hd_keywords {virtual machine instructions} {VDBE} {virtual machine} \ {opcodes} {bytecode engine} {bytecodes} {bytecode}

Executive Summary

SQLite works by translating SQL statements into bytecode and then running that bytecode in a virtual machine. This document describes how how the bytecode engine works.

This document describes SQLite internals. The information provided here is not needed for routine application development using SQLite. This document is intended for people who want to delve more deeply into the internal operation of SQLite.

The bytecode engine is not an API of SQLite. Details about the bytecode engine change from one release of SQLite to the next. Applications that use SQLite should not depend on any of the details found in this document. set uuid {} catch { exec fossil sha3sum $::SRC/src/vdbe.c } uuid set uuid [lindex $uuid 0] set fd [open $::SRC/src/vdbe.c r] set file [read $fd] close $fd set current_op {} unset -nocomplain Opcode unset -nocomplain OpcodeList foreach line [split $file \n] { set line [string trim $line] if {[string index $line 1]!="*"} { set current_op {} continue } if {[regexp {^\*\* Synopsis:} $line]} continue if {[regexp {^/\* Opcode: } $line]} { set current_op [lindex $line 2] set txt [lrange $line 3 end] regsub -all {>} $txt {\>} txt regsub -all {<} $txt {\<} txt set Opcode($current_op:args) $txt lappend OpcodeList $current_op set pend {} set pstart {} continue } if {$current_op==""} continue if {[regexp {^\*/} $line]} { set current_op {} continue } set line [string trim [string range $line 3 end]] if {$line==""} { append Opcode($current_op:text) $pend set pend {} set pstart {

} } else { if {![regexp {} $line {\>} line regsub -all {<} $line {\<} line } append Opcode($current_op:text) \n$pstart$line set pstart {} set pend "

\n" } } unset file set fd [open $::SRC/VERSION r] set SQLITE_VERSION [string trim [read $fd]] close $fd set fd [open $::SRC/manifest.uuid r] set SQLITE_SHA1_VERSION [string trim [read $fd]] set SQLITE_HASH [string range $SQLITE_SHA1_VERSION 0 12] close $fd set fd [open $::SRC/manifest r] set file [read $fd] close $fd set SQLITE_CKIN_DATE {unknown} regexp {\nD (\d\d\d\d-\d\d-\d\d)} $file all SQLITE_CKIN_DATE unset file # Scan $txt and replace every opcode name with a link to its documentation # proc LinkOpcodeNames {txt} { global Opcode set out {} while {[regexp {^(.*?)(OP_[A-Z][a-z])\y(.*)$} $txt \ all pre op tail] || [regexp {^(.*?)\y((OP_)?[A-Z][A-Za-z][A-Za-z0-9]+)\y(.*)$} $txt \ all pre op opx tail]} { hd_resolve $pre if {($pre=="" || [regexp {> *$} $pre]) && ![regexp {^OP_} $op] || [regexp {^object} [string trim $tail]]} { hd_puts $op } else { regsub {^OP_} $op {} key if {[info exists Opcode($key:text)]} { hd_puts "$key" } else { hd_puts $op } } set txt $tail } hd_resolve $txt }


SQLite works by translating each SQL statement into bytecode and then running that bytecode. A [prepared statement] in SQLite is mostly just the bytecode needed to implement the corresponding SQL. The [sqlite3_prepare_v2()] interface is a compiler that translates SQL into bytecode. The [sqlite3_step()] interface is the virtual machine that runs the bytecode contained within the [prepared statement].

The bytecode virtual machine is the heart of SQLite. Programmers who want to understand how SQLite operates internally must be familiar with the bytecode engine.

Historically, the bytecode engine in SQLite is called the "Virtual DataBase Engine" or "VDBE". This website uses the terms "bytecode engine", "VDBE", "virtual machine", and "bytecode virtual machine" interchangeably, as they all mean the same thing.

This article also uses the terms "bytecode program" and "prepared statement" interchangeably, as they are mostly the same thing.

VDBE Source Code

The source code to the bytecode engine is in the [http://www.sqlite.org/src/file/src/vdbe.c | vdbe.c] source file. The [opcode definitions] in this document are derived from comments in that source file. The source code comments are the canonical source of information about the bytecode engine. When in doubt, refer to the source code.

In addition to the primary vdbe.c source code file, there are other helper code files in the source tree, all of whose names begin with "vdbe" - short for "Virtual DataBase Engine".

