This document outlines the chief goals and objectives of the SQLite library. All of the features and capabilities of SQLite (that is to say, the requirements of SQLite) can ultimately be traced back to one of the broad and general requirements specified here.
SQLite is not a complete system and so the requirements provided by this document are not system requirements in the strict sense. Perhaps "sub-system requirements" would be a better term, since SQLite is always a sub-component of a larger system.
This document is targeted primarily at developers who are working within a waterfall development model that uses detailed requirements written with the modal auxiliary verb "shall". However, this document is also useful as a general-purpose high-level description of SQLite for developers who are not working under different development paradigms.
S10000: The SQLite library shall translate high-level SQL statements into low-level I/O calls to persistent storage.
SQLite is an SQL database engine. And the fundamental task of every SQL database engine it to translate the abstract SQL statements readily understood by humans into sequences of I/O operations readily understood by computer hardware. This requirement expresses the essense of SQLite.
S10100: The SQLite library shall accepts a well-defined dialect of SQL that conforms to published SQL standards.
SQL is one of the worlds most widely known programming languages, but it is also one of the most ill-defined. There are various SQL standards documents available. However the SQL standards documents are obtuse to the point of being incomprehensible. And the standards allow for so much "implementation defined" behavior that there exist two SQL database engines understand exactly the same language.
SQLite does not attempt to obtain strict compliance with any one of the various SQL standards. Instead, SQLite tries to be as compatible as possible with other SQL database engines. SQLite attempts to operate on the principle of least surprise. That is to say, experienced SQL programmers should find SQLite's dialect intuitive and natural.
SQLite may omit some obscure features of SQL. And the SQL dialect that SQLite understands might contain some enhancements not found in some standards documents. Nevertheless, applications written for other SQL database engines should be portable to SQLite with little to no change. And programmers writing code for SQLite should not encounter anything unexpected.
S10120: The SQLite library shall implement the standard SQL interpretation of NULL values.
In cases where published standards are ambiguous, SQLite will follow the practice of other popular database engines.
S10200: The SQLite library shall communicate directly with database files in persistent storage.
Most other database engines implement a client/server model in which a small client library is linked with the application and the client communicates with a separate server process using interprocess communication (IPC). SQLite avoids the complication of having a separate server process by doing direct I/O directly to the underlying filesystem.
S10300: The SQLite library shall implement ACID transactions.
In the database world, "ACID" is an acronym for Atomic, Consistent, Isolated, and Durable. Atomic means that a change to the database happens either entirely or not at all. Consistent means that if the database file is well-formed before the start of a transaction then it is guaranteed to be well-formed after the transaction commits. Isolated means that when two or more threads are processes are working with the same database, uncommitted changes made by one are not visible to the other. Durable means that once a transaction commits, it stays committed even if there is a subsequent software crash or power failure.
S10500: The SQLite library shall implement transactions that are robust across application crashes, operating-system crashes, and power failures.
An operating system crash or an unexpected power loss can sometimes damage the underlying persistent storage in ways that no software can defend against. (For example, the content of a disk drive might be completely erased and become unrecoverable.) Nevertheless, software can take steps to defend against the kinds of damage that typically occurs following operating system crashes and power failures. The usual damage is that some writes are missing or incomplete and that writes have occurred out of order. We say that software is "robust" if it defines against the common kinds of damage seen following an operating system crash or power loss.
S10600: The SQLite library shall support simultaneous access to multiple database files on the same database connection.
Many applications benefit from being about to access multiple database file using the same database connection, so that information can be transfered from from one database to another atomically, or so that queries can join data across multiple databases.
S20000: The SQLite library shall be extensible and configurable.
SQLite is intended to be an embedded database that functions well in resource-limited systems. For that reason we desire to keep the size of the library small. That choices argues against a large default function set. Instead of having many built-in features, SQLite is designed to be extensible at compile-time and run-time with new application-defined functions and behaviors.
S20100: The SQLite library shall provide interfaces that permit the application to override interfaces to the platform on which the application is running.
SQLite works on common workstations and in embedded systems. Sometimes these devices, particularly embedded systems, have odd and unusual operating systems. In order to support this level of portability, SQLite allows the interface to the operating system to be defined at run-time.
S20200: The SQLite library shall provide interfaces that permit the application to create new SQL functions.
Most SQL database engines support a rich set of SQL functions. SQLite, in contrast, supports only a select few SQL functions. But SQLite makes up for its dearth of built-in SQL functions by allowing the application to create new SQL function easily.
