cuda_compiler_version is "None" even in cuda builds

cuda_compiler_version correctly gets inserted correctly into the CI jobs of staged-recipes:

++ cat /home/conda/staged-recipes-copy/.ci_support/linux64_cuda110.yaml

c_compiler:
- gcc
cxx_compiler:
- gxx
fortran_compiler:
- gfortran
c_compiler_version:
- 9
cxx_compiler_version:
- 9
fortran_compiler_version:
- 9
cudnn:
- 8
cdt_name:
- cos7
target_platform:
- linux-64
channel_sources:
- conda-forge
docker_image:
- quay.io/condaforge/linux-anvil-cuda:11.0
cuda_compiler_version:
- 11.0

but when the meta.yaml is rendered, the value isn’t there.

I puzzled for a while why

  skip: true  # [not linux]
  skip: true  # [cuda_compiler_version == "None"]

ended up skipping all cuda builds, until I removed the second line and ran into resolver errors:

Mamba failed to solve:
 - nvcc_linux-64 None.*
 - python 3.8.* *_cpython
 - glog
 - gxx_linux-64 9.*
 - pybind11
 - gflags 2.2.*
 - gflags
 - python
 - cmake >=3.12
 - glog 0.5.*

[h-vetinari]

Edit: You should also be able to choose a blank issue if nothing fits.

Never noticed that one, cool thanks!

And also for adding the bug template! 🙃

[h-vetinari]

How about “bugs in staged-recipes” or “staged-recipes CI issues”?