v0.5.1

[0.5.1] - 2026-11-09: cookbook for the geometric checks, SPDX 3.0 SBOM

Patch release. No API or behavior change.

Changed

• README: copy-paste examples for HasAngleBetweenApproximately, the scale-invariant IsParallelTo, IsNormalized (vector and quaternion), and IsRotation; a note that only PoseRenderer needs using MathAssertions.Render;; and a corrected sentence that still called per-axis delta rendering future work, although it shipped in 0.2.0.
• SBOM: both packages embed an SPDX 3.0 software bill of materials (SPDX 2.2 before), like the other packages of the family.
• TUnit.Assertions and TUnit.Core minimum versions raised from 1.54.0 to 1.81.10, and System.Numerics.Tensors from 10.0.9 to 10.0.14.
• Package metadata: the project URL of both packages now points to https://mathassertions.dev, and the NuGet tags now include testing and dotnet.
• README: now says that adapters for other test frameworks (NUnit, xUnit, MSTest) are not planned; the family targets TUnit.
• Core README: now says that adapters for other test frameworks (NUnit, xUnit, MSTest) are not planned; the family targets TUnit.
• README (namespaces): now lists the real namespaces (MathAssertions.TUnit, TUnit.Assertions.Extensions, MathAssertions, MathAssertions.Geometry3D, MathAssertions.Render).
• README (using directives): now says that the geometry samples need using MathAssertions.Geometry3D;, and no longer says that System.Numerics is the only extra using. The package README says the same.
• README (adapter classes): now counts thirteen adapter classes instead of twelve, in the README and the package README, and no longer calls the geometry surface "complete".
• README (IsCloseTo samples): the Vector3.Distance and per-axis samples now cast the float values to double so they compile.
• README (protobuf sample): the projection now casts its fields to float.
• README (pose projection): now warns that the cast from double to float can turn distinct positions into the same Vector3, so a pose assertion can pass despite a position error, and says to compare the original double fields when that precision matters.
• README (failure diagnostics): now describes what the failure messages contain, per compound type, instead of showing an exact layout. The package README says the same.
• README (failure diagnostics, TUnit wrapper): now says that TUnit prints Expected ... and but ... around the message, with "to satisfy" plus the method as the expectation.
• README (NaN scope): the NaN and infinity claims are now scoped to the tolerance comparisons, and the README says that pose position, pointcloud on-plane and on-sphere, convex-hull containment, and monotonic-sequence checks do not reject NaN. The package and core READMEs say the same.
• README (NaN table): now documents NaN vs NaN as true under any valid tolerance, in the README and the package README.
• README (infinity table): now says that the generic ReadOnlyTensorSpan<T> overload does not follow the infinity rows: under a tolerance of positive infinity it returns true for opposite-sign infinities and for an infinity against a finite value. The core README says the same.
• README (tolerance precision): now says that the double tolerance is kept and the component difference is computed in double, instead of claiming that a cast to float would round 1e-9 to zero.
• README (scalar assertions): now says that ScalarAssertions has fluent IsApproximatelyEqualTo for double and float.
• Package README (ScalarAssertions): now says that only IsApproximatelyEqualTo and IsCloseInUlpsTo have float overloads.
• README (IsParallelTo): now says that the scale-invariant check applies to vectors longer than the tolerance, that a shorter vector counts as parallel to anything, and that the check holds only while both lengths and their product stay below about 1.8e19, because the lengths and the cross product are computed in float.
• README (IsRotation): the cookbook comment now says that two checks must both pass, M * M^T = I and determinant +1, each within the tolerance: a reflection fails under any tolerance below 2; in an affine matrix (last column 0, 0, 0, 1) a translation or a scale factor other than +1 or -1 fails unless both checks stay within the tolerance; CreateScale(-1, -1, 1) passes as a 180-degree rotation; and the whole 4x4 matrix is checked, not the affine form, so a 4D rotation that mixes W with X, Y or Z passes.
