- 1. Overview
- 2. Recommended Compile-time Options
- 3. Platform Configuration
- 4. Options To Set Default Parameter Values
- 5. Options To Set Size Limits
- 6. Options To Control Operating Characteristics
- 7. Options To Enable Features Normally Turned Off
- 8. Options To Disable Features Normally Turned On
- 9. Options To Omit Features
For most purposes, SQLite can be built just fine using the default compilation options. However, if required, the compile-time options documented below can be used to omit SQLite features (resulting in a smaller compiled library size) or to change the default values of some parameters.
Every effort has been made to ensure that the various combinations of compilation options work harmoniously and produce a working library. Nevertheless, it is strongly recommended that the SQLite test-suite be executed to check for errors before using an SQLite library built with non-standard compilation options.
2. Recommended Compile-time Options
The following compile-time options are recommended for applications that are able to use them, in order to minimized the number of CPU cycles and the bytes of memory used by SQLite. Not all of these compile-time options are usable by every application. For example, the SQLITE_THREADSAFE=0 option is only usable by applications that never access SQLite from more than one thread at a time. And the SQLITE_OMIT_PROGRESS_CALLBACK option is only usable by applications that do not use the sqlite3_progress_handler() interface. And so forth.
It is impossible to test every possible combination of compile-time options for SQLite. But the following set of compile-time options is one configuration that is always fully tested.
SQLITE_THREADSAFE=0. Setting -DSQLITE_THREADSAFE=0 causes all of the mutex and thread-safety logic in SQLite to be omitted. This is the single compile-time option causes SQLite to run about 2% faster and also reduces the size of the library by about 2%. But the downside is that using the compile-time option means that SQLite can never be used by more than a single thread at a time, even if each thread has its own database connection.
SQLITE_DEFAULT_MEMSTATUS=0. This setting causes the sqlite3_status() interfaces that track memory usage to be disabled. This helps the sqlite3_malloc() routines run much faster, and since SQLite uses sqlite3_malloc() internally, this helps to make the entire library faster.
SQLITE_DEFAULT_WAL_SYNCHRONOUS=1. For maximum database safety following a power loss, the setting of PRAGMA synchronous=FULL is recommended. However, in WAL mode, complete database integrity is guaranteed with PRAGMA synchronous=NORMAL. With PRAGMA synchronous=NORMAL in WAL mode, recent changes to the database might be rolled back by a power loss, but the database will not be corrupted. Furthermore, transaction commit is much faster in WAL mode using synchronous=NORMAL than with the default synchronous=FULL. For these reasons, it is recommended that the synchronous setting be changed from FULL to NORMAL when switching to WAL mode. This compile-time option will accomplish that.
SQLITE_LIKE_DOESNT_MATCH_BLOBS. Historically, SQLite has allowed BLOB operands to the LIKE and GLOB operators. But having a BLOB as an operand of LIKE or GLOB complicates and slows the LIKE optimization. When this option is set, it means that the LIKE and GLOB operators always return FALSE if either operand is a BLOB. That simplifies the implementation of the LIKE optimization and allows queries that use the LIKE optimization to run faster.
SQLITE_MAX_EXPR_DEPTH=0. Setting the maximum expression parse-tree depth to zero disables all checking of the expression parse-tree depth, which simplifies the code resulting in faster execution, and helps the parse tree to use less memory.
SQLITE_OMIT_DECLTYPE. By omitting the (seldom-needed) ability to return the declared type of columns from the result set of query, prepared statements can be made to consume less memory.
SQLITE_OMIT_DEPRECATED. Omitting deprecated interfaces and features will not help SQLite to run any faster. It will reduce the library footprint, however. And it is the right thing to do.
SQLITE_OMIT_PROGRESS_CALLBACK. The progress handler callback counter must be checked in the inner loop of the bytecode engine. By omitting this interface, a single conditional is removed from the inner loop of the bytecode engine, helping SQL statements to run slightly faster.
SQLITE_OMIT_SHARED_CACHE. Omitting the possibility of using shared cache allows many conditionals in performance-critical sections of the code to be eliminated. This can give a noticeable improvement in performance.
SQLITE_USE_ALLOCA. Make use of alloca() for dynamically allocating temporary stack space for use within a single function, on systems that support alloca(). Without this option, temporary space is allocated from the heap.
SQLITE_OMIT_AUTOINIT. The SQLite library needs to be initialized using a call to sqlite3_initialize() before certain interfaces are used. This initialization normally happens automatically the first time it is needed. However, with the SQLITE_OMIT_AUTOINIT option, the automatic initialization is omitted. This helps many API calls to run a little faster (since they do not have to check to see if initialization has already occurred and then run initialization if it has not previously been invoked) but it also means that the application must call sqlite3_initialize() manually. If SQLite is compiled with -DSQLITE_OMIT_AUTOINIT and a routine like sqlite3_malloc() or sqlite3_vfs_find() or sqlite3_open() is invoked without first calling sqlite3_initialize(), the likely result will be a segfault.
When all of the recommended compile-time options above are used, the SQLite library will be approximately 3% smaller and use about 5% fewer CPU cycles. So these options do not make a huge difference. But in some design situations, every little bit helps.
3. Platform Configuration
If the _HAVE_SQLITE_CONFIG_H macro is defined then the SQLite source code will attempt to #include a file named «config.h». The «config.h» file usually contains other configuration options, especially «HAVE_INTERFACE» type options generated by autoconf scripts.
If the HAVE_FDATASYNC compile-time option is true, then the default VFS for unix systems will attempt to use fdatasync() instead of fsync() where appropriate. If this flag is missing or false, then fsync() is always used.
If the HAVE_GMTIME_R option is true and if SQLITE_OMIT_DATETIME_FUNCS is true, then the CURRENT_TIME, CURRENT_DATE, and CURRENT_TIMESTAMP keywords will use the threadsafe «gmtime_r()» interface rather than «gmtime()». In the usual case where SQLITE_OMIT_DATETIME_FUNCS is not defined or is false, then the built-in date and time functions are used to implement the CURRENT_TIME, CURRENT_DATE, and CURRENT_TIMESTAMP keywords and neither gmtime_r() nor gmtime() is ever called.
