15.4. Release 0.126
General Changes
- Add error location information (line and column number) for semantic errors.
- Fix a CLI crash during tab-completion when no schema is currently selected.
- Fix reset of session properties in CLI when running USE.
- Fix occasional query planning failure due to a bug in the projection push down optimizer.
- Fix a parsing issue when expressions contain the form POSITION(x in (y)).
- Add a new version of approx_percentile() that takes an accuracy parameter.
- Allow specifying columns names in INSERT queries.
- Add field_length table property to blackhole connector to control the size of generated VARCHAR and VARBINARY fields.
- Bundle Teradata functions plugin in server package.
- Improve handling of physical properties which can increase performance for queries involving window functions.
- Add ability to control whether index join lookups and caching are shared within a task. This allows us to optimize for index cache hits or for more CPU parallelism. This option is toggled by the task.share-index-loading config property or the task_share_index_loading session property.
- Improve performance of queries that use an IN expression with a large list of constant values.
- Enable connector predicate push down for all comparable and equatable types.
- Fix query planning failure when using certain operations such as GROUP BY, DISTINCT, etc. on the output columns of UNNEST.
- In ExchangeClient set maxResponseSize to be slightly smaller than the configured value. This reduces the possibility of encountering PageTooLargeException.
- Fix memory leak in coordinator.
- Add validation for names of table properties.
Hive Changes
- Fix reading structural types containing nulls in Parquet.
- Fix writing DATE type when timezone offset is negative. Previous versions would write the wrong date (off by one day).
- Fix an issue where VARCHAR columns added to an existing table could not be queried.
- Fix over-creation of initial splits.
- Fix hive.immutable-partitions config property to also apply to unpartitioned tables.
- Allow non-VARCHAR columns in DELETE query.
- Support DATE columns as partition columns in parquet tables.
- Improve error message for cases where partition columns are also table columns.