test_codegen.py can be executed on CI as it does not require the rest of pyflamegpu / libcuda.so to be available at runtime.
This is of very little use however, as it doesn't test that the generated code is actually valid as that would require nvrtc, and implicitly libcuda.so.
I.e it hasn't caught the python 3.8 bug introduced in #1147 (without a test case adding with that particular shape?)
[ ] Check that conftest still behaves correctly for normal pyflamegpu test runs (i.e. sensible telemetry behaviour).
test_codegen.py can be executed on CI as it does not require the rest of pyflamegpu / libcuda.so to be available at runtime.
This is of very little use however, as it doesn't test that the generated code is actually valid as that would require nvrtc, and implicitly libcuda.so.
I.e it hasn't caught the python 3.8 bug introduced in #1147 (without a test case adding with that particular shape?)