[HIVE-4884] ORC TimestampTreeReader.nextVector() off by a second when time in fractional

• Type: Sub-task
• Status: Resolved
• Priority: Major
• Resolution: Fixed
• Affects Version/s: vectorization-branch
• Fix Version/s: vectorization-branch, 0.13.0
• Component/s: File Formats
• Release Note: Adjust milliseconds down when encountering a negative second value with a fractional second stored as nanoseconds.
• Assignee: Gopal Vijayaraghavan
• Reporter: Gopal Vijayaraghavan
• Created: 18/Jul/13 20:23
• Updated: 23/Oct/13 21:59
• Resolved: 23/Jul/13 19:07

Description

When using negative timestamps, the ORC vectorized Timestampreader does the following

result.vector[i] = (result.vector[i] * 1000000) + nanoVector.vector[i];

This suffers from inaccuracies because the nanoseconds are always positive and the result.vector[i] is in effect

seconds = (getTime() / 1000)
nanos = (getNanos())
// so -42001 ms is written as
// seconds = -42
// nanos = 999000000
// (-42001 / 1000) * 1000 == -42000
// + 999 ms (from nanos)
// which is -42999 ms

This needs to be adjusted down if nanos has been zero adjusted, to result in -42001 as the vector[i] value.

Issue Links

• is related to: HIVE-4681 Fix ORC TimestampTreeReader.nextVector() to handle milli-nano math corectly
• relates to: HIVE-4971 Unit test failure in TestVectorTimestampExpressions