Skip to content

fix: MoveTool drag translate keeps three-point arcs - #59

Draft
grootstebozewolf wants to merge 2 commits into
feature/sfa-curve-rgrfrom
cursor/translate-curve-honesty-b6ab
Draft

fix: MoveTool drag translate keeps three-point arcs#59
grootstebozewolf wants to merge 2 commits into
feature/sfa-curve-rgrfrom
cursor/translate-curve-honesty-b6ab

Conversation

@grootstebozewolf

@grootstebozewolf grootstebozewolf commented Aug 16, 2026

Copy link
Copy Markdown
Owner

HOLD merge until PO. UX SIGNs the MoveTool fix from the next JAR after this PR lands a tip — do not merge.

TestBuilder-only honesty fix on #7 (cbc45d4a). Do not join #55 #52 #38. Do not revert #54 #57 #58. No new WKB / type number. Type set stays 1–7 + 8–12. Type 19 stays named fallback. No Bézier I/O type. Off #7 as a core type.

Do not retip pin JAR 61eb3377. Testers still shoot Function-tree AffineTranslation (10, 8) on that pin. Do not rebuild the guides JAR.

Canvas miss is MoveTool drag

Java2D draws arcs with Béziers (hidden canvas controls). Those points are a draw path, not the JTS geometry. The miss is MoveTool drag on logoLines in A, not the Function-tree AffineTranslation (10, 8).

MoveTool.execute (whole-geom, not Ctrl-component):

AffineTransformation trans = AffineTransformation.translationInstance(dx, dy);
GeometryComponentTransformer.transform(geom, trans)

which is geom.copy(); geom.apply(trans).

After drag, type stayed MULTICURVE but arcs lied: each later CIRCULARSTRING start was untranslated while mid/end moved.

Same apply as Function-tree — verified, not a Function-tree SIGN

Function-tree AffineTransformation.translate is:

AffineTransformation.translationInstance(dx, dy).transform(g)

and AffineTransformation.transform is also copy(); apply(this).

Same apply. Stated. Not a UX SIGN that Function-tree AffineTranslation (10, 8) is the canvas miss. Testers still shoot that function on pin JAR 61eb3377. Do not retip.

Cause

CompoundCurve.apply inherited LineString.apply, which walks only the concatenated sequence. concatMembers drops each later member’s start. Those starts are distinct Coordinate objects. Mid+end moved; start did not.

No PathIterator / ShapeWriter / CurveShapeWriter round-trip bakes cubicTo into the geom. MoveTool does not reconstruct from a Shape.

CurvePolygon inherited Polygon.apply on the flat rings. A COMPOUNDCURVE shell/hole had the same skipped-start miss.

Fix

  • CompoundCurve.apply visits every member (including circular starts), then syncs the concatenated sequence in place.
  • CurvePolygon.apply visits structural rings when they are not the same object as the flat views.
  • MoveTool.execute javadoc: canvas drag translate; same apply as AffineTransformation.transform; translate only.

Affine translate is the signed type-honest case. This does not sign that shear or non-uniform scale still describes a circular arc.

Unchanged: Geometry.buffer / logoBuffer (still toLinear + BufferOp CHORD-PATH). Named fallback stays named. Never isApproximate()=false. OverlayNGCurve not touched.

Tests

  • JTSFunctionsLogoLinesTranslateTest: MoveTool path on logoLines +(10, 8). Types kept. Every CIRCULARSTRING still 3 control points. Every control point, including member starts, is +(10, 8). Same-apply check vs AffineTransformationFunctions.translate (verification, not a Function-tree SIGN).
  • CompoundCurveTranslateTest: apply honesty on COMPOUNDCURVE / MULTICURVE / CURVEPOLYGON shell and hole.

Architect SIGN: translate only. UX SIGN: MoveTool from the next JAR after this tip. HOLD merge.

