SQLite Release 3.36.0 On 2021-06-18
Improvement to the EXPLAIN QUERY PLAN output to make it easier to understand.
Byte-order marks at the start of a token are skipped as if they were whitespace.
An error is raised on any attempt to access the rowid of a VIEW or subquery. Formerly, the rowid of a VIEW would be indeterminate and often would be NULL. The -DSQLITE_ALLOW_ROWID_IN_VIEW compile-time option is available to restore the legacy behavior for applications that need it.
The sqlite3_deserialize() and sqlite3_serialize() interfaces are now enabled by default. The -DSQLITE_ENABLE_DESERIALIZE compile-time option is no longer required. Instead, there is is a new -DSQLITE_OMIT_DESERIALIZE compile-time option to omit those interfaces.
The "memdb" VFS now allows the same in-memory database to be shared among multiple database connections in the same process as long as the database name begins with "/".
Back out the EXISTS-to-IN optimization (item 8b in the SQLite 3.35.0 change log) as it was found to slow down queries more often than speed them up.
Improve the constant-propagation optimization so that it works on non-join queries.
The REGEXP extension is now included in CLI builds.
version 3.35.5:
Fix defects in the new ALTER TABLE DROP COLUMN feature that could corrupt the database file.
Fix an obscure query optimizer problem that might cause an incorrect query result.
3.35.4
Fix a defect in the query planner optimization identified by item 8b above.
Fix a defect in the new RETURNING syntax.
Fix the new RETURNING feature so that it raises an error if one of the terms in the RETURNING clause references a unknown table, instead of silently ignoring that error.
Fix an assertion associated with aggregate function processing that was incorrectly triggered by the push-down optimization.
changes in version 3.35.3
Enhance the OP_OpenDup opcode of the bytecode engine so that it works even if the cursor being duplicated itself came from OP_OpenDup. Fix for ticket bb8a9fd4a9b7fce5. This problem only came to light due to the recent MATERIALIZED hint enhancement.
When materializing correlated common table expressions, do so separately for each use case, as that is required for correctness. This fixes a problem that was introduced by the MATERIALIZED hint enhancement.
Fix a problem in the filename normalizer of the unix VFS.
Fix the "box" output mode in the CLI so that it works with statements that returns one or more rows of zero columns (such as PRAGMA incremental_vacuum). Forum post afbbcb5b72.
Improvements to error messages generated by faulty common table expressions. Forum post aa5a0431c99e.
Fix some incorrect assert() statements.
Fix to the SELECT statement syntax diagram so that the FROM clause syntax is shown correctly. Forum post 9ed02582fe.
Fix the EBCDIC character classifier so that it understands newlines as whitespace. Forum post 58540ce22dcd.
Improvements the xBestIndex method in the implementation of the (unsupported) wholenumber virtual table extension so that it does a better job of convincing the query planner to avoid trying to materialize a table with an infinite number of rows.
Changes in version 3.35.2:
Fix a problem in the appendvfs.c extension that was introduced into version 3.35.0.
Ensure that date/time functions with no arguments (which generate responses that depend on the current time) are treated as non-deterministic functions. Ticket 2c6c8689fb5f3d2f
Fix a problem in the sqldiff utility program having to do with unusual whitespace characters in a virtual table definition.
Limit the new UNION ALL optimization described by item 8c in the 3.35.0 release so that it does not try to make too many new subqueries. See forum thread 140a67d3d2 for details.
3.35.1:
Fix a bug in the new DROP COLUMN feature when used on columns that are indexed and that are quoted in the index definition.
Documentation improvements.
SQLite Release 3.35.0 On 2021-03-12
Added built-in SQL math functions(). (Requires the -DSQLITE_ENABLE_MATH_FUNCTIONS compile-time option.)
Added support for ALTER TABLE DROP COLUMN.
Generalize UPSERT:
Allow multiple ON CONFLICT clauses that are evaluated in order,
The final ON CONFLICT clause may omit the conflict target and yet still use DO UPDATE.
Add support for the RETURNING clause on DELETE, INSERT, and UPDATE statements.
Use less memory when running VACUUM on databases containing very large TEXT or BLOB values. It is no longer necessary to hold the entire TEXT or BLOB in memory all at once.
Add support for the MATERIALIZED and NOT MATERIALIZED hints when specifying common table expressions. The default behavior was formerly NOT MATERIALIZED, but is now changed to MATERIALIZED for CTEs that are used more than once.
The SQLITE_DBCONFIG_ENABLE_TRIGGER and SQLITE_DBCONFIG_ENABLE_VIEW settings are modified so that they only control triggers and views in the main database schema or in attached database schemas and not in the TEMP schema. TEMP triggers and views are always allowed.
Query planner/optimizer improvements:
Enhancements to the min/max optimization so that it works better with the IN operator and the OP_SeekScan optimization of the previous release.
Attempt to process EXISTS operators in the WHERE clause as if they were IN operators, in cases where this is a valid transformation and seems likely to improve performance.
Allow UNION ALL sub-queries to be flattened even if the parent query is a join.
Use an index, if appropriate, on IS NOT NULL expressions in the WHERE clause, even if STAT4 is disabled.
Expressions of the form "x IS NULL" or "x IS NOT NULL" might be converted to simply FALSE or TRUE, if "x" is a column that has a "NOT NULL" constraint and is not involved in an outer join.
Avoid checking foreign key constraints on an UPDATE statement if the UPDATE does not modify any columns associated with the foreign key.
Allow WHERE terms to be pushed down into sub-queries that contain window functions, as long as the WHERE term is made up of entirely of constants and copies of expressions found in the PARTITION BY clauses of all window functions in the sub-query.
CLI enhancements:
Enhance the ".stats" command to accept new arguments "stmt" and "vmstep", causing prepare statement statistics and only the virtual-machine step count to be shown, respectively.
Add the ".filectrl data_version" command.
Enhance the ".once" and ".output" commands so that if the destination argument begins with "|" (indicating that output is redirected into a pipe) then the argument does not need to be quoted.
Bug fixes:
Fix a potential NULL pointer dereference when processing a syntactically incorrect SELECT statement with a correlated WHERE clause and a "HAVING 0" clause. (Also fixed in the 3.34.1 patch release.)
Fix a bug in the IN-operator optimization of version 3.33.0 that can cause an incorrect answer.
Fix incorrect answers from the LIKE operator if the pattern ends with "%" and there is an "ESCAPE '_'" clause.
Changes in version 3.34.1:
Fix a potential use-after-free bug when processing a a subquery with both a correlated WHERE clause and a "HAVING 0" clause and where the parent query is an aggregate.
Fix documentation typos
Fix minor problems in extensions.
SQLite Release 3.34.0
Added the sqlite3_txn_state() interface for reporting on the current transaction state of the database connection.
Enhance recursive common table expressions to support two or more recursive terms as is done by SQL Server, since this helps make queries against graphs easier to write and faster to execute.
Improved error messages on CHECK constraint failures.
CLI enhancements:
The .read dot-command now accepts a pipeline in addition to a filename.
Added options --data-only and --nosys to the .dump dot-command.
