Google Git
Sign in
chromium / chromium / src / 29dcc87a4d431025a5c8a6a4b5498172804cc9f9 / . / third_party / sqlite / src / ext / fts3 / fts3.c
blob: d11572ac4cc57d2ddae7087a7e9eed71d93fdd68 [file] [log] [blame]
/*
** 2006 Oct 10
**
** The author disclaims copyright to this source code. In place of
** a legal notice, here is a blessing:
**
** May you do good and not evil.
** May you find forgiveness for yourself and forgive others.
** May you share freely, never taking more than you give.
**
******************************************************************************
**
** This is an SQLite module implementing full-text search.
*/
/*
** The code in this file is only compiled if:
**
** * The FTS3 module is being built as an extension
** (in which case SQLITE_CORE is not defined), or
**
** * The FTS3 module is being built into the core of
** SQLite (in which case SQLITE_ENABLE_FTS3 is defined).
*/
/* The full-text index is stored in a series of b+tree (-like)
** structures called segments which map terms to doclists. The
** structures are like b+trees in layout, but are constructed from the
** bottom up in optimal fashion and are not updatable. Since trees
** are built from the bottom up, things will be described from the
** bottom up.
**
**
**** Varints ****
** The basic unit of encoding is a variable-length integer called a
** varint. We encode variable-length integers in little-endian order
** using seven bits * per byte as follows:
**
** KEY:
** A = 0xxxxxxx 7 bits of data and one flag bit
** B = 1xxxxxxx 7 bits of data and one flag bit
**
** 7 bits - A
** 14 bits - BA
** 21 bits - BBA
** and so on.
**
** This is similar in concept to how sqlite encodes "varints" but
** the encoding is not the same. SQLite varints are big-endian
** are are limited to 9 bytes in length whereas FTS3 varints are
** little-endian and can be up to 10 bytes in length (in theory).
**
** Example encodings:
**
** 1: 0x01
** 127: 0x7f
** 128: 0x81 0x00
**
**
**** Document lists ****
** A doclist (document list) holds a docid-sorted list of hits for a
** given term. Doclists hold docids and associated token positions.
** A docid is the unique integer identifier for a single document.
** A position is the index of a word within the document. The first
** word of the document has a position of 0.
**
** FTS3 used to optionally store character offsets using a compile-time
** option. But that functionality is no longer supported.
**
** A doclist is stored like this:
**
** array {
** varint docid;
** array { (position list for column 0)
** varint position; (2 more than the delta from previous position)
** }
** array {
** varint POS_COLUMN; (marks start of position list for new column)
** varint column; (index of new column)
** array {
** varint position; (2 more than the delta from previous position)
** }
** }
** varint POS_END; (marks end of positions for this document.
** }
**
** Here, array { X } means zero or more occurrences of X, adjacent in
** memory. A "position" is an index of a token in the token stream
** generated by the tokenizer. Note that POS_END and POS_COLUMN occur
** in the same logical place as the position element, and act as sentinals
** ending a position list array. POS_END is 0. POS_COLUMN is 1.
** The positions numbers are not stored literally but rather as two more
** than the difference from the prior position, or the just the position plus
** 2 for the first position. Example:
**
** label: A B C D E F G H I J K
** value: 123 5 9 1 1 14 35 0 234 72 0
**
** The 123 value is the first docid. For column zero in this document
** there are two matches at positions 3 and 10 (5-2 and 9-2+3). The 1
** at D signals the start of a new column; the 1 at E indicates that the
** new column is column number 1. There are two positions at 12 and 45
** (14-2 and 35-2+12). The 0 at H indicate the end-of-document. The
** 234 at I is the next docid. It has one position 72 (72-2) and then
** terminates with the 0 at K.
**
** A "position-list" is the list of positions for multiple columns for
** a single docid. A "column-list" is the set of positions for a single
** column. Hence, a position-list consists of one or more column-lists,
** a document record consists of a docid followed by a position-list and
** a doclist consists of one or more document records.
**
** A bare doclist omits the position information, becoming an
** array of varint-encoded docids.
