Tupl Versions Save

Main changes:

• The Table.load, insert, replace, update, merge, and delete methods now throw an exception instead of returning a boolean value. The original methods are supported in the form of newly added "try" variants: tryLoad, tryInsert, tryReplace, tryUpdate, tryMerge, and tryDelete.
• Defined a new Query interface which directly represents a runnable query. It can be used as an alternative to passing query strings to the Table class all the time.
• The Table methods for obtaining QueryPlan instances have been moved to the new Query interface.
• Added new Table methods for supporting custom filtering, transformation, aggregation, grouping, joins, and views.
• Various bug fixes and performance improvements.

See changelog for more details.

Main changes:

• Removed the EntryConsumer, EntryFunction, RowScanner, and RowUpdater interfaces.
• Repurposed the Scanner and Updater interfaces to only work with tables.
• Reading from a closed/deleted index now throws ClosedIndexException or DeletedIndexException.
• Support remote database access.
• Support table query ordering, which selects an appropriate index or performs a sort.

See changelog for more details.

• Fix race condition which can cause the JVM to crash when the database is closed.
• Fix deadlock during compaction caused by releasing the wrong node latch.
• Fix improper lock release when writing cache primer.
• Fix improper lock release/acquire when when opening indexes.
• Fix BoundedCursor such that it doesn't observe uncommitted deletes when initially positioning it.
• Fix possible reporting of negative cursor count in the database stats.
• Fix safety of using updaters that require lock mode upgrades.
• Improve utility of the suspendCheckpoints method by waiting for an in-progress checkpoint to complete.
• Don't reset checkpoint duration stat when thresholds aren't met.
• Depends on Java 17.
• Add an API for supporting relational collections of persistent rows.

• Ensure that index passed to verification observer is unmodifiable when called from an unmodifiable index.

• Fix storage leaks caused by fragmented keys.
• Fix NPE when obtaining stats on a closed database.
• Compaction scan now finds fragmented keys.
• Suppress uncaught exception from background sorter thread when database is closed.

• Fix potential race condition when concurrently deleting and truncating the checkpoint undo log.

• Avoid holding global commit lock when deleting an index.
• Add a method to check if transaction is bogus.

• Newly entered transaction scopes always select the UPGRADABLE_READ lock mode instead of inheriting the mode from the parent scope.
• Use stronger random number for database id.

• Add more utilities for decoding short integers and unsigned integers.
• Ignore interrupts during critical operations, preventing database panics.
• Prune empty nodes when deleting ghost entries. If this isn't done, then an index whose entries have been fully deleted can claim to be non-empty.

• Support fully mapped databases when the page size doesn't match the OS page size.
• Use a faster hash implementation for the lock manager.
• Fix edge case where a non-stored database would prematurely run out of space.