[
https://issues.apache.org/jira/browse/SPARK-56844?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Wenchen Fan reassigned SPARK-56844:
-----------------------------------
Assignee: Matt Zhang
> Support ArrayType / MapType / StructType in ConstantColumnVector and
> FileSourceMetadataAttribute
> ------------------------------------------------------------------------------------------------
>
> Key: SPARK-56844
> URL: https://issues.apache.org/jira/browse/SPARK-56844
> Project: Spark
> Issue Type: Improvement
> Components: SQL
> Affects Versions: 4.2.0
> Reporter: Matt Zhang
> Assignee: Matt Zhang
> Priority: Minor
> Labels: pull-request-available
>
> `FileSourceMetadataAttribute.isSupportedType` currently rejects ArrayType,
> MapType, and StructType, citing `ColumnVectorUtils.populate` as the limiting
> factor. As a result, file source constant metadata columns cannot use complex
> types, even though all the underlying machinery (`RowToColumnConverter`,
> `OffHeapColumnVector` array/map layout) supports them.
> This issue tracks broadening that gate, implementing the missing populate and
> scatter paths in `ColumnVectorUtils.populate` and
> `ConstantColumnVector.writeToOffHeapColumnVector`, and enabling complex
> constants end to end.
> Specifically:
> - `FileSourceMetadataAttribute.isSupportedType` allows array/map/struct
> recursively, contingent on element types also being supported.
> - `ColumnVectorUtils.populate` gains struct/array/map branches. Array and map
> allocate a one-row off-heap backing vector and reuse the existing
> `RowToColumnConverter` to write the constant value with full recursive type
> support.
> - `ConstantColumnVector` gains optional ownership of a backing
> `WritableColumnVector` (closed by `close()`), and
> `writeToOffHeapColumnVector` gains array/map branches that copy the M
> constant elements once into the target child vector and write per-row
> `(offset=0, length=M)`.
> No user-facing API changes.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