If the HAVE_ISNAN option is true, then SQLite invokes the system library isnan() function to determine if a double-precision floating point value is a NaN. If HAVE_ISNAN is undefined or false, then SQLite substitutes its own home-grown implementation of isnan().
If the HAVE_LOCALTIME_R option is true, then SQLite uses the threadsafe localtime_r() library routine instead of localtime() to help implement the localtime modifier to the built-in date and time functions.
If the HAVE_LOCALTIME_S option is true, then SQLite uses the threadsafe localtime_s() library routine instead of localtime() to help implement the localtime modifier to the built-in date and time functions.
If the HAVE_MALLOC_USABLE_SIZE option is true, then SQLite tries uses the malloc_usable_size() interface to find the size of a memory allocation obtained from the standard-library malloc() or realloc() routines. This option is only applicable if the standard-library malloc() is used. On Apple systems, «zone malloc» is used instead, and so this option is not applicable. And, of course, if the application supplies its own malloc implementation using SQLITE_CONFIG_MALLOC then this option has no effect.
If the HAVE_MALLOC_USABLE_SIZE option is omitted or is false, then SQLite uses a wrapper around system malloc() and realloc() that enlarges each allocation by 8 bytes and writes the size of the allocation in the initial 8 bytes, and then SQLite also implements its own home-grown version of malloc_usable_size() that consults that 8-byte prefix to find the allocation size. This approach works but it is suboptimal. Applications are encouraged to use HAVE_MALLOC_USABLE_SIZE whenever possible.
If the HAVE_STRCHRNUL option is true, then SQLite uses the strchrnul() library function. If this option is missing or false, then SQLite substitutes its own home-grown implementation of strchrnul().
If the HAVE_USLEEP option is true, then the default unix VFS uses the usleep() system call to implement the xSleep method. If this option is undefined or false, then xSleep on unix is implemented using sleep() which means that sqlite3_sleep() will have a minimum wait interval of 1000 milliseconds regardless of its argument.
If the HAVE_UTIME option is true, then the built-in but non-standard «unix-dotfile» VFS will use the utime() system call, instead of utimes(), to set the last access time on the lock file.
SQLite needs to know if the native byte order of the target CPU is big-endian or little-ending. The SQLITE_BYTEORDER preprocessor is set to 4321 for big-endian machines and 1234 for little-endian machines, or it can be 0 to mean that the byte order must be determined at run-time. There are #ifdefs in the code that set SQLITE_BYTEORDER automatically for all common platforms and compilers. However, it may be advantageous to set SQLITE_BYTEORDER appropriately when compiling SQLite for obscure targets. If the target byte order cannot be determined at compile-time, then SQLite falls back to doing run-time checks, which always work, though with a small performance penalty.
4. Options To Set Default Parameter Values
This macro determines the initial setting for PRAGMA automatic_index for newly opened database connections. For all versions of SQLite through 3.7.17, automatic indices are normally enabled for new database connections if this compile-time option is omitted. However, that might change in future releases of SQLite.
This macro determines if SQLite creates databases with the auto_vacuum flag set by default to OFF (0), FULL (1), or INCREMENTAL (2). The default value is 0 meaning that databases are created with auto-vacuum turned off. In any case the compile-time default may be overridden by the PRAGMA auto_vacuum command.
This macro sets the default maximum size of the page-cache for each attached database. A positive value means that the limit is N page. If N is negative that means to limit the cache size to -N*1024 bytes. The suggested maximum cache size can be overridden by the PRAGMA cache_size command. The default value is -2000, which translates into a maximum of 2048000 bytes per cache.
The default schema format number used by SQLite when creating new database files is set by this macro. The schema formats are all very similar. The difference between formats 1 and 4 is that format 4 understands descending indices and has a tighter encoding for boolean values.
All versions of SQLite since 3.3.0 (2006-01-10) can read and write any schema format between 1 and 4. But older versions of SQLite might not be able to read formats greater than 1. So that older versions of SQLite will be able to read and write database files created by newer versions of SQLite, the default schema format was set to 1 for SQLite versions through 3.7.9 (2011-11-01). Beginning with version 3.7.10 (2012-01-16), the default schema format is 4.
The schema format number for a new database can be set at runtime using the PRAGMA legacy_file_format command.
The default numeric file permissions for newly created database files under unix. If not specified, the default is 0644 which means that the files is globally readable but only writable by the creator.
This macro determines whether enforcement of foreign key constraints is enabled or disabled by default for new database connections. Each database connection can always turn enforcement of foreign key constraints on and off and run-time using the foreign_keys pragma. Enforcement of foreign key constraints is normally off by default, but if this compile-time parameter is set to 1, enforcement of foreign key constraints will be on by default.
This macro sets the default limit on the amount of memory that will be used for memory-mapped I/O for each open database file. If the N is zero, then memory mapped I/O is disabled by default. This compile-time limit and the SQLITE_MAX_MMAP_SIZE can be modified at start-time using the sqlite3_config(SQLITE_CONFIG_MMAP_SIZE) call, or at run-time using the mmap_size pragma.
This option sets the size limit on rollback journal files in persistent journal mode and exclusive locking mode and on the size of the write-ahead log file in WAL mode. When this compile-time option is omitted there is no upper bound on the size of the rollback journals or write-ahead logs. The journal file size limit can be changed at run-time using the journal_size_limit pragma.
If set to 1, then the default locking_mode is set to EXCLUSIVE. If omitted or set to 0 then the default locking_mode is NORMAL.
Sets the default size of the lookaside memory allocator memory pool to N entries of SZ bytes each. This setting can be modified at start-time using sqlite3_config(SQLITE_CONFIG_LOOKASIDE) and/or as each database connection is opened using sqlite3_db_config(db, SQLITE_DBCONFIG_LOOKASIDE).
This macro is used to determine whether or not the features enabled and disabled using the SQLITE_CONFIG_MEMSTATUS argument to sqlite3_config() are available by default. The default value is 1 (SQLITE_CONFIG_MEMSTATUS related features enabled).
The sqlite3_memory_used() and sqlite3_memory_highwater() interfaces, the sqlite3_status64(SQLITE_STATUS_MEMORY_USED) interface, and the SQLITE_MAX_MEMORY compile-time option are all non-functional when memory usage tracking is disabled.
