*** DRAFT ***

SQLite Virtual Machine Opcodes

Introduction

In order to execute an SQL statement, the SQLite library first parses the SQL, analyzes the statement, then generates a short program to execute the statement. The program is generated for a "virtual machine" implemented by the SQLite library. This document describes the operation of that virtual machine.

This document is intended as a reference, not a tutorial. A separate Virtual Machine Tutorial is available. If you are looking for a narrative description of how the virtual machine works, you should read the tutorial and not this document. Once you have a basic idea of what the virtual machine does, you can refer back to this document for the details on a particular opcode. Unfortunately, the virtual machine tutorial was written for SQLite version 1.0. There are substantial changes in the virtual machine for version 2.0 and and again for version 3.0.0 and again for version 3.5.5 and the document has not been updated. But the basic concepts behind the virtual machine still apply.

The source code to the virtual machine is in the vdbe.c source file. All of the opcode definitions further down in this document are contained in comments in the source file. In fact, the opcode table in this document was generated by scanning the vdbe.c source file and extracting the necessary information from comments. So the source code comments are really the canonical source of information about the virtual machine. When in doubt, refer to the source code.

Each instruction in the virtual machine consists of an opcode and up to five operands named P1, P2 P3, P4, and P5. P1, P2, and P3 are 32-bit signed integers. These operands often refer to registers. P2 is always the jump destination in any operation that might cause a jump. 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 unsigned character normally used as a flag. Some operators use all five operands. Some use one or two. Some operators use none of the operands.

The virtual machine begins execution on instruction number 0. Execution continues until a Halt instruction is seen, or the program counter becomes one greater than the address of last instruction, or there is an execution error. When the virtual machine 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.

The virtual machine can have zero or more cursors. Each cursor is a pointer into a single table or index within the database. There can be multiple cursors pointing at 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 (OpenRead or OpenWrite), read data from a cursor (Column), advance the cursor to the next entry in the table (Next) or index (NextIdx), and many other operations. All cursors are automatically closed when the virtual machine terminates.

The virtual machine contains an arbitrary number of registers locations with addresses beginning at one and growing upward. Each memory location can hold an arbitrary string. The registers hold all intermediate results of a calculation.

Viewing Programs Generated By SQLite

Every SQL statement that SQLite interprets results in a program for the virtual machine. But if you precede the SQL statement with the keyword EXPLAIN the virtual machine will not execute the program. Instead, the instructions of the program will be returned like a query result. This feature is useful for debugging and for learning how the virtual machine operates.

You can use the sqlite3.exe command-line interface (CLI) tool to see the instructions generated by an SQL statement. The following is an example:

sqlite3 ex1.db
sqlite> .explain
sqlite> explain delete from tbl1 where two<20;
addr  opcode         p1    p2    p3    p4         p5  comment
----  -------------  ----  ----  ----  ---------  --  -------
0     Trace          0     0     0     explain..  00         
1     Goto           0     20    0                00         
2     OpenRead       0     2     0                00  tbl    
3     SetNumColumns  0     2     0                00         
4     Rewind         0     11    0                00         
5     Column         0     1     2                00  tbl.two
6     Integer        20    3     0                00         
7     Ge             3     10    2     cs(BINARY) 6a         
8     Rowid          0     1     0                00         
9     FifoWrite      1     0     0                00         
10    Next           0     5     0                00         
11    Close          0     0     0                00         
12    OpenWrite      0     2     0                00  tbl    
13    SetNumColumns  0     2     0                00         
14    FifoRead       1     18    0                00         
15    NotExists      0     17    1                00         
16    Delete         0     1     0     tbl        00         
17    Goto           0     14    0                00         
18    Close          0     0     0                00         
19    Halt           0     0     0                00         
20    Transaction    0     1     0                00         
21    VerifyCookie   0     1     0                00         
22    TableLock      -1    2     0     tbl        00         
23    Goto           0     2     0                00

All you have to do is add the EXPLAIN keyword to the front of the SQL statement. But if you use the ".explain" command in the CLI, it will set up the output mode to make the program more easily viewable.

Depending on compile-time options, you can put the SQLite virtual machine in a mode where it will trace its execution by writing messages to standard output. The non-standard SQL "PRAGMA" comments can be used to turn tracing on and off. To turn tracing on, enter:

PRAGMA vdbe_trace=on;

You can turn tracing back off by entering a similar statement but changing the value "on" to "off".

The Opcodes

There are currently 141 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 vdbe.c.

Opcode NameDescription

Add

Add the value in register P1 to the value in register P2 and store the result in register P3. If either input is NULL, the result is NULL.

AddImm

Add the constant P2 to the value in register P1. The result is always an integer.

To force any register to be an integer, just add 0.

Affinity

Apply affinities to a range of P2 registers starting with P1.

P4 is a string that is P2 characters long. The nth character of the string indicates the column affinity that should be used for the nth memory cell in the range.

AggFinal

Execute the finalizer function for an aggregate. P1 is the memory location that is the accumulator for the aggregate.

P2 is the number of arguments that the step function takes and P4 is a pointer to the FuncDef for this function. The P2 argument is not used by this opcode. It is only there to disambiguate functions that can take varying numbers of arguments. The P4 argument is only needed for the degenerate case where the step function was not previously called.

AggStep

Execute the step function for an aggregate. The function has P5 arguments. P4 is a pointer to the FuncDef structure that specifies the function. Use register P3 as the accumulator.

The P5 arguments are taken from register P2 and its successors.

And

Take the logical AND of the values in registers P1 and P2 and write the result into register P3.

If either P1 or P2 is 0 (false) then the result is 0 even if the other input is NULL. A NULL and true or two NULLs give a NULL output.

AutoCommit

Set the database auto-commit flag to P1 (1 or 0). If P2 is true, roll back any currently active btree transactions. If there are any active VMs (apart from this one), then a ROLLBACK fails. A COMMIT fails if there are active writing VMs or active VMs that use shared cache.