**
**** Segment leaf nodes ****
** Segment leaf nodes store terms and doclists, ordered by term. Leaf
** nodes are written using LeafWriter, and read using LeafReader (to
** iterate through a single leaf node's data) and LeavesReader (to
** iterate through a segment's entire leaf layer). Leaf nodes have
** the format:
**
** varint iHeight; (height from leaf level, always 0)
** varint nTerm; (length of first term)
** char pTerm[nTerm]; (content of first term)
** varint nDoclist; (length of term's associated doclist)
** char pDoclist[nDoclist]; (content of doclist)
** array {
** (further terms are delta-encoded)
** varint nPrefix; (length of prefix shared with previous term)
** varint nSuffix; (length of unshared suffix)
** char pTermSuffix[nSuffix];(unshared suffix of next term)
** varint nDoclist; (length of term's associated doclist)
** char pDoclist[nDoclist]; (content of doclist)
** }
**
** Here, array { X } means zero or more occurrences of X, adjacent in
** memory.
**
** Leaf nodes are broken into blocks which are stored contiguously in
** the %_segments table in sorted order. This means that when the end
** of a node is reached, the next term is in the node with the next
** greater node id.
**
** New data is spilled to a new leaf node when the current node
** exceeds LEAF_MAX bytes (default 2048). New data which itself is
** larger than STANDALONE_MIN (default 1024) is placed in a standalone
** node (a leaf node with a single term and doclist). The goal of
** these settings is to pack together groups of small doclists while
** making it efficient to directly access large doclists. The
** assumption is that large doclists represent terms which are more
** likely to be query targets.
**
** TODO(shess) It may be useful for blocking decisions to be more
** dynamic. For instance, it may make more sense to have a 2.5k leaf
** node rather than splitting into 2k and .5k nodes. My intuition is
** that this might extend through 2x or 4x the pagesize.
**
**
**** Segment interior nodes ****
** Segment interior nodes store blockids for subtree nodes and terms
** to describe what data is stored by the each subtree. Interior
** nodes are written using InteriorWriter, and read using
** InteriorReader. InteriorWriters are created as needed when
** SegmentWriter creates new leaf nodes, or when an interior node
** itself grows too big and must be split. The format of interior
** nodes:
**
** varint iHeight; (height from leaf level, always >0)
** varint iBlockid; (block id of node's leftmost subtree)
** optional {
** varint nTerm; (length of first term)
** char pTerm[nTerm]; (content of first term)
** array {
** (further terms are delta-encoded)
** varint nPrefix; (length of shared prefix with previous term)
** varint nSuffix; (length of unshared suffix)
** char pTermSuffix[nSuffix]; (unshared suffix of next term)
** }
** }
**
** Here, optional { X } means an optional element, while array { X }
** means zero or more occurrences of X, adjacent in memory.
**
** An interior node encodes n terms separating n+1 subtrees. The
** subtree blocks are contiguous, so only the first subtree's blockid
** is encoded. The subtree at iBlockid will contain all terms less
** than the first term encoded (or all terms if no term is encoded).
** Otherwise, for terms greater than or equal to pTerm[i] but less
** than pTerm[i+1], the subtree for that term will be rooted at
** iBlockid+i. Interior nodes only store enough term data to
** distinguish adjacent children (if the rightmost term of the left
** child is "something", and the leftmost term of the right child is
** "wicked", only "w" is stored).
**
** New data is spilled to a new interior node at the same height when
** the current node exceeds INTERIOR_MAX bytes (default 2048).
** INTERIOR_MIN_TERMS (default 7) keeps large terms from monopolizing
** interior nodes and making the tree too skinny. The interior nodes
** at a given height are naturally tracked by interior nodes at
** height+1, and so on.
**
**
**** Segment directory ****
** The segment directory in table %_segdir stores meta-information for
** merging and deleting segments, and also the root node of the
** segment's tree.
**
** The root node is the top node of the segment's tree after encoding
** the entire segment, restricted to ROOT_MAX bytes (default 1024).
