Move callback wrappers to Python layer

[vzhurba01]

close#531

[copy-pr-bot]

This pull request requires additional validation before any workflows can run on NVIDIA's runners.

Pull request vetters can view their responsibilities here.

Contributors can view more details about this message here.

[vzhurba01]

/ok to test

[leofang]

I have reviewed 3 out of 4 files (haven't had time to cover runtime.pyx.in yet).

Q: I assume so far we have not supported passing a pure Python function as callback to any of the bindings directly, without using ctypes or other means as a trampoline?

[leofang]

Q: Should this be a 1-tuple?

Suggested change

	return CUresult.CUDA_ERROR_OUT_OF_MEMORY
	return (CUresult.CUDA_ERROR_OUT_OF_MEMORY,)

[vzhurba01]

Good catch. Fixed.

[kkraus14]

It would be good to have type annotations or something else as a way to prevent issues like this from sneaking in.

[leofang]

It seems currently Cython cannot raise a compile-time error for either case, only a run-time error (which is not desired I think).

cpdef tuple f():
return42defg() -> tuple:
return 42

Output:

>>>importtest_tuple>>>test_tuple.f()
Traceback (mostrecentcalllast):
File"<stdin>", line1, in<module>File"test_tuple.pyx", line1, intest_tuple.fcpdeftuplef():
File"test_tuple.pyx", line3, intest_tuple.freturn42TypeError: Expectedtuple, gotint>>>test_tuple.g()
Traceback (mostrecentcalllast):
File"<stdin>", line1, in<module>File"test_tuple.pyx", line7, intest_tuple.greturn42TypeError: Expectedtuple, gotint

[vzhurba01]

Q: I assume so far we have not supported passing a pure Python function as callback to any of the bindings directly, without using ctypes or other means as a trampoline?

Correct.

[vzhurba01]

/ok to test

[leofang]

LGTM!

[github-actions]

Doc Preview CI
Preview removed because the pull request was closed or merged.