[jira] [Created] (FLINK-39433) Consolidate scattered Hive reflection calls into HiveShim for type safety

[Jayesh Lalwani (Jira)]

Jayesh Lalwani created FLINK-39433:
--------------------------------------

             Summary:  Consolidate scattered Hive reflection calls into 
HiveShim for type safety 
                 Key: FLINK-39433
                 URL: https://issues.apache.org/jira/browse/FLINK-39433
             Project: Flink
          Issue Type: Improvement
          Components: Connectors / Hive
            Reporter: Jayesh Lalwani


The HiveShim abstraction exists to isolate version-specific Hive API 
differences behind typed methods. However, 38 raw reflection calls 
(getMethod/invoke/Class.forName) are scattered across 13 files outside 
HiveShim, bypassing the shim and losing compile-time type checking.

 

This makes Hive version upgrades harder — breakages are only discovered at 
runtime when reflection fails. For example, during Hive 4 support work 
(FLINK-35515), several API changes (HiveFileFormatUtils.getRecordWriter 
signature change, SerDeUtils.initializeSerDe removal) were only caught by 
running tests because the calls used raw reflection instead of typed HiveShim 
methods.

 

Affected files:

- HiveWriterFactory.java

- HiveMapredSplitReader.java

- HiveTableSink.java

- HiveSessionState.java

- HiveParserUtils.java

- HiveReflectionUtils.java

- MRSplitsGetter.java

- HiveTableMetaStoreFactory.java

- ThriftObjectConversions.java

- HiveSimpleUDF.java

- HiveFunctionDefinitionFactory.java

- HiveCatalogUtils.java

- HiveScriptTransformOperator.java

 

Proposal: Move these reflection calls into typed HiveShim methods with proper 
return types, so version-specific differences are caught at compile time and 
callers get type safety.



--
This message was sent by Atlassian Jira
(v8.20.10#820010)