Open
Description
Describe the bug
A build error is thrown when building the Dockerfile. It is caused by these lines, which access conda environment variables.
To fix this behavior
if is_conda_env:
# Set CUDA_INC_PATH from the appropriate conda target cross-compilation platform directory
# NOTE: This strategy will potentially have to change when compiling for different platforms but by then we will likely not be using setuptools…
target_platform_include_dir = (
Path(os.getenv("CONDA_PREFIX")) / "targets" / f"{platform.machine()}-{platform.system().lower()}" / "include"
)
# The cuda-toolkit headers (and other '-dev' package headers) from the packages on the `conda-forge` channel are installed in the `targets` directory
# which is to support cross-compilation for different platforms. The headers are installed in the appropriate target platform directory.
if (target_platform_include_dir / "cuda.h").exists():
os.environ["CUDA_INC_PATH"] = str(target_platform_include_dir)
To Reproduce
Steps to reproduce the behavior:
- Build with
docker build -t fvdb/prod .
Expected behavior
(A clear and concise description of what you expected to happen.)
Additional context
(Add any other context about the problem here.)