This instruction causes the VM to halt.

BitAnd

Take the bit-wise AND of the values in register P1 and P2 and store the result in register P3. If either input is NULL, the result is NULL.

BitNot

Interpret the content of register P1 as an integer. Store the ones-complement of the P1 value into register P2. If P1 holds a NULL then store a NULL in P2.

BitOr

Take the bit-wise OR of the values in register P1 and P2 and store the result in register P3. If either input is NULL, the result is NULL.

Blob

P4 points to a blob of data P1 bytes long. Store this blob in register P2. This instruction is not coded directly by the compiler. Instead, the compiler layer specifies an OP_HexBlob opcode, with the hex string representation of the blob as P4. This opcode is transformed to an OP_Blob the first time it is executed.

Checkpoint

Checkpoint database P1. This is a no-op if P1 is not currently in WAL mode.

Clear

Delete all contents of the database table or index whose root page in the database file is given by P1. But, unlike Destroy, do not remove the table or index from the database file.

The table being clear is in the main database file if P2==0. If P2==1 then the table to be clear is in the auxiliary database file that is used to store tables create using CREATE TEMPORARY TABLE.

If the P3 value is non-zero, then the table referred to must be an intkey table (an SQL table, not an index). In this case the row change count is incremented by the number of rows in the table being cleared. If P3 is greater than zero, then the value stored in register P3 is also incremented by the number of rows in the table being cleared.

See also: Destroy

Close

Close a cursor previously opened as P1. If P1 is not currently open, this instruction is a no-op.

CollSeq

P4 is a pointer to a CollSeq struct. If the next call to a user function or aggregate calls sqlite3GetFuncCollSeq(), this collation sequence will be returned. This is used by the built-in min(), max() and nullif() functions.

The interface used by the implementation of the aforementioned functions to retrieve the collation sequence set by this opcode is not available publicly, only to user functions defined in func.c.

Column

Interpret the data that cursor P1 points to as a structure built using the MakeRecord instruction. (See the MakeRecord opcode for additional information about the format of the data.) Extract the P2-th column from this record. If there are less that (P2+1) values in the record, extract a NULL.

The value extracted is stored in register P3.

If the column contains fewer than P2 fields, then extract a NULL. Or, if the P4 argument is a P4_MEM use the value of the P4 argument as the result.

If the OPFLAG_CLEARCACHE bit is set on P5 and P1 is a pseudo-table cursor, then the cache of the cursor is reset prior to extracting the column. The first OP_Column against a pseudo-table after the value of the content register has changed should have this bit set.

Compare

Compare to vectors of registers in reg(P1)..reg(P1+P3-1) (all this one "A") and in reg(P2)..reg(P2+P3-1) ("B"). Save the result of the comparison for use by the next OP_Jump instruct.

P4 is a KeyInfo structure that defines collating sequences and sort orders for the comparison. The permutation applies to registers only. The KeyInfo elements are used sequentially.

The comparison is a sort comparison, so NULLs compare equal, NULLs are less than numbers, numbers are less than strings, and strings are less than blobs.

Concat

Add the text in register P1 onto the end of the text in register P2 and store the result in register P3. If either the P1 or P2 text are NULL then store NULL in P3.

P3 = P2 || P1

It is illegal for P1 and P3 to be the same register. Sometimes, if P3 is the same register as P2, the implementation is able to avoid a memcpy().

Copy

Make a copy of register P1 into register P2.

This instruction makes a deep copy of the value. A duplicate is made of any string or blob constant. See also OP_SCopy.

Count

Store the number of entries (an integer value) in the table or index opened by cursor P1 in register P2

CreateIndex

Allocate a new index in the main database file if P1==0 or in the auxiliary database file if P1==1 or in an attached database if P1>1. Write the root page number of the new table into register P2.

See documentation on OP_CreateTable for additional information.

CreateTable

Allocate a new table in the main database file if P1==0 or in the auxiliary database file if P1==1 or in an attached database if P1>1. Write the root page number of the new table into register P2

The difference between a table and an index is this: A table must have a 4-byte integer key and can have arbitrary data. An index has an arbitrary key but no data.

See also: CreateIndex

Delete

Delete the record at which the P1 cursor is currently pointing.

The cursor will be left pointing at either the next or the previous record in the table. If it is left pointing at the next record, then the next Next instruction will be a no-op. Hence it is OK to delete a record from within an Next loop.

If the OPFLAG_NCHANGE flag of P2 is set, then the row change count is incremented (otherwise not).

P1 must not be pseudo-table. It has to be a real table with multiple rows.

If P4 is not NULL, then it is the name of the table that P1 is pointing to. The update hook will be invoked, if it exists. If P4 is not NULL then the P1 cursor must have been positioned using OP_NotFound prior to invoking this opcode.

Destroy

Delete an entire database table or index whose root page in the database file is given by P1.

The table being destroyed is in the main database file if P3==0. If P3==1 then the table to be clear is in the auxiliary database file that is used to store tables create using CREATE TEMPORARY TABLE.

If AUTOVACUUM is enabled then it is possible that another root page might be moved into the newly deleted root page in order to keep all root pages contiguous at the beginning of the database. The former value of the root page that moved - its value before the move occurred - is stored in register P2. If no page movement was required (because the table being dropped was already the last one in the database) then a zero is stored in register P2. If AUTOVACUUM is disabled then a zero is stored in register P2.

See also: Clear

Divide

Divide the value in register P1 by the value in register P2 and store the result in register P3 (P3=P2/P1). If the value in register P1 is zero, then the result is NULL. If either input is NULL, the result is NULL.

DropIndex

Remove the internal (in-memory) data structures that describe the index named P4 in database P1. This is called after an index is dropped in order to keep the internal representation of the schema consistent with what is on disk.