This macro determines the number of pages initially allocated by the page cache module when SQLITE_CONFIG_PAGECACHE configuration option is not use and memory for the page cache is obtained from sqlite3_malloc() instead. The number of pages set by this macro are allocated in a single allocation, which reduces the load on the memory allocator.
This macro is used to set the default page-size used when a database is created. The value assigned must be a power of 2. The default value is 4096. The compile-time default may be overridden at runtime by the PRAGMA page_size command.
This macro determines the default value of the PRAGMA synchronous setting. If not overridden at compile-time, the default setting is 2 (FULL).
This macro determines the default value of the PRAGMA synchronous setting for database files that open in WAL mode. If not overridden at compile-time, this value is the same as SQLITE_DEFAULT_SYNCHRONOUS.
If SQLITE_DEFAULT_WAL_SYNCHRONOUS differs from SQLITE_DEFAULT_SYNCHRONOUS, and if the application has not modified the synchronous setting for the database file using the PRAGMA synchronous statement, then the synchronous setting is changed to value defined by SQLITE_DEFAULT_WAL_SYNCHRONOUS when the database connection switches into WAL mode for the first time. If the SQLITE_DEFAULT_WAL_SYNCHRONOUS value is not overridden at compile-time, then it will always be the same as SQLITE_DEFAULT_SYNCHRONOUS and so no automatic synchronous setting changes will ever occur.
This macro sets the default page count for the WAL automatic checkpointing feature. If unspecified, the default page count is 1000.
This macro sets the default value for the SQLITE_LIMIT_WORKER_THREADS parameter. The SQLITE_LIMIT_WORKER_THREADS parameter sets the maximum number of auxiliary threads that a single prepared statement will launch to assist it with a query. If not specified, the default maximum is 0. The value set here cannot be more than SQLITE_MAX_WORKER_THREADS.
This macro determines the default values for SQLITE_DBCONFIG_DQS_DDL and SQLITE_DBCONFIG_DQS_DML, which in turn how SQLite handles each double-quoted string literal. The «DQS» name stands for «Double-Quoted String». The N argument should be an integer 0, 1, 2, or 3.
SQLITE_DQS Double-Quoted Strings Allowed Remarks In DDL In DML 3 yes yes default 2 yes no 1 no yes 0 no no recommended
The recommended setting is 0, meaning that double-quoted strings are disallowed in all contexts. However, the default setting is 3 for maximum compatibility with legacy applications.
The SQLITE_EXTRA_DURABLE compile-time option that used to cause the default PRAGMA synchronous setting to be EXTRA, rather than FULL. This option is no longer supported. Use SQLITE_DEFAULT_SYNCHRONOUS=3 instead.
This macro sets the maximum depth of the search tree that corresponds to the right-hand side of the MATCH operator in an FTS3 or FTS4 full-text index. The full-text search uses a recursive algorithm, so the depth of the tree is limited to prevent using too much stack space. The default limit is 12. This limit is sufficient for up to 4095 search terms on the right-hand side of the MATCH operator and it holds stack space usage to less than 2000 bytes.
For ordinary FTS3/FTS4 queries, the search tree depth is approximately the base-2 logarithm of the number of terms in the right-hand side of the MATCH operator. However, for phrase queries and NEAR queries the search tree depth is linear in the number of right-hand side terms. So the default depth limit of 12 is sufficient for up to 4095 ordinary terms on a MATCH, it is only sufficient for 11 or 12 phrase or NEAR terms. Even so, the default is more than enough for most application.
This compile-time option causes the LIKE operator to always return False if either operand is a BLOB. The default behavior of LIKE is that BLOB operands are cast to TEXT before the comparison is done.
This compile-time option makes SQLite run more efficiently when processing queries that use the LIKE operator, at the expense of breaking backwards compatibility. However, the backwards compatibility break may be only a technicality. There was a long-standing bug in the LIKE processing logic (see https://www.sqlite.org/src/info/05f43be8fdda9f) that caused it to misbehavior for BLOB operands and nobody observed that bug in nearly 10 years of active use. So for more users, it is probably safe to enable this compile-time option and thereby save a little CPU time on LIKE queries.
This compile-time option affects the SQL LIKE operator only and has no impact on the sqlite3_strlike() C-language interface.
This option limits the total amount of memory that SQLite will request from malloc() to N bytes. Any attempt by SQLite to allocate new memory that would cause the sum of all allocations held by SQLite to exceed N bytes will result in an out-of-memory error. This is a hard upper limit. See also the sqlite3_soft_heap_limit() interface.
This limit is only functional if memory usage statistics are available via the sqlite3_memory_used() and sqlite3_status64(SQLITE_STATUS_MEMORY_USED) interfaces. Without that memory usage information, SQLite has no way of knowing when it is about to go over the limit, and thus is unable to prevent the excess memory allocation. Memory usage tracking is turned on by default, but can be disabled at compile-time using the SQLITE_DEFAULT_MEMSTATUS option, or at start-time using sqlite3_config(SQLITE_CONFIG_MEMSTATUS).
This macro sets a hard upper bound on the amount of address space that can be used by any single database for memory-mapped I/O. Setting this value to 0 completely disables memory-mapped I/O and causes logic associated with memory-mapped I/O to be omitted from the build. This option does change the default memory-mapped I/O address space size (set by SQLITE_DEFAULT_MMAP_SIZE or sqlite3_config(SQLITE_CONFIG_MMAP_SIZE) or the run-time memory-mapped I/O address space size (set by sqlite3_file_control(SQLITE_FCNTL_MMAP_SIZE) or PRAGMA mmap_size) as long as those other settings are less than the maximum value defined here.
Whenever the database schema changes, prepared statements are automatically reprepared to accommodate the new schema. There is a race condition here in that if one thread is constantly changing the schema, another thread might spin on reparses and repreparations of a prepared statement and never get any real work done. This parameter prevents an infinite loop by forcing the spinning thread to give up after a fixed number of attempts at recompiling the prepared statement. The default setting is 50 which is more than adequate for most applications.
Set an upper bound on the sqlite3_limit(db,SQLITE_LIMIT_WORKER_THREADS,N) setting that determines the maximum number of auxiliary threads that a single prepared statement will use to aid with CPU-intensive computations (mostly sorting). See also the SQLITE_DEFAULT_WORKER_THREADS options.
