[
https://issues.apache.org/jira/browse/GEOMETRY-163?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Aryan Ganjoo updated GEOMETRY-163:
----------------------------------
Description:
Applying a shear with EmbeddingPlane.transform produces a non-orthogonal basis,
so points on the plane no longer round-trip through toSubspace and toSpace.
This transform is finite and invertible (determinant 1).
{code:java}
var precision = Precision.doubleEquivalenceOfEpsilon(1e-10);
var plane = Planes.fromPointAndPlaneVectors(
Vector3D.ZERO, Vector3D.Unit.PLUS_X, Vector3D.Unit.PLUS_Y, precision);
var shear = AffineTransformMatrix3D.of(
1, 1, 0, 0,
0, 1, 0, 0,
0, 0, 1, 0);
var transformed = plane.transform(shear);
var point = Vector3D.of(1, 0, 0);
System.out.println(transformed.getU().dot(transformed.getV()));
System.out.println(transformed.toSpace(transformed.toSubspace(point)));
{code}
The dot product is about 0.7071 and the round-trip point is (1.5, 0.5, 0.0),
rather than (1.0, 0.0, 0.0). Using plane.subspaceTransform(shear).getPlane()
gives an orthogonal basis and the expected round trip.
Reproduced on master at 08664182d6c8f8a69da67f03c98e4c7bf6aa193b
(1.1-SNAPSHOT). I am working on a fix with regression tests and plan to submit
a PR. Please coordinate here before starting a parallel fix.
Report prepared with AI assistance.
was:
Applying a shear with EmbeddingPlane.transform produces a non-orthogonal basis,
so points on the plane no longer round-trip through toSubspace and toSpace.
This transform is finite and invertible (determinant 1).
{code:java}
var precision = Precision.doubleEquivalenceOfEpsilon(1e-10);
var plane = Planes.fromPointAndPlaneVectors(
Vector3D.ZERO, Vector3D.Unit.PLUS_X, Vector3D.Unit.PLUS_Y, precision);
var shear = AffineTransformMatrix3D.of(
1, 1, 0, 0,
0, 1, 0, 0,
0, 0, 1, 0);
var transformed = plane.transform(shear);
var point = Vector3D.of(1, 0, 0);
System.out.println(transformed.getU().dot(transformed.getV()));
System.out.println(transformed.toSpace(transformed.toSubspace(point)));
{code}
The dot product is about 0.7071 and the round-trip point is (1.5, 0.5, 0.0),
rather than (1.0, 0.0, 0.0). Using plane.subspaceTransform(shear).getPlane()
gives an orthogonal basis and the expected round trip.
Reproduced on master at 08664182d6c8f8a69da67f03c98e4c7bf6aa193b
(1.1-SNAPSHOT). Happy to contribute a fix and regression tests.
> EmbeddingPlane.transform loses orthogonality under shear
> --------------------------------------------------------
>
> Key: GEOMETRY-163
> URL: https://issues.apache.org/jira/browse/GEOMETRY-163
> Project: Commons Geometry
> Issue Type: Bug
> Environment: Java 17, macOS arm64; Commons Geometry 1.1-SNAPSHOT at
> 08664182d6c8f8a69da67f03c98e4c7bf6aa193b.
> Reporter: Aryan Ganjoo
> Priority: Major
>
> Applying a shear with EmbeddingPlane.transform produces a non-orthogonal
> basis, so points on the plane no longer round-trip through toSubspace and
> toSpace. This transform is finite and invertible (determinant 1).
> {code:java}
> var precision = Precision.doubleEquivalenceOfEpsilon(1e-10);
> var plane = Planes.fromPointAndPlaneVectors(
> Vector3D.ZERO, Vector3D.Unit.PLUS_X, Vector3D.Unit.PLUS_Y, precision);
> var shear = AffineTransformMatrix3D.of(
> 1, 1, 0, 0,
> 0, 1, 0, 0,
> 0, 0, 1, 0);
> var transformed = plane.transform(shear);
> var point = Vector3D.of(1, 0, 0);
> System.out.println(transformed.getU().dot(transformed.getV()));
> System.out.println(transformed.toSpace(transformed.toSubspace(point)));
> {code}
> The dot product is about 0.7071 and the round-trip point is (1.5, 0.5, 0.0),
> rather than (1.0, 0.0, 0.0). Using plane.subspaceTransform(shear).getPlane()
> gives an orthogonal basis and the expected round trip.
> Reproduced on master at 08664182d6c8f8a69da67f03c98e4c7bf6aa193b
> (1.1-SNAPSHOT). I am working on a fix with regression tests and plan to
> submit a PR. Please coordinate here before starting a parallel fix.
> Report prepared with AI assistance.
--
This message was sent by Atlassian Jira
(v10.3.26#10030026)
