CNA-BUG-026: Matrix::CreatePerspectiveFieldOfView accepts a field of view of exactly MathHelper::Pi

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

The upper guard compares against the literal 3.141593f, which rounds to the float one step above MathHelper::Pi, so a field of view of exactly Pi passes validation and builds a degenerate projection instead of throwing as XNA does.

Identifier

CNA-BUG-026

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

Matrix::CreatePerspectiveFieldOfView(float fieldOfView, float aspectRatio, float nearPlaneDistance, float farPlaneDistance) and its output-parameter form; C ABI cna_matrix_create_perspective_field_of_view

Expected behaviour

A field of view must be strictly between 0 and π. XNA 4.0's guard compares against 3.14159274f, which is exactly MathHelper.Pi, and throws for fieldOfView >= Pi (read from the IL of the genuine XNA 4.0 Microsoft.Xna.Framework assembly; not executed). CNA's own message states the same rule: "fieldOfView <= 0 or >= PI".

Actual behaviour at TARGET

Matrix.cpp tests fieldOfView >= 3.141593f. The decimal literal rounds to the single-precision value 3.14159298 (0x40490FDC), one step above MathHelper::Pi = 3.14159274 (0x40490FDB), so CreatePerspectiveFieldOfView(MathHelper::Pi, …) passes. 1.0f / std::tan(fieldOfView * 0.5f) then evaluates the tangent just past π/2, a large negative number, so M22 and M11 become tiny negative values (about -4.4e-8): the projection flips and collapses the image instead of throwing. Exactly one float value is admitted; FNA uses the same literal.

Separately, the thrown type is std::invalid_argument where XNA throws ArgumentOutOfRangeException; that is part of the mixed exception families recorded in its own entry.

Source locations

• modules/math/src/Matrix.cpp — Matrix::CreatePerspectiveFieldOfView(float, float, float, float, Matrix&): the 3.141593f guard
• modules/math/include/Microsoft/Xna/Framework/MathHelper.hpp — MathHelper::Pi = 3.14159274f
• modules/math/tests/Microsoft/Xna/Framework/MatrixTests.cpp — MatrixTest.CreatePerspectiveFieldOfViewThrowsBadFov tests only a zero field of view

Evidence

Checked by reading at 009d40f5; nothing was built or executed for this entry. The IEEE-754 single-precision rounding of the literal and of π, and the tangent value quoted above, were computed outside CNA (in double precision from the float argument); CNA itself was not built or run. The XNA constant was read from its IL, where the same assembly declares MathHelper.Pi as 3.14159274.

Focused reproduction

Illustrative; not compiled or run for this entry.

Matrix m;
EXPECT_THROW(Matrix::CreatePerspectiveFieldOfView(MathHelper::Pi, 1.0f, 1.0f, 100.0f, m),
             std::invalid_argument);          // fails at TARGET: nothing is thrown

Current tests

MatrixTest.CreatePerspectiveFieldOfViewThrowsBadFov in MatrixTests.cpp covers only fieldOfView = 0; no test touches the upper bound. The C ABI smoke test uses valid values only.

Regression test

Expect the exception at MathHelper::Pi and at std::nextafter(MathHelper::Pi, 4.0f), and no exception at std::nextafter(MathHelper::Pi, 0.0f). The fix is to compare against MathHelper::Pi itself.

Blast radius

Only CreatePerspectiveFieldOfView (both overloads) and its C ABI route, and only for the single input value π; a caller passing it gets a degenerate matrix instead of an error.

Workaround

Keep the field of view below MathHelper::Pi before calling; a field of view that wide is unusable anyway.