GH-41803: [MATLAB] Add C Data Interface format import/export functionality for arrow.tabular.RecordBatch

[sgilmore10]

Rationale for this change

This pull requests adds two new APIs for importing and exporting arrow.tabular.RecordBatch instances using the C Data Interface format.

Example:

>> T = table((1:3)', ["A"; "B"; "C"]);
>> expected = arrow.recordBatch(T)

expected = 

  Arrow RecordBatch with 3 rows and 2 columns:

    Schema:

        Var1: Float64 | Var2: String

    First Row:

        1 | "A"

>> cArray = arrow.c.Array();
>> cSchema = arrow.c.Schema();

% Export the RecordBatch to C Data Interface Format
>> expected.export(cArray.Address, cSchema.Address);

% Import the RecordBatch from C Data Interface Format
>> actual = arrow.tabular.RecordBatch.import(cArray, cSchema)

actual = 

  Arrow RecordBatch with 3 rows and 2 columns:

    Schema:

        Var1: Float64 | Var2: String

    First Row:

        1 | "A"

% The RecordBatch is the same after round-tripping to the C Data Interface format
>> isequal(actual, expected)

ans =

  logical

   1

What changes are included in this PR?

1. Added a new method arrow.tabular.RecordBatch.export for exporting RecordBatch objects to the C Data Interface format.
2. Added a new static method arrow.tabular.RecordBatch.import for importing RecordBatch objects from the C Data Interface format.
3. Added a new internal class arrow.c.internal.RecordBatchImporter for importing RecordBatch objects from the C Data Interface format.

Are these changes tested?

Yes.

1. Added a new test file matlab/test/arrow/c/tRoundtripRecordBatch.m which has basic round-trip tests for importing and exporting RecordBatch objects.

Are there any user-facing changes?

Yes.

1. Two new user-facing methods were added to arrow.tabular.RecordBatch. The first is arrow.tabular.RecordBatch.export(cArrowArrayAddress, cArrowSchemaAddress). The second is arrow.tabular.RecordBatch.import(cArray, cSchema). These APIs can be used to export/import RecordBatch objects using the C Data Interface format.

Future Directions

1. Add integration tests for sharing data between MATLAB/mlarrow and Python/pyarrow running in the same process using the MATLAB interface to Python.
2. Add support for the Arrow C stream interface format.

Notes

1. Thanks to @kevingurney for the help with this feature!

• GitHub Issue: [MATLAB] Add C Data Interface format import/export functionality for arrow.tabular.RecordBatch #41803

[kou]

+1

We need RecordBatchImporter because we can't define custom constructor (something like static libmexclass::proxyMakeResult arrow::matlab::tabular::proxy::RecordBatch::import(....)), right?

[sgilmore10]

+1

We need RecordBatchImporter because we can't define custom constructor (something like static libmexclass::proxyMakeResult arrow::matlab::tabular::proxy::RecordBatch::import(....)), right?

Yes, that's correct. @kevingurney and I have talked about extending mathworks/libmexclass to support either static proxy methods or free-floating functions at some point. We just haven't been able to get around to adding that functionality yet.

[conbench-apache-arrow]

After merging your PR, Conbench analyzed the 5 benchmarking runs that have been run so far on merge-commit fe2d926.

There were no benchmark performance regressions. 🎉

The full Conbench report has more details. It also includes information about 13 possible false positives for unstable benchmarks that are known to sometimes produce them.