Set the default size limit (in bytes) for in-memory databases created using sqlite3_deserialize(). This is just the default. The limit can be changed at start-time using sqlite3_config(SQLITE_CONFIG_MEMDB_MAXSIZE,N) or at run-time for individual databases using the SQLITE_FCNTL_SIZE_LIMIT file-control. If no default is specified, 1073741824 is used.
The unix VFS will never use a file descriptor less than N. The default value of N is 3.
Avoiding the use of low-numbered file descriptors is a defense against accidental database corruption. If a database file was opened using file descriptor 2, for example, and then an assert() failed and invoked write(2. ), that would likely cause database corruption by overwriting part of the database file with the assertion error message. Using only higher-valued file descriptors avoids this potential problem. The protection against using low-numbered file descriptors can be disabled by setting this compile-time option to 0.
This option changes the default assumption about powersafe overwrite for the underlying filesystems for the unix and windows VFSes. Setting SQLITE_POWERSAFE_OVERWRITE to 1 causes SQLite to assume that application-level writes cannot changes bytes outside the range of bytes written even if the write occurs just before a power loss. With SQLITE_POWERSAFE_OVERWRITE set to 0, SQLite assumes that other bytes in the same sector with a written byte might be changed or damaged by a power loss.
This option limits the maximum width and precision of substitutions for the printf() SQL function and the other C-language string formatting functions such as sqlite3_mprintf() and sqlite3_str_appendf(). This is turn can prevent a hostile or malfunctioning script from using excessive memory by invoking a format such as: «printf('%*s',2147483647,'hi')«. A value for N of around 100000 is normally sufficient.
The printf() SQL function is subject to the SQLITE_LIMIT_LENGTH limit of sqlite3_limit(). Hence any printf() result with a width or precision more than the SQLITE_LIMIT_LENGTH will cause an SQLITE_TOOBIG error. However, the low-level formatting for the printf() function is done by a subroutine that does not have access to SQLITE_LIMIT_LENGTH. So the low-level formatting is done into a memory allocation that might be considerably larger than SQLITE_LIMIT_LENGTH and the SQLITE_LIMIT_LENGTH check is only performed after all formatting is complete. Thus there might be a transient buffer that exceeds SQLITE_LIMIT_LENGTH. The SQLITE_PRINTF_PRECISION_LIMIT option is an additional check that prevents excess sizes for the transient buffer used inside the low-level formatting subroutine, prior to the SQLITE_LIMIT_LENGTH check.
Be careful not to set SQLITE_PRINTF_PRECISION_LIMIT too low. SQLite uses its built-in printf() functionality to format the text of CREATE statements stored in the sqlite_schema table. So SQLITE_PRINTF_PRECISION_LIMIT should be at least as big as the largest table, index, view, or trigger definition that you are likely to encounter.
No error is raised if a width or precision exceeds SQLITE_PRINTF_PRECISION_LIMIT. Instead, the large width or precision is silently truncated.
The default value for SQLITE_PRINTF_PRECISION_LIMIT is 2147483647 (0x7fffffff).
As part of the query planning process, SQLite enumerates all usable combinations of indexes and WHERE-clause constraints. For certain pathological queries, the number of these index-and-constraint combinations can be very large, resulting in slow performance by the query planner. The SQLITE_QUERY_PLANNER_LIMIT value (in conjunction with the related SQLITE_QUERY_PLANNER_LIMIT_INCR setting) limits the number of index-and-constraint combinations that the query planner will consider, in order to prevent the query planner from using excess CPU time. The default value for SQLITE_QUERY_PLANNER_LIMIT is set high enough so that is never reached for real-world queries. The query planner search limit only applies to queries that are deliberately crafted to use excess planning time.
The SQLITE_QUERY_PLANNER_LIMIT option sets an initial baseline value for the maximum number of index-and-constraint combinations that the query planner consider. The baseline query planner limit is increased by SQLITE_QUERY_PLANNER_LIMIT_INCR prior to processing each table of a join so that each table is guaranteed to be able to propose at least some index-and-constraint combinations to the optimizer even if prior tables of the join have exhausted the baseline limit. The default value for both this compile-time option and the SQLITE_QUERY_PLANNER_LIMIT option are set high enough so that they should never be reached for real-world queries.
This option causes the PRAGMA reverse_unordered_selects setting to be enabled by default. When enabled, SELECT statements that lack an ORDER BY clause will run in reverse order.
This option is useful for detecting when applications (incorrectly) assume that the order of rows in a SELECT without an ORDER BY clause will always be the same.
If multi-threaded processing is enabled via the PRAGMA threads setting, then sort operations will attempt to start helper threads when the amount of content to be sorted exceeds the minimum of the cache_size and PMA Size determined by the SQLITE_CONFIG_PMASZ start-time option. This compile-time option sets the default value for the SQLITE_CONFIG_PMASZ start-time option. The default value is 250.
The SQLITE_STMTJRNL_SPILL compile-time option determines the default setting of the SQLITE_CONFIG_STMTJRNL_SPILL start-time setting. That setting determines the size threshold above which statement journals are moved from memory to disk.
This option enables the use of the Windows Heap API functions for memory allocation instead of the standard library malloc() and free() routines.
This macro sets the maximum depth of the LALR(1) stack used by the SQL parser within SQLite. The default value is 100. A typical application will use less than about 20 levels of the stack. Developers whose applications contain SQL statements that need more than 100 LALR(1) stack entries should seriously consider refactoring their SQL as it is likely to be well beyond the ability of any human to comprehend.
5. Options To Set Size Limits
There are compile-time options that will set upper bounds on the sizes of various structures in SQLite. The compile-time options normally set a hard upper bound that can be changed at run-time on individual database connections using the sqlite3_limit() interface.
The compile-time options for setting upper bounds are documented separately. The following is a list of the available settings:
6. Options To Control Operating Characteristics
On most systems, the malloc() system call returns a buffer that is aligned to an 8-byte boundary. But on some systems (ex: windows) malloc() returns 4-byte aligned pointer. This compile-time option must be used on systems that return 4-byte aligned pointers from malloc().
If this option is present, then the built-in LIKE operator will be case sensitive. This same effect can be achieved at run-time using the case_sensitive_like pragma.
When this option is present, content contained in overflow pages of the database file is read directly from disk, bypassing the page cache, during read transactions. In applications that do a lot of reads of large BLOBs, this option might improve read performance.
