Removes context.compile_internal where easy

[guilhermeleobas]

Removes context.compile_internal where easy

• np.ravel
• array.flatten
• array.sort
• round
• not in
• math.gcd
• np.argsort

[kc611]

Looks Good to me!

Few minor nitpicks below.

[guilhermeleobas]

Hi @kc611, thanks for the review. I have addressed all your comments.

[stuartarchibald]

Thanks for doing this refactoring @guilhermeleobas. A couple of comments to resolve else looks good.

[gmarkall]

Thanks for the update - this looks good to me from a CUDA perspective now.

[gmarkall]

(Note that the gpuCI fail is just an issue inherited from main, not this PR)

[gmarkall]

gpuci run tests

(just to make everything look nice and green now #8537 is merged)

[stuartarchibald]

@guilhermeleobas please could you resolve the conflicts against main? Many thanks.

[guilhermeleobas]

I'm seeing some failures in this PR after migrating np.ravel to @overload:
https://gist.github.com/guilhermeleobas/91fc7a4137b4fa27b95ff4b8830d0952

[stuartarchibald]

I'm seeing some failures in this PR after migrating np.ravel to @overload: https://gist.github.com/guilhermeleobas/91fc7a4137b4fa27b95ff4b8830d0952

Public CI is passing, is this still an issue or was it resolved by the latest two commits (or just missed by public CI)?

[guilhermeleobas]

I'm seeing some failures in this PR after migrating np.ravel to @overload: https://gist.github.com/guilhermeleobas/91fc7a4137b4fa27b95ff4b8830d0952

Public CI is passing, is this still an issue or was it resolved by the latest two commits (or just missed by public CI)?

This is still an "issue".

[gmarkall]

This message in the traceback: Only accept returning of array passed into the function as argument I see popping up when NRT is not available (usually in the CUDA target, but these messages indicate nopython mode) - is there something disabling NRT going on there?

[stuartarchibald]

This message in the traceback: Only accept returning of array passed into the function as argument I see popping up when NRT is not available (usually in the CUDA target, but these messages indicate nopython mode) - is there something disabling NRT going on there?

I think this is indeed the root cause of the problem. In the test this is present:

numba/numba/tests/test_array_manipulation.py

Line 547 in 036d956

def test_ravel_array(self, flags=enable_pyobj_flags):

which is set via this:

numba/numba/tests/test_array_manipulation.py

Lines 15 to 16 in 036d956

	enable_pyobj_flags = Flags()
	enable_pyobj_flags.enable_pyobject = True

The Flags() ctor won't enable the NRT by default.

This problem exists on main too, it's just far less "noisy" as the split typing/lowering API doesn't have such tracebacks. I think the test needs updating to run with the NRT enabled.

[github-actions]

This pull request is marked as stale as it has had no activity in the past 3 months. Please respond to this comment if you're still interested in working on this. Many thanks!

[guilhermeleobas]

I should probably give another try to this PR sometime in the future.