Remember that the names and meanings of opcodes often change from one release of SQLite to the next. So if you are studying the [EXPLAIN] output from SQLite, you should reference the version of this document (or the vdbe.c source code) that corresponds to the version of SQLite that ran the [EXPLAIN]. Otherwise, the description of the opcodes may not be accurate. This document is derived from SQLite hd_puts " version $SQLITE_VERSION check-in\n" set url https://www.sqlite.org/src/timeline?c=$SQLITE_HASH hd_puts "$SQLITE_HASH dated $SQLITE_CKIN_DATE.\n\n"

Instruction Format

LinkOpcodeNames {

A bytecoded program in SQLite consists of one or more instructions. Each instruction has an opcode and five operands named P1, P2 P3, P4, and P5. The P1, P2, and P3 operands are 32-bit signed integers. These operands often refer to registers. For instructions that operate on b-tree cursors, the P1 operand is usually the cursor number. For jump instructions, P2 is usually the jump destination. P4 may be a 32-bit signed integer, a 64-bit signed integer, a 64-bit floating point value, a string literal, a Blob literal, a pointer to a collating sequence comparison function, or a pointer to the implementation of an application-defined SQL function, or various other things. P5 is an 16-bit unsigned integer normally used to hold flags. Bits of the P5 flag can sometimes affect the opcode in subtle ways. For example, if the SQLITE_NULLEQ (0x0080) bit of the P5 operand is set on the OP_Eq opcode, then the NULL values compare equal to one another. Otherwise NULL values compare different from one another. }

Some opcodes use all five operands. Some opcodes use one or two. Some opcodes use none of the operands.

LinkOpcodeNames {

The bytecode engine begins execution on instruction number 0. Execution continues until a OP_Halt instruction is seen, or until the program counter becomes greater than the address of last instruction, or until there is an error. When the bytecode engine halts, all memory that it allocated is released and all database cursors it may have had open are closed. If the execution stopped due to an error, any pending transactions are terminated and changes made to the database are rolled back.

LinkOpcodeNames {

The OP_ResultRow opcode causes the bytecode engine to pause and the corresponding [sqlite3_step()] call to return [SQLITE_ROW]. Before invoking OP_ResultRow, the bytecoded program will have loaded the results for a single row of a query into a series of registers. C-language APIs such as [sqlite3_column_int()] or [sqlite3_column_text()] extract the query results from those registers. The bytecode engine resumes with the next instruction after the OP_ResultRow on the next call to [sqlite3_step()]. }


LinkOpcodeNames {

Every bytecode program has a fixed (but potentially large) number of registers. A single register can hold a variety of objects:

  • A NULL value
  • A signed 64-bit integer
  • An IEEE double-precision (64-bit) floating point number
  • An arbitrary length strings
  • An arbitrary length BLOB
  • A RowSet object (See the OP_RowSetAdd, OP_RowSetRead, and OP_RowSetTest opcodes)
  • A Frame object (Used by [subprograms] - see OP_Program)

A register can also be "Undefined" meaning that it holds no value at all. Undefined is different from NULL. Depending on compile-time options, an attempt to read an undefined register will usually cause a run-time error. If the code generator ([sqlite3_prepare_v2()]) ever generates a [prepared statement] that reads an Undefined register, that is a bug in the code generator.

Registers are numbered beginning with 0. Most opcodes refer to at least one register.

The number of registers in a single prepared statement is fixed at compile-time. The content of all registers is cleared when a prepared statement is [sqlite3_reset()|reset] or [sqlite3_finalize()|finalized].

The internal Mem object stores the value for a single register. The abstract [sqlite3_value] object that is exposed in the API is really just a Mem object or register.

B-Tree Cursors

LinkOpcodeNames {

A prepared statement can have zero or more open cursors. Each cursor is identified by a small integer, which is usually the P1 parameter to the opcode that uses the cursor. There can be multiple cursors open on the same index or table. All cursors operate independently, even cursors pointing to the same indices or tables. The only way for the virtual machine to interact with a database file is through a cursor. Instructions in the virtual machine can create a new cursor (ex: OP_OpenRead or OP_OpenWrite), read data from a cursor (OP_Column), advance the cursor to the next entry in the table (ex: OP_Next or OP_Prev), and so forth. All cursors are automatically closed when the prepared statement is [sqlite3_reset()|reset] or [sqlite3_finalize()|finalized]. } hd_fragment subprog subprograms

Subroutines, Coroutines, and Subprograms

The bytecode engine has no stack on which to store the return address of a subroutine. Return addresses must be stored in registers. Hence, bytecode subroutines are not reentrant. LinkOpcodeNames {

The OP_Gosub opcode stores the current program counter into register P1 then jumps to address P2. The OP_Return opcode jumps to address P1+1. Hence, every subroutine is associated with two integers: the address of the entry point in the subroutine and the register number that is used to hold the return address.

