feat(csharp): improve handling of StructArrays #2587
Conversation
davidhcoe
commented
Mar 7, 2025
davidhcoe
commented
- improves the handling of structs to return objects or JsonString (defaults to JsonString to not break existing callers)
- additional testing for each return type
- updates to the ADO.NET wrapper to support both struct types
- fixes csharp: ValueAt extension causes error when StringArray length = 0 #2586
davidhcoe
changed the title
There was a problem hiding this comment.
Thanks for the change! I've left a few comments and questions to consider, but nothing I'd think of as seriously blocking.
In hindsight, I think this is arguably the wrong approach to dealing with "nonstandard" values whether they're structured or decimal. It would have been better to keep all conversions in Arrow "vector" space instead of dealing with them one-by-one in a "get scalar" function. That way, if I'm a consumer who wants to deal with the results as an array but I don't want to have to handle values one at a time I can say "convert this struct array into a string array" and then it's just a regular Arrow string vector and I can keep going in vector space. For full generality, this might require a change to the C# Arrow implementation to support a common interface between C# arrays and Arrow arrays, but that's probably worth doing or at least thinking about.
(And we can obviously move in those directions over time.)
| @@ -76,7 +83,9 @@ public static class IArrowArrayExtensions | |||
| case ArrowTypeId.Int64: | |||
| return ((Int64Array)arrowArray).GetValue(index); | |||
| case ArrowTypeId.String: | |||
| return ((StringArray)arrowArray).GetString(index); | |||
| StringArray sArray = (StringArray)arrowArray; | |||
| if (sArray.Length == 0) { return null; } | |||
There was a problem hiding this comment.
How can we get here? Why is this not an error, and why does it impact only StringArray and not other arrays?
csharp/src/Apache.Arrow.Adbc/Extensions/IArrowArrayExtensions.cs
Outdated
Show resolved
Hide resolved
| Assert.True(ctv.ExpectedValue.Equals(value), Utils.FormatMessage($"Expected value [{ctv.ExpectedValue}] does not match actual value [{value}] for {ctv.Name} for query [{query}]", environmentName)); | ||
| bool areEqual = false; | ||
|
|
||
| if (value is ExpandoObject) |
There was a problem hiding this comment.
Is this missing a case for when the result is an array?
There was a problem hiding this comment.
No. There is a check above if (type.BaseType?.Name.Contains("PrimitiveArray") == false) that puts us in a different path for comparing arrays.
| return (array, index) => ((StringArray)array).GetString(index); | ||
| return (array, index) => | ||
| { | ||
| StringArray? sArray = array as StringArray; |
There was a problem hiding this comment.
This gives up some of the performance benefit of the approach. Instead of having to add a check to each invocation, can we have the caller tell us in advance what the expected source type is and return one of two different delegates?
There was a problem hiding this comment.
I couldnt figure out an elegant way to have two separate delegates so I went with one and checking the DataType of the array that's passed in.
csharp/src/Apache.Arrow.Adbc/Extensions/IArrowArrayExtensions.cs
Outdated
Show resolved
Hide resolved
