[gitRasheed]

Summary

Vortex exports string columns to Arrow as Utf8View, so StringViewAccessor serves every string column Spark reads, and it copied each value into a new byte[]. StringAccessor already returns Utf8 values in place.

Changes

StringViewAccessor.getUTF8String reads each value into a reused NullableViewVarCharHolder and returns UTF8String.fromAddress over the holder's buffer, the same way StringAccessor and LargeStringAccessor do. Arrow's holder overload points at the view itself for values of up to 12 bytes and at the data buffer and offset the view names for longer ones. A value is valid until its batch is released, like the values of Spark's own column vectors.

Tests

VortexArrowColumnVectorTest.stringViewValues reads inline, out-of-line, empty and multi-byte values spread over two data buffers, one at a nonzero offset, and checks each value is off-heap. It fails on develop, where every value is a copied byte[].

Checked locally: :vortex-spark-3.5_2.12:test, :vortex-spark-3.5_2.13:test, :vortex-spark-4.0_2.13:test, spark41CompatTest, spark42CompatTest and spotlessCheck.

I used AI tools to help write and review this change.

[gitRasheed]

I just saw #11109 does this too: it reads string views in place, so I'm closing this one. The difference is that this PR goes through Arrow's NullableViewVarCharHolder, while #11109 reads the view buffers directly. #11109 has no string-view test. This PR's stringViewValues passes on it unchanged (inline values at the 12/13-byte boundary, out-of-line values at a nonzero offset and in a second data buffer, and an off-heap check), so I'm happy to move it over if it's useful.