S20300: The SQLite library shall provide interfaces that permit the application to create new text collating sequences.
By default, SQLite only understands ASCII text. The tables needed to do proper comparisons and case folding of full unicode text are huge - much larger than the SQLite library itself. And, any application that is dealing with unicode already probably already has those tables built in. For SQLite to include unicode comparison tables would be redundant and wasteful. As a compromise, SQLite allows the application to specify alternative collating sequences for things such as unicode text, so that for applications that need such comparison sequences can have them easily while other applications that are content with ASCII are not burdened with unnecessary tables.
S20400: The SQLite library shall provide interfaces that permit the application to create new classes of virtual SQL tables.
A virtual table is an SQL object that appears to be an ordinary SQL table for the purposes of INSERT, UPDATE, DELETE, and SELECT statements. But instead of being backed by persistent storage, the virtual table is an object that responds programmatically to INSERT, UPDATE, DELETE, and SELECT requests. Virtual tables have been used to implement full-text search and R-Tree indices, among other things.
S20500: The SQLite library shall provide interfaces that permit the application to load extensions at run-time using shared libraries.
Some applications choose to package extensions in separate shared library files and load those extensions at run-time on an as-needed basis. Depending on the nature of the application, this can be an aid to configuration management, since it allows the extension to be updated without having to replace the core application.
S20600: The SQLite library shall provide interfaces that permit the application to dynamically query and modify size limits.
SQLite has finite limits. For example, there is a maximum size BLOB or CLOB that SQLite will store, a maximum size to a database file, a maximum number of columns in a table or query, and a maximum depth of an expression parse tree. All of these have default values that are sufficiently large that a typical application is very unlikely to ever reach the limits. But some applications (for example, applications that process content from untrusted and possibly hostile sources) might want to define much lower limits on some database connections for the purpose of preventing denial-of-service attacks. Or, an application might want to select much lower limits in order to prevent over-utilization of limited resources on an embedded device. Whatever the rationale, SQLite permits limits to be queried and set at run-time.
S30000: The SQLite library shall be safe for use in long-running, low-resource, high-reliability applications.
SQLite is designed to work well within embedded devices with very limited resources. To this end, it expects to confront situations where memory is unavailable and where I/O operations fail and it is designed to handle such situations with ease and grace. SQLite also avoids aggravating low-resource situations by correctly freeing rather than leaking resources it uses itself.
S30100: The SQLite library shall release all system resources it holds when it is properly shutdown.
A "Proper shutdown" means that all resources that the application has allocated from SQLite have been released by the application. The leak-free operation guarantee of SQLite applies even if there have been memory allocation errors or I/O errors during operation.
S30200: The SQLite library shall be configurable so that it is guaranteed to never fail a memory allocation as long as the application does not request resources in excess of reasonable and published limits.
Safety-critical systems typically disallow the use of malloc() and free() because one never knows when they might fail due to memory fragmentation. However, SQLite makes extensive use of dynamic objects and so it must be able to allocate and deallocate memory to hold those objects.
In order to be acceptable for use in safety critical systems, SQLite can be configured to use its own internal memory allocator which, subject to proper usage by the application, guarantees that memory allocation will never fail either due to memory fragmentation or any other cause. The proof of correctness is due to J. M. Robson: "Bounds for Some Functions Concerning Dynamic Storage Allocations", Journal of the ACM, Volume 21, Number 3, July 1974.
The internal memory allocator is seeded with a large contiguous block of memory at application start. SQLite makes all of its internal memory allocations from this initial seed. The Robson proof depends on SQLite being coupled to a well-behaved application. The application must not try to use more than a precomputed fraction of the available memory - that fraction depending on the size ratio between the largest and smallest memory allocations. Additional details are provided elsewhere.
S30210: The SQLite library shall be provide instrumentation that can alert the application when its resource usages nears or exceeds the limits of the memory breakdown guarantee.
To help insure that an application never fails a memory allocation call, SQLite provides interfaces that can inform the application if its memory usage is growing close to or has exceeded the critical Robson limits. In practice, the memory used by an application can exceed the limits of the Robson proof by a wide margin with no harmful effect. There is plenty of safety margin. But the Robson proof does break down once the limits are exceeded and the guarantee that no memory allocation will fail is lost. Hence it is important to be able to track how close an application has come to reaching critical limits.
S30220: The SQLite library shall be provide facilities to automatically recycle memory when usage nears preset limits.