Added the --nosys option to the .schema dot-command.
Table name quoting works correctly for the .import dot-command.
The generate_series(START,END,STEP) table-valued function extension is now built into the CLI.
The .databases dot-command now show the status of each database file as determined by sqlite3_db_readonly() and sqlite3_txn_state().
Added the --tabs command-line option that sets .mode tabs.
The --init option reports an error if the file named as its argument cannot be opened. The --init option also now honors the --bail option.
Query planner improvements:
Improved estimates for the cost of running a DISTINCT operator.
When doing an UPDATE or DELETE using a multi-column index where only a few of the earlier columns of the index are useful for the index lookup, postpone doing the main table seek until after all WHERE clause constraints have been evaluated, in case those constraints can be covered by unused later terms of the index, thus avoiding unnecessary main table seeks.
The new OP_SeekScan opcode is used to improve performance of multi-column index look-ups when later columns are constrained by an IN operator.
The BEGIN IMMEDIATE and BEGIN EXCLUSIVE commands now work even if one or more attached database files are read-only.
Enhanced FTS5 to support trigram indexes.
Improved performance of WAL mode locking primitives in cases where there are hundreds of connections all accessing the same database file at once.
Enhanced the carray() table-valued function to include a single-argument form that is bound using the auxiliary sqlite3_carray_bind() interface.
The substr() SQL function can now also be called "substring()" for compatibility with SQL Server.
The syntax diagrams are now implemented as Pikchr scripts and rendered as SVG for improved legibility and ease of maintenance.
SQLite Release 3.33.0
Support for UPDATE FROM following the PostgreSQL syntax.
Increase the maximum size of database files to 281 TB.
Extended the PRAGMA integrity_check statement so that it can optionally be limited to verifying just a single table and its indexes, rather than the entire database file.
Added the decimal extension for doing arbitrary-precision decimal arithmetic.
Enhancements to the ieee754 extension for working with IEEE 754 binary64 numbers.
CLI enhancements:
Added four new output modes: "box", "json", "markdown", and "table".
The "column" output mode automatically expands columns to contain the longest output row and automatically turns ".header" on if it has not been previously set.
The "quote" output mode honors ".separator"
The decimal extension and the ieee754 extension are built-in to the CLI
Query planner improvements:
Add the ability to find a full-index-scan query plan for queries using INDEXED BY which previously would fail with "no query solution".
Do a better job of detecting missing, incomplete, and/or dodgy sqlite_stat1 data and generates good query plans in spite of the misinformation.
Improved performance of queries like "SELECT min(x) FROM t WHERE y IN (?,?,?)" assuming an index on t(x,y).
In WAL mode, if a writer crashes and leaves the shm file in an inconsistent state, subsequent transactions are now able to recover the shm file even if there are active read transactions. Before this enhancement, shm file recovery that scenario would result in an SQLITE_PROTOCOL error.
Changes in version 3.32.1:
Fix two long-standing bugs that allow malicious SQL statements to crash the process that is running SQLite. These bugs were announced by a third-party approximately 24 hours after the 3.32.0 release but are not specific to the 3.32.0 release.
Other minor compiler-warning fixes and whatnot.
SQLite Release 3.32.0:
Added support for approximate ANALYZE using the PRAGMA analysis_limit command.
Added the bytecode virtual table.
Add the checksum VFS shim to the set of run-time loadable extensions included in the source tree.
Added the iif() SQL function.
INSERT and UPDATE statements now always apply column affinity before computing CHECK constraints. This bug fix could, in theory, cause problems for legacy databases with unorthodox CHECK constraints the require the input type for an INSERT is different from the declared column type. See ticket 86ba67afafded936 for more information.
Added the sqlite3_create_filename(), sqlite3_free_filename(), and sqlite3_database_file_object() interfaces to better support of VFS shim implementations.
Increase the default upper bound on the number of parameters from 999 to 32766.
Added code for the UINT collating sequence as an optional loadable extension.
Enhancements to the CLI:
Add options to the .import command: --csv, --ascii, --skip
The .dump command now accepts multiple LIKE-pattern arguments and outputs the union of all matching tables.
Add the .oom command in debugging builds
Add the --bom option to the .excel, .output, and .once commands.
Enhance the .filectrl command to support the --schema option.
The UINT collating sequence extension is automatically loaded
The ESCAPE clause of a LIKE operator now overrides wildcard characters, so that the behavior now matches what PostgreSQL does.
SQLite Release 3.31.1:
Revert the data layout for an internal-use-only SQLite data structure. Applications that use SQLite should never reference internal SQLite data structures, but some do anyhow, and a change to one such data structure in 3.30.0 broke a popular and widely-deployed application. Reverting that change in SQLite, at least temporarily, gives developers of misbehaving applications time to fix their code.
Fix a typos in the sqlite3ext.h header file that prevented the sqlite3_stmt_isexplain() and sqlite3_value_frombind() interfaces from being called from run-time loadable extensions.
SQLite Release 3.31.0:
Add support for generated columns.
Add the sqlite3_hard_heap_limit64() interface and the corresponding PRAGMA hard_heap_limit command.
Enhance the function_list pragma to show the number of arguments on each function, the type of function (scalar, aggregate, window), and the function property flags SQLITE_DETERMINISTIC, SQLITE_DIRECTONLY, SQLITE_INNOCUOUS, and/or SQLITE_SUBTYPE.
Add the aggregated mode feature to the DBSTAT virtual table.
Add the SQLITE_OPEN_NOFOLLOW option to sqlite3_open_v2() that prevents SQLite from opening symbolic links.
Added the "#-N" array notation for JSON function path arguments.
Added the SQLITE_DBCONFIG_TRUSTED_SCHEMA connection setting which is also controllable via the new trusted_schema pragma and at compile-time using the -DSQLITE_TRUSTED_SCHEMA compile-time option.
Added APIs sqlite3_filename_database(), sqlite3_filename_journal(), and sqlite3_filename_wal() which are useful for specialized extensions.
Add the sqlite3_uri_key() interface.
Upgraded the sqlite3_uri_parameter() function so that it works with the rollback journal or WAL filename in addition to the database filename.
Provide the ability to tag application-defined SQL functions with new properties SQLITE_INNOCUOUS or SQLITE_DIRECTONLY.
Add new verbs to sqlite3_vtab_config() so that the xConnect method of virtual tables can declare the virtual table as SQLITE_VTAB_INNOCUOUS or SQLITE_VTAB_DIRECTONLY.
Faster response to sqlite3_interrupt().
Added the uuid.c extension module implementing functions for processing RFC-4122 UUIDs.
The lookaside memory allocator is enhanced to support two separate memory pools with different sized allocations in each pool. This allows more memory allocations to be covered by lookaside while at the same time reducing the heap memory usage to 48KB per connection, down from 120KB.
The legacy_file_format pragma is deactivated. It is now a no-op. In its place, the SQLITE_DBCONFIG_LEGACY_FILE_FORMAT option to sqlite3_db_config() is provided. The legacy_file_format pragma is deactivated because (1) it is rarely useful and (2) it is incompatible with VACUUM in schemas that have tables with both generated columns and descending indexes.
