Arrow: Fix int96 timestamp offset in arrow dictionary decode

[sungwy]

Issue: #13485

This PR fixes the packed dictionary INT96 timestamp decode path in the VectorizedParquetDefinitionLevelReader to write using byte offsets rather than row indexes.

The unit test constructs a dictionary with expected values and verifies that the decoded Arrow buffer contains the expected timestamp values, and verifies that writing multiple rows of values did not corrupt the data as a result of the values being written into the wrong offsets.

The fix is the same as the earlier proposed change discussed in #13486 , and mirrors the offset handling used by other readers in the same class:

iceberg/arrow/src/main/java/org/apache/iceberg/arrow/vectorized/parquet/VectorizedParquetDefinitionLevelReader.java

Line 302 in 8e7ab3c

.setLong((long) idx * typeWidth, dict.decodeToLong(reader.readInteger()));

Disclosure:

• AI-assisted analysis and test drafting were used while investigating this issue. The final code and test were reviewed, edited, and validated manually. Refer: https://iceberg.apache.org/contribute/#guidelines-for-ai-assisted-contributions

[Copilot]

Pull request overview

Fixes incorrect INT96 timestamp decoding for Parquet dictionary-encoded packed data in Iceberg’s Arrow vectorized reader by writing to Arrow buffers using byte offsets (instead of value indices), preventing buffer corruption and wrong timestamps.

Changes:

• Correct TimestampInt96Reader.nextDictEncodedVal (PACKED mode) to use (long) idx * typeWidth when calling setLong.
• Add a unit test that reproduces the corruption scenario and verifies correct decoded timestamps across multiple rows.

Reviewed changes

Copilot reviewed 2 out of 2 changed files in this pull request and generated no comments.

File	Description
arrow/src/main/java/org/apache/iceberg/arrow/vectorized/parquet/VectorizedParquetDefinitionLevelReader.java	Fixes packed dictionary INT96 decode to write to Arrow buffers using byte offsets, matching other readers and preventing data corruption.
arrow/src/test/java/org/apache/iceberg/arrow/vectorized/parquet/TestVectorizedParquetDefinitionLevelReader.java	Adds a regression test ensuring packed dictionary INT96 decoding writes correct per-row values without overwriting adjacent entries.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.