When SQLite comes under memory pressure, it can be configured to recycle memory from one use to another, thus helping to reduce the pressure. "Memory pressure" means that memory available for allocation is becoming less plentiful. In a safety-critical application, memory pressure might mean that the amount of allocated memory is getting close to the point where the Robson proof breaks down. On a workstation, memory pressure might mean that available virtual memory is running low.
S30230: The SQLite library shall be permit BLOB and CLOB objects to be read and written incrementally using small memory buffers.
SQLite provides the ability to read and write megabyte or gigabyte blobs and text strings without having to allocate enough memory to hold the entire blob and string in memory all at once. This enables SQLite to read and write BLOBs that are actually larger than the available memory on the device. It also helps reduce the size of the maximum memory allocation which helps keep memory usage below Robson limits and thus helps to guarantee failure-free memory allocation.
S30300: When a memory allocation fails, SQLite shall either silently make due without the requested memory or else it shall report the error back to the application.
Memory allocation problems do not cause SQLite to fail catastrophically. SQLite recognizes all memory allocation failures and either works around them, or cleanly aborts what it is doing and returns to the application with an error that indicates insufficient memory was available. Assuming new memory becomes available, SQLite is able to continue operating normally after a memory allocation failure.
S30400: When a I/O operation fails, SQLite shall either silently recover or else it shall report the error back to the application.
SQLite responses sanely to disk I/O errors. If it is unable to work around the problem, SQLite might have to report the error back up to the application. In either case, SQLite is able to continue functioning, assuming of course that the I/O error was transient.
S30500: SQLite shall provide the capability to monitor the progress and interrupt the evaluation of a long-running query.
SQLite is able to cleanly abort an operation in progress and afterwards continue functioning normally without any memory or other resource leaks. An example of where this functionality is used occurs in the command-line interface (CLI) program for SQLite. If the user enters a query that has millions of result rows, those rows begin pouring out onto the screen. The operator can then hit the interrupt key sequence (which varies from one operating system to another but it often Control-C) which causes the query to be aborted.
S30600: All unused portions of a well-formed SQLite database file shall be available for reuse.
When information is deleted from an SQLite database, the default action is for SQLite to mark the space as unused and then to reuse the space at the next opportune INSERT. On devices where persistent storage is scarce, however, it is sometime desirable to return the unused space back to the operating system. SQLite supports this.
S30800: SQLite shall provide the interfaces that support testing and validation of the library code in an as-delivered configuration.
In consumer-grade software, it is often acceptable to run tests on an instrumented version of the code. But for high-reliability systems, it is better to test the code exactly as it is deployed. The saying at NASA is "test what you fly and fly what you test." In support of this goal, SQLite includes interfaces whose only purpose is to observe internal state and to place SQLite into internal states for the testing.
S30900: SQLite shall provide the ability for separate database connections within the same process to share resources.
On resource-constrained devices, it is desirable to get double-duty out of resources where possible.
S40000: The SQLite library shall be safe for use in applications that make concurrent access to the underlying database from different threads and/or processes.
In nearly all modern digital systems, there are many things happening at once. And many of those things involve SQLite.
S40100: The SQLite library shall be configurable to operate correctly in a multi-threaded application.
The developers of SQLite believe that "thread-safe" is a self contradiction. No application that includes multiple threads of control within the same address space is every truly "safe". And yet it is recognized that many developers want to create multithreaded applications and to use SQLite in those applications. Therefore, SQLite is engineered to be "thread-safe".
S40300: The SQLite library shall automatically control access to common databases from different connections in different threads or processes.
SQLite uses both internal mutexes and external file locking to ensure that two or more threads or processes working on the same database file play nicely with one another.
S40410: The SQLite library shall provide interfaces to assist the application in responding appropriately when an operation can not be completed due to concurrent access constraints.
If an SQL statement cannot be completed because another process is holding a lock on the database, then the application needs to be able to take corrective action, such waiting for the lock to clear.
S50000: The SQLite library shall be cross-platform.
Cross-platform in this context means that the SQLite can be used on a wide variety of operating systems and processors, ranging from small, special-purpose embedded systems, to workstations, to servers. Platforms can be 32- or 64-bit, big-endian or little-endian. Cross-platform refers to the source code. Obviously the SQLite would need to be recompiled in order to run on processors with different instruction sets.
S50100: The SQLite library shall be implemented in ANSI-C.
C has been called the "universal assembly language". Nearly all computer systems accept code written in C. Thus, to help make SQLite cross-platform:
S50300: SQLite database files shall be processor and byte-order independent.
