[HIVE-9711] ORC Vectorization DoubleColumnVector.isRepeating=false if all entries are NaN

• Type: Bug
• Status: Closed
• Priority: Major
• Resolution: Fixed
• Affects Version/s: 1.2.0
• Fix Version/s: 1.2.0
• Component/s: File Formats, Vectorization
• Release Note: Fix isRepeating checks for NaN in Float and Double vectorized readers
• Assignee: Gopal Vijayaraghavan Assign to me
• Reporter: Gopal Vijayaraghavan
• Created: 17/Feb/15 21:56
• Updated: 13/Feb/18 18:45
• Resolved: 22/Apr/15 23:01

Description

The isRepeating=true check uses Java equality, which results in NaN != NaN comparison operations.

The noNulls case needs the current check folded into the previous loop, while the hasNulls case needs a logical AND of the isNull[] field instead of == comparisons.

Issue Links

• causes: ORC-285 Empty vector batches of floats or doubles get java.io.EOFException