** This could be either a leaf node or an interior node. If the top
** node requires more than ROOT_MAX bytes, it is flushed to %_segments
** and a new root interior node is generated (which should always fit
** within ROOT_MAX because it only needs space for 2 varints, the
** height and the blockid of the previous root).
**
** The meta-information in the segment directory is:
** level - segment level (see below)
** idx - index within level
** - (level,idx uniquely identify a segment)
** start_block - first leaf node
** leaves_end_block - last leaf node
** end_block - last block (including interior nodes)
** root - contents of root node
**
** If the root node is a leaf node, then start_block,
** leaves_end_block, and end_block are all 0.
**
**
**** Segment merging ****
** To amortize update costs, segments are grouped into levels and
** merged in batches. Each increase in level represents exponentially
** more documents.
**
** New documents (actually, document updates) are tokenized and
** written individually (using LeafWriter) to a level 0 segment, with
** incrementing idx. When idx reaches MERGE_COUNT (default 16), all
** level 0 segments are merged into a single level 1 segment. Level 1
** is populated like level 0, and eventually MERGE_COUNT level 1
** segments are merged to a single level 2 segment (representing
** MERGE_COUNT^2 updates), and so on.
**
** A segment merge traverses all segments at a given level in
** parallel, performing a straightforward sorted merge. Since segment
** leaf nodes are written in to the %_segments table in order, this
** merge traverses the underlying sqlite disk structures efficiently.
** After the merge, all segment blocks from the merged level are
** deleted.
**
** MERGE_COUNT controls how often we merge segments. 16 seems to be
** somewhat of a sweet spot for insertion performance. 32 and 64 show
** very similar performance numbers to 16 on insertion, though they're
** a tiny bit slower (perhaps due to more overhead in merge-time
** sorting). 8 is about 20% slower than 16, 4 about 50% slower than
** 16, 2 about 66% slower than 16.
**
** At query time, high MERGE_COUNT increases the number of segments
** which need to be scanned and merged. For instance, with 100k docs
** inserted:
**
** MERGE_COUNT segments
** 16 25
** 8 12
** 4 10
** 2 6
**
** This appears to have only a moderate impact on queries for very
** frequent terms (which are somewhat dominated by segment merge
** costs), and infrequent and non-existent terms still seem to be fast
** even with many segments.
**
** TODO(shess) That said, it would be nice to have a better query-side
** argument for MERGE_COUNT of 16. Also, it is possible/likely that
** optimizations to things like doclist merging will swing the sweet
** spot around.
**
**
**
**** Handling of deletions and updates ****
** Since we're using a segmented structure, with no docid-oriented
** index into the term index, we clearly cannot simply update the term
** index when a document is deleted or updated. For deletions, we
** write an empty doclist (varint(docid) varint(POS_END)), for updates
** we simply write the new doclist. Segment merges overwrite older
** data for a particular docid with newer data, so deletes or updates
** will eventually overtake the earlier data and knock it out. The
** query logic likewise merges doclists so that newer data knocks out
** older data.
**
** TODO(shess) Provide a VACUUM type operation to clear out all
** deletions and duplications. This would basically be a forced merge
** into a single segment.
*/
#define CHROMIUM_FTS3_CHANGES 1
#if !defined(SQLITE_CORE) || defined(SQLITE_ENABLE_FTS3)
#if defined(SQLITE_ENABLE_FTS3) && !defined(SQLITE_CORE)
# define SQLITE_CORE 1
#endif
#include "fts3Int.h"
#include <assert.h>
#include <stdlib.h>
#include <stddef.h>
#include <stdio.h>
#include <string.h>
#include <stdarg.h>
#include "fts3.h"
#ifndef SQLITE_CORE
# include "sqlite3ext.h"
SQLITE_EXTENSION_INIT1
#endif
/*
** Write a 64-bit variable-length integer to memory starting at p[0].
** The length of data written will be between 1 and FTS3_VARINT_MAX bytes.
** The number of bytes written is returned.
*/
int sqlite3Fts3PutVarint(char *p, sqlite_int64 v){
unsigned char *q = (unsigned char *) p;
sqlite_uint64 vu = v;
do{
*q++ = (unsigned char) ((vu & 0x7f) | 0x80);
vu >>= 7;
}while( vu!=0 );
q[-1] &= 0x7f; /* turn off high bit in final byte */
assert( q - (unsigned char *)p <= FTS3_VARINT_MAX );
return (int) (q - (unsigned char *)p);
