allow zero sized empty fixed

[smiklos]

Which issue does this PR close?

Closes #4623.

Rationale for this change

Technically it's valid to construct fixed sized lists with 0 length values.

What changes are included in this PR?

The constructor will not panic if 0 is passed for value size. FixedsizeList's length will be the length of values array in this case.

Are there any user-facing changes?

No

No

[smiklos]

@tustvold

My concern with this is that FixedSizeList::value_offset will always return 0 if value_length is 0.
This could be an issue if callers expect FixedSizeList::value to panic if the index is out of bound, which it won't rn.

[tustvold]

I don't see an issue with adding a length check to FixedSizeList::value, ultimately it is doing a dyn dispatch followed by potentially multiple atomic ref increments / allocations, so the additional branch is not going to make any difference. The value APIs are not an efficient way to interact with nested data, but are purely there for ergonomics

[smiklos]

Looking at it more I think the values array should be always empty if value_length is empty. After all what would we even store there in that case? This is actually validated for in the builder variant but not in FixedSizeListArray::try_new.

I think the right solution is to add this validation to try_new such that we check that values.len() = value_length * len.

Then I believe ::value(i) for any i should just blow up if value_length is 0 coz the values array is than empty, and can't be indexed into

[tustvold]

In general we only verify that buffers and children are at least as long as required, and don't error if they are longer than necessary. We could change this, but we should probably change that consistently if we do so.

[smiklos]

Then perhaps it's enough what is already implemented in this pr?

[tustvold]

Thank you