SQLite Release 3.30.1:
Fix a bug in the query flattener that might cause a segfault for nested queries that use the new FILTER clause on aggregate functions.
Cherrypick fixes for other obscure problems found since the 3.30.0 release.
SQLite Release 3.30.0
Add support for the FILTER clause on aggregate functions.
Add support for the NULLS FIRST and NULLS LAST syntax in ORDER BY clauses.
The index_info and index_xinfo pragmas are enhanced to provide information about the on-disk representation of WITHOUT ROWID tables.
Add the sqlite3_drop_modules() interface, allowing applications to disable automatically loaded virtual tables that they do not need.
Improvements to the .recover dot-command in the CLI so that it recovers more content from corrupt database files.
Enhance the RBU extension to support indexes on expressions.
Change the schema parser so that it will error out if any of the type, name, and tbl_name columns of the sqlite_master table have been corrupted and the database connection is not in writable_schema mode.
The PRAGMA function_list, PRAGMA module_list, and PRAGMA pragma_list commands are now enabled in all builds by default. Disable them using -DSQLITE_OMIT_INTROSPECTION_PRAGMAS.
Add the SQLITE_DBCONFIG_ENABLE_VIEW option for sqlite3_db_config().
Added the TCL Interface config method in order to be able to disable SQLITE_DBCONFIG_ENABLE_VIEW as well as control other sqlite3_db_config() options from TCL.
Added the SQLITE_DIRECTONLY flag for application-defined SQL functions to prevent those functions from being used inside triggers and views.
The legacy SQLITE_ENABLE_STAT3 compile-time option is now a no-op.
SQLite Release 3.29.0:
Added the SQLITE_DBCONFIG_DQS_DML and SQLITE_DBCONFIG_DQS_DDL actions to sqlite3_db_config() for activating and deactivating the double-quoted string literal misfeature. Both default to "on" for legacy compatibility, but developers are encouraged to turn them "off", perhaps using the -DSQLITE_DQS=0 compile-time option.
-DSQLITE_DQS=0 is now a recommended compile-time option.
Improvements to the query planner:
Improved optimization of AND and OR operators when one or the other operand is a constant.
Enhancements to the LIKE optimization for cases when the left-hand side column has numeric affinity.
Added the "sqlite_dbdata" virtual table for extracting raw low-level content from an SQLite database, even a database that is corrupt.
Enhancements to the CLI:
Add the ".recover" command which tries to recover as much content as possible from a corrupt database file.
Add the ".filectrl" command useful for testing.
Add the long-standing ".testctrl" command to the ".help" menu.
Added the ".dbconfig" command
SQLite Release 3.28.0:
Enhanced window functions:
Add support the EXCLUDE clause.
Add support for window chaining.
Add support for GROUPS frames.
Add support for "<expr> PRECEDING" and "<expr> FOLLOWING" boundaries in RANGE frames.
Added the new sqlite3_stmt_isexplain(S) interface for determining whether or not a prepared statement is an EXPLAIN.
Enhanced VACUUM INTO so that it works for read-only databases.
New query optimizations:
Enable the LIKE optimization for cases when the ESCAPE keyword is present and PRAGMA case_sensitive_like is on.
In queries that are driven by a partial index, avoid unnecessary tests of the constraint named in the WHERE clause of the partial index, since we know that constraint must always be true.
Enhancements to the TCL Interface:
Added the -returntype option to the function method.
Added the new bind_fallback method.
Enhancements to the CLI:
Added support for bound parameters and the .parameter command.
Fix the readfile() function so that it returns an empty BLOB rather than throwing an out-of-memory error when reading an empty file.
Fix the writefile() function so that when it creates new directories along the path of a new file, it gives them umask permissions rather than the same permissions as the file.
Change --update option in the .archive command so that it skips files that are already in the archive and are unchanged. Add the new --insert option that works like --update used to work.
Added the fossildelta.c extension that can create, apply, and deconstruct the Fossil DVCS file delta format that is used by the RBU extension.
Added the SQLITE_DBCONFIG_WRITABLE_SCHEMA verb for the sqlite3_db_config() interface, that does the same work as PRAGMA writable_schema without using the SQL parser.
Added the sqlite3_value_frombind() API for determining if the argument to an SQL function is from a bound parameter.
Security and compatibilities enhancements to fts3_tokenizer():
The fts3_tokenizer() function always returns NULL unless either the legacy application-defined FTS3 tokenizers interface are enabled using the sqlite3_db_config(SQLITE_DBCONFIG_ENABLE_FTS3_TOKENIZER) setting, or unless the first argument to fts3_tokenizer() is a bound parameter.
The two-argument version of fts3_tokenizer() accepts a pointer to the tokenizer method object even without the sqlite3_db_config(SQLITE_DBCONFIG_ENABLE_FTS3_TOKENIZER) setting if the second argument is a bound parameter
Improved robustness against corrupt database files.
Miscellaneous performance enhancements
SQLite Release 3.27.2:
Fix a bug in the IN operator that was introduced by an attempted optimization in version 3.27.0.
Fix a bug causing a crash when a window function is misused.
Fix various documentation typos
SQLite Release 3.27.1:
Fix a bug in the query optimizer: an adverse interaction between the OR optimization and the optimization that tries to use values read directly from an expression index instead of recomputing the expression.
Hashes:
Changes carried forward from version 3.27.0:
Added the VACUUM INTO command
Issue an SQLITE_WARNING message on the error log if a double-quoted string literal is used.
The sqlite3_normalized_sql() interface works on any prepared statement created using sqlite3_prepare_v2() or sqlite3_prepare_v3(). It is no longer necessary to use sqlite3_prepare_v3() with SQLITE_PREPARE_NORMALIZE in order to use sqlite3_normalized_sql().
Added the remove_diacritics=2 option to FTS3 and FTS5.
Added the SQLITE_PREPARE_NO_VTAB option to sqlite3_prepare_v3(). Use that option to prevent circular references to shadow tables from causing resource leaks.
Enhancements to the sqlite3_deserialize() interface:
Add the SQLITE_FCNTL_SIZE_LIMIT file-control for setting an upper bound on the size of the in-memory database created by sqlite3_deserialize. The default upper bound is 1GiB, or whatever alternative value is specified by sqlite3_config(SQLITE_CONFIG_MEMDB_MAXSIZE) and/or SQLITE_MEMDB_DEFAULT_MAXSIZE.
Honor the SQLITE_DESERIALIZE_READONLY flag, which was previously described in the documentation, but was previously a no-op.
Enhance the "deserialize" command of the TCL Interface to give it new "--maxsize N" and "--readonly BOOLEAN" options.
Enhancements to the CLI, mostly to support testing and debugging of the SQLite library itself:
Add support for ".open --hexdb". The "dbtotxt" utility program used to generate the text for the "hexdb" is added to the source tree.
Add support for the "--maxsize N" option on ".open --deserialize".
Add the "--memtrace" command-line option, to show all memory allocations and deallocations.
