GH-45639: [C++][Statistics] Add ARROW:average_byte_width:exact and ARROW:average_byte_width:approximate statistics attributes to arrow::ArrayStatistics

[andishgar]

Rationale for this change

Add ARROW:average_byte_width:exact and ARROW:average_byte_width:approximate statistics attributes to arrow::ArrayStatistics

What changes are included in this PR?

Add average_byte_width and is_average_byte_width_exact member variables to arrow::ArrayStatistics

Are these changes tested?

Yes, I run the relevant unit tests

Are there any user-facing changes?

Yes

• GitHub Issue: [C++] implement Statistic Schema attribute in C++ #45639

[github-actions]

⚠️ GitHub issue #45639 has been automatically assigned in GitHub to PR creator.

[andishgar]

@kou, I’ve converted this pull request to a draft in order to clarify whether any modifications are required on the Parquet side.

[kou]

We should not use == for floating point numbers.
We need something like

arrow/cpp/src/arrow/compare.cc

Lines 85 to 94 in 5989387

	if (x == y) {
	return Flags::signed_zeros_equal || (std::signbit(x) == std::signbit(y));
	}
	if (Flags::nans_equal && std::isnan(x) && std::isnan(y)) {
	return true;
	}
	if (Flags::approximate && (fabs(x - y) <= epsilon)) {
	return true;
	}
	return false;

instead.

[andishgar]

I have the following questions regarding the implementation:

1-Should I first create a separate issue to address the min and max member variables, given that they can be double values?

2- I plan to add arrow::EqualOptions to the Equal method. This would make the operator== redundant. Is it acceptable to remove operator== in this case?

3-When comparing two arrow::Array objects for equality, should their arrow::ArrayStatistics also be checked?( the current implementation does not check ArrayStatistics)

4-Should I implement a separate ApproximateEqual method to handle EqualOptions::atol, or should this logic be integrated into the Equal method?

[kou]

1. Ah, I didn't notice it. Yes.
2. Yes.
3. (I think) No. I think arrow::Array::Equals() should check only data. For example, metadata in its data type isn't used. But we may want to discuss this in a separated issue.
4. Yes or no. Let's discuss it as a separated issue. I think that adding floating_approximate to EqualOptions and using only Equals is better but others may think differ.

[andishgar]

Okay, I start to create an issue to solve the problem of min and max first

[andishgar]

@kou thank you for your feedback, I will look into it soon