Proper way to initialize an ListArray.Builder

[mobiusklein]

I've been using this library for a while now, but I have found a few places where I miss an invariant over how to initialize a ListArray.Builder.

arrow-dotnet/src/Apache.Arrow/Arrays/ListArray.cs

Lines 46 to 52 in 3e235ea

	internal Builder(ListType dataType)
	{
	ValueBuilder = ArrowArrayBuilderFactory.Build(dataType.ValueDataType);
	ValueOffsetsBufferBuilder = new ArrowBuffer.Builder<int>();
	ValidityBufferBuilder = new ArrowBuffer.BitmapBuilder();
	DataType = dataType;
	}

When initializing the builder the ValueOffsetsBufferBuilder is empty. When I am attentive, I remember to immediately call the Append method which adds 0 to the ValueBuilder.Length attribute being 0. This ensures that the offset to the start of the first list in the array is 0, as shown in the specification https://arrow.apache.org/docs/format/Intro.html#list. Otherwise trying to read this array later will skip over everything before the first call to Append.

I recently spent a time running down a bug that was effectively that I forgot to call Append after Clearing the ListArray.Builder.

Is there a reason an empty ListArray.Builder shouldn't always initialize with ValueOffsetsBufferBuilder = [0] and do similar when cleared?

Thank you

[CurtHagenlocher]

I'm not sure anyone actively involved in the project today could tell you why the code is the way it is; it seems to date back a little over six years. A better question would be whether there are any improvements that could be made that would be compatible with existing code, and I think that's a pretty hard constraint to meet. Perhaps ListArrayBuilder.Build could throw an exception if the first offset is anything other than zero? The only programs which would be broken by this should be those that are already producing invalid arrays.

I often think it would be nice to redesign the builder APIs.