Open in Web Open in Cursor 

cursoragent and others added 2 commits August 16, 2026 20:54
Affine translate on CompoundCurve/CurvePolygon must visit every
member control point, including circular starts omitted from the
concatenated sequence. Keep ISO/IEC 13249-3 types and 3-point arcs.
Translate only is type-honest; no Bézier I/O type.

Co-authored-by: Jeroen Bloemscheer <grootstebozewolf@users.noreply.github.com>
Canvas miss is MoveTool, not Function-tree AffineTranslation.
Whole-geom drag is copy(); apply(translation) — verified the same
apply as AffineTransformation.translate. Testers still shoot pin
JAR 61eb337. Do not retip. Do not rebuild the guides JAR.

Co-authored-by: Jeroen Bloemscheer <grootstebozewolf@users.noreply.github.com>
@cursor cursor Bot changed the title fix: translation should not warp curves fix: MoveTool drag translate keeps three-point arcs Aug 16, 2026
cursor Bot pushed a commit that referenced this pull request Aug 17, 2026
Fold #59 apply honesty onto MMF tip: Affine/MoveTool translation no
longer drops later CircularString starts omitted from concatMembers.
LogoLines translate tests green.

Co-authored-by: Jeroen Bloemscheer <grootstebozewolf@users.noreply.github.com>
grootstebozewolf added a commit that referenced this pull request Aug 17, 2026
* feat: Option B SegmentKind + linearization strategy + WKB 18–21 codes

MMF spine for locationtech#1195 on fork #7:
- Option B: SegmentKind (LINEARIZED/ARC/CERTIFIED), CircularNodedSegmentString,
  default methods on SegmentString, IntersectionAdder mayCollapseToChord,
  EdgeNodingBuilder/OverlayNG prepared exact edges (from #27 seed).
- CurveLinearizationStrategy: default LINEARIZED always warns; PRESERVE keeps type.
- WKBConstants zoo SIGN: Clothoid=18, Bezier=19, Ellipse=20, Nurbs=21.

Co-authored-by: Jeroen Bloemscheer <grootstebozewolf@users.noreply.github.com>

* feat: WKB 18 CRV-CLOTHOID greenfield round-trip

Signed zoo type 18: GeometryFactory.createClothoid stub,
CurveGeometryFactory + ClothoidSegment, WKBReader layout
(startXY[ZM] + tangent + κ0 + κ1 + L), CurveWKBWriter emit,
core writer refuses flatten. ISO 1018 type word covered.

Co-authored-by: Jeroen Bloemscheer <grootstebozewolf@users.noreply.github.com>

* feat: OverlayNG-for-circles R-OV (Option B) + curve segment bridge

Fold #27 OverlayNGCircle / CircleNoder, CurveSegmentString/Noder/Faces,
BiteVsHole, TwoHoleOverlay onto #7 tip. OverlayNGCurve R-OV dispatches
H-SHELL-N-MIXED via CircularNodedSegmentString. 112 overlay/strategy
tests green including OverlayNGCircleTest 7/7.

Co-authored-by: Jeroen Bloemscheer <grootstebozewolf@users.noreply.github.com>

* feat: OFF concentric OffsetCurve for single-arc CircularString

Public OffsetCurve.getCurve lasers three-point CircularString to a
concentric arc (left-of-direction sign: CCW +d inward). Collapse to
empty when r'<=0. Multi-arc stays chordsaw. CurveWKTReader defaults to
CurveGeometryFactory. Ship: drop CurveAwarenessSpecTest#test_OFF_*.

Co-authored-by: Jeroen Bloemscheer <grootstebozewolf@users.noreply.github.com>

* feat: WKB 19–21 BezierCurve EllipseCurve NurbsCurve greenfield

PRF-BEZIER / PRF-ELLIPSE / CRV-NURBS types with factory stubs,
WKBReader/CurveWKBWriter layouts, Linearizable.toLinear, ISO type
words, and round-trip tests. Core writer refuses flatten.