Add the ".eqp trace" option on builds with SQLITE_DEBUG, to enable bytecode program listing with indentation and PRAGMA vdbe_trace all in one step.
Add the ".progress" command for accessing the sqlite3_progress_handler() interface.
Add the "--async" option to the ".backup" command.
Add options "--expanded", "--normalized", "--plain", "--profile", "--row", "--stmt", and "--close" to the ".trace" command.
Increased robustness against malicious SQL that is run against a maliciously corrupted database.
Bug fixes:
Do not use a partial index to do a table scan on an IN operator.
Fix the query flattener so that it works on queries that contain subqueries that use window functions.
Ensure that ALTER TABLE modifies table and column names embedded in WITH clauses that are part of views and triggers.
Fix a parser bug that prevented the use of parentheses around table-valued functions.
Fix a problem with the OR optimization on indexes on expressions.
Fix a problem with the LEFT JOIN strength reduction optimization in which the optimization was being applied inappropriately due to an IS NOT NULL operator.
Fix the REPLACE command so that it is no longer able to sneak a NULL value into a NOT NULL column even if the NOT NULL column has a default value of NULL.
Fix a problem with the use of window functions used within correlated subqueries.
Fix the ALTER TABLE RENAME COLUMN command so that it works for tables that have redundant UNIQUE constraints.
Fix a bug that caused zeroblob values to be truncated when inserted into a table that uses an expression index.
SQLite Release 3.26.0:
Optimization: When doing an UPDATE on a table with indexes on expressions, do not update the expression indexes if they do not refer to any of the columns of the table being updated.
Allow the xBestIndex() method of virtual table implementations to return SQLITE_CONSTRAINT to indicate that the proposed query plan is unusable and should not be given further consideration.
Added the SQLITE_DBCONFIG_DEFENSIVE option which disables the ability to create corrupt database files using ordinary SQL.
Added support for read-only shadow tables when the SQLITE_DBCONFIG_DEFENSIVE option is enabled.
Added the PRAGMA legacy_alter_table command, which if enabled causes the ALTER TABLE command to behave like older version of SQLite (prior to version 3.25.0) for compatibility.
Added PRAGMA table_xinfo that works just like PRAGMA table_info except that it also shows hidden columns in virtual tables.
Added the explain virtual table as a run-time loadable extension.
Add a limit counter to the query planner to prevent excessive sqlite3_prepare() times for certain pathological SQL inputs.
Added support for the sqlite3_normalized_sql() interface, when compiling with SQLITE_ENABLE_NORMALIZE.
Enhanced triggers so that they can use table-valued functions that exist in schemas other than the schema where the trigger is defined.
Enhancements to the CLI:
- Improvements to the ".help" command.
- The SQLITE_HISTORY environment variable, if it exists, specifies the name of the command-line editing history file
- The --deserialize option associated with opening a new database cause the database file to be read into memory and accessed using the sqlite3_deserialize() API. This simplifies running tests on a database without modifying the file on disk.
Enhancements to the geopoly extension:
- Aways stores polygons using the binary format, which is faster and uses less space.
- Added the geopoly_regular() function.
- Added the geopoly_ccw() function.
Enhancements to the session extension:
- Added the SQLITE_CHANGESETAPPLY_INVERT flag
- Added the sqlite3changeset_start_v2() interface and the SQLITE_CHANGESETSTART_INVERT flag.
- Added the changesetfuzz.c test-case generator utility.
SQLite Release 3.25.3:
Disallow the use of window functions in the recursive part of a CTE.
Fix the behavior of typeof() and length() on virtual tables.
Strengthen defenses against deliberately corrupted database files.
Fix a problem in the query planner that results when a row-value expression is used with a PRIMARY KEY with redundant columns.
Fix the query planner so that it works correctly for IS NOT NULL operators in the ON clause of a LEFT JOIN with the SQLITE_ENABLE_STAT4 compile-time option.
SQLite Release 3.25.2:
Add the PRAGMA legacy_alter_table=ON command that causes the "ALTER TABLE RENAME" command to behave as it did in SQLite versions 3.24.0 and earlier: references to the renamed table inside the bodies of triggers and views are not updated. This new pragma provides a compatibility work around for older programs that expected the older, wonky behavior of ALTER TABLE RENAME.
Fix a problem with the new window functions implementation that caused a malfunction when complicated expressions involving window functions were used inside of a view.
Fixes for various other compiler warnings and minor problems associated with obscure configurations.
SQLite Release 3.25.1:
Extra sanity checking added to ALTER TABLE in the 3.25.0 release sometimes raises a false-positive when the table being modified has a trigger that updates a virtual table. The false-positive caused the ALTER TABLE to rollback, thus leaving the schema unchanged.
The fix in the 3.25.0 release for the endless-loop in the byte-code associated with the ORDER BY LIMIT optimization did not work for some queries involving window functions. An additional correction is required.
SQLite Release 3.25.0:
Add support for window functions
Enhancements the ALTER TABLE command:
- Add support for renaming columns within a table using ALTER TABLE table RENAME COLUMN oldname TO newname.
- Fix table rename feature so that it also updates references to the renamed table in triggers and views.
Query optimizer improvements:
- Avoid unnecessary loads of columns in an aggregate query that are not within an aggregate function and that are not part of the GROUP BY clause.
- The IN-early-out optimization: When doing a look-up on a multi-column index and an IN operator is used on a column other than the left-most column, then if no rows match against the first IN value, check to make sure there exist rows that match the columns to the right before continuing with the next IN value.
- Use the transitive property to try to propagate constant values within the WHERE clause. For example, convert "a=99 AND b=a" into "a=99 AND b=99".
Use a separate mutex on every inode in the unix VFS, rather than a single mutex shared among them all, for slightly better concurrency in multi-threaded environments.
Enhance the PRAGMA integrity_check command for improved detection of problems on the page freelist.
Output infinity as 1e999 in the ".dump" command of the command-line shell.
Added the SQLITE_FCNTL_DATA_VERSION file-control.
Added the Geopoly module
Bug fixes:
The ORDER BY LIMIT optimization might have caused an infinite loop in the byte code of the prepared statement under very obscure circumstances, due to a confluence of minor defects in the query optimizer.
On an UPSERT when the order of constraint checks is rearranged, ensure that the affinity transformations on the inserted content occur before any of the constraint checks.
Avoid using a prepared statement for ".stats on" command of the CLI after it has been closed by the ".eqp full" logicc.
The LIKE optimization was generating incorrect byte-code and hence getting the wrong answer if the left-hand operand has numeric affinity and the right-hand-side pattern is '/%' or if the pattern begins with the ESCAPE character.
SQLite Release 3.24.0:
Add support for PostgreSQL-style UPSERT.
Add support for auxiliary columns in r-tree tables.
Add C-language APIs for discovering SQL keywords used by SQLite: sqlite3_keyword_count(), sqlite3_keyword_name(), and sqlite3_keyword_check().
Add C-language APIs for dynamic strings based on the sqlite3_str object.
Enhance ALTER TABLE so that it recognizes "true" and "false" as valid arguments to DEFAULT.
