Rework warning when compiler is installed in environment but activation scripts aren't

[kkraus14]

Solution to issue cannot be found in the documentation.

• I checked the documentation.

Issue

Currently, if you use the cuda-nvcc_{platform} package in an environment, it will try to detect if you don't have host compilers installed and warn you. This happens here:

cuda-nvcc-feedstock/recipe/activate.sh

Lines 33 to 35 in a1a69ae

	if [ -z "${CXX+x}" ]; then
	echo 'cuda-nvcc: Please add the `compiler("c")` and `compiler("cxx")` packages to the environment.'
	ERROR=true

.

This makes sense when in the context of a conda build but not for a user in an environment. For a user in an environment, we should probably point them towards the c-compiler and cxx-compiler packages to maintain the current behavior.

Additionally, I intentionally avoid the c-compiler and cxx-compiler packages in my development environment because they set a bunch of compiler flags that I don't necessarily want (see https://github.com/conda-forge/ctng-compiler-activation-feedstock/blob/main/recipe/conda_build_config.yaml and https://github.com/conda-forge/ctng-compiler-activation-feedstock/blob/main/recipe/activate-gcc.sh). I do have the gcc and gxx packages installed which give me a set of host compilers to use.

I've worked around this for now by setting the CXX and CC environment variables to an activation script in my environment with something that looks like:

export CC=`which gcc`
export CXX=`which g++`

Installed packages

n/a

Environment info

n/a

[jakirkham]

Thanks for raising this Keith! 🙏

Since this was opened, we added cuda-compiler for this purpose, which depends on c-compiler and cxx-compiler

Alternatively there is a lighter weight cuda-nvcc, which includes the host C and C++ compilers. This may be closer to your workflow

Please let us know what you think of these options 🙂