If this option is present, then SQLite will use the isnan() function from the system math library. This is an alias for the HAVE_ISNAN configuration option.
The option causes SQLite to omit its built-in operating system interfaces for Unix, Windows, and OS/2. The resulting library will have no default operating system interface. Applications must use sqlite3_vfs_register() to register an appropriate interface before using SQLite. Applications must also supply implementations for the sqlite3_os_init() and sqlite3_os_end() interfaces. The usual practice is for the supplied sqlite3_os_init() to invoke sqlite3_vfs_register(). SQLite will automatically invoke sqlite3_os_init() when it initializes.
This option is typically used when building SQLite for an embedded platform with a custom operating system.
This compile-time option changes the default setting of the secure_delete pragma. When this option is not used, secure_delete defaults to off. When this option is present, secure_delete defaults to on.
The secure_delete setting causes deleted content to be overwritten with zeros. There is a small performance penalty since additional I/O must occur. On the other hand, secure_delete can prevent fragments of sensitive information from lingering in unused parts of the database file after it has been deleted. See the documentation on the secure_delete pragma for additional information.
This option controls whether or not code is included in SQLite to enable it to operate safely in a multithreaded environment. The default is SQLITE_THREADSAFE=1 which is safe for use in a multithreaded environment. When compiled with SQLITE_THREADSAFE=0 all mutexing code is omitted and it is unsafe to use SQLite in a multithreaded program. When compiled with SQLITE_THREADSAFE=2, SQLite can be used in a multithreaded program so long as no two threads attempt to use the same database connection (or any prepared statements derived from that database connection) at the same time.
To put it another way, SQLITE_THREADSAFE=1 sets the default threading mode to Serialized. SQLITE_THREADSAFE=2 sets the default threading mode to Multi-threaded. And SQLITE_THREADSAFE=0 sets the threading mode to Single-threaded.
The value of SQLITE_THREADSAFE can be determined at run-time using the sqlite3_threadsafe() interface.
When SQLite has been compiled with SQLITE_THREADSAFE=1 or SQLITE_THREADSAFE=2 then the threading mode can be altered at run-time using the sqlite3_config() interface together with one of these verbs:
Note that when SQLite is compiled with SQLITE_THREADSAFE=0, the code to make SQLite threadsafe is omitted from the build. When this occurs, it is impossible to change the threading mode at start-time or run-time.
See the threading mode documentation for additional information on aspects of using SQLite in a multithreaded environment.
This option controls whether temporary files are stored on disk or in memory. The meanings for various settings of this compile-time option are as follows:
|Always use temporary files|
|1||Use files by default but allow the PRAGMA temp_store command to override|
|2||Use memory by default but allow the PRAGMA temp_store command to override|
|3||Always use memory|
The default setting is 1. Additional information can be found in tempfiles.html.
If this macro is defined to a positive integer N, then the length of strings and BLOB that are expanded into parameters in the output of sqlite3_trace() is limited to N bytes.
This macro determines the default value for the SQLITE_DBCONFIG_TRUSTED_SCHEMA and PRAGMA trusted_schema setting. If no alternative is specified, the trusted-schema setting defaults to ON (a value of 1) for legacy compatibility. However, for best security, systems that implement application-defined SQL functions and/or virtual tables should consider changing the default to OFF.
This option causes the URI filename process logic to be enabled by default.
7. Options To Enable Features Normally Turned Off
URI filenames normally throws an error if the authority section is not either empty or «localhost». However, if SQLite is compiled with the SQLITE_ALLOW_URI_AUTHORITY compile-time option, then the URI is converted into a Uniform Naming Convention (UNC) filename and passed down to the underlying operating system that way.
Some future versions of SQLite may change to enable this feature by default.
This C-preprocess macro determines the default setting of the SQLITE_CONFIG_COVERING_INDEX_SCAN configuration setting. It defaults to 1 (on) which means that covering indices are used for full table scans where possible, in order to reduce I/O and improve performance. However, the use of a covering index for a full scan will cause results to appear in a different order from legacy, which could cause some (incorrectly-coded) legacy applications to break. Hence, the covering index scan option can be disabled at compile-time on systems that what to minimize their risk of exposing errors in legacy applications.
If this C-preprocessor macro is defined, then extra code is included that allows SQLite to function on a filesystem that only support 8+3 filenames. If the value of this macro is 1, then the default behavior is to continue to use long filenames and to only use 8+3 filenames if the database connection is opened using URI filenames with the «8_3_names=1» query parameter. If the value of this macro is 2, then the use of 8+3 filenames becomes the default but may be disabled on using the 8_3_names=0 query parameter.
When defined, this C-preprocessor macro activates extra code that attempts to detect misuse of the SQLite API, such as passing in NULL pointers to required parameters or using objects after they have been destroyed.
If this C-preprocessor macro is defined and if the xDeviceCharacteristics method of sqlite3_io_methods object for a database file reports (via one of the SQLITE_IOCAP_ATOMIC bits) that the filesystem supports atomic writes and if a transaction involves a change to only a single page of the database file, then the transaction commits with just a single write request of a single page of the database and no rollback journal is created or written. On filesystems that support atomic writes, this optimization can result in significant speed improvements for small updates. However, few filesystems support this capability and the code paths that check for this capability slow down write performance on systems that lack atomic write capability, so this feature is disabled by default.
This compile-time option enables SQLite to take advantage batch atomic write capabilities in the underlying filesystem. As of SQLite version 3.21.0 (2017-10-24) this is only supported on F2FS. However, the interface is implemented generically, using sqlite3_file_control() with SQLITE_FCNTL_BEGIN_ATOMIC_WRITE and SQLITE_FCNTL_COMMIT_ATOMIC_WRITE so the capability can be added to other filesystem times in the future. When this option is enabled, SQLite automatically detects that the underlying filesystem supports batch atomic writes, and when it does so it avoids writing the rollback journal for transaction control. This can make transactions over twice as fast, while simultaneously reducing wear on SSD storage devices.
Future versions of SQLite might enable the batch-atomic-write capability by default, at which point this compile-time option will become superfluous.