Add the sorter-reference optimization as a compile-time option. Only available if compiled with SQLITE_ENABLE_SORTER_REFERENCES.
Improve the format of the EXPLAIN QUERY PLAN raw output, so that it gives better information about the query plan and about the relationships between the various components of the plan.
Added the SQLITE_DBCONFIG_RESET_DATABASE option to the sqlite3_db_config() API.
CLI Enhancements:
Automatically intercepts the raw EXPLAIN QUERY PLAN output and reformats it into an ASCII-art graph.
Lines that begin with "#" and that are not in the middle of an SQL statement are interpreted as comments.
Added the --append option to the ".backup" command.
Added the ".dbconfig" command.
Performance:
UPDATE avoids unnecessary low-level disk writes when the contents of the database file do not actually change. For example, "UPDATE t1 SET x=25 WHERE y=?" generates no extra disk I/O if the value in column x is already 25. Similarly, when doing UPDATE on records that span multiple pages, only the subset of pages that actually change are written to disk. This is a low-level performance optimization only and does not affect the behavior of TRIGGERs or other higher level SQL structures.
Queries that use ORDER BY and LIMIT now try to avoid computing rows that cannot possibly come in under the LIMIT. This can greatly improve performance of ORDER BY LIMIT queries, especially when the LIMIT is small relative to the number of unrestricted output rows.
The OR optimization is allowed to proceed even if the OR expression has also been converted into an IN expression. Uses of the OR optimization are now also more clearly shown in the EXPLAIN QUERY PLAN output.
The query planner is more aggressive about using automatic indexes for views and subqueries for which it is not possible to create a persistent index.
Make use of the one-pass UPDATE and DELETE query plans in the R-Tree extension where appropriate.
Performance improvements in the LEMON-generated parser.
Bug fixes:
For the right-hand table of a LEFT JOIN, compute the values of expressions directly rather than loading precomputed values out of an expression index as the expression index might not contain the correct value.
Do not attempt to use terms from the WHERE clause to enable indexed lookup of the right-hand table of a LEFT JOIN.
Fix a memory leak that can occur following a failure to open error in the CSV virtual table
Fix a long-standing problem wherein a corrupt schema on the sqlite_sequence table used by AUTOINCREMENT can lead to a crash.
Fix the json_each() function so that it returns valid results on its "fullkey" column when the input is a simple value rather than an array or object.
SQLite Release 3.23.1:
Fix two problems in the new LEFT JOIN strength reduction optimization.
Fix misbehavior of the FTS5 xBestIndex method.
Fix a harmless reference to an uninitialized virtual machine register.
Fix the CLI so that it builds with -DSQLITE_UNTESTABLE
Fix the eval.c extension so that it works with PRAGMA empty_result_callbacks=ON.
Fix the generate_series virtual table so that it correctly returns no rows if any of its constraints are NULL.
Performance enhancements in the parser.
SQLite Release 3.23.0:
Add the sqlite3_serialize() and sqlite3_deserialize() interfaces when the SQLITE_ENABLE_DESERIALIZE compile-time option is used.
Recognize TRUE and FALSE as constants. (For compatibility, if there exist columns named "true" or "false", then the identifiers refer to the columns rather than Boolean constants.)
Support operators IS TRUE, IS FALSE, IS NOT TRUE, and IS NOT FALSE.
Added the SQLITE_DBSTATUS_CACHE_SPILL option to sqlite3_db_status() for reporting the number of cache spills that have occurred.
The "alternate-form-2" flag ("!") on the built-in printf implementation now causes string substitutions to measure the width and precision in characters instead of bytes.
If the xColumn method in a virtual table implementation returns an error message using sqlite3_result_error() then give that error message preference over internally-generated messages.
Added the -A command-line option to the CLI to make it easier to manage SQLite Archive files.
Add support for INSERT OR REPLACE, INSERT OR IGNORE, and UPDATE OR REPLACE in the Zipfile virtual table.
Enhance the sqlite3changeset_apply() interface so that it is hardened against attacks from deliberately corrupted changeset objects.
Added the sqlite3_normalize() extension function.
Query optimizer enhancements:
Improve the omit-left-join optimization so that it works in cases where the right-hand table is UNIQUE but not necessarily NOT NULL.
Improve the push-down optimization so that it works for many LEFT JOINs.
Add the LEFT JOIN strength reduction optimization that converts a LEFT JOIN into an ordinary JOIN if there exist terms in the WHERE clause that would prevent the extra all-NULL row of the LEFT JOIN from appearing in the output set.
Avoid unnecessary writes to the sqlite_sequence table when an AUTOINCREMENT table is updated with an rowid that is less than the maximum.
Bug fixes:
Fix the parser to accept valid row value syntax.
Fix the query planner so that it takes into account dependencies in the arguments to table-valued functions in subexpressions in the WHERE clause.
Fix incorrect result with complex OR-connected WHERE and STAT4.
Fix potential corruption in indexes on expressions due to automatic datatype conversions.
Assertion fault in FTS4.
Incorrect result on the less-than operator in row values.
Always interpret non-zero floating-point values as TRUE, even if the integer part is zero.
Fix an issue in the fsdir(PATH) table-valued function to the fileio.c extension, that caused a segfault if the fsdir() table was used as the inner table of a join.
Issue an error rather instead of an assertion-fault or null-pointer dereference when the sqlite_master table is corrupted so that the sqlite_sequence table root page is really a btree-index page.
Fix the ANALYZE command so that it computes statistics on tables whose names begin with "sqlite".
Additional fixes for issues detected by OSSFuzz:
Fix a possible infinite loop on VACUUM for corrupt database files.
Disallow parameters in the WITH clause of triggers and views.
Fix a potential memory leak in row value processing.
Improve the performance of the replace() SQL function for cases where there are many substitutions on megabyte-sized strings, in an attempt to avoid OSSFuzz timeouts during testing.
Provide an appropriate error message when the sqlite_master table contains a CREATE TABLE AS statement. Formerly this caused either an assertion fault or null pointer dereference. Problem found by OSSFuzz on the GDAL project.
Incorrect assert() statement removed.
Fix a problem with using the LIKE optimization on an INTEGER PRIMARY KEY.
Release 3.22.0:
The output of sqlite3_trace_v2() now shows each individual SQL statement run within a trigger.
Add the ability to read from WAL mode databases even if the application lacks write permission on the database and its containing directory, as long as the -shm and -wal files exist in that directory.
Added the rtreecheck() scalar SQL function to the R-Tree extension.
Added the sqlite3_vtab_nochange() and sqlite3_value_nochange() interfaces to help virtual table implementations optimize UPDATE operations.
Added the sqlite3_vtab_collation() interface.
Added support for the "^" initial token syntax in FTS5.
New extensions:
The Zipfile virtual table can read and write a ZIP Archive.
Added the fsdir(PATH) table-valued function to the fileio.c extension, for listing the files in a directory.
The sqlite_btreeinfo eponymous virtual table for introspecting and estimating the sizes of the btrees in a database.
