- File www/prog-intro.wiki — part of check-in [a4782fccbe] at 2013-02-23 14:16:46 on branch trunk — Add the "Programmers Introduction To SQLite4" document outlining the desired API for SQLite4. (user: drh size: 14916) [more...]
Introduction For Programmers
This document provides a broad overview of the Application Programming Interface (API) for SQLite4. If you are new to SQLite and are wondering how to write a program that uses SQLite or how to incorporate SQLite into your existing program, this document is a good starting point.
2.0 Implementation Language
SQLite is written in ANSI-C. The C programming language is used because it is the universal assembly language - it can run on just about any hardware and on just about any operating system. No matter what programming language is used for the application code, it can usually interface easily with a library written in C.
SQLite is written in an Object-Oriented (OO) style. This may comes as a surprise to readers who have been trained to think that C++ is necessary for OO programming. While it is true that C++ has lots of syntactic sugar designed to make OO-style programming easier, there is nothing missing from generic ANSI-C that disqualifies it from using OO ideas and practices.
In this document, there will be much talk of "objects", which are implemented as ordinary C structures, of course. There is also talk of "methods" which are really just procedures which take a pointer to the object as their first parameter. SQLite also uses object inheritance, though inheritance is only used in fringe areas of the API that do not fall within the scope of this document.
3.0 Principal Objects
Here are four object classes that every programmer who uses SQLite4 needs to be aware of:
- sqlite4_mm - A memory allocator
- sqlite4_env - A run-time environment
- sqlite4 - A database connection
- sqlite4_stmt - A prepared statement
The most important of these are the latter two, the database connection and prepared statement objects. SQLite4 provides default instances of a memory allocator and run-time environment which are adequate for most system programming tasks.
3.1 The Memory Allocator Object
A memory allocator object is used to allocate and deallocate bulk memory. SQLite4 supplies a default memory allocator object (obtained by calling the sqlite4_mm_default() function) which is just a wrapper around the system malloc()/realloc()/free() routines. Most applications will get along fine using just this default. But for applications with more advanced needs, other memory allocator objects are available, including:
- Memory allocators that use fixed pool of memory (provided at start time) rather than the heap.
- Meta-memory allocators that use one memory allocator first but then fail-over to a second one if the first one runs out of memory or is otherwise unable to satisfy an allocation.
- Memory allocators operating out of a fixed pool of memory and that provide robust guarantees against memory fragmentation.
- Memory allocators that provide lots of information about how much memory is being used.
- Memory allocators that can simulate out-of-memory (OOM) faults for testing purposes.
- Memory allocators that aggressively check for misuse of the memory allocation system, by (for example) overwriting memory as it is freed, and initializing newly allocated memory to a random bit pattern.
- Completely customized and application-specific memory allocators.
The various memory allocators mentioned above are implemented by subclassing the top-level memory allocator object. Unless you are implementing your own customized memory allocator, you do not really need to know about this subclassing. At the application programming level, you will only encounter pointers to the sqlite4_mm object base class.
There are various methods on the sqlite4_mm object, including:
- sqlite4_mm_malloc() - allocate bulk memory
- sqlite4_mm_free() - release memory previously allocated
- sqlite4_mm_realloc() - change the size of a prior allocation
- sqlite4_mm_msize() - report the size of an allocation
- sqlite4_mm_stat() - report usage statistics on a memory allocator
A memory allocator object is also passed as parameter to certain interfaces in order to be used as a destructor. For example, when binding a string to a parameter in a prepared statement using the sqlite4_bind_text() method on the prepared statement object, there is a parameter which is a pointer to a memory allocator object. If that parameter is not NULL (and is not one of a handful of other values with special meanings) then SQLite4 will invoke the sqlite3_mm_free() method of the memory allocator on the string when it has finished using the string.
3.2 The Run-Time Environment Object
An instance of the run-time environment object defines how SQLite4 interfaces to the rest of the system. Each run-time environment object includes a pointer to a memory allocator object which is the default memory allocator for that environment. The run-time environment also includes information such as:
- How to create and use mutexes
- Which storage engines are available for use on the backend
- A library of SQL functions available for use by all database connections
- Methods for obtaining the current time and for obtaining randomness
- State information that is shared across multiple database connections
- Other "global" configuration settings and start-time settings
SQLite4 supplies a default run-time environment object (obtained by calling the sqlite4_env_default() function) that is suitable for use in most applications. In fact, for many routines that require a pointer to a run-time environment object, you can pass in a NULL pointer and the NULL pointer will be automatically converted into a pointer to the default run-time environment.
Usually the default run-time environment object is adequate, but some applications, especially those running on custom platforms or on small resource-limited devices, may want to create and use a new run-time environment object with non-standard settings. A reasonable design is to create a single instance of an appropriate run-time environment object, then store a pointer to that object in a global variable where it is readily accessible to all parts of the application.
It is possible to use two or more run-time environment objects within the same application. But use caution here. Due to a long-standing bug in the design of Posix Advisor Locks (NB: the bug is in posix, not in SQLite) if you open two or more database connections to the same database using different run-time environment objects, then file locks will be broken in unexpected ways, which can lead to database corruption. Opening connections with different run-time environments to different databases is harmless and safe. Opening multiple connections to the same database using the same run-time environment is also harmless and safe. But opening multiple connections to the same database from different run-time environments can cause problems under posix. Therefore, if you are programming for a posix system (iOS, Android, QNX, Linux, MacOS, etc.) you are advised to stick to using a single run-time environment for your application. Windows does not have the posix advisory locking bug and so you are free to mix and match run-time environment objects, database connections, and databases with wild abandon there, if you are willing to sacrifice portability.