An SQLite database file can be freely moved between machine with different operating systems, different processors, different size integers, and different byte orders. The same database file should work on any machine.
S60000: The SQLite library shall provide introspection capabilities to the application.
Some applications need to be able to discover characteristics of their environment at run-time and to make appropriate adjustments to their processing to accommodate the environment they find themselves in. SQLite attempts to support this need.
S60100: The SQLite library shall provide interfaces that an application can use to discover fixed, compile-time characteristics of the SQLite library.
Some applications are designed to work with different versions of SQLite which may or may not enable selected features. For example, SQLite can be compiled to be threadsafe or not. The threadsafe version works in multi-threaded applications. The non-threadsafe build runs faster. When an application is using an unknown version of SQLite it is important that it be able to determine the characteristics of the particular SQLite build it is using.
S60600: The SQLite library shall provide interfaces that allow an application to discover relationships between SQLite objects.
SQLite objects are often related. For example, every prepared statement is associated with a database connection. And every function context is associated with a prepared statement. Applications and extensions frequently find it useful to be able to discover these relationships at runtime.
S70000: The SQLite library shall provide interfaces that promote the safe construction and processing of SQL statements and data from untrusted sources.
Many applications need to be able to safely process data or even SQL statements that are received from untrusted sources. An "SQL Injection Attack" occurs when an adversary intentionally introduces data that is designed to have undesirable side effects on the database files. For example, suppose an application generates an INSERT statement as follows:snprintf(z, n, "INSERT INTO table1 VALUES('%s')", zUserData);
If a hostile user supplies data that reads:beginning'); DELETE FROM table1; INSERT INTO table1 VALUES('
Then the constructed INSERT statement would be transformed into three statements, the second of which is an undesired deletion of all prior content from the table. SQLite contains interfaces that are designed to help applications avoid SQL injection attacks and similar problems.
S70100: The SQLite library shall provide the application means by which the application can test and enforce compliance with database access policies for any particular SQL statement.
Some applications (for example CVSTrac and Fossil) will run SELECT statements entered by anonymous users on the internet. Such applications want to be able to guarantee that a hostile users does not access restricted tables (such as the PASSWORD column of the USER table) or modify the database in any way. SQLite supports the ability to analyze an arbitrary SQL statement to insure that it does not perform undesired operations.
S70200: The SQLite library shall provide interfaces that test to see if an SQL statement being received incrementally is complete.
Applications such as the command-line interface (CLI) for SQLite will prompt the user to enter SQL statements and will evaluate those statements as they are entered. But sometimes an SQL statement spans multiple lines. The CLI needs to know to issue a continuation prompt and await additional input if the input received so far is incomplete. SQLite supports interfaces that allow the CLI and similar applications to know if the input it has gathered so far is complete or if it needs to await additional input before processing the SQL.
S70300: The SQLite library shall support prepared statement objects with late parameter binding
The concept of a "prepared statement" allows an SQL statement to be parsed and compiled once and then reused many times. This is a performance advantage in many applications. In addition, binding values to variables in the prepared statement is safer than embedding values as literals because bound values do not need to be quoted in order to avoid an SQL injection attack.
S80000: SQLite shall exhibit ductile failure characteristics
A common characteristic of digital systems (as opposed to analog systems) is that digital systems tend to be brittle. In other words, digital systems tend to work perfectly with no sign of stress until they fail utterly and completely. The behavior is like a physical object that holds its shape as external loads increase, until it shatters without warning.
In most circumstances, ductile failure is preferred over brittle failure. A ductile device begins showing signs of trouble well in advance of failure. Physical objects bend and/or crack, providing operators with warnings of overload and an opportunity to take corrective action, while continuing to function for as long as possible.
Digital systems have a reputation for being brittle, yet brittleness is not an intrinsic property of digital systems. Digital systems can be designed to continuing functioning outside their design parameters while providing operators with warning of possible trouble. But there most be focused effort on the part of the designers to make digital systems ductile. With analog systems, the ductileness tends to be inherent in the medium, but with digital systems ductileness needs to be explicitly added.
S80100: SQLite shall make anomalies visible to the application
SQLite strives to deal gracefully with anomalous behavior by the application or by its own internal subsystems. Yet graceful handling of out-of-band inputs is of no value if the anomaly goes unreported. The problems must be visible to the application so that warnings and alarms can be propagated to operators. The useful aspect of ductile failure is that it gives advance warning. Ductile behavior is of no use to anyone if nobody can see the part bending.
*** DRAFT ***