The Append VFS is a VFS shim that allows an SQLite database to be appended to some other file. This allows (for example) a database to be appended to an executable that then opens and reads the database.
Query planner enhancements:
The optimization that uses an index to quickly compute an aggregate min() or max() is extended to work with indexes on expressions.
The decision of whether to implement a FROM-clause subquery as a co-routine or using query flattening now considers whether the result set of the outer query is "complex" (if it contains functions or expression subqueries). A complex result set biases the decision toward the use of co-routines.
The planner avoids query plans that use indexes with unknown collating functions.
The planner omits unused LEFT JOINs even if they are not the right-most joins of a query.
Other performance optimizations:
A smaller and faster implementation of text to floating-point conversion subroutine: sqlite3AtoF().
The Lemon parser generator creates a faster parser.
Use the strcspn() C-library routine to speed up the LIKE and GLOB operators.
Improvements to the command-line shell:
The ".schema" command shows the structure of virtual tables.
Added support for reading and writing SQL Archive files using the .archive command.
Added the experimental .expert command
Added the ".eqp trigger" variant of the ".eqp" command
Enhance the ".lint fkey-indexes" command so that it works with WITHOUT ROWID tables.
If the filename argument to the shell is a ZIP archive rather than an SQLite database, then the shell automatically opens that ZIP archive using the Zipfile virtual table.
Added the edit() SQL function.
Added the .excel command to simplify exporting database content to a spreadsheet.
Databases are opened using Append VFS when the --append flag is used on the command line or with the .open command.
Enhance the SQLITE_ENABLE_UPDATE_DELETE_LIMIT compile-time option so that it works for WITHOUT ROWID tables.
Provide the sqlite_offset(X) SQL function that returns the byte offset into the database file to the beginning of the record holding value X, when compiling with -DSQLITE_ENABLE_OFFSET_SQL_FUNC.
Bug fixes
Release 3.21.0:
Take advantage of the atomic-write capabilities in the F2FS filesystem when available, for greatly reduced transaction overhead. This currently requires the SQLITE_ENABLE_BATCH_ATOMIC_WRITE compile-time option.
Allow ATTACH and DETACH commands to work inside of a transaction.
Allow WITHOUT ROWID virtual tables to be writable if the PRIMARY KEY contains exactly one column.
The "fsync()" that occurs after the header is written in a WAL reset now uses the sync settings for checkpoints. This means it will use a "fullfsync" on macs if PRAGMA checkpoint_fullfsync set on.
The sqlite3_sourceid() function tries to detect if the source code has been modified from what is checked into version control and if there are modifications, the last four characters of the version hash are shown as "alt1" or "alt2". The objective is to detect accidental and/or careless edits. A forger can subvert this feature.
Improved de-quoting of column names for CREATE TABLE AS statements with an aggregate query on the right-hand side.
Fewer "stat()" system calls issued by the unix VFS.
Enhanced the LIKE optimization so that it works with an ESCAPE clause.
Enhanced PRAGMA integrity_check and PRAGMA quick_check to detect obscure row corruption that they were formerly missing. Also update both pragmas so that they return error text rather than SQLITE_CORRUPT when encountering corruption in records.
The query planner now prefers to implement FROM-clause subqueries using co-routines rather using the query flattener optimization. Support for the use of co-routines for subqueries may no longer be disabled.
Pass information about !=, IS, IS NOT, NOT NULL, and IS NULL constraints into the xBestIndex method of virtual tables.
Enhanced the CSV virtual table so that it accepts the last row of input if the final new-line character is missing.
Remove the rarely-used "scratch" memory allocator. Replace it with the SQLITE_CONFIG_SMALL_MALLOC configuration setting that gives SQLite a hint that large memory allocations should be avoided when possible.
Added the swarm virtual table to the existing union virtual table extension.
Added the sqlite_dbpage virtual table for providing direct access to pages of the database file. The source code is built into the amalgamation and is activated using the -DSQLITE_ENABLE_DBPAGE_VTAB compile-time option.
Add a new type of fts5vocab virtual table - "instance" - that provides direct access to an FTS5 full-text index at the lowest possible level.
Remove a call to rand_s() in the Windows VFS since it was causing problems in Firefox on some older laptops.
The src/shell.c source code to the command-line shell is no longer under version control. That file is now generated as part of the build process.
Miscellaneous microoptimizations reduce CPU usage by about 2.1%.
Bug fixes
Fix more bugs in the LEFT JOIN flattening optimization. Ticket 7fde638e94287d2c.
Changes carried forward from version 3.19.1 (2017-05-24):
Fix a bug in the LEFT JOIN flattening optimization. Ticket cad1ab4cb7b0fc.
Remove a surplus semicolon that was causing problems for older versions of MSVC.
---------------------
The SQLITE_READ authorizer callback is invoked once with a column name that is an empty string for every table referenced in a query from which no columns are extracted.
When using an index on an expression, try to use expression values already available in the index, rather than loading the original columns and recomputing the expression.
Enhance the flattening optimization so that it is able to flatten views on the right-hand side of a LEFT JOIN.
Use replace() instead of char() for escaping newline and carriage-return characters embedded in strings in the .dump output from the command-line shell.
Avoid unnecessary foreign key processing in UPDATE statements that do not touch the columns that are constrained by the foreign keys.
On a DISTINCT query that uses an index, try to skip ahead to the next distinct entry using the index rather than stepping through rows, when an appropriate index is available.
Avoid unnecessary invalidation of sqlite3_blob handles when making changes to unrelated tables.
Transfer any terms of the HAVING clause that use only columns mentioned in the GROUP BY clause over to the WHERE clause for faster processing.
Reuse the same materialization of a VIEW if that VIEW appears more than once in the same query.
Enhance PRAGMA integrity_check so that it identifies tables that have two or more rows with the same rowid.
Enhance the FTS5 query syntax so that column filters may be applied to arbitrary expressions.
Enhance the json_extract() function to cache and reuse parses of JSON input text.
Added the anycollseq.c loadable extension that allows a generic SQLite database connection to read a schema that contains unknown and/or application-specific collating sequences.
Bug Fixes:
----------
Fix a problem in REPLACE that can result in a corrupt database containing two or more rows with the same rowid.
Fix a problem in PRAGMA integrity_check that was causing a subsequent VACUUM to behave suboptimally.
Fix the PRAGMA foreign_key_check command so that it works correctly with foreign keys on WITHOUT ROWID tables.
Disallow leading zeros in numeric constants in JSON.
Disallow control characters inside of strings in JSON.
Limit the depth of recursion for JSON objects and arrays in order to avoid excess stack usage in the recursive descent parser.
Added the PRAGMA optimize command
The SQLite version identifier returned by the sqlite_source_id() SQL function and the sqlite3_sourceid() C API and found in the SQLITE_SOURCE_ID macro is now a 64-digit SHA3-256 hash instead of a 40-digit SHA1 hash.
Added the json_patch() SQL function to the JSON1 extension.
Enhance the LIKE optimization so that it works for arbitrary expressions on the left-hand side as long as the LIKE pattern on the right-hand side does not begin with a digit or minus sign.