Co-authored-by: Jeroen Bloemscheer <grootstebozewolf@users.noreply.github.com>

* feat: BUF-1/BUF-NEG open-arc buffer corridor laser

Single-arc CircularString.buffer(d>0) returns CurvePolygon with
CompoundCurve shell (outer, end-cap, inner-rev, start-cap). BUF-NEG:
d<0 and |d|>=R yields EMPTY. Ship meter methods for BUF-1 and BUF-NEG.

Co-authored-by: Jeroen Bloemscheer <grootstebozewolf@users.noreply.github.com>

* feat: fold HP.2/HP.3 CurveHotPixel + snap headings onto MMF tip

CurveHotPixel (arc ∩ pixel, PM scale) and CurveHotPixelSnap (shared
snapped ray stamp) from the HP pin stack. Option B index lie stays
keyed to PrecisionModel scale. Six green pins.

Co-authored-by: Jeroen Bloemscheer <grootstebozewolf@users.noreply.github.com>

* spec: retip epic to MMF Option B + WKB 18-21 SIGN

Co-authored-by: Jeroen Bloemscheer <grootstebozewolf@users.noreply.github.com>

* spec: epic field contract names Option B and no silent linearize

Co-authored-by: Jeroen Bloemscheer <grootstebozewolf@users.noreply.github.com>

* feat: fold CurveSegmentDcel (P2.5.7) onto MMF tip

Package-private curve DCEL for PLG/COV consumers. Half-edges /
twins / next / face stay in jts-curve. Ten green Dcel tests.

Co-authored-by: Jeroen Bloemscheer <grootstebozewolf@users.noreply.github.com>

* fix: TB-FN #60 Exec stays on selected function, not Buffer

getMetaFunction prefers currentFunc over tree selection so focusing
the shared dX/Distance param field cannot re-bind Exec to
Buffer.buffer. Null-safe functionChanged; param fields request focus
on press. Refs #60.

Co-authored-by: Jeroen Bloemscheer <grootstebozewolf@users.noreply.github.com>

* feat: BUF-N stadium CompoundCurve buffer subset

Certified stadium CurvePolygon (and closed stadium CompoundCurve)
dilates with arc-preserving caps at r+d and re-joined sides. Open
mixed CompoundCurve corridors stay named miss / chainsaw.

Co-authored-by: Jeroen Bloemscheer <grootstebozewolf@users.noreply.github.com>

* fix: VBF honesty densify curve inputs via CurveOps.linearise

VariableBuffer samples chord vertices; curve A/B densify with the
explicit strategy warning instead of silent control-polygon flatten.
Full arc-length VBF TAG meter stays red.

Co-authored-by: Jeroen Bloemscheer <grootstebozewolf@users.noreply.github.com>

* spec: epic §4.0 MMF Option B landed table + MMF_OPTION_B.md

Document Option B, strategy, WKB 18–21, OFF/BUF, R-OV, HP, DCEL,
TB-FN #60, VBF honesty, and #56 test locks. README points at the
quality-gate note. CurvePolygonToolTest 16/16 green on tip.

Co-authored-by: Jeroen Bloemscheer <grootstebozewolf@users.noreply.github.com>

* doc: README cites WKB 18-21 zoo and MMF Option B gate

Co-authored-by: Jeroen Bloemscheer <grootstebozewolf@users.noreply.github.com>

* fix: CompoundCurve/CurvePolygon translate visits all member controls

Fold #59 apply honesty onto MMF tip: Affine/MoveTool translation no
longer drops later CircularString starts omitted from concatMembers.
LogoLines translate tests green.

Co-authored-by: Jeroen Bloemscheer <grootstebozewolf@users.noreply.github.com>

* feat: BUF-N open two-member line+arc CompoundCurve corridor

Buffer of COMPOUNDCURVE (LineString, CircularString) builds left/right
parallels with round joins and end caps as a CurvePolygon shell.
Ships BUF-N meter (stadium subset already landed).

