROSS / H-SHELL-HOLE-X	Named miss. Holes stay null (P2.3 / P2.4).

What this is not

• Not P2.3 bite vs hole (H-SHELL-HOLE-CROSS)
• Not P2.4 two-hole (H-SHELL-HOLE-X)
• Not P2.5 / N-SS (public SegmentString / Noder in jts-core)
• Not a snap-rounding noder
• Not densify-then-node flagged as exact
• Not a merge into feature/sfa-curve-rgr / PR SFA/SQL-MM curve awareness: Curve rename, arc-aware operations, OverlayNGCurve ratchet (#1195 RGR series) #7 / locationtech master
• Not a general face walk. Overlay kits still assemble CAP/CUP/SUB/XOR.

Verify

mvn -pl modules/curve -am test -DfailIfNoTests=false -Dcheckstyle.skip=true -Dpmd.skip=true \
  -Dtest=CurveSegmentStringTest,CompoundCurveShellOverlayTest,CircularArcOverlayTest,CircularDiscOverlayTest,CircularDiscPolygonOverlayTest,CircularLineOverlayTest,OverlayNGCurvePhase0Test,OverlayNGCurveRatchetTest,OverlayNGCurvePerfGateTest
bash dev/check-no-curved.sh
mvn -pl modules/curve checkstyle:check
mvn -pl modules/curve pmd:check

All green after this revision:

• CurveSegmentStringTest 16 / 0 (TOUCH-ext pin added)
• CircularDiscOverlayTest 12 / 0
• OverlayNGCurveRatchetTest 44 / 0
• OverlayNGCurvePerfGateTest 78 / 0 (15% slack)
• Naming-guard OK; Checkstyle 0 violations; PMD check on jts-curve succeeded

Slack stays 15%. Overlay name is OverlayNGCurve. No new public API. No noder was written in jts-core. Overlay kits were not changed.