Added the sqlite3_set_last_insert_rowid() interface and use the new interface in the FTS3, FTS4, and FTS5 extensions to ensure that the sqlite3_last_insert_rowid() interface always returns reasonable values.
Enhance PRAGMA integrity_check and PRAGMA quick_check so that they verify CHECK constraints.
Enhance the query plans for joins to detect empty tables early and halt without doing unnecessary work.
Enhance the sqlite3_mprintf() family of interfaces and the printf SQL function to put comma separators at the thousands marks for integers, if the "," format modifier is used in between the "%" and the "d" (example: "%,d").
Added the -DSQLITE_MAX_MEMORY=N compile-time option.
Added the .sha3sum dot-command and the .selftest dot-command to the command-line shell
Begin enforcing SQLITE_LIMIT_VDBE_OP. This can be used, for example, to prevent excessively large prepared statements in systems that accept SQL queries from untrusted users.
Various performance improvements.
Bug Fixes:
----------
Ensure that indexed expressions with collating sequences are handled correctly. Fix for ticket eb703ba7b50c1a5.
Fix a bug in the 'start of ...' modifiers for the date and time functions. Ticket 6097cb92745327a1
Fix a potential segfault in complex recursive triggers, resulting from a bug in the OP_Once opcode introduced as part of a performance optimization in version 3.15.0. Ticket 06796225f59c057c
In the RBU extension, add extra sync operations to avoid the possibility of corruption following a power failure.
The sqlite3_trace_v2() output for nested SQL statements should always begin with a "--" comment marker.
* Approximately 25% better performance from the R-Tree extension.
* Uses compiler built-ins (ex: __builtin_bswap32() or _byteswap_ulong()) for byteswapping when available.
* Uses the sqlite3_blob key/value access object instead of SQL for pulling content out of R-Tree nodes
* Other miscellaneous enhancements such as loop unrolling.
* Add the SQLITE_DEFAULT_LOOKASIDE compile-time option.
* Increase the default lookaside size from 512,125 to 1200,100 as this provides better performance while only adding 56KB of extra memory per connection. Memory-sensitive applications can restore the old default at compile-time, start-time, or run-time.
* Use compiler built-ins __builtin_sub_overflow(), __builtin_add_overflow(), and __builtin_mul_overflow() when available. (All compiler built-ins can be omitted with the SQLITE_DISABLE_INTRINSIC compile-time option.)
* Added the SQLITE_ENABLE_NULL_TRIM compile-time option, which can result in significantly smaller database files for some applications, at the risk of being incompatible with older versions of SQLite.
* Change SQLITE_DEFAULT_PCACHE_INITSZ from 100 to 20, for improved performance.
* Added the SQLITE_UINT64_TYPE compile-time option as an analog to SQLITE_INT64_TYPE.
* Perform some UPDATE operations in a single pass instead of in two passes.
* Enhance the session extension to support WITHOUT ROWID tables.
* Fixed performance problems and potential stack overflows when creating views from multi-row VALUES clauses with hundreds of thousands of rows.
* Added the sha1.c extension.
* In the command-line shell, enhance the ".mode" command so that it restores the default column and row separators for modes "line", "list", "column", and "tcl".
* Enhance the SQLITE_DIRECT_OVERFLOW_READ option so that it works in WAL mode as long as the pages being read are not in the WAL file.
* Enhance the LEMON parser generator so that it can store the parser object as a stack variable rather than allocating space from the heap and make use of that enhancement in the amalgamation.
* Other performance improvements. Uses about 6.5% fewer CPU cycles.
Bug Fixes:
* Throw an error if the ON clause of a LEFT JOIN references tables to the right of the ON clause. This is the same behavior as PostgreSQL. Formerly, SQLite silently converted the LEFT JOIN into an INNER JOIN.
* Use the correct affinity for columns of automatic indexes.
* Ensure that the sqlite3_blob_reopen() interface can correctly handle short rows.
Fix the REPLACE statement for WITHOUT ROWID tables that lack secondary indexes so that it works correctly with triggers and foreign keys. This was a new bug caused by performance optimizations added in version 3.16.0. Ticket 30027b613b4
Fix the sqlite3_value_text() interface so that it correctly translates content generated by zeroblob() into a string of all 0x00 characters. This is a long-standing issue discovered after the 3.16.1 release by OSS-Fuzz
Fix the bytecode generator to deal with a subquery in the FROM clause that is itself a UNION ALL where one side of the UNION ALL is a view that contains an ORDER BY. This is a long-standing issue that was discovered after the release of 3.16.1. See ticket 190c2507.
Adjust the sqlite3_column_count() API so it more often returns the same values for PRAGMA statements as it did in prior releases, to minimize disruption to applications that might be using that interface in unexpected ways.
Fix a bug concerning the use of row values within triggers (see ticket 8c9458e7) that was in version 3.15.0 but was not reported until moments after the 3.16.0 release was published.
Uses 9% fewer CPU cycles. (See the CPU performance measurement report for details on how the this performance increase was computed.)
Added experimental support for PRAGMA functions.
Added the SQLITE_DBCONFIG_NO_CKPT_ON_CLOSE option to sqlite3_db_config().
Enhance the date and time functions so that the 'unixepoch' modifier works for the full span of supported dates.
Changed the default configuration of the lookaside memory allocator from 500 slots of 128 bytes each into 125 slots of 512 bytes each.
Enhanced "WHERE x NOT NULL" partial indexes so that they are usable if the "x" column appears in a LIKE or GLOB operator.
Enhanced sqlite3_interrupt() so that it interrupts checkpoint operations that are in process.
Enhanced the LIKE and GLOB matching algorithm to be faster for cases when the pattern contains multiple wildcards.
Added the SQLITE_FCNTL_WIN32_GET_HANDLE file control opcode.
Added ".mode quote" to the command-line shell.
Added ".lint fkey-indexes" to the command-line shell.
Added the .imposter dot-command to the command-line shell.
Added the remember(V,PTR) SQL function as a loadable extension.
Rename the SQLITE_OMIT_BUILTIN_TEST compile-time option to SQLITE_UNTESTABLE to better reflect the implications of using it.
Bug Fixes:
Fix a long-standing bug in the query planner that caused incorrect results on a LEFT JOIN where the left-hand table is a subquery and the join constraint is a bare column name coming from the left-hand subquery. Ticket 2df0107b.
Correctly handle the integer literal -0x8000000000000000 in the query planner.
* Added SQLITE_FCNTL_WIN32_GET_HANDLE file control opcode.
Bug Fixes:
* Fix the VACUUM command so that it spills excess content to disk rather than holding everything in memory, and possible causing an out-of-memory error for larger database files. This fixes an issue introduced by version 3.15.0.
* Fix a case (present since 3.8.0 - 2013-08-26) where OR-connected terms in the ON clause of a LEFT JOIN might cause incorrect results.
* Fix a case where the use of row values in the ON clause of a LEFT JOIN might cause incorrect results.
Added support for row values.
Allow deterministic SQL functions in the WHERE clause of a partial index.