The OP_Yield opcode swaps the value of the program counter with the integer value in register P1. This opcode is used to implement coroutines. Coroutines are often used to implement subqueries from which content is pulled on an as-needed basis. }

[CREATE TRIGGER|Triggers] need to be reentrant. LinkOpcodeNames { Since bytecode subroutines are not reentrant a different mechanism must be used to implement triggers. Each trigger is implemented using a separate bytecode program with its own opcodes, program counter, and register set. The OP_Program opcode invokes the trigger subprogram. The OP_Program instruction allocates and initializes a fresh register set for each invocation of the subprogram, so subprograms can be reentrant and recursive. The OP_Param opcode is used by subprograms to access content in registers of the calling bytecode program. }

Self-Altering Code

LinkOpcodeNames {

Some opcodes are self-altering. For example, the OP_Init opcode (which is always the first opcode run a bytecode program) increments its P1 operand. Subsequent OP_Once opcodes compare their P1 operands to the P1 value for the OP_Init opcode in order to determine if the one-time initialization code that follows should be skipped. Another example is the OP_String8 opcode which converts its P4 operand from UTF-8 into the correct database string encoding, then converts itself into a OP_String opcode. }

Viewing The Bytecode

Every SQL statement that SQLite interprets results in a program for the virtual machine. But if the SQL statement begins with the keyword [EXPLAIN] the virtual machine will not execute the program. Instead, the instructions of the program will be returned, one instruction per row, like a query result. This feature is useful for debugging and for learning how the virtual machine operates. For example:

proc Code {body} { hd_puts {
  regsub -all {&} [string trim $body] {\&} body
  regsub -all {>} $body {\>} body
  regsub -all {<} $body {\<} body
  regsub -all {\(\(\(} $body {} body
  regsub -all {\)\)\)} $body {} body
  regsub -all { } $body {\ } body
  hd_puts $body
  hd_puts {
} } Code { $ (((sqlite3 ex1.db))) sqlite> (((explain delete from tbl1 where two<20;))) addr opcode p1 p2 p3 p4 p5 comment ---- ------------- ---- ---- ---- ------------- -- ------------- 0 Init 0 12 0 00 Start at 12 1 Null 0 1 0 00 r[1]=NULL 2 OpenWrite 0 2 0 3 00 root=2 iDb=0; tbl1 3 Rewind 0 10 0 00 4 Column 0 1 2 00 r[2]=tbl1.two 5 Ge 3 9 2 (BINARY) 51 if r[2]>=r[3] goto 9 6 Rowid 0 4 0 00 r[4]=rowid 7 Once 0 8 0 00 8 Delete 0 1 0 tbl1 02 9 Next 0 4 0 01 10 Noop 0 0 0 00 11 Halt 0 0 0 00 12 Transaction 0 1 1 0 01 usesStmtJournal=0 13 TableLock 0 2 1 tbl1 00 iDb=0 root=2 write=1 14 Integer 20 3 0 00 r[3]=20 15 Goto 0 1 0 00 }

Any application can run an [EXPLAIN] query to get output similar to the above. However, indentation to show the loop structure is not generated by the SQLite core. The [command-line shell] contains extra logic for indenting loops. Also, the "comment" column in the [EXPLAIN] output is only provided if SQLite is compiled with the [-DSQLITE_ENABLE_EXPLAIN_COMMENTS] options.

When SQLite is compiled with the [SQLITE_DEBUG] compile-time option, extra [PRAGMA] commands are available that are useful for debugging and for exploring the operation of the VDBE. For example the [vdbe_trace] pragma can be enabled to cause a disassembly of each VDBE opcode to be printed on standard output as the opcode is executed. These debugging pragmas include:

  • [PRAGMA parser_trace]
  • [PRAGMA vdbe_addoptrace]
  • [PRAGMA vdbe_debug]
  • [PRAGMA vdbe_listing]
  • [PRAGMA vdbe_trace]

The Opcodes

There are currently hd_puts [llength $OpcodeList] opcodes defined by the virtual machine. All currently defined opcodes are described in the table below. This table was generated automatically by scanning the source code from the file if {$uuid==""} { hd_puts "vdbe.c.\n" } else { hd_puts "vdbe.c.\n" }

Remember: The VDBE opcodes are not part of the interface definition for SQLite. The number of opcodes and their names and meanings change from one release of SQLite to the next. The opcodes shown in the table below are valid for SQLite hd_puts " version $SQLITE_VERSION check-in\n" hd_puts "$SQLITE_HASH dated $SQLITE_CKIN_DATE.\n\n" hd_fragment codes {list of current bytecodes} {opcode definitions} hd_puts {

} foreach op [lsort -dictionary $OpcodeList] { hd_puts {\n" } hd_resolve {
Opcode NameDescription
} hd_puts "\n$op\n" regsub -all {\[(P[0-9+]+)\]} $Opcode($op:text) {\[\1\]} txt hd_puts "" set txt [string trim $txt] if {[string match

* $txt]} {set txt [string range $txt 3 end]} LinkOpcodeNames $txt hd_puts "