Co-authored-by: Jeroen Bloemscheer <grootstebozewolf@users.noreply.github.com>

* spec: epic notes BUF-N open corridor ship

Co-authored-by: Jeroen Bloemscheer <grootstebozewolf@users.noreply.github.com>

* feat: BUF-N open CompoundCurve corridor for N members

Generalize open mixed buffer to any count of single-segment /
single-arc members with round joins between pieces and end caps.

Co-authored-by: Jeroen Bloemscheer <grootstebozewolf@users.noreply.github.com>

* feat: VBF arc-length densify via CurveOps.lineariseArcLength

CircularString/CompoundCurve feed VariableBuffer after equal-arc-length
sampling (not control-chord length). Strategy warning retained; meter
stays for arc-preserving variable offsets.

Co-authored-by: Jeroen Bloemscheer <grootstebozewolf@users.noreply.github.com>

* feat: DSF Densifier→toLinear + TRI-DT/VR curve site densify

Densifier.densify routes jts-curve Linearizable types through
toLinear so samples lie on the arc. Delaunay/Voronoi setSites use
the same path for curve-package geometries. Ship DSF, TRI-DT, TRI-VR
meters.

Co-authored-by: Jeroen Bloemscheer <grootstebozewolf@users.noreply.github.com>

* spec: epic notes DSF and TRI-DT/VR ships

Co-authored-by: Jeroen Bloemscheer <grootstebozewolf@users.noreply.github.com>

* feat: ship green meters + AT-NS densify + public CurveIntersection

Ship M-LEN-CS/CC, M-AREA-CP, M-DIM, F-MC/MS, B-CC, H-CV, R-EQ, AT-S,
AT-NS, D-PT, N-AA, N-AL with CurveAwarenessGreenMetersTest. Non-similarity
AffineTransformation densifies curve types. CurveIntersection wraps
arc-arc / arc-line solvers.

Co-authored-by: Jeroen Bloemscheer <grootstebozewolf@users.noreply.github.com>

* feat: F-CP compound rings, LRF-LEN arc extractPoint, ship B-CP/B-MS

CurveWKTReader parses ISO CURVEPOLYGON ((CIRCULARSTRING, ...)) as
CompoundCurve shell. LengthIndexedLine uses CurveOps analytical arc
length for curve types. Ship F-CP, B-CP, B-MS, LRF-LEN meters.

Co-authored-by: Jeroen Bloemscheer <grootstebozewolf@users.noreply.github.com>

* feat: C-LIN centroid, S-* preserve CS, ship OV/D-*/R-CONT/F-RD/TB-T

CircularString arc-length centroid (2R/π). Simplifiers keep 3-point
CircularString identity. Ship F-RD, D-AA, D-OP, R-CONT, OV, C-LIN,
C-AREA, V-CS, S-DP/VW/TP, TB-T meters with green verification.

Co-authored-by: Jeroen Bloemscheer <grootstebozewolf@users.noreply.github.com>

* feat: LRF-LOC CompoundCurve members, C-IP densify, PRC-SN preserve

LinearIterator treats CompoundCurve members as components (LRF-LOC).
InteriorPointArea densifies curve-package polygons before scanning (C-IP).
GeometryPrecisionReducer preserves CircularString when snapped centre
lies on the FIXED grid, else densifies then reduces (PRC-SN).

Co-authored-by: Jeroen Bloemscheer <grootstebozewolf@users.noreply.github.com>

* feat: V-CP densify validity, N-SS/R-PR meters, TB-FN badges

CurvePolygon.isValid densifies structural rings before IsValidOp.
Ship N-SS (CircularNodedSegmentString ARC) and R-PR DE-9IM meters.
Metadata.curveAwareness drives ●/◯/✕ badges in the function tree.