}
/*
** Read a 64-bit variable-length integer from memory starting at p[0].
** Return the number of bytes read, or 0 on error.
** The value is stored in *v.
*/
int sqlite3Fts3GetVarint(const char *p, sqlite_int64 *v){
const unsigned char *q = (const unsigned char *) p;
sqlite_uint64 x = 0, y = 1;
while( (*q&0x80)==0x80 && q-(unsigned char *)p<FTS3_VARINT_MAX ){
x += y * (*q++ & 0x7f);
y <<= 7;
}
x += y * (*q++);
*v = (sqlite_int64) x;
return (int) (q - (unsigned char *)p);
}
/*
** Similar to sqlite3Fts3GetVarint(), except that the output is truncated to a
** 32-bit integer before it is returned.
*/
int sqlite3Fts3GetVarint32(const char *p, int *pi){
sqlite_int64 i;
int ret = sqlite3Fts3GetVarint(p, &i);
*pi = (int) i;
return ret;
}
/*
** Return the number of bytes required to encode v as a varint
*/
int sqlite3Fts3VarintLen(sqlite3_uint64 v){
int i = 0;
do{
i++;
v >>= 7;
}while( v!=0 );
return i;
}
/*
** Convert an SQL-style quoted string into a normal string by removing
** the quote characters. The conversion is done in-place. If the
** input does not begin with a quote character, then this routine
** is a no-op.
**
** Examples:
**
** "abc" becomes abc
** 'xyz' becomes xyz
** [pqr] becomes pqr
** `mno` becomes mno
**
*/
void sqlite3Fts3Dequote(char *z){
char quote; /* Quote character (if any ) */
quote = z[0];
if( quote=='[' || quote=='\'' || quote=='"' || quote=='`' ){
int iIn = 1; /* Index of next byte to read from input */
int iOut = 0; /* Index of next byte to write to output */
/* If the first byte was a '[', then the close-quote character is a ']' */
if( quote=='[' ) quote = ']';
while( ALWAYS(z[iIn]) ){
if( z[iIn]==quote ){
if( z[iIn+1]!=quote ) break;
z[iOut++] = quote;
iIn += 2;
}else{
z[iOut++] = z[iIn++];
}
}
z[iOut] = '\0';
}
}
/*
** Read a single varint from the doclist at *pp and advance *pp to point
** to the first byte past the end of the varint. Add the value of the varint
** to *pVal.
*/
static void fts3GetDeltaVarint(char **pp, sqlite3_int64 *pVal){
sqlite3_int64 iVal;
*pp += sqlite3Fts3GetVarint(*pp, &iVal);
*pVal += iVal;
}
/*
** As long as *pp has not reached its end (pEnd), then do the same
** as fts3GetDeltaVarint(): read a single varint and add it to *pVal.
** But if we have reached the end of the varint, just set *pp=0 and
** leave *pVal unchanged.
*/
static void fts3GetDeltaVarint2(char **pp, char *pEnd, sqlite3_int64 *pVal){
if( *pp>=pEnd ){
*pp = 0;
}else{
fts3GetDeltaVarint(pp, pVal);
}
}
/*
** The xDisconnect() virtual table method.
*/
static int fts3DisconnectMethod(sqlite3_vtab *pVtab){
Fts3Table *p = (Fts3Table *)pVtab;
int i;
assert( p->nPendingData==0 );
assert( p->pSegments==0 );
/* Free any prepared statements held */
for(i=0; i<SizeofArray(p->aStmt); i++){
sqlite3_finalize(p->aStmt[i]);
}
sqlite3_free(p->zSegmentsTbl);
sqlite3_free(p->zReadExprlist);
sqlite3_free(p->zWriteExprlist);
/* Invoke the tokenizer destructor to free the tokenizer. */
p->pTokenizer->pModule->xDestroy(p->pTokenizer);
sqlite3_free(p);
return SQLITE_OK;