3.3 The Database Connection Object
The most important object is the database connection object: sqlite4. Each database connection object represents an open connection to one or more database files. (The database connection starts out associated with just a single database file, but more databases can be added using the ATTACH statement.) Many applications will have just a single database connection object, since they are only talking to a single database. But there are no arbitrary restrictions on the number of database objects that can be open at once. (The number of open database connections might be constrained by system resource limitations such as the maximum number of open file descriptors.) Multiple database connections can be opened on the same database file, if desired.
Every database connection object is associated with a single run-time environment object. The environment of a database connection determines how the database connection interacts with the operating system. The run-time environment for the database connection is specified when the database connection is opened and cannot be changed. In fact, the procedure for creating (for "opening") a database connection object is really a method on the run-time environment object.
The life-cycle of a database connection object usually goes something like this:
- Create the database connection object (also referred to as "opening" the database connection) using the sqlite4_open() method of the sqlite4_env object.
- Run SQL statements against the database connection using either the direct interface (sqlite4_exec()) or by creating prepared statement objects, one for each SQL statement, and evaluating the prepared statement objects.
- Destroy the database connection object (also referred to as "closing" the database connection) using the sqlite4_close() method of the sqlite4 object.
Much SQLite programming centers around the use of database connection objects and programmers who use the SQLite library will quickly become familar with this object. It is the traditional (at least among the SQLite developers themselves) to use the name "db" for variable that are pointers to a database connection object.
3.4 The Prepared Statement Object
A prepared statement object, sqlite4_stmt, represents a single statement of SQL that is to be run against a particular database connection. Every prepared statement object is associated with a particular database connection object. All prepared statements for a database connection must be destroyed (or "finalized") before the database connection can be destroyed (or "closed").
The use of prepared statements is not strictly necessary in SQLite4. One can do any type of SQL statement desired using the sqlite4_exec() method of the database connection object. However, prepared statements are very handy and every programmer who wants to make serious use of the SQLite4 library needs to be familiar with them.
A prepared statement object is created using the sqlite4_prepare() method on the database connection object. The sqlite4_prepare() method takes a string as input which contains the text of the SQL statement, and outputs a pointer to a prepared statement object. Think of the SQL statement text as source code and the prepared statement object as object code and the sqlite4_prepare() method as the compiler.
A prepared statement is evaluated or executed using the sqlite4_step() method on the prepared statement object. Each call to sqlite4_step() advances the prepared statement until it either (1) determines the next row of output or (2) completely finishes evaluating the SQL statement or (3) encounters an error. The integer return value from sqlite4_step() lets the application know which of these three outcomes occurred. For an SQL statement like an INSERT, UPDATE, or DELETE that does not return a result, the sqlite4_step() statement is usually called just a single time and reports completion of the SQL statement after the first step. For a SELECT statement that returns multiple rows of output, on the other hand, the sqlite4_step() statement will be called multiple times, once for each row of output, then one last time to complete the evaluation.
At any time, though typically after the prepared statement completes, the prepared statement object can be destroyed (or "finalized") by calling the sqlite4_finalize() method. The perpared statement can also be reset back to the beginning of its execution program at any time by using the sqlite4_reset() method. A common paradigm is to prepare a prepared statement once, then run the same statement many times using calls to sqlite4_step() for each execution with calls to sqlite4_reset() to rewind the program back to the beginning whenever it completes, followed by a single call to sqlite4_finalize() to destroy the prepared statement object at the very end.
The SQL text used to create a prepared statement might include "parameter" - placeholders for unknown values. For example:
INSERT INTO tab_xyz(x1,y2,z3) VALUES(:x, :y, :z);
In the INSERT statement above, the three values to be inserted as a new row into tab_xyz are specified as parameters ":x", ":y", and ":z". Parameters can also be named as just a question mark ("?") or as a question mark followed by a small integer (ex: "?17") or as identifiers that come after "$" or "@" (ex: "$var1" or "@xyz").
When a prepared statement contains parameters, values (strings, blobs, integers, floating point numbers) can be "bound" to those parameters using methods on the prepared statement object. For example, the sqlite4_bind_text() method will bind a UTF8 string to a parameter. The sqlite4_bind_int64() method will bind a signed 64-bit integer to a parameter. The sqlite4_bind_blob() method will bind a BLOB to the parameter. And so forth. A parameter can be bound multiple times. Each binding overrides the previous.
The use of parameters has important advantages:
- Prepared statements can be compiled once and then reused many times, by resetting the prepared statement and rebinding new values into the parameters. This reduces the number of calls to sqlite4_prepare() and thus improves application performance.
- Values bound to parameters do not need to be escaped or encoded in any way. One binds raw content. This is both more efficient than having to encode the value as SQL text, and it completely eliminates the possibility of an SQL injection attack.
It is stated above that the use of prepared statements is not required since all SQL statements can be run using the sqlite4_exec() method on the database connection object. That statement is true, but does not give the whole story. In reality, the use of a prepared statement is the only way to evaluate SQL statements in SQLite4. The sqlite4_exec() method is just a convenience wrapper that does all of the low-level work of preparing, stepping, and finalizing the prepared statements automatically, and out of sight. So even though an application can avoid having to use prepared statements directly by using just sqlite4_exec(), prepared statement objects are still being used behind the scenes.
4.0 Usage Example