fix: support total_size=0 in byte_range_to_row_range

[aalexandrov]

Summary

In certain cases DataFusion might decide to create a ranges for empty files that have zero rows. In order to avoid hitting a division by zero error in the

let average_row = total_size / row_count;

line, add an early exit check to VortexOpener::open to immediately return an empty stream if the file contains zero rows:

if vxf.row_count() == 0 {
    let empty_stream = stream::iter(vec![]).boxed();
    return Ok(empty_stream);
}

The enforced row_count > 0 invariant for the byte_range_to_row_range call is now also called out with a debug_assert! macro.

Testing

Add a new test_open_empty_file test case.

[AdamGS]

If the file has 0 rows, shouldn't VortexOpener::open just return an empty stream?

[aalexandrov]

If the file has 0 rows, shouldn't VortexOpener::open just return an empty stream?

Fair point. I updated the PR and added a test_open_empty_file to cover the refined VortexOpener::open behavior.

Nit: The changes are in a fixup commit, so they need to be squashed before the PR is merged.

[AdamGS]

just a nit - I think using the buildin utility is nicer

Suggested change

	let empty_stream = stream::iter(vec![]).boxed();
	return Ok(empty_stream);
	return Ok(futures::stream::empty().boxed());

[AdamGS]

overall LGTM, just running CI to see everything is good

[AdamGS]

btw - you need to sign-off on your commits for us to merge it, we'll also squash it before merging.

[aalexandrov]

btw - you need to sign-off on your commits for us to merge it, we'll also squash it before merging.

Done. I used the occasion to also squashed the commits and update the commit message to align with the PR description.

[AdamGS]

thank you for contributing this fix!