}
/*
** Construct one or more SQL statements from the format string given
** and then evaluate those statements. The success code is written
** into *pRc.
**
** If *pRc is initially non-zero then this routine is a no-op.
*/
static void fts3DbExec(
int *pRc, /* Success code */
sqlite3 *db, /* Database in which to run SQL */
const char *zFormat, /* Format string for SQL */
... /* Arguments to the format string */
){
va_list ap;
char *zSql;
if( *pRc ) return;
va_start(ap, zFormat);
zSql = sqlite3_vmprintf(zFormat, ap);
va_end(ap);
if( zSql==0 ){
*pRc = SQLITE_NOMEM;
}else{
*pRc = sqlite3_exec(db, zSql, 0, 0, 0);
sqlite3_free(zSql);
}
}
/*
** The xDestroy() virtual table method.
*/
static int fts3DestroyMethod(sqlite3_vtab *pVtab){
int rc = SQLITE_OK; /* Return code */
Fts3Table *p = (Fts3Table *)pVtab;
sqlite3 *db = p->db;
/* Drop the shadow tables */
fts3DbExec(&rc, db, "DROP TABLE IF EXISTS %Q.'%q_content'", p->zDb, p->zName);
fts3DbExec(&rc, db, "DROP TABLE IF EXISTS %Q.'%q_segments'", p->zDb,p->zName);
fts3DbExec(&rc, db, "DROP TABLE IF EXISTS %Q.'%q_segdir'", p->zDb, p->zName);
fts3DbExec(&rc, db, "DROP TABLE IF EXISTS %Q.'%q_docsize'", p->zDb, p->zName);
fts3DbExec(&rc, db, "DROP TABLE IF EXISTS %Q.'%q_stat'", p->zDb, p->zName);
/* If everything has worked, invoke fts3DisconnectMethod() to free the
** memory associated with the Fts3Table structure and return SQLITE_OK.
** Otherwise, return an SQLite error code.
*/
return (rc==SQLITE_OK ? fts3DisconnectMethod(pVtab) : rc);
}
/*
** Invoke sqlite3_declare_vtab() to declare the schema for the FTS3 table
** passed as the first argument. This is done as part of the xConnect()
** and xCreate() methods.
**
** If *pRc is non-zero when this function is called, it is a no-op.
** Otherwise, if an error occurs, an SQLite error code is stored in *pRc
** before returning.
*/
static void fts3DeclareVtab(int *pRc, Fts3Table *p){
if( *pRc==SQLITE_OK ){
int i; /* Iterator variable */
int rc; /* Return code */
char *zSql; /* SQL statement passed to declare_vtab() */
char *zCols; /* List of user defined columns */
/* Create a list of user columns for the virtual table */
zCols = sqlite3_mprintf("%Q, ", p->azColumn[0]);
for(i=1; zCols && i<p->nColumn; i++){
zCols = sqlite3_mprintf("%z%Q, ", zCols, p->azColumn[i]);
}
/* Create the whole "CREATE TABLE" statement to pass to SQLite */
zSql = sqlite3_mprintf(
"CREATE TABLE x(%s %Q HIDDEN, docid HIDDEN)", zCols, p->zName
);
if( !zCols || !zSql ){
rc = SQLITE_NOMEM;
}else{
rc = sqlite3_declare_vtab(p->db, zSql);
}
sqlite3_free(zSql);
sqlite3_free(zCols);
*pRc = rc;
