CNA-BUG-074: Vector, Matrix, Quaternion and CurveKey hash codes use raw float bits, so +0.0f and -0.0f components compare equal but hash differently

CNA snapshot 009d40f5  ·  Known Issues › Current bugs  ·  source links pinned to 009d40f5

FloatHash returns the IEEE bit pattern, so Equals and GetHashCode disagree for values with a negative-zero component; .NET's Single.GetHashCode returns 0 for both zeros, keeping XNA's hashes consistent with Equals.

Identifier

CNA-BUG-074

Category

Bug

Subsystem

Math & geometry

Status

Open

Verified against

CNA 009d40f5 (009d40f5dd085c4e674d3479675fac84b12b3e0a)

Severity

Low (a triage suggestion, not a project priority)

Evidence basis

Source-verified: read at TARGET, not executed

Tests touching this area

Yes: see Current tests

Affected contract

GetHashCode() of Vector2, Vector3, Vector4, Matrix, Quaternion and CurveKey

Expected behaviour

The Equals/GetHashCode contract: equal values have equal hashes. .NET Framework's Single.GetHashCode returns 0 when the value is 0, so XNA's component-sum hashes agree for +0 and -0.

Actual behaviour at TARGET

Each of Vector3.cpp, Vector2.cpp, Vector4.cpp, Matrix.cpp, Quaternion.cpp and CurveKey.cpp has a local FloatHash that memcpys the float into a uint32_t. Vector3(0,0,0) == Vector3(-0.0f,0,0) is true, yet the hashes differ by 0x80000000.

Source locations

• modules/math/src/Vector3.cpp — FloatHash and Vector3::GetHashCode
• modules/math/src/Matrix.cpp — FloatHash and Matrix::GetHashCode
• modules/math/src/Quaternion.cpp — FloatHash and Quaternion::GetHashCode

Evidence

Checked by reading the named sources at 009d40f5; nothing was built or executed for this entry. .NET behaviour from the mscorlib implementation of Single.GetHashCode.

Focused reproduction

// Illustrative; not compiled or run for this entry.
Vector3 a(0.0f, 1.0f, 2.0f), b(-0.0f, 1.0f, 2.0f);
bool eq = a.Equals(b);                       // true
bool sameHash = a.GetHashCode() == b.GetHashCode();   // false in CNA, true in XNA

Current tests

ObjectEqualityContractTests.cpp exists, but no case uses a negative zero.

Regression test

Add a signed-zero pair to ObjectEqualityContractTests for each type.

Blast radius

Hash containers keyed on vectors or matrices (for example a vertex-welding map) that can receive -0.0f from negation or rounding.

Workaround

Normalise -0.0f to 0.0f before hashing.