Co-authored-by: Jeroen Bloemscheer <grootstebozewolf@users.noreply.github.com>

* feat: VBF VariableBuffer densifies curves via arc-length

VariableBuffer.buffer reflects into CurveOps.lineariseArcLength so
distance interpolation follows arc parameter (strategy + warn).

Co-authored-by: Jeroen Bloemscheer <grootstebozewolf@users.noreply.github.com>

* feat: H-CC densify ConcaveHull sites; PLG densify Polygonizer edges

ConcaveHull densifies curve-package inputs at hull fraction 1e-4 so
sites follow the arc. Polygonizer.add densifies CompoundCurve /
CircularString linework before graph insert (faces remain Polygon).

Co-authored-by: Jeroen Bloemscheer <grootstebozewolf@users.noreply.github.com>

* feat: mark HullFunctions concaveHull length-ratio as curve passthrough

Co-authored-by: Jeroen Bloemscheer <grootstebozewolf@users.noreply.github.com>

* docs: refresh MMF_OPTION_B shipped meters and HOLD

Co-authored-by: Jeroen Bloemscheer <grootstebozewolf@users.noreply.github.com>

* feat: COV CurveCoverageUnion keeps exterior CIRCULARSTRINGs

Dissolve shared shell members across a hole-free CurvePolygon
coverage; stitch remaining edges into CurvePolygon/MultiSurface.
CoverageUnion reflects into CurveCoverageUnion before densify path.

Co-authored-by: Jeroen Bloemscheer <grootstebozewolf@users.noreply.github.com>

* docs: mark COV shipped in MMF_OPTION_B

Co-authored-by: Jeroen Bloemscheer <grootstebozewolf@users.noreply.github.com>

* feat: TestBuilder curve strategy picker + warn sink to Log tab

CurveLinearizationStrategy.WarnSink bridges densify warnings into the
TestBuilder Log panel. Edit menu toggles LINEARIZED vs PRESERVE.
Retip epic §4.0 with COV/PLG/H-CC meter ship status.

Co-authored-by: Jeroen Bloemscheer <grootstebozewolf@users.noreply.github.com>

* docs: retip epic to bd1dba0

Co-authored-by: Jeroen Bloemscheer <grootstebozewolf@users.noreply.github.com>

* docs: tip SHA + MMF TestBuilder WarnSink note

Co-authored-by: Jeroen Bloemscheer <grootstebozewolf@users.noreply.github.com>

* docs: sync epic tip SHA to 6f9ae15

Co-authored-by: Jeroen Bloemscheer <grootstebozewolf@users.noreply.github.com>

* feat: D-HF densify curve package inputs on general Hausdorff path

Certified apex/disc pairs keep closed form. Other curve operands
densify before the vertex path so control chords are not the sample.
Full TAG meter remains fail() by design.

Co-authored-by: Jeroen Bloemscheer <grootstebozewolf@users.noreply.github.com>

* docs: note D-HF densify honesty + folded #52/#55 pins

Co-authored-by: Jeroen Bloemscheer <grootstebozewolf@users.noreply.github.com>

* feat: fold #40 ClothoidOverlay kit + #26 HP.1 pin docs/tests

ClothoidOverlay answers 0-node identity/disjoint/nest cells; Fresnel
misses return null (never chord-flatten). Wire through
CompoundCurveShellOverlay; refuse clothoid members in TwoNodeClip /
CircularLineOverlay flatten. HP.1 wrong-ring walk pin docs + tests.

Co-authored-by: Jeroen Bloemscheer <grootstebozewolf@users.noreply.github.com>

* docs: note #40 ClothoidOverlay and #26 HP.1 fold

Co-authored-by: Jeroen Bloemscheer <grootstebozewolf@users.noreply.github.com>

* docs: cherry-pick #47 LaTeX guide sources onto MMF tip

Bring doc/latex/ (Makefile, three manuals, figures, sty) without
replacing conflicting Amyuni PDF binaries. Point MMF_OPTION_B at
the rebuild path; retip latex README to PR #61.

