Buggy GCC 14.3.0 bounds-check #1812
Description
I updated my system (nixos) and am now unable to build.
[ 0%] Building CXX object _deps/jolt-build/CMakeFiles/Jolt.dir/__/Jolt/Physics/Body/MassProperties.cpp.o
In file included from build/_deps/jolt-src/Build/../Jolt/Math/Matrix.h:7,
from build/_deps/jolt-src/Jolt/Physics/Body/MassProperties.cpp:8:
In member function ‘float& JPH::Vector<Rows>::operator[](JPH::uint) [with unsigned int Rows = 3]’,
inlined from ‘bool JPH::MassProperties::DecomposePrincipalMomentsOfInertia(JPH::Mat44&, JPH::Vec3&) const’ at build/_deps/jolt-src/Jolt/Physics/Body/MassProperties.cpp:44:51:
build/_deps/jolt-src/Build/../Jolt/Math/Vector.h:48:28: error: array subscript 3 is above array bounds of ‘float [3]’ [-Werror=array-bounds=]
48 | return mF32[inCoordinate];
| ~~~~^
build/_deps/jolt-src/Build/../Jolt/Math/Vector.h: In member function ‘bool JPH::MassProperties::DecomposePrincipalMomentsOfInertia(JPH::Mat44&, JPH::Vec3&) const’:
build/_deps/jolt-src/Build/../Jolt/Math/Vector.h:208:57: note: while referencing ‘JPH::Vector<3>::mF32’
208 | float mF32[Rows];
| ^~~~
I solved the static analysis issue locally by adding a manual bounds check, but this is clearly a GCC bug.
diff --git a/Jolt/Physics/Body/MassProperties.cpp b/Jolt/Physics/Body/MassProperties.cpp
index 91df6b0d..eaeeba86 100644
--- a/Jolt/Physics/Body/MassProperties.cpp
+++ b/Jolt/Physics/Body/MassProperties.cpp
@@ -40,6 +40,8 @@ bool MassProperties::DecomposePrincipalMomentsOfInertia(Mat44 &outRotation, Vec3
outRotation = Mat44::sIdentity();
for (int i = 0; i < 3; ++i)
{
+ if (indices[i] >= 3 || indices[i] < 0)
+ continue; // unreachable, but gcc bounds checking is buggy
outRotation.SetColumn3(i, Vec3(reinterpret_cast<Float3 &>(eigen_vec.GetColumn(indices[i]))));
outDiagonal.SetComponent(i, eigen_val[indices[i]]);
}I am not sure how to proceed, but decided to post this here anyway, in case someone else comes across this issue too.
edit: Also I could have sworn I saw this bug in a c++ talk video recently...