When this C-preprocessor macro is defined, SQLite includes some additional APIs that provide convenient access to meta-data about tables and queries. The APIs that are enabled by this option are:
This option was formerly used to enable the sqlite3_serialize() and sqlite3_deserialize() interfaces. However, as of SQLite 3.36.0 (2021-06-18) those interfaces are enabled by default and a new compile-time option SQLITE_OMIT_DESERIALIZE is added to omit them.
This option adds extra logic to SQLite that inserts comment text into the output of EXPLAIN. These extra comments use extra memory, thus making prepared statements larger and very slightly slower, and so they are turned off by default and in most application. But some applications, such as the command-line shell for SQLite, value clarity of EXPLAIN output over raw performance and so this compile-time option is available to them. The SQLITE_ENABLE_EXPLAIN_COMMENTS compile-time option is also enabled automatically if SQLITE_DEBUG is enabled.
When this option is defined in the amalgamation, version 3 of the full-text search engine is added to the build automatically.
This option modifies the query pattern parser in FTS3 such that it supports operators AND and NOT (in addition to the usual OR and NEAR) and also allows query expressions to contain nested parenthesis.
This option enables the two-argument version of the fts3_tokenizer() interface. The second argument to fts3_tokenizer() is suppose to be a pointer to a function (encoded as a BLOB) that implements an application defined tokenizer. If hostile actors are able to run the two-argument version of fts3_tokenizer() with an arbitrary second argument, they could use crash or take control of the process.
Because of security concerns, the two-argument fts3_tokenizer() feature was disabled beginning with Version 3.11.0 (2016-02-15) unless this compile-time option is used. Version 3.12.0 (2016-03-29) added the sqlite3_db_config(db,SQLITE_DBCONFIG_ENABLE_FTS3_TOKENIZER,1,0) interface that activates the two-argument version of fts3_tokenizer() for a specific database connection at run-time.
When this option is defined in the amalgamation, versions 3 and 4 of the full-text search engine is added to the build automatically.
When this option is defined in the amalgamation, versions 5 of the full-text search engine (fts5) is added to the build automatically.
When this option is defined in the amalgamation, the Geopoly extension is included in the build.
This option causes the International Components for Unicode or «ICU» extension to SQLite to be added to the build.
When both the SQLite core and the Command Line Interface (CLI) are both compiled with this option, then the CLI provides an extra command named «.iotrace» that provides a low-level log of I/O activity. This option is experimental and may be discontinued in a future release.
This macro enables the built-in SQL math functions. This option is automatically added to the Makefile by the configure script on unix platforms, unless the —disable-math option is used. This option is also included on Windows builds using the «Makefile.msc» makefile for nmake.
This compile-time option is a no-op. Prior to SQLite version 3.38.0 (2022-02-22), it was necessary to compile with this option in order to include the JSON SQL functions in the build. However, beginning with SQLite version 3.38.0, those functions are included by default. Use the -DSQLITE_OMIT_JSON option to omit them.
This option enables additional logic in the OS interface layer for Mac OS X. The additional logic attempts to determine the type of the underlying filesystem and choose and alternative locking strategy that works correctly for that filesystem type. Five locking strategies are available:
- POSIX locking style. This is the default locking style and the style used by other (non Mac OS X) Unixes. Locks are obtained and released using the fcntl() system call.
AFP locking style. This locking style is used for network file systems that use the AFP (Apple Filing Protocol) protocol. Locks are obtained by calling the library function _AFPFSSetLock().
Flock locking style. This is used for file-systems that do not support POSIX locking style. Locks are obtained and released using the flock() system call.
Dot-file locking style. This locking style is used when neither flock nor POSIX locking styles are supported by the file system. Database locks are obtained by creating and entry in the file-system at a well-known location relative to the database file (a «dot-file») and relinquished by deleting the same file.
Additionally, five extra VFS implementations are provided as well as the default. By specifying one of the extra VFS implementations when calling sqlite3_open_v2(), an application may bypass the file-system detection logic and explicitly select one of the above locking styles. The five extra VFS implementations are called «unix-posix», «unix-afp», «unix-flock», «unix-dotfile» and «unix-none».
This option adds extra logic to SQLite that allows it to release unused memory upon request. This option must be enabled in order for the sqlite3_release_memory() interface to work. If this compile-time option is not used, the sqlite3_release_memory() interface is a no-op.
This option includes code in SQLite that implements an alternative memory allocator. This alternative memory allocator is only engaged when the SQLITE_CONFIG_HEAP option to sqlite3_config() is used to supply a large chunk of memory from which all memory allocations are taken. The MEMSYS3 memory allocator uses a hybrid allocation algorithm patterned after dlmalloc(). Only one of SQLITE_ENABLE_MEMSYS3 and SQLITE_ENABLE_MEMSYS5 may be enabled at once.
This option includes code in SQLite that implements an alternative memory allocator. This alternative memory allocator is only engaged when the SQLITE_CONFIG_HEAP option to sqlite3_config() is used to supply a large chunk of memory from which all memory allocations are taken. The MEMSYS5 module rounds all allocations up to the next power of two and uses a first-fit, buddy-allocator algorithm that provides strong guarantees against fragmentation and breakdown subject to certain operating constraints.
This option enables an optimization that omits NULL columns at the ends of rows, for a space savings on disk.
Databases generated with this option enabled are not readable by SQLite version 3.1.6 (2005-03-17) and earlier. Also, databases generated with this option enabled are prone to triggering the e6e962d6b0f06f46 bug in the sqlite3_blob_reopen() interface. For those reasons, this optimization is disabled by default. However, this optimization may be enabled by default in a future release of SQLite.
This option enables support for the sqlite_offset(X) SQL function.
The sqlite_offset(X) SQL function requires a new interface on the B-tree storage engine, a new opcode in the virtual machine that runs SQL statements, and a new conditional in a critical path of the code generator. To avoid that overhead in applications that do not need the utility of sqlite_offset(X), the function is disabled by default.
This option enables several new APIs that provide callbacks prior to any change to a rowid table. The callbacks can be used to record the state of the row before the change occurs.
The action of the preupdate hook is similar to the update hook except that the callback is invoked before the change, not afterwards, and the preupdate hook interfaces are omitted unless this compile-time option is used.
The preupdate hook interfaces were originally added to support the session extension.
This option causes the query planner stability guarantee (QPSG) to be on by default. Normally the QPSG is off and must be activated at run-time using the SQLITE_DBCONFIG_ENABLE_QPSG option to the sqlite3_db_config() interface.
