src/core/cuda_env.cpp

This file is the native CUDA-toolkit probe. It looks for `nvcc` on PATH first, falls back to `CUDA_HOME`, `CUDA_PATH`, `/usr/local/cuda`, `/opt/cuda`, and versioned sibling directories, then parses the CUDA release string from `nvcc --version` output.

cuda_env.hpp process.cpp configure.cpp

Behavior

Collects candidate roots from environment variables and common install directories.
Searches for `bin/nvcc` under those roots when PATH discovery fails.
Parses `release X.Y` from the `nvcc --version` output.
Reports `missing` when `nvcc` cannot be found and `issue` when the command exists but cannot be parsed cleanly.

Caveats

This implementation proves that a toolkit is visible, not that the toolkit can target the local GPU architecture or run kernels successfully. That deeper validation still belongs to future validation and architecture-aware logic.

include/core/cuda_env.hpp

include/core/configure.hpp

Related files

Core contract

include/core/cuda_env.hpp

Declares the toolkit detection probe plus helper functions that locate `nvcc` and derive a toolkit root.

Core implementation

src/core/process.cpp

Implements PATH scanning, shell command execution, output trimming, and exit code decoding.

Core implementation

src/core/configure.cpp

Writes `.cuda-doctor.env` with `CUDA_HOME`, `PATH`, and `LD_LIBRARY_PATH` for the discovered toolkit root.