NanoVDB HDDA.h triggers -Wmaybe-uninitialized with GCC 14.2.1

Summary

Building with GCC 14.2.1 (Red Hat gcc-toolset-14 on Rocky Linux 8) and -Werror produces a -Wmaybe-uninitialized warning in nanovdb/math/HDDA.h at line 171.

Error

nanovdb/nanovdb/math/HDDA.h:171:28: error: '*(nanovdb::math::Vec3<float>*)((char*)&it + offsetof(fvdb::HDDASegmentIterator<nanovdb::ReadAccessor<nanovdb::ValueOnIndex, 0, 1, 2>, float>,fvdb::HDDASegmentIterator<nanovdb::ReadAccessor<nanovdb::ValueOnIndex, 0, 1, 2>, float>::mHdda.nanovdb::math::HDDA<nanovdb::math::Ray<float>, nanovdb::math::Coord>::mDelta)).nanovdb::math::Vec3<float>::mVec[0]' may be used uninitialized [-Werror=maybe-uninitialized]
  171 |         mNext[ axis] += mDim * mDelta[axis];

The mDelta member of HDDA is flagged as potentially uninitialized when used in this expression.

Environment

• Compiler: GCC 14.2.1 (gcc-toolset-14 on Rocky Linux 8)
• NanoVDB version: v32.9.1 (commit 0c4c067bd5f23f0c1b05ab224faa0ad89f73a4b9 via CPM)
• Build type: Release (with -Wall -Werror)
• CUDA: NVIDIA nvcc 13.0.88 with GCC 14.2.1 as host compiler (via -Xcompiler=-Wall,-Werror)

Notes

We have confirmed that conda-forge's GCC 14.2.0 does not trigger this warning on the same code, so it appears specific to the 14.2.1 patch level. -Wmaybe-uninitialized false positives are a known class of GCC issue.

Workaround

We are suppressing with -Wno-maybe-uninitialized in our downstream build system.

Suggested Fix

Consider either:

1. Explicitly initializing mDelta in the HDDA constructor/init() to silence the warning
2. Adding a targeted #pragma GCC diagnostic ignored "-Wmaybe-uninitialized" around the affected code

Start with nanovdb/math/HDDA.h around line 171 and inspect how mDelta is initialized in the HDDA constructor or init(). Reproduce the warning with GCC 14.2.1, Release, and -Wall -Werror, including the CUDA host-compiler setup if relevant. Done means the affected build no longer emits -Wmaybe-uninitialized without relying on the downstream suppression.