This option causes SQLite to include support for the R*Tree index extension.
This option activates an optimization that reduces the memory required by the sorter at the cost of doing additional B-tree lookups after the sort has occurred.
The default sorting procedure is to gather all information that will ultimately be output into a «record» and pass that complete record to the sorter. But in some cases, for example if some of the output columns consists of large BLOB values, the size of the each record can be large, which means that the sorter has to either use more memory, and/or write more content to temporary storage.
When SQLITE_ENABLE_SORTER_REFERENCES is enabled, the records passed to the sorter often contain only a ROWID value. Such records are much smaller. This means the sorter has much less «payload» to deal with and can run faster. After sorting has occurred, the ROWID is used to look up the output column values in the original table. That requires another search into the table, and could potentially result in a slowdown. Or, it might be a performance win, depending on how large the values are.
Even when the SQLITE_ENABLE_SORTER_REFERENCES compile-time option is on, sorter references are still disabled by default. To use sorter references, the application must set a sorter reference size threshold using the sqlite3_config(SQLITE_CONFIG_SORTERREF_SIZE) interface at start-time.
Because the SQLite developers do not know whether the SQLITE_ENABLE_SORTER_REFERENCES option will help or hurt performance, it is disabled by default at this time (2018-05-04). It might be enabled by default in some future release, depending on what is learned about its impact on performance.
This option enables the sqlite3_stmt_scanstatus() interface. The sqlite3_stmt_scanstatus() interface is normally omitted from the build because it imposes a small performance penalty, even on statements that do not use the feature.
This compile-time option is deprecated and untested.
This option enables extra code (especially the SQLITE_CONFIG_SQLLOG option to sqlite3_config()) that can be used to create logs of all SQLite processing performed by an application. These logs can be useful in doing off-line analysis of the behavior of an application, and especially for performance analysis. In order for the SQLITE_ENABLE_SQLLOG option to be useful, some extra code is required. The «test_sqllog.c» source code file in the SQLite source tree is a working example of the required extra code. On unix and windows systems, a developer can append the text of the «test_sqllog.c» source code file to the end of an «sqlite3.c» amalgamation, recompile the application using the -DSQLITE_ENABLE_SQLLOG option, then control logging using environment variables. See the header comment on the «test_sqllog.c» source file for additional detail.
This option used to cause the ANALYZE command to collect index histogram data in the sqlite_stat2 table. But that functionality was superceded by SQLITE_ENABLE_STAT3 as of SQLite version 3.7.9 (2011-11-01). The SQLITE_ENABLE_STAT2 compile-time option is now a no-op.
This option used to cause the ANALYZE command to collect index histogram data in the sqlite_stat3 table. But that functionality was superceded by SQLITE_ENABLE_STAT4 as of SQLite version 3.8.1 (2013-10-17). The SQLITE_ENABLE_STAT3 compile-time option continued to be supported through version 3.29.0 (2019-07-10) but has now become a no-op.
This option adds additional logic to the ANALYZE command and to the query planner that can help SQLite to chose a better query plan under certain situations. The ANALYZE command is enhanced to collect histogram data from all columns of every index and store that data in the sqlite_stat4 table. The query planner will then use the histogram data to help it make better index choices. The downside of this compile-time option is that it violates the query planner stability guarantee making it more difficult to ensure consistent performance in mass-produced applications.
SQLITE_ENABLE_STAT4 is an enhancement of SQLITE_ENABLE_STAT3. STAT3 only recorded histogram data for the left-most column of each index whereas the STAT4 enhancement records histogram data from all columns of each index. The SQLITE_ENABLE_STAT3 compile-time option has become a no-op.
This compile-time option is no longer used.
This option enables an optional ORDER BY and LIMIT clause on UPDATE and DELETE statements.
If this option is defined, then it must also be defined when using the Lemon parser generator tool to generate a parse.c file. Because of this, this option may only be used when the library is built from source, not from the amalgamation or from the collection of pre-packaged C files provided for non-Unix like platforms on the website.
When the SQLITE_ENABLE_UNKNOWN_SQL_FUNCTION compile-time option is activated, SQLite will suppress «unknown function» errors when running an EXPLAIN or EXPLAIN QUERY PLAN. Instead of throwing an error, SQLite will insert a substitute no-op function named «unknown()». The substitution of «unknown()» in place of unrecognized functions only occurs on EXPLAIN and EXPLAIN QUERY PLAN, not on ordinary statements.
When used in the command-line shell, the SQLITE_ENABLE_UNKNOWN_SQL_FUNCTION feature allows SQL text that contains application-defined functions to be pasted into the shell for analysis and debugging without having to create and load an extension that implements the application-defined functions.
This option enables the sqlite3_unlock_notify() interface and its associated functionality. See the documentation titled Using the SQLite Unlock Notification Feature for additional information.
This option is obsolete. It used to enable some extra some extra PRAGMA statements such as PRAGMA function_list, PRAGMA module_list, and PRAGMA pragma_list, but those pragmas are now all enabled by default. See SQLITE_OMIT_INTROSPECTION_PRAGMAS.
If this option is enabled, then the alloca() memory allocator will be used in a few situations where it is appropriate. This results in a slightly smaller and faster binary. The SQLITE_USE_ALLOCA compile-time only works, of course, on systems that support alloca().
This option causes SQLite to issue extra SQLITE_FCNTL_TRACE file controls to provide supplementary information to the VFS. The «vfslog.c» extension makes use of this to provide enhanced logs of VFS activity.
This option causes some extensions to link against the zlib compression library.
This option has no effect on the SQLite core. It is only used by extensions. This is option is necessary for the compression and decompression functions that are part of SQL Archive support in the command-line shell.
When compiling with this option, it will normally be necessary to add a linker option to include the zlib library in the build. Normal this option is «-lz» but might be different on different systems.
When building with MSVC on Windows systems, one can put the zlib source code in the compat/zlib subdirectory of the source tree and then add the USE_ZLIB=1 option to the nmake command to cause the The Makefile.msc to automatically build and use an appropriate zlib library implementation.
This option causes the LALR(1) parser stack depth to be tracked and reported using the sqlite3_status(SQLITE_STATUS_PARSER_STACK. ) interface. SQLite’s LALR(1) parser has a fixed stack depth (determined at compile-time using the YYSTACKDEPTH options). This option can be used to help determine if an application is getting close to exceeding the maximum LALR(1) stack depth.
