[IMPALA-3000] BitReader::Reset does not initialize buffered_values_

• Type: Bug
• Status: Resolved
• Priority: Blocker
• Resolution: Fixed
• Affects Version/s: Impala 2.3.0
• Fix Version/s: Impala 2.5.0, Impala 2.3.4
• Component/s: Backend
• Target Version: Impala 2.5.0, Impala 2.3.4
• Assignee: Tim Armstrong
• Reporter: Wes McKinney
• Created: 15/Feb/16 21:54
• Updated: 27/Mar/16 10:51
• Resolved: 19/Feb/16 00:54

Description

Observed while working on PARQUET-515 on the ASF JIRA.

This method is used both in the deprecated BIT_PACKED repetition/definition level encoding and RLE encoding, however the buffered_values_ variable has data copied into it incidentally as part of other BitReader methods invoked by RleDecoder.

If a user were to read a Parquet file using BIT_PACKED level encoding (unclear if this is seen in the while) containing multiple data pages in a row group, the results could possibly be incorrect (there is also a separate MSB / LSB bit ordering question between BIT_PACKED and RLE encoding styles – it's possible the BIT_PACKED implementation does not match parquet-format at the moment).