GH-48470: [Python] Construct UuidArray from list of UuidScalars

[tadeja]

Rationale for this change

Fixes #48470. Also fixes all extension types, not just UUID.

What changes are included in this PR?

An extension scalar is unwrapped to its storage type when building arrays.

Are these changes tested?

Yes, new test_array_from_extension_scalars covers builtin (uuid, bool8, json_, opaque) and custom types across all storage types (int, float, bool, string, binary, large string/binary, decimal, fixed-size binary, struct, timestamp, duration, date).

Are there any user-facing changes?

Now user can run such an example to get the output below instead of ArrowInvalid message.
This now works for any extension type, not just UUID.

import pyarrow as pa
pa.array([pa.scalar(b'1'*16, type=pa.uuid())], type=pa.uuid())

<pyarrow.lib.UuidArray object at 0x128186970>
[
  31313131313131313131313131313131
]

• GitHub Issue: [Python] Cannot construct UuidArray from list of UuidScalars #48470

[github-actions]

⚠️ GitHub issue #48470 has been automatically assigned in GitHub to PR creator.

[rok]

Looks good. Some minor change and more extensive tests suggestions.

[rok]

How about using Result here?

Suggested change

	// Helper function to unwrap extension scalar to its storage scalar
	const Scalar& GetStorageScalar(const Scalar& scalar) {
	if (scalar.type->id() == Type::EXTENSION) {
	return *checked_cast<const ExtensionScalar&>(scalar).value;
	}
	return scalar;
	}
	// Helper function to unwrap extension scalar to its storage scalar
	Result<const Scalar*> GetStorageScalar(const Scalar& scalar) {
	if (scalar.type->id() != Type::EXTENSION) {
	return &scalar;
	}
	const auto& extension_scalar = checked_cast<const ExtensionScalar&>(scalar);
	return extension_scalar.value.get();
	}

[tadeja]

Perhaps Result is not needed here - if its kind of failure path is unlikely then const Scalar& might be better?
Looking at the line 757 check if (PyValue::IsNull(this->options_, value)) { ...
( For an (unlikely?) non-None extension scalar with internal storage .value of a null shared_ptr both approaches don't avoid it? )

[pitrou]

Result is probably not needed indeed, since no error can be returned here. As you prefer @tadeja.

[tadeja]

For the C++ part of changes, @pitrou would you have a chance to take a look?

[pitrou]

Can move this stdlib import to the top of file.

[pitrou]

Neat @tadeja ! Just minor suggestions below.

[rok]

Thanks for sanity check @pitrou! Merging.

[conbench-apache-arrow]

After merging your PR, Conbench analyzed the 3 benchmarking runs that have been run so far on merge-commit d08d5e6.

There were no benchmark performance regressions. 🎉

The full Conbench report has more details.