8. Options To Disable Features Normally Turned On
If this C-preprocessor macro is defined, large file support is disabled.
If this C-preprocessor macro is defined, directory syncs are disabled. SQLite typically attempts to sync the parent directory when a file is deleted to ensure the directory entries are updated immediately on disk.
If this C-preprocessor macro is defined, the unicode61 tokenizer in FTS3 is omitted from the build and is unavailable to applications.
If this C-preprocessor macro disables the «deferred token» optimization in FTS4. The «deferred token» optimization avoids loading massive posting lists for terms that are in most documents of the collection and instead simply scans for those tokens in the document source. FTS4 should get exactly the same answer both with and without this optimization.
This option disables the use of compiler-specific built-in functions such as __builtin_bswap32() and __builtin_add_overflow() in GCC and Clang, or _byteswap_ulong() and _ReadWriteBarrier() with MSVC.
9. Options To Omit Features
The following options can be used to reduce the size of the compiled library by omitting unused features. This is probably only useful in embedded systems where space is especially tight, as even with all features included the SQLite library is relatively small. Don’t forget to tell your compiler to optimize for binary size! (the -Os option if using GCC). Telling your compiler to optimize for size usually has a much larger impact on library footprint than employing any of these compile-time options. You should also verify that debugging options are disabled.
The macros in this section do not require values. The following compilation switches all have the same effect:
If any of these options are defined, then the same set of SQLITE_OMIT_* options must also be defined when using the Lemon parser generator tool to generate the parse.c file and when compiling the ‘mkkeywordhash’ tool which generates the keywordhash.h file. Because of this, these options may only be used when the library is built from canonical source, not from the amalgamation. Some SQLITE_OMIT_* options might work, or appear to work, when used with the amalgamation. But this is not guaranteed. In general, always compile from canonical sources in order to take advantage of SQLITE_OMIT_* options.
Important Note: The SQLITE_OMIT_* options may not work with the amalgamation. SQLITE_OMIT_* compile-time options usually work correctly only when SQLite is built from canonical source files.
Special versions of the SQLite amalgamation that do work with a predetermined set of SQLITE_OMIT_* options can be generated. To do so, make a copy of the Makefile.linux-gcc makefile template in the canonical source code distribution. Change the name of your copy to simply «Makefile». Then edit «Makefile» to set up appropriate compile-time options. Then type:
The resulting «sqlite3.c» amalgamation code file (and its associated header file «sqlite3.h») can then be moved to a non-unix platform for final compilation using a native compiler.
The SQLITE_OMIT_* options are unsupported. By this we mean that an SQLITE_OMIT_* option that omits code from the build in the current release might become a no-op in the next release. Or the other way around: an SQLITE_OMIT_* that is a no-op in the current release might cause code to be excluded in the next release. Also, not all SQLITE_OMIT_* options are tested. Some SQLITE_OMIT_* options might cause SQLite to malfunction and/or provide incorrect answers.
Important Note: The SQLITE_OMIT_* compile-time options are mostly unsupported.
The following are the available OMIT options:
When this option is defined, the ALTER TABLE command is not included in the library. Executing an ALTER TABLE statement causes a parse error.
When this option is defined, the ANALYZE command is omitted from the build.
When this option is defined, the ATTACH and DETACH commands are omitted from the build.
Defining this option omits the authorization callback feature from the library. The sqlite3_set_authorizer() API function is not present in the library.
This option is omits the AUTOINCREMENT feature. When this is macro is defined, columns declared as «INTEGER PRIMARY KEY AUTOINCREMENT» behave in the same way as columns declared as «INTEGER PRIMARY KEY» when a NULL is inserted. The sqlite_sequence system table is neither created, nor respected if it already exists.
For backwards compatibility with older versions of SQLite that lack the sqlite3_initialize() interface, the sqlite3_initialize() interface is called automatically upon entry to certain key interfaces such as sqlite3_open(), sqlite3_vfs_register(), and sqlite3_mprintf(). The overhead of invoking sqlite3_initialize() automatically in this way may be omitted by building SQLite with the SQLITE_OMIT_AUTOINIT C-preprocessor macro. When built using SQLITE_OMIT_AUTOINIT, SQLite will not automatically initialize itself and the application is required to invoke sqlite3_initialize() directly prior to beginning use of the SQLite library.
By default, the sqlite3_step() interface will automatically invoke sqlite3_reset() to reset the prepared statement if necessary. This compile-time option changes that behavior so that sqlite3_step() will return SQLITE_MISUSE if it called again after returning anything other than SQLITE_ROW, SQLITE_BUSY, or SQLITE_LOCKED unless there was an intervening call to sqlite3_reset().
In SQLite version 22.214.171.124 (2010-03-26) and earlier, sqlite3_step() used to always return SQLITE_MISUSE if it was invoked again after returning anything other than SQLITE_ROW without an intervening call to sqlite3_reset(). This caused problems on some poorly written smartphone applications which did not correctly handle the SQLITE_LOCKED and SQLITE_BUSY error returns. Rather than fix the many defective smartphone applications, the behavior of SQLite was changed in 126.96.36.199 to automatically reset the prepared statement. But that changed caused issues in other improperly implemented applications that were actually looking for an SQLITE_MISUSE return to terminate their query loops. (Anytime an application gets an SQLITE_MISUSE error code from SQLite, that means the application is misusing the SQLite interface and is thus incorrectly implemented.) The SQLITE_OMIT_AUTORESET interface was added to SQLite version 3.7.5 (2011-02-01) in an effort to get all of the (broken) applications to work again without having to actually fix the applications.
If this option is defined, the library cannot create or write to databases that support auto_vacuum. Executing a PRAGMA auto_vacuum statement is not an error (since unknown PRAGMAs are silently ignored), but does not return a value or modify the auto-vacuum flag in the database file. If a database that supports auto-vacuum is opened by a library compiled with this option, it is automatically opened in read-only mode.
This option disables the use of indices with WHERE clause terms that employ the BETWEEN operator.
When this option is defined, it is not possible to specify a blob in an SQL statement using the X’ABCD’ syntax.
This option is no longer used for anything. It is a no-op.