12.78. Release 0.168

General Changes

Fix correctness issues for certain JOIN queries that require implicit coercions for terms in the join criteria.
Fix invalid “No more locations already set” error.
Fix invalid “No more buffers already set” error.
Temporarily revert empty join short-circuit optimization due to issue with hanging queries.
Improve performance of DECIMAL type and operators.
Optimize window frame computation for empty frames.
json_extract() and json_extract_scalar() now support escaping double quotes or backslashes using a backslash with a JSON path subscript. This changes the semantics of any invocation using a backslash, as backslashes were previously treated as normal characters.
Improve performance of filter() and map_filter() lambda functions.
Add Memory Connector.
Add arrays_overlap() and array_except() functions.
Allow concatenating more than two arrays with concat() or maps with map_concat().
Add a time limit for the iterative optimizer. It can be adjusted via the iterative_optimizer_timeout session property or experimental.iterative-optimizer-timeout configuration option.
ROW types are now orderable if all of the field types are orderable. This allows using them in comparison expressions, ORDER BY and functions that require orderable types (e.g., max()).

Fix reading decimals for RCFile text format using non-optimized reader.
Fix bug which prevented the file based metastore from being used.
Enable optimized RCFile reader by default.
Common user errors are now correctly categorized.
Add new, experimental, RCFile writer optimized for Presto. The new writer can be enabled with the rcfile_optimized_writer_enabled session property or the hive.rcfile-optimized-writer.enabled Hive catalog property.

ConnectorIndex now returns ConnectorPageSource instead of RecordSet. Existing connectors that support index join can use the RecordPageSource to adapt to the new API.