CUDA: Remove catch of NotImplementedError in target.py

[gmarkall]

Maybe this is more of a question than a PR, but I saw it added in #6948 and I'm not sure what it does. Removing it seems not to cause a problem - all tests still seem to pass without it.

[gmarkall]

From OOB discussion, it seems like this might have initially been added due to #701, which is no longer relevant because NumPy 1.6 is no longer supported, and it also appears that there is no way to trigger the NotImplementedError at import time. The PR has been updated to remove the try from the CPU target's import of numba.np.npdatetime as well.

[stuartarchibald]

Thanks for the patch. I've tried a few ways to trigger the noted exception and it seems like something else is always an issue before a NotImplementedError is hit. Further, it appears on inspection that the exception cannot be raised at import time as the control flow required to do so is always in a closure held by a lowering implementation and not actually executed as part of an import sequence.

This patch should get a CPU and CUDA smoke test and assuming both come back fine I think that would further confirm its validity.

Thanks again!

[stuartarchibald]

Waiting on merge of #7003

[stuartarchibald]

Buildfarm ID: numba_smoketest_cpu_yaml_30.
Buildfarm ID: numba_smoketest_cuda_yaml_62.

[stuartarchibald]

Buildfarm ID: numba_smoketest_cuda_yaml_62.

Passed.

[stuartarchibald]

Buildfarm ID: numba_smoketest_cpu_yaml_30.

Passed