FindCUDA : Latest builds break support for per configuration options with CUDA_ADD_LIBRARY
As per the FindCUDA documentation:
The arguments passed in after OPTIONS are extra command line options to give to nvcc. You can also specify per configuration options by specifying the name of the configuration followed by the options.
This used to behave as expected on older versions (known to work on 3.8), but has regressed on latest versions (bug found on nightly build cmake-3.10.20180114-g3265b ).
Per-configuration options are now incorrectly appended to the CUDA_NVCC_FLAGS variable in generated intermediate CMake file, along with configuration name.
I've attached a minimal example to reproduce the error.