String decode using 'new String' is slow #1532
Description
There are three implementations of the Binary class and only one is using the faster 'UTF8.decode' in the 'toStringUsingUTF8' method. This fixes them to all use the faster UTF8.decode.
As noted in the comments, the 'new String' approach creates a new decoder each time, which is slower than the cached instance used by 'UTF8.decode'.
#1. ByteArraySliceBackedBinary <-- UTF8.decode
#2. ByteArrayBackedBinary <-- new String
#3. ByteBufferBackedBinary <-- new String
https://github.com/apache/incubator-parquet-mr/pull/40
Reporter: Daniel Weeks / @danielcweeks
Assignee: Daniel Weeks / @danielcweeks
Related issues:
- Use thread local decoder cache in Binary toStringUsingUTF8() (is duplicated by)
Note: This issue was originally created as PARQUET-75. Please see the migration documentation for further details.