• README (number theory): now says that IsDivisibleBy(long.MinValue, -1) throws OverflowException, next to the long.MinValue-aware GCD and LCM claim.
• Core README (static classes): now lists PoseRenderer (MathAssertions.Render) next to the five static classes, and no longer says "six static classes" followed by five names.
• Core README (stability): now says that breaking changes before 1.0 bump the minor version, like the README.
• Core README (failure messages): now describes the failure-message renderer as internal and unstable.
• MathTolerance.IsApproximatelyEqual: the Vector3 XML docs now say that the double tolerance is kept and the difference is computed in double.
• MathTolerance.IsApproximatelyEqual<T> (ReadOnlyTensorSpan<T>): the XML docs now say that infinities are compared through T.Abs(a - b) > tolerance instead of the scalar infinity rules, which differs from the scalar overloads under a tolerance of positive infinity.
• MathTolerance.IsCloseInUlps: the XML docs now say that the distance counts steps between adjacent representable values (adjacent values are distance 1) and that values of opposite sign, even double.Epsilon and its negation, return false.
• MathTolerance.RotationAngleDegrees: the XML docs now say that a quaternion compared with itself yields zero or a value within about 1e-14 degrees, instead of exactly zero.
• MathTolerance.ExtractAxisAngle: the XML docs now say that the extracted angle lies in [-180, 180] degrees and that its sign is aligned by the axis dot product. The internal failure-message renderer docs say the same.
• Sequences.IsArithmeticProgression: the XML docs, and those of the adapter method, now say that every difference is compared with the first one, not with the adjacent one.
• Sequences.IsGeometricProgression: the XML docs, and those of the adapter method, now say that every ratio is compared with the first one, not with the adjacent one.
• LinearAlgebra.AreParallel: the XML docs now say that the check is scale-invariant only for vectors longer than the tolerance, and only while both lengths and their product stay below about 1.8e19 (float range).
• LinearAlgebra.IsRotation and LinearAlgebra.IsOrthogonal: the XML docs, and those of the adapter method IsRotation, now say that both checks of IsRotation must pass within the tolerance, that a reflection fails only under a tolerance below 2, that in an affine matrix a translation or a scale fails only once one of the checks exceeds the tolerance, that a scale of -1 keeps a matrix orthogonal (so CreateScale(-1, -1, 1) is a rotation), and that the affine form is not checked, so a 4D rotation that mixes W with X, Y or Z passes.
• NumberTheory.IsDivisibleBy: the XML docs now say that long.MinValue with a divisor of -1 throws OverflowException.
• Intersection (class): the XML docs now say that only ray-plane and ray-triangle have an out float t overload, that ray-sphere and ray-AABB are boolean only, and that t is the ray parameter of the hit point origin + t * direction, a distance only for a unit direction.
• Intersection.Intersects(Ray3D, Plane, out float): the XML docs now say that t is 0 for a parallel ray and the computed negative t when the plane is behind the ray.
• Intersection.Intersects(Ray3D, Triangle3D, out float): the XML docs now say that the hit requires the ray parameter t > 1e-7 (a threshold on the parameter, not on a distance), so an origin on the triangle is not a hit, and what t is in each false case.
• Intersection.Intersects(Ray3D, Sphere): the XML docs now say that the ray direction must be a unit vector.
• Ray3D: the XML docs now say that ray-sphere needs a unit direction and that the parallel tests use an absolute epsilon of 1e-7 (ray-plane on |direction . normal|, ray-triangle on |direction . (edge2 x edge1)|, ray-AABB per direction component, where a near-zero component only requires the origin to lie inside that axis's slab).
• PoseAssertions: the XML docs now say that the Matrix4x4 overload assumes unit scale instead of saying that scale is ignored.
• Containment.ConvexHullContains: the XML docs no longer promise automatic triangulation in 0.2.0.
• Statistics.HasPercentileApproximately: the XML docs now describe the rank p/100 * (N - 1) (NumPy "linear", Excel PERCENTILE.INC) instead of citing NIST/SEMATECH.