Runtime failures due to LLVM hardening asserts when compiled with LLVM

Polytope fails at the run-time when compiled with LLVM because you access elements past the end of arrays.

Program received signal SIGILL, Illegal instruction.
Privileged opcode.
std::__1::vector<double, std::__1::allocator<double> >::operator[][abi:se190107](unsigned long) (this=0x7fffffffe248, __n=1338) at /usr/include/c++/v1/vector:1436
1436      _LIBCPP_ASSERT_VALID_ELEMENT_ACCESS(__n < size(), "vector[] index out of bounds");
(gdb) up
#1  0x00000008004b63cd in polytope::MeshEditor<2, double>::cleanMesh (this=0x7fffffffe1f0, cellMap=std::vector of length 279 = {...}, faceMap=std::vector of length 948 = {...}, nodeMap=std::vector of length 669 = {...})
    at /usr/ports/science/polytope/work/polytope-0.7.3-18-gcf83206/src/MeshEditor.cc:664
664           std::copy(&mMesh.nodes[Dimension*i], &mMesh.nodes[Dimension*(i+1)], back_inserter(newNodes));

 660│   std::vector<RealType> newNodes;
 661│   newNodes.reserve(Dimension*nnodes1);
 662│   for (unsigned i = 0; i != nnodes0; ++i) {
 663│     if (mNodeMask[i] == 1) {
 664├─────> std::copy(&mMesh.nodes[Dimension*i], &mMesh.nodes[Dimension*(i+1)], back_inserter(newNodes));
 665│     }
 666│   }
 667│   mMesh.nodes = newNodes;

We had to use this workaround to get rid of these failures:

CXXFLAGS+=      -D_LIBCPP_HARDENING_MODE=_LIBCPP_HARDENING_MODE_NONE # prevent hardening issues with libc++

llvm-19
FreeBSD 15 STABLE

Reproduce the failure with LLVM 19 on FreeBSD 15 and inspect src/MeshEditor.cc around line 664 in MeshEditor::cleanMesh. Check the node-mask and mesh-node sizes used by the copy, then verify the runtime no longer fails with libc++ hardening enabled and the workaround is unnecessary.