DropTable

Remove the internal (in-memory) data structures that describe the table named P4 in database P1. This is called after a table is dropped in order to keep the internal representation of the schema consistent with what is on disk.

DropTrigger

Remove the internal (in-memory) data structures that describe the trigger named P4 in database P1. This is called after a trigger is dropped in order to keep the internal representation of the schema consistent with what is on disk.

Eq

This works just like the Lt opcode except that the jump is taken if the operands in registers P1 and P3 are equal. See the Lt opcode for additional information.

If SQLITE_NULLEQ is set in P5 then the result of comparison is always either true or false and is never NULL. If both operands are NULL then the result of comparison is true. If either operand is NULL then the result is false. If neither operand is NULL the the result is the same as it would be if the SQLITE_NULLEQ flag were omitted from P5.

Expire

Cause precompiled statements to become expired. An expired statement fails with an error code of SQLITE_SCHEMA if it is ever executed (via sqlite3_step()).

If P1 is 0, then all SQL statements become expired. If P1 is non-zero, then only the currently executing statement is affected.

FkCounter

Increment a "constraint counter" by P2 (P2 may be negative or positive). If P1 is non-zero, the database constraint counter is incremented (deferred foreign key constraints). Otherwise, if P1 is zero, the statement counter is incremented (immediate foreign key constraints).

FkIfZero

This opcode tests if a foreign key constraint-counter is currently zero. If so, jump to instruction P2. Otherwise, fall through to the next instruction.

If P1 is non-zero, then the jump is taken if the database constraint-counter is zero (the one that counts deferred constraint violations). If P1 is zero, the jump is taken if the statement constraint-counter is zero (immediate foreign key constraint violations).

Found

If P4==0 then register P3 holds a blob constructed by MakeRecord. If P4>0 then register P3 is the first of P4 registers that form an unpacked record.

Cursor P1 is on an index btree. If the record identified by P3 and P4 is a prefix of any entry in P1 then a jump is made to P2 and P1 is left pointing at the matching entry.

Function

Invoke a user function (P4 is a pointer to a Function structure that defines the function) with P5 arguments taken from register P2 and successors. The result of the function is stored in register P3. Register P3 must not be one of the function inputs.

P1 is a 32-bit bitmask indicating whether or not each argument to the function was determined to be constant at compile time. If the first argument was constant then bit 0 of P1 is set. This is used to determine whether meta data associated with a user function argument using the sqlite3_set_auxdata() API may be safely retained until the next invocation of this opcode.

See also: AggStep and AggFinal

Ge

This works just like the Lt opcode except that the jump is taken if the content of register P3 is greater than or equal to the content of register P1. See the Lt opcode for additional information.

Gosub

Write the current address onto register P1 and then jump to address P2.

Goto

An unconditional jump to address P2. The next instruction executed will be the one at index P2 from the beginning of the program.

Gt

This works just like the Lt opcode except that the jump is taken if the content of register P3 is greater than the content of register P1. See the Lt opcode for additional information.

Halt

Exit immediately. All open cursors, etc are closed automatically.

P1 is the result code returned by sqlite3_exec(), sqlite3_reset(), or sqlite3_finalize(). For a normal halt, this should be SQLITE_OK (0). For errors, it can be some other value. If P1!=0 then P2 will determine whether or not to rollback the current transaction. Do not rollback if P2==OE_Fail. Do the rollback if P2==OE_Rollback. If P2==OE_Abort, then back out all changes that have occurred during this execution of the VDBE, but do not rollback the transaction.

If P4 is not null then it is an error message string.

There is an implied "Halt 0 0 0" instruction inserted at the very end of every program. So a jump past the last instruction of the program is the same as executing Halt.

HaltIfNull

Check the value in register P3. If is is NULL then Halt using parameter P1, P2, and P4 as if this were a Halt instruction. If the value in register P3 is not NULL, then this routine is a no-op.

IdxDelete

The content of P3 registers starting at register P2 form an unpacked index key. This opcode removes that entry from the index opened by cursor P1.

IdxGE

The P4 register values beginning with P3 form an unpacked index key that omits the ROWID. Compare this key value against the index that P1 is currently pointing to, ignoring the ROWID on the P1 index.

If the P1 index entry is greater than or equal to the key value then jump to P2. Otherwise fall through to the next instruction.

If P5 is non-zero then the key value is increased by an epsilon prior to the comparison. This make the opcode work like IdxGT except that if the key from register P3 is a prefix of the key in the cursor, the result is false whereas it would be true with IdxGT.

IdxInsert

Register P2 holds a SQL index key made using the MakeRecord instructions. This opcode writes that key into the index P1. Data for the entry is nil.

P3 is a flag that provides a hint to the b-tree layer that this insert is likely to be an append.

This instruction only works for indices. The equivalent instruction for tables is OP_Insert.

IdxLT

The P4 register values beginning with P3 form an unpacked index key that omits the ROWID. Compare this key value against the index that P1 is currently pointing to, ignoring the ROWID on the P1 index.

If the P1 index entry is less than the key value then jump to P2. Otherwise fall through to the next instruction.

If P5 is non-zero then the key value is increased by an epsilon prior to the comparison. This makes the opcode work like IdxLE.

IdxRowid

Write into register P2 an integer which is the last entry in the record at the end of the index key pointed to by cursor P1. This integer should be the rowid of the table entry to which this index entry points.

See also: Rowid, MakeRecord.

If

Jump to P2 if the value in register P1 is true. The value is is considered true if it is numeric and non-zero. If the value in P1 is NULL then take the jump if P3 is true.

IfNeg

If the value of register P1 is less than zero, jump to P2.

It is illegal to use this instruction on a register that does not contain an integer. An assertion fault will result if you try.

IfNot

Jump to P2 if the value in register P1 is False. The value is is considered true if it has a numeric value of zero. If the value in P1 is NULL then take the jump if P3 is true.

IfPos

If the value of register P1 is 1 or greater, jump to P2.

It is illegal to use this instruction on a register that does not contain an integer. An assertion fault will result if you try.

IfZero

The register P1 must contain an integer. Add literal P3 to the value in register P1. If the result is exactly 0, jump to P2.

It is illegal to use this instruction on a register that does not contain an integer. An assertion fault will result if you try.

IncrVacuum

Perform a single step of the incremental vacuum procedure on the P1 database. If the vacuum has finished, jump to instruction P2. Otherwise, fall through to the next instruction.

Insert

Write an entry into the table of cursor P1. A new entry is created if it doesn't already exist or the data for an existing entry is overwritten. The data is the value MEM_Blob stored in register number P2. The key is stored in register P3. The key must be a MEM_Int.

If the OPFLAG_NCHANGE flag of P5 is set, then the row change count is incremented (otherwise not). If the OPFLAG_LASTROWID flag of P5 is set, then rowid is stored for subsequent return by the sqlite3_last_insert_rowid() function (otherwise it is unmodified).

If the OPFLAG_USESEEKRESULT flag of P5 is set and if the result of the last seek operation (OP_NotExists) was a success, then this operation will not attempt to find the appropriate row before doing the insert but will instead overwrite the row that the cursor is currently pointing to. Presumably, the prior OP_NotExists opcode has already positioned the cursor correctly. This is an optimization that boosts performance by avoiding redundant seeks.

If the OPFLAG_ISUPDATE flag is set, then this opcode is part of an UPDATE operation. Otherwise (if the flag is clear) then this opcode is part of an INSERT operation. The difference is only important to the update hook.

Parameter P4 may point to a string containing the table-name, or may be NULL. If it is not NULL, then the update-hook (sqlite3.xUpdateCallback) is invoked following a successful insert.

(WARNING/TODO: If P1 is a pseudo-cursor and P2 is dynamically allocated, then ownership of P2 is transferred to the pseudo-cursor and register P2 becomes ephemeral. If the cursor is changed, the value of register P2 will then change. Make sure this does not cause any problems.)

This instruction only works on tables. The equivalent instruction for indices is OP_IdxInsert.

InsertInt

This works exactly like OP_Insert except that the key is the integer value P3, not the value of the integer stored in register P3.

Int64

P4 is a pointer to a 64-bit integer value. Write that value into register P2.

Integer

The 32-bit integer value P1 is written into register P2.

IntegrityCk

Do an analysis of the currently open database. Store in register P1 the text of an error message describing any problems. If no problems are found, store a NULL in register P1.

The register P3 contains the maximum number of allowed errors. At most reg(P3) errors will be reported. In other words, the analysis stops as soon as reg(P1) errors are seen. Reg(P1) is updated with the number of errors remaining.

The root page numbers of all tables in the database are integer stored in reg(P1), reg(P1+1), reg(P1+2), .... There are P2 tables total.

If P5 is not zero, the check is done on the auxiliary database file, not the main database file.

This opcode is used to implement the integrity_check pragma.

IsNull

Jump to P2 if the value in register P1 is NULL.

IsUnique

Cursor P1 is open on an index b-tree - that is to say, a btree which no data and where the key are records generated by OP_MakeRecord with the list field being the integer ROWID of the entry that the index entry refers to.

The P3 register contains an integer record number. Call this record number R. Register P4 is the first in a set of N contiguous registers that make up an unpacked index key that can be used with cursor P1. The value of N can be inferred from the cursor. N includes the rowid value appended to the end of the index record. This rowid value may or may not be the same as R.

If any of the N registers beginning with register P4 contains a NULL value, jump immediately to P2.

Otherwise, this instruction checks if cursor P1 contains an entry where the first (N-1) fields match but the rowid value at the end of the index entry is not R. If there is no such entry, control jumps to instruction P2. Otherwise, the rowid of the conflicting index entry is copied to register P3 and control falls through to the next instruction.

See also: NotFound, NotExists, Found

JournalMode

Change the journal mode of database P1 to P3. P3 must be one of the PAGER_JOURNALMODE_XXX values. If changing between the various rollback modes (delete, truncate, persist, off and memory), this is a simple operation. No IO is required.

If changing into or out of WAL mode the procedure is more complicated.

Write a string containing the final journal-mode to register P2.

Jump

Jump to the instruction at address P1, P2, or P3 depending on whether in the most recent OP_Compare instruction the P1 vector was less than equal to, or greater than the P2 vector, respectively.

Last

The next use of the Rowid or Column or Next instruction for P1 will refer to the last entry in the database table or index. If the table or index is empty and P2>0, then jump immediately to P2. If P2 is 0 or if the table or index is not empty, fall through to the following instruction.

Le

This works just like the Lt opcode except that the jump is taken if the content of register P3 is less than or equal to the content of register P1. See the Lt opcode for additional information.

LoadAnalysis

Read the sqlite_stat1 table for database P1 and load the content of that table into the internal index hash table. This will cause the analysis to be used when preparing all subsequent queries.

Lt

Compare the values in register P1 and P3. If reg(P3)<reg(P1) then jump to address P2.

If the SQLITE_JUMPIFNULL bit of P5 is set and either reg(P1) or reg(P3) is NULL then take the jump. If the SQLITE_JUMPIFNULL bit is clear then fall thru if either operand is NULL.

The SQLITE_AFF_MASK portion of P5 must be an affinity character - SQLITE_AFF_TEXT, SQLITE_AFF_INTEGER, and so forth. An attempt is made to coerce both inputs according to this affinity before the comparison is made. If the SQLITE_AFF_MASK is 0x00, then numeric affinity is used. Note that the affinity conversions are stored back into the input registers P1 and P3. So this opcode can cause persistent changes to registers P1 and P3.

Once any conversions have taken place, and neither value is NULL, the values are compared. If both values are blobs then memcmp() is used to determine the results of the comparison. If both values are text, then the appropriate collating function specified in P4 is used to do the comparison. If P4 is not specified then memcmp() is used to compare text string. If both values are numeric, then a numeric comparison is used. If the two values are of different types, then numbers are considered less than strings and strings are considered less than blobs.

If the SQLITE_STOREP2 bit of P5 is set, then do not jump. Instead, store a boolean result (either 0, or 1, or NULL) in register P2.

MakeRecord

Convert P2 registers beginning with P1 into a single entry suitable for use as a data record in a database table or as a key in an index. The details of the format are irrelevant as long as the OP_Column opcode can decode the record later. Refer to source code comments for the details of the record format.

P4 may be a string that is P2 characters long. The nth character of the string indicates the column affinity that should be used for the nth field of the index key.

The mapping from character to affinity is given by the SQLITE_AFF_ macros defined in sqliteInt.h.

If P4 is NULL then all index fields have the affinity NONE.

MemMax

P1 is a register in the root frame of this VM (the root frame is different from the current frame if this instruction is being executed within a sub-program). Set the value of register P1 to the maximum of its current value and the value in register P2.

This instruction throws an error if the memory cell is not initially an integer.

Move

Move the values in register P1..P1+P3-1 over into registers P2..P2+P3-1. Registers P1..P1+P1-1 are left holding a NULL. It is an error for register ranges P1..P1+P3-1 and P2..P2+P3-1 to overlap.

Multiply

Multiply the value in register P1 by the value in register P2 and store the result in register P3. If either input is NULL, the result is NULL.

MustBeInt

Force the value in register P1 to be an integer. If the value in P1 is not an integer and cannot be converted into an integer without data loss, then jump immediately to P2, or if P2==0 raise an SQLITE_MISMATCH exception.

Ne

This works just like the Lt opcode except that the jump is taken if the operands in registers P1 and P3 are not equal. See the Lt opcode for additional information.

If SQLITE_NULLEQ is set in P5 then the result of comparison is always either true or false and is never NULL. If both operands are NULL then the result of comparison is false. If either operand is NULL then the result is true. If neither operand is NULL the the result is the same as it would be if the SQLITE_NULLEQ flag were omitted from P5.

NewRowid

Get a new integer record number (a.k.a "rowid") used as the key to a table. The record number is not previously used as a key in the database table that cursor P1 points to. The new record number is written written to register P2.

If P3>0 then P3 is a register in the root frame of this VDBE that holds the largest previously generated record number. No new record numbers are allowed to be less than this value. When this value reaches its maximum, a SQLITE_FULL error is generated. The P3 register is updated with the ' generated record number. This P3 mechanism is used to help implement the AUTOINCREMENT feature.

Next

Advance cursor P1 so that it points to the next key/data pair in its table or index. If there are no more key/value pairs then fall through to the following instruction. But if the cursor advance was successful, jump immediately to P2.

The P1 cursor must be for a real table, not a pseudo-table.

If P5 is positive and the jump is taken, then event counter number P5-1 in the prepared statement is incremented.

See also: Prev

Noop

Do nothing. This instruction is often useful as a jump destination.

Not

Interpret the value in register P1 as a boolean value. Store the boolean complement in register P2. If the value in register P1 is NULL, then a NULL is stored in P2.

NotExists

Use the content of register P3 as a integer key. If a record with that key does not exist in table of P1, then jump to P2. If the record does exist, then fall thru. The cursor is left pointing to the record if it exists.

The difference between this operation and NotFound is that this operation assumes the key is an integer and that P1 is a table whereas NotFound assumes key is a blob constructed from MakeRecord and P1 is an index.

See also: Found, NotFound, IsUnique

NotFound

If P4==0 then register P3 holds a blob constructed by MakeRecord. If P4>0 then register P3 is the first of P4 registers that form an unpacked record.

Cursor P1 is on an index btree. If the record identified by P3 and P4 is not the prefix of any entry in P1 then a jump is made to P2. If P1 does contain an entry whose prefix matches the P3/P4 record then control falls through to the next instruction and P1 is left pointing at the matching entry.

See also: Found, NotExists, IsUnique

NotNull

Jump to P2 if the value in register P1 is not NULL.

Null

Write a NULL into register P2.

NullRow

Move the cursor P1 to a null row. Any OP_Column operations that occur while the cursor is on the null row will always write a NULL.

OpenAutoindex

This opcode works the same as OP_OpenEphemeral. It has a different name to distinguish its use. Tables created using by this opcode will be used for automatically created transient indices in joins.

OpenEphemeral

Open a new cursor P1 to a transient table. The cursor is always opened read/write even if the main database is read-only. The ephemeral table is deleted automatically when the cursor is closed.

P2 is the number of columns in the ephemeral table. The cursor points to a BTree table if P4==0 and to a BTree index if P4 is not 0. If P4 is not NULL, it points to a KeyInfo structure that defines the format of keys in the index.

This opcode was once called OpenTemp. But that created confusion because the term "temp table", might refer either to a TEMP table at the SQL level, or to a table opened by this opcode. Then this opcode was call OpenVirtual. But that created confusion with the whole virtual-table idea.

OpenPseudo

Open a new cursor that points to a fake table that contains a single row of data. The content of that one row in the content of memory register P2. In other words, cursor P1 becomes an alias for the MEM_Blob content contained in register P2.

A pseudo-table created by this opcode is used to hold a single row output from the sorter so that the row can be decomposed into individual columns using the OP_Column opcode. The OP_Column opcode is the only cursor opcode that works with a pseudo-table.

P3 is the number of fields in the records that will be stored by the pseudo-table.

OpenRead

Open a read-only cursor for the database table whose root page is P2 in a database file. The database file is determined by P3. P3==0 means the main database, P3==1 means the database used for temporary tables, and P3>1 means used the corresponding attached database. Give the new cursor an identifier of P1. The P1 values need not be contiguous but all P1 values should be small integers. It is an error for P1 to be negative.

If P5!=0 then use the content of register P2 as the root page, not the value of P2 itself.

There will be a read lock on the database whenever there is an open cursor. If the database was unlocked prior to this instruction then a read lock is acquired as part of this instruction. A read lock allows other processes to read the database but prohibits any other process from modifying the database. The read lock is released when all cursors are closed. If this instruction attempts to get a read lock but fails, the script terminates with an SQLITE_BUSY error code.

The P4 value may be either an integer (P4_INT32) or a pointer to a KeyInfo structure (P4_KEYINFO). If it is a pointer to a KeyInfo structure, then said structure defines the content and collating sequence of the index being opened. Otherwise, if P4 is an integer value, it is set to the number of columns in the table.

See also OpenWrite.

OpenWrite

Open a read/write cursor named P1 on the table or index whose root page is P2. Or if P5!=0 use the content of register P2 to find the root page.

The P4 value may be either an integer (P4_INT32) or a pointer to a KeyInfo structure (P4_KEYINFO). If it is a pointer to a KeyInfo structure, then said structure defines the content and collating sequence of the index being opened. Otherwise, if P4 is an integer value, it is set to the number of columns in the table, or to the largest index of any column of the table that is actually used.

This instruction works just like OpenRead except that it opens the cursor in read/write mode. For a given table, there can be one or more read-only cursors or a single read/write cursor but not both.

See also OpenRead.

Or

Take the logical OR of the values in register P1 and P2 and store the answer in register P3.

If either P1 or P2 is nonzero (true) then the result is 1 (true) even if the other input is NULL. A NULL and false or two NULLs give a NULL output.

Pagecount

Write the current number of pages in database P1 to memory cell P2.

Param

This opcode is only ever present in sub-programs called via the OP_Program instruction. Copy a value currently stored in a memory cell of the calling (parent) frame to cell P2 in the current frames address space. This is used by trigger programs to access the new.* and old.* values.

The address of the cell in the parent frame is determined by adding the value of the P1 argument to the value of the P1 argument to the calling OP_Program instruction.

ParseSchema

Read and parse all entries from the SQLITE_MASTER table of database P1 that match the WHERE clause P4. P2 is the "force" flag. Always do the parsing if P2 is true. If P2 is false, then this routine is a no-op if the schema is not currently loaded. In other words, if P2 is false, the SQLITE_MASTER table is only parsed if the rest of the schema is already loaded into the symbol table.

This opcode invokes the parser to create a new virtual machine, then runs the new virtual machine. It is thus a re-entrant opcode.

Permutation

Set the permutation used by the OP_Compare operator to be the array of integers in P4.

The permutation is only valid until the next OP_Permutation, OP_Compare, OP_Halt, or OP_ResultRow. Typically the OP_Permutation should occur immediately prior to the OP_Compare.

Prev

Back up cursor P1 so that it points to the previous key/data pair in its table or index. If there is no previous key/value pairs then fall through to the following instruction. But if the cursor backup was successful, jump immediately to P2.

The P1 cursor must be for a real table, not a pseudo-table.

If P5 is positive and the jump is taken, then event counter number P5-1 in the prepared statement is incremented.

Program

Execute the trigger program passed as P4 (type P4_SUBPROGRAM).

P1 contains the address of the memory cell that contains the first memory cell in an array of values used as arguments to the sub-program. P2 contains the address to jump to if the sub-program throws an IGNORE exception using the RAISE() function. Register P3 contains the address of a memory cell in this (the parent) VM that is used to allocate the memory required by the sub-vdbe at runtime.

P4 is a pointer to the VM containing the trigger program.

ReadCookie

Read cookie number P3 from database P1 and write it into register P2. P3==1 is the schema version. P3==2 is the database format. P3==3 is the recommended pager cache size, and so forth. P1==0 is the main database file and P1==1 is the database file used to store temporary tables.

There must be a read-lock on the database (either a transaction must be started or there must be an open cursor) before executing this instruction.

Real

P4 is a pointer to a 64-bit floating point value. Write that value into register P2.

RealAffinity

If register P1 holds an integer convert it to a real value.

This opcode is used when extracting information from a column that has REAL affinity. Such column values may still be stored as integers, for space efficiency, but after extraction we want them to have only a real value.

Remainder

Compute the remainder after integer division of the value in register P1 by the value in register P2 and store the result in P3. If the value in register P2 is zero the result is NULL. If either operand is NULL, the result is NULL.

ResetCount

The value of the change counter is copied to the database handle change counter (returned by subsequent calls to sqlite3_changes()). Then the VMs internal change counter resets to 0. This is used by trigger programs.

ResultRow

The registers P1 through P1+P2-1 contain a single row of results. This opcode causes the sqlite3_step() call to terminate with an SQLITE_ROW return code and it sets up the sqlite3_stmt structure to provide access to the top P1 values as the result row.

Return

Jump to the next instruction after the address in register P1.

Rewind

The next use of the Rowid or Column or Next instruction for P1 will refer to the first entry in the database table or index. If the table or index is empty and P2>0, then jump immediately to P2. If P2 is 0 or if the table or index is not empty, fall through to the following instruction.

RowData

Write into register P2 the complete row data for cursor P1. There is no interpretation of the data. It is just copied onto the P2 register exactly as it is found in the database file.

If the P1 cursor must be pointing to a valid row (not a NULL row) of a real table, not a pseudo-table.

Rowid

Store in register P2 an integer which is the key of the table entry that P1 is currently point to.

P1 can be either an ordinary table or a virtual table. There used to be a separate OP_VRowid opcode for use with virtual tables, but this one opcode now works for both table types.

RowKey

Write into register P2 the complete row key for cursor P1. There is no interpretation of the data. The key is copied onto the P3 register exactly as it is found in the database file.

If the P1 cursor must be pointing to a valid row (not a NULL row) of a real table, not a pseudo-table.

RowSetAdd

Insert the integer value held by register P2 into a boolean index held in register P1.

An assertion fails if P2 is not an integer.

RowSetRead

Extract the smallest value from boolean index P1 and put that value into register P3. Or, if boolean index P1 is initially empty, leave P3 unchanged and jump to instruction P2.

RowSetTest

Register P3 is assumed to hold a 64-bit integer value. If register P1 contains a RowSet object and that RowSet object contains the value held in P3, jump to register P2. Otherwise, insert the integer in P3 into the RowSet and continue on to the next opcode.

The RowSet object is optimized for the case where successive sets of integers, where each set contains no duplicates. Each set of values is identified by a unique P4 value. The first set must have P4==0, the final set P4=-1. P4 must be either -1 or non-negative. For non-negative values of P4 only the lower 4 bits are significant.

This allows optimizations: (a) when P4==0 there is no need to test the rowset object for P3, as it is guaranteed not to contain it, (b) when P4==-1 there is no need to insert the value, as it will never be tested for, and (c) when a value that is part of set X is inserted, there is no need to search to see if the same value was previously inserted as part of set X (only if it was previously inserted as part of some other set).

Savepoint

Open, release or rollback the savepoint named by parameter P4, depending on the value of P1. To open a new savepoint, P1==0. To release (commit) an existing savepoint, P1==1, or to rollback an existing savepoint P1==2.

SCopy

Make a shallow copy of register P1 into register P2.

This instruction makes a shallow copy of the value. If the value is a string or blob, then the copy is only a pointer to the original and hence if the original changes so will the copy. Worse, if the original is deallocated, the copy becomes invalid. Thus the program must guarantee that the original will not change during the lifetime of the copy. Use OP_Copy to make a complete copy.

Seek

P1 is an open table cursor and P2 is a rowid integer. Arrange for P1 to move so that it points to the rowid given by P2.

This is actually a deferred seek. Nothing actually happens until the cursor is used to read a record. That way, if no reads occur, no unnecessary I/O happens.

SeekGe

If cursor P1 refers to an SQL table (B-Tree that uses integer keys), use the value in register P3 as the key. If cursor P1 refers to an SQL index, then P3 is the first in an array of P4 registers that are used as an unpacked index key.

Reposition cursor P1 so that it points to the smallest entry that is greater than or equal to the key value. If there are no records greater than or equal to the key and P2 is not zero, then jump to P2.

See also: Found, NotFound, Distinct, SeekLt, SeekGt, SeekLe

SeekGt

If cursor P1 refers to an SQL table (B-Tree that uses integer keys), use the value in register P3 as a key. If cursor P1 refers to an SQL index, then P3 is the first in an array of P4 registers that are used as an unpacked index key.

Reposition cursor P1 so that it points to the smallest entry that is greater than the key value. If there are no records greater than the key and P2 is not zero, then jump to P2.

See also: Found, NotFound, Distinct, SeekLt, SeekGe, SeekLe

SeekLe

If cursor P1 refers to an SQL table (B-Tree that uses integer keys), use the value in register P3 as a key. If cursor P1 refers to an SQL index, then P3 is the first in an array of P4 registers that are used as an unpacked index key.

Reposition cursor P1 so that it points to the largest entry that is less than or equal to the key value. If there are no records less than or equal to the key and P2 is not zero, then jump to P2.

See also: Found, NotFound, Distinct, SeekGt, SeekGe, SeekLt

SeekLt

If cursor P1 refers to an SQL table (B-Tree that uses integer keys), use the value in register P3 as a key. If cursor P1 refers to an SQL index, then P3 is the first in an array of P4 registers that are used as an unpacked index key.

Reposition cursor P1 so that it points to the largest entry that is less than the key value. If there are no records less than the key and P2 is not zero, then jump to P2.

See also: Found, NotFound, Distinct, SeekGt, SeekGe, SeekLe

Sequence

Find the next available sequence number for cursor P1. Write the sequence number into register P2. The sequence number on the cursor is incremented after this instruction.

SetCookie

Write the content of register P3 (interpreted as an integer) into cookie number P2 of database P1. P2==1 is the schema version. P2==2 is the database format. P2==3 is the recommended pager cache size, and so forth. P1==0 is the main database file and P1==1 is the database file used to store temporary tables.

A transaction must be started before executing this opcode.

ShiftLeft

Shift the integer value in register P2 to the left by the number of bits specified by the integer in regiser P1. Store the result in register P3. If either input is NULL, the result is NULL.

ShiftRight

Shift the integer value in register P2 to the right by the number of bits specified by the integer in register P1. Store the result in register P3. If either input is NULL, the result is NULL.

Sort

This opcode does exactly the same thing as OP_Rewind except that it increments an undocumented global variable used for testing.

Sorting is accomplished by writing records into a sorting index, then rewinding that index and playing it back from beginning to end. We use the OP_Sort opcode instead of OP_Rewind to do the rewinding so that the global variable will be incremented and regression tests can determine whether or not the optimizer is correctly optimizing out sorts.

String

The string value P4 of length P1 (bytes) is stored in register P2.

String8

P4 points to a nul terminated UTF-8 string. This opcode is transformed into an OP_String before it is executed for the first time.

Subtract

Subtract the value in register P1 from the value in register P2 and store the result in register P3. If either input is NULL, the result is NULL.

TableLock

Obtain a lock on a particular table. This instruction is only used when the shared-cache feature is enabled.

P1 is the index of the database in sqlite3.aDb[] of the database on which the lock is acquired. A readlock is obtained if P3==0 or a write lock if P3==1.

P2 contains the root-page of the table to lock.

P4 contains a pointer to the name of the table being locked. This is only used to generate an error message if the lock cannot be obtained.

ToBlob

Force the value in register P1 to be a BLOB. If the value is numeric, convert it to a string first. Strings are simply reinterpreted as blobs with no change to the underlying data.

A NULL value is not changed by this routine. It remains NULL.

ToInt

Force the value in register P1 be an integer. If The value is currently a real number, drop its fractional part. If the value is text or blob, try to convert it to an integer using the equivalent of atoi() and store 0 if no such conversion is possible.

A NULL value is not changed by this routine. It remains NULL.

ToNumeric

Force the value in register P1 to be numeric (either an integer or a floating-point number.) If the value is text or blob, try to convert it to an using the equivalent of atoi() or atof() and store 0 if no such conversion is possible.

A NULL value is not changed by this routine. It remains NULL.

ToReal

Force the value in register P1 to be a floating point number. If The value is currently an integer, convert it. If the value is text or blob, try to convert it to an integer using the equivalent of atoi() and store 0.0 if no such conversion is possible.

A NULL value is not changed by this routine. It remains NULL.

ToText

Force the value in register P1 to be text. If the value is numeric, convert it to a string using the equivalent of printf(). Blob values are unchanged and are afterwards simply interpreted as text.

A NULL value is not changed by this routine. It remains NULL.

Trace

If tracing is enabled (by the sqlite3_trace()) interface, then the UTF-8 string contained in P4 is emitted on the trace callback.

Transaction

Begin a transaction. The transaction ends when a Commit or Rollback opcode is encountered. Depending on the ON CONFLICT setting, the transaction might also be rolled back if an error is encountered.

P1 is the index of the database file on which the transaction is started. Index 0 is the main database file and index 1 is the file used for temporary tables. Indices of 2 or more are used for attached databases.

If P2 is non-zero, then a write-transaction is started. A RESERVED lock is obtained on the database file when a write-transaction is started. No other process can start another write transaction while this transaction is underway. Starting a write transaction also creates a rollback journal. A write transaction must be started before any changes can be made to the database. If P2 is 2 or greater then an EXCLUSIVE lock is also obtained on the file.

If a write-transaction is started and the Vdbe.usesStmtJournal flag is true (this flag is set if the Vdbe may modify more than one row and may throw an ABORT exception), a statement transaction may also be opened. More specifically, a statement transaction is opened iff the database connection is currently not in autocommit mode, or if there are other active statements. A statement transaction allows the affects of this VDBE to be rolled back after an error without having to roll back the entire transaction. If no error is encountered, the statement transaction will automatically commit when the VDBE halts.

If P2 is zero, then a read-lock is obtained on the database file.

Vacuum

Vacuum the entire database. This opcode will cause other virtual machines to be created and run. It may not be called from within a transaction.

Variable

Transfer the values of bound parameter P1 into register P2

If the parameter is named, then its name appears in P4 and P3==1. The P4 value is used by sqlite3_bind_parameter_name().

VBegin

P4 may be a pointer to an sqlite3_vtab structure. If so, call the xBegin method for that table.

Also, whether or not P4 is set, check that this is not being called from within a callback to a virtual table xSync() method. If it is, the error code will be set to SQLITE_LOCKED.

VColumn

Store the value of the P2-th column of the row of the virtual-table that the P1 cursor is pointing to into register P3.

VCreate

P4 is the name of a virtual table in database P1. Call the xCreate method for that table.

VDestroy

P4 is the name of a virtual table in database P1. Call the xDestroy method of that table.

VerifyCookie

Check the value of global database parameter number 0 (the schema version) and make sure it is equal to P2. P1 is the database number which is 0 for the main database file and 1 for the file holding temporary tables and some higher number for auxiliary databases.

The cookie changes its value whenever the database schema changes. This operation is used to detect when that the cookie has changed and that the current process needs to reread the schema.

Either a transaction needs to have been started or an OP_Open needs to be executed (to establish a read lock) before this opcode is invoked.

VFilter

P1 is a cursor opened using VOpen. P2 is an address to jump to if the filtered result set is empty.

P4 is either NULL or a string that was generated by the xBestIndex method of the module. The interpretation of the P4 string is left to the module implementation.

This opcode invokes the xFilter method on the virtual table specified by P1. The integer query plan parameter to xFilter is stored in register P3. Register P3+1 stores the argc parameter to be passed to the xFilter method. Registers P3+2..P3+1+argc are the argc additional parameters which are passed to xFilter as argv. Register P3+2 becomes argv[0] when passed to xFilter.

A jump is made to P2 if the result set after filtering would be empty.

VNext

Advance virtual table P1 to the next row in its result set and jump to instruction P2. Or, if the virtual table has reached the end of its result set, then fall through to the next instruction.

VOpen

P4 is a pointer to a virtual table object, an sqlite3_vtab structure. P1 is a cursor number. This opcode opens a cursor to the virtual table and stores that cursor in P1.

VRename

P4 is a pointer to a virtual table object, an sqlite3_vtab structure. This opcode invokes the corresponding xRename method. The value in register P1 is passed as the zName argument to the xRename method.

VUpdate

P4 is a pointer to a virtual table object, an sqlite3_vtab structure. This opcode invokes the corresponding xUpdate method. P2 values are contiguous memory cells starting at P3 to pass to the xUpdate invocation. The value in register (P3+P2-1) corresponds to the p2th element of the argv array passed to xUpdate.

The xUpdate method will do a DELETE or an INSERT or both. The argv[0] element (which corresponds to memory cell P3) is the rowid of a row to delete. If argv[0] is NULL then no deletion occurs. The argv[1] element is the rowid of the new row. This can be NULL to have the virtual table select the new rowid for itself. The subsequent elements in the array are the values of columns in the new row.

If P2==1 then no insert is performed. argv[0] is the rowid of a row to delete.

P1 is a boolean flag. If it is set to true and the xUpdate call is successful, then the value returned by sqlite3_last_insert_rowid() is set to the value of the rowid for the row just inserted.

Yield

Swap the program counter with the value in register P1.

*** DRAFT ***