src/commands/check.cpp

The `check` command is intentionally small. It detects the platform first, then collects driver, CUDA, and GPU probes, and finally collapses those probe statuses into a single `overall` status.

check.hpp platform.cpp driver.cpp cuda_env.cpp gpu.cpp

Behavior

Runs the probes in a fixed order: platform, driver, CUDA, GPU.
Marks the whole report unsupported immediately if any probe returns `unsupported`.
Otherwise prefers `issue` over `missing`, and `missing` over `ok` when collapsing the final status.

Why it matters

`run_check()` is the foundation under `doctor`. That keeps `doctor` free to focus on repair logic and next-step generation instead of reimplementing base probe collection.

include/commands/check.hpp

include/commands/doctor.hpp

Related files

Command contract

include/commands/check.hpp

Declares the `run_check()` entrypoint that collects the baseline platform, driver, CUDA, and GPU probes.

Core implementation

src/core/platform.cpp

Determines OS and CPU architecture at compile time and marks macOS as unsupported for NVIDIA CUDA runtime use.

Core implementation

src/core/driver.cpp

Checks for `nvidia-smi` and, when available, queries the GPU driver version.

Core implementation

src/core/cuda_env.cpp

Finds `nvcc`, searches common CUDA install roots, parses `nvcc --version`, and derives the toolkit root directory.

Core implementation

src/core/gpu.cpp

Uses `nvidia-smi` to ask for the GPU name and treats the driver tool as a prerequisite, not as proof that no GPU exists.