Co-authored-by: Jeroen Bloemscheer <grootstebozewolf@users.noreply.github.com>

* docs: epic notes doc/latex cherry-pick from #47

Co-authored-by: Jeroen Bloemscheer <grootstebozewolf@users.noreply.github.com>

* docs: retip latex README to MMF PR #61

Co-authored-by: Jeroen Bloemscheer <grootstebozewolf@users.noreply.github.com>

* feat: WKB 18-21 inspect labels + curve strategy status strip

GeometryLocationsWriter labels Bezier/Ellipse/NURBS for mouse inspect.
Edit-menu strategy picker also updates the bottom status bar; frame
startup shows LINEARIZED.

Co-authored-by: Jeroen Bloemscheer <grootstebozewolf@users.noreply.github.com>

* feat: show Curve strategy LINEARIZED on TestBuilder status at startup

Co-authored-by: Jeroen Bloemscheer <grootstebozewolf@users.noreply.github.com>

* test: WKB 18-21 ISO Z round-trips + Option B PM kind stability

Lock Phase 4 accept (XY + ISO Z for 18–21). Assert PrecisionModel
coordinate snap does not rename SegmentKind.ARC (allowed index lie).

Co-authored-by: Jeroen Bloemscheer <grootstebozewolf@users.noreply.github.com>

* docs: Phase 4 ISO-Z and PM kind-stability accept criteria

Co-authored-by: Jeroen Bloemscheer <grootstebozewolf@users.noreply.github.com>

* docs+fix: Option B prose in guides; clothoid ISO M/ZM honesty

Update technical-specs and developer-guide for SegmentKind, no-silent
linearize, and WKB 18-21. Preserve Z/M on ClothoidSegment start so ISO
2018/3018 type words round-trip.

Co-authored-by: Jeroen Bloemscheer <grootstebozewolf@users.noreply.github.com>

* docs: user-guide Option B prose + MMF verification walkthrough

Document TestBuilder strategy/badges/WKB 18-21 in user-guide.tex.
Add doc/MMF_WALKTHROUGH.md with smoke evidence and HOLD checklist.

Co-authored-by: Jeroen Bloemscheer <grootstebozewolf@users.noreply.github.com>

* docs: retip epic MMF ship status + explicit named-miss list

Correct stale OFF/BUF/H-CC/N-*/PLG/COV 'still open' claims against
tip meters. List leftover named misses for Phase 2 accept.

Co-authored-by: Jeroen Bloemscheer <grootstebozewolf@users.noreply.github.com>

* docs: sync epic tip SHA to 0cf0eb3

Co-authored-by: Jeroen Bloemscheer <grootstebozewolf@users.noreply.github.com>

* fix: CI red — SqlMmTypes HOLD 15-17, CurveWKTWriter owns CurvePolygon, BiteVsHole honesty

- Allow Triangle/Tin/PolyhedralSurface through WKTWriter (HOLD 15-17).
- CurveWKTWriter always formats CurvePolygon (EMPTY / all-linear).
- BiteVsHole refuses densified outer CAP; update H-SHELL-HOLE-*
  assertions to certified arc-preserving bites.

Co-authored-by: Jeroen Bloemscheer <grootstebozewolf@users.noreply.github.com>

* fix: restore logoBuffer JOIN_MITRE + mitre limit (CI app tests)

Co-authored-by: Jeroen Bloemscheer <grootstebozewolf@users.noreply.github.com>

* fix: headless-safe AppCursors for CI drawing-tool tests

Co-authored-by: Jeroen Bloemscheer <grootstebozewolf@users.noreply.github.com>

---------

Co-authored-by: Cursor Agent <cursoragent@cursor.com>
Co-authored-by: Jeroen Bloemscheer <grootstebozewolf@users.noreply.github.com>
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants