Convert CONCATENATE and aggregate functions to it.
Fix some bugs:
- select sum(n) over (order by n2), sum(n) over (), n, n2 from x1;
- select (select 1 from x1 b where list(1, a.n) = 1) from x1 a;
Fixed the layering for WITH LOCK and (partially) ANY/ALL predicates.
Cleaned up the outdated code that never worked (mapping DISTINCT to an index, some VMS remainings).
Wiped out a lot of the pre-ODS11 optimizer logic. Some minor adjustments there.
Re-implemented the full outer join from scratch. This resolves CORE-2678 (full outer join cannot use available indices).
Resolved CORE-2796: DB_KEY is always zero for external tables.
Implemented the core part of the scrollable PSQL cursors. Implementation is still incomplete, but ready for testing.
Some other changes I don't recall at the moment ;-)
WARNING! The engine may be unstable. Please email me about any issues found.
- Plugins API.
- Feature CORE-2700 - UDR (User Defined Routines) Engine - C++ API supporting functions, triggers and stored procedures.
- Feature CORE-2470 - Support for alternate format of strings literals.
- Feature CORE-2310 - DDL triggers.
- Feature CORE-2312 - PSQL Packages.
- Feature CORE-1209 - CONTINUE statement.
- Feature CORE-1180 - DDL syntax to change (not) nullable state of columns.
- Feature CORE-2090 - Support OVER () clause with current aggregate functions.
- Fixed CORE-2699 - Common table expression context could be used with parameters.
- Introduce ODS 12.0.
- Work in progress in type-safe parser.
- Refactor some DDL commands (procedures and triggers) from DYN to DdlNodes.
- Refactor virtual tables to use a class hierarchy instead of namespaces. This is basic thing, not based on the changes done in Vulcan. Window functions is based on this work.
- Refactor COMMENT ON and DROP FUNCTION from DYN to DdlNodes. COMMENT ON do not use GDML anymore, it uses DSQL with PreparedStatement class.
- Refactor EXECUTE BLOCK to StmtNodes.
- Refactor the IUDF to SysFunctions. That eliminates RDB$GET_CONTEXT and RDB$SET_CONTEXT from RDB$FUNCTIONS.
2. Fixed minor bug - use of MetaName in PAR_symbol_to_gdscode() was illegal
cause maximum length of error name can be up to 32 symbols,
which does not fit into MetaName limit (31). And we already have 4 such names!
CORE-501 - Optimization problem with COALESCE
CORE-1343 - Bug with a simple case and a subquery
CORE-2041 - update or insert with gen_id() with wrong generator value
And should fix this one (snapshot build bug):
CORE-2051 - don't work subquery in COALESCE
2) Throw an error when the relation cannot be locked because of being already dropped.
3) Don't release the existence lock before RDB$RELATIONS is cleaned up.
This is the first part of the changes intended to fix CORE-1938.
Allow to specify EXECUTE STATEMENToptional clauses in any (not fixed) order.
Use INTERNAL_PROVIDER for access to the current database by another user too.
2) Creation of StmtNode
3) Conversion of IN AUTONOMOUS TRANSACTION to InAutonomousTransactionNode (inherited from StmtNode) - it makes this feature almost self-contained
Please do a boot build or at least run blr_table.
1) Make memory counters aggregated.
2) Add attachment pool and allocate appropriate resources out of this pool.
3) Always release attachments explicitly (via destructor).
4) Always delete user requests prior to attachment deletion.
5) Introduce memory usage counters per every monitoring object.
6) Misc refactoring.
Some pieces are still incomplete (although everything basically works), but I'd like to get feedback and testing sooner rather than later.
and CORE-1079: Every attach of fbclient/fbembed library leaks 64KB of memory
To do it:
- created new template GlobalPtr, controlling access to destructors of global objects.
- applied it to global objects.
Also some related cleanup.