Added the "modeof=filename" URI parameter on the unix VFS
Added support for SQLITE_DBCONFIG_MAINDBNAME.
Added the ability to VACUUM an ATTACH-ed database.
Enhancements to the command-line shell:
Add the ".testcase" and ".check" dot-commands.
Added the --new option to the ".open" dot-command, causing any prior content in the database to be purged prior to opening.
Enhance the fts5vocab virtual table to handle "ORDER BY term" efficiently.
Miscellaneous micro-optimizations reduce CPU usage by more than 7% on common workloads. Most optimization in this release has been on the front-end (sqlite3_prepare_v2()).
Bug Fixes:
The multiply operator now correctly detects 64-bit integer overflow and promotes to floating point in all corner-cases.
Correct handling of columns with redundant unique indexes when those columns are used on the LHS of an IN operator.
Skip NULL entries on range queries in indexes on expressions.
Ensure that the AUTOINCREMENT counters in the sqlite_sequence table are initialized doing "Xfer Optimization" on "INSERT ... SELECT" statements.
Make sure the ORDER BY LIMIT optimization (from check-in 559733b09e) works with IN operators on INTEGER PRIMARY KEYs.
* Improved support for using the STDCALL calling convention in winsqlite3.dll.
* Fix the sqlite3_trace_v2() interface so that it is disabled if either the callback or the mask arguments are zero, in accordance with the documentation.
* Fix commenting errors and improve the comments generated on EXPLAIN listings when the -DSQLITE_ENABLE_EXPLAIN_COMMENTS compile-time option is used.
* Fix the ".read" command in the command-line shell so that it understands that its input is not interactive.
* Correct affinity computations for a SELECT on the RHS of an IN operator.
* The ORDER BY LIMIT optimization is not valid unless the inner-most IN operator loop is actually used by the query plan. Fix for ticket 0c4df46116e90f92.
* Fix an internal code generator problem that was causing some DELETE operations to no-op.
* A performance enhancement to the page-cache "truncate" operation reduces COMMIT time by dozens of milliseconds on systems with a large page cache.
* Fix to the --rbu option of sqldiff.
Added support for WITHOUT ROWID virtual tables.
Improved the query planner so that the OR optimization can be used on virtual tables even if one or more of the disjuncts use the LIKE, GLOB, REGEXP, MATCH operators.
Added the CSV virtual table for reading RFC 4180 formatted comma-separated value files.
Added the carray() table-valued function extension.
Enabled persistent loadable extensions using the new SQLITE_OK_LOAD_PERMANENTLY return code from the extension entry point.
Added the SQLITE_DBSTATUS_CACHE_USED_SHARED option to sqlite3_db_status().
Add the vfsstat.c loadable extension - a VFS shim that measures I/O together with an eponymous virtual table that provides access to the measurements.
Improved algorithm for running queries with both an ORDER BY and a LIMIT where only the inner-most loop naturally generates rows in the correct order.
Enhancements to Lemon parser generator, so that it generates a faster parser.
The PRAGMA compile_options command now attempts to show the version number of the compiler that generated the library.
Enhance PRAGMA table_info so that it provides information about eponymous virtual tables.
Added the "win32-none" VFS, analogous to the "unix-none" VFS, that works like the default "win32" VFS except that it ignores all file locks.
The query planner uses a full scan of a partial index instead of a full scan of the main table, in cases where that makes sense.
Allow table-valued functions to appear on the right-hand side of an IN operator.
Created the dbhash.exe command-line utility.
Added two new C-language interfaces: sqlite3_expanded_sql() and sqlite3_trace_v2(). These new interfaces subsume the functions of sqlite3_trace() and sqlite3_profile() which are now deprecated.
Added the json_quote() SQL function to the json1 extension.
Disable the authorizer callback while reparsing the schema.
Added the SQLITE_ENABLE_UNKNOWN_SQL_FUNCTION compile-time option and turned that option on by default when building the command-line shell.
Bug Fixes:
Fix the ALTER TABLE command so that it does not corrupt descending indexes when adding a column to a legacy file format database.
Fix a NULL-pointer dereference/crash that could occurs when a transitive WHERE clause references a non-existent collating sequence.
Improved the cost estimation for an index scan which includes a WHERE clause that can be partially or fully evaluated using columns in the index and without having to do a table lookup. This fixes a performance regression that occurred for some obscure queries following the ORDER BY LIMIT optimization introduced in version 3.12.0.
Postpone I/O associated with TEMP files for as long as possible, with the hope that the I/O can ultimately be avoided completely.
Merged the session extension into trunk.
Added the ".auth ON|OFF" command to the command-line shell.
Added the "--indent" option to the ".schema" and ".fullschema" commands of the command-line shell, to turn on pretty-printing.
Added the ".eqp full" option to the command-line shell, that does both EXPLAIN and EXPLAIN QUERY PLAN on each statement that is evaluated.
Improved unicode filename handling in the command-line shell on Windows.
Improved resistance against goofy query planner decisions caused by incomplete or incorrect modifications to the sqlite_stat1 table by the application.
Added the sqlite3_db_config(db,SQLITE_DBCONFIG_ENABLE_LOAD_EXTENSION) interface which allows the sqlite3_load_extension() C-API to be enabled while keeping the load_extension() SQL function disabled for security.
Change the temporary directory search algorithm on Unix to allow directories with write and execute permission, but without read permission, to serve as temporary directories. Apply this same standard to the "." fallback directory.
Bug Fixes:
----------
Fix a problem with the multi-row one-pass DELETE optimization that was causing it to compute incorrect answers with a self-referential subquery in the WHERE clause. Fix for ticket dc6ebeda9396087
Fix a possible segfault with DELETE when table is a rowid table with an INTEGER PRIMARY KEY and the WHERE clause contains a OR and the table has one or more indexes that are able to trigger the OR optimization, but none of the indexes reference any table columns other than the INTEGER PRIMARY KEY. Ticket 16c9801ceba49.
When checking for the WHERE-clause push-down optimization, verify that all terms of the compound inner SELECT are non-aggregate, not just the last term. Fix for ticket f7f8c97e97597.
Fix a locking race condition in Windows that can occur when two or more processes attempt to recover the same hot journal at the same time.
Fix a backwards compatibility problem in version 3.12.0 and 3.12.1: Columns declared as "INTEGER" PRIMARY KEY (with quotes around the datatype keyword) where not being recognized as an INTEGER PRIMARY KEY, which resulted in an incompatible database file.
Fix a bug (present since version 3.9.0) that can cause the DELETE operation to miss rows if PRAGMA reverse_unordered_selects is turned on.
Fix a bug in the code generator that can causes incorrect results if two or more virtual tables are joined and the virtual table used in outer loop of the join has an IN operator constraint.
Correctly interpret negative "PRAGMA cache_size" values when determining the cache size used for sorting large amounts of data.
Fix a boundary condition error introduced by version 3.12.0 that can result in a crash during heavy SAVEPOINT usage.
Fix views so that they inherit column datatypes from the table that they are defined against, when possible.
Fix the query planner so that IS and IS NULL operators are able to drive an index on a LEFT OUTER JOIN.
