[HIVE-4681] Fix ORC TimestampTreeReader.nextVector() to handle milli-nano math corectly

• Type: Sub-task
• Status: Resolved
• Priority: Major
• Resolution: Fixed
• Fix Version/s: vectorization-branch, 0.13.0
• Labels: vectorization
• Release Note: Fix ORC TimestampTreeReader to sum the second and nanosecond fraction timestamp vectors correctly
• Tags: vectorization
• Assignee: Gopal Vijayaraghavan
• Reporter: Gopal Vijayaraghavan
• Created: 07/Jun/13 00:06
• Updated: 23/Oct/13 21:59
• Resolved: 07/Jun/13 21:33

Description

The current TimestampTreeReader implementation is broken when fed only milliseconds.

If TS = 222ms.

   result.vector[i] = ... // == 0;
   nanoVector.vector[i] = ... // == 222000000
   if (result.vector[i] >= 0) {
     result.vector[i] += nanoVector.vector[i] / 1000000; // 222
   }
...
   result.vector[i] = (result.vector[i] * 1000000)
                + nanoVector.vector[i];
   // result.vector[i] == 444000000

which is broken.

This is not a problem in the Timestamp case, because

ts.setTime(222);        // {time=0; nanos=222000000}
ts.setNanos(222000000); // {time=0; nanos=222000000}

Since that is not additive, the error does not occur in the regular ORC reader.

Issue Links

• relates to: HIVE-4884 ORC TimestampTreeReader.nextVector() off by a second when time in fractional