}
}
/*
** Create the backing store tables (%_content, %_segments and %_segdir)
** required by the FTS3 table passed as the only argument. This is done
** as part of the vtab xCreate() method.
**
** If the p->bHasDocsize boolean is true (indicating that this is an
** FTS4 table, not an FTS3 table) then also create the %_docsize and
** %_stat tables required by FTS4.
*/
static int fts3CreateTables(Fts3Table *p){
int rc = SQLITE_OK; /* Return code */
int i; /* Iterator variable */
char *zContentCols; /* Columns of %_content table */
sqlite3 *db = p->db; /* The database connection */
/* Create a list of user columns for the content table */
zContentCols = sqlite3_mprintf("docid INTEGER PRIMARY KEY");
for(i=0; zContentCols && i<p->nColumn; i++){
char *z = p->azColumn[i];
zContentCols = sqlite3_mprintf("%z, 'c%d%q'", zContentCols, i, z);
}
if( zContentCols==0 ) rc = SQLITE_NOMEM;
/* Create the content table */
fts3DbExec(&rc, db,
"CREATE TABLE %Q.'%q_content'(%s)",
p->zDb, p->zName, zContentCols
);
sqlite3_free(zContentCols);
/* Create other tables */
fts3DbExec(&rc, db,
"CREATE TABLE %Q.'%q_segments'(blockid INTEGER PRIMARY KEY, block BLOB);",
p->zDb, p->zName
);
fts3DbExec(&rc, db,
"CREATE TABLE %Q.'%q_segdir'("
"level INTEGER,"
"idx INTEGER,"
"start_block INTEGER,"
"leaves_end_block INTEGER,"
"end_block INTEGER,"
"root BLOB,"
"PRIMARY KEY(level, idx)"
");",
p->zDb, p->zName
);
if( p->bHasDocsize ){
fts3DbExec(&rc, db,
"CREATE TABLE %Q.'%q_docsize'(docid INTEGER PRIMARY KEY, size BLOB);",
p->zDb, p->zName
);
}
if( p->bHasStat ){
fts3DbExec(&rc, db,
"CREATE TABLE %Q.'%q_stat'(id INTEGER PRIMARY KEY, value BLOB);",
p->zDb, p->zName
);
}
return rc;
}
/*
** Store the current database page-size in bytes in p->nPgsz.
**
** If *pRc is non-zero when this function is called, it is a no-op.
** Otherwise, if an error occurs, an SQLite error code is stored in *pRc
** before returning.
*/
static void fts3DatabasePageSize(int *pRc, Fts3Table *p){
if( *pRc==SQLITE_OK ){
int rc; /* Return code */
char *zSql; /* SQL text "PRAGMA %Q.page_size" */
sqlite3_stmt *pStmt; /* Compiled "PRAGMA %Q.page_size" statement */
zSql = sqlite3_mprintf("PRAGMA %Q.page_size", p->zDb);
if( !zSql ){
rc = SQLITE_NOMEM;
}else{
rc = sqlite3_prepare(p->db, zSql, -1, &pStmt, 0);
if( rc==SQLITE_OK ){
sqlite3_step(pStmt);
p->nPgsz = sqlite3_column_int(pStmt, 0);
rc = sqlite3_finalize(pStmt);
}else if( rc==SQLITE_AUTH ){
p->nPgsz = 1024;
rc = SQLITE_OK;
}
}
assert( p->nPgsz>0 || rc!=SQLITE_OK );
sqlite3_free(zSql);
*pRc = rc;
}
}
/*
** "Special" FTS4 arguments are column specifications of the following form:
**
** <key> = <value>
**
** There may not be whitespace surrounding the "=" character. The <value>
** term may be quoted, but the <key> may not.
*/
static int fts3IsSpecialColumn(
const char *z,
int *pnKey,
char **pzValue
){
char *zValue;
const char *zCsr = z;
while( *zCsr!='=' ){
if( *zCsr=='\0' ) return 0;
zCsr++;
}
*pnKey = (int)(zCsr-z);
zValue = sqlite3_mprintf("%s", &zCsr[1]);
if( zValue ){
sqlite3Fts3Dequote(zValue);
}
*pzValue = zValue;
return 1;
}
/*
** Append the output of a printf() style formatting to an existing string.
*/
static void fts3Appendf(
int *pRc, /* IN/OUT: Error code */
char **pz, /* IN/OUT: Pointer to string buffer */
const char *zFormat, /* Printf format string to append */
... /* Arguments for printf format string */
){
if( *pRc==SQLITE_OK ){
va_list ap;
char *z;
va_start(ap, zFormat);
z = sqlite3_vmprintf(zFormat, ap);
if( z && *pz ){
char *z2 = sqlite3_mprintf("%s%s", *pz, z);
sqlite3_free(z);
z = z2;
}
if( z==0 ) *pRc = SQLITE_NOMEM;
sqlite3_free(*pz);
*pz = z;
}
}
/*
** Return a copy of input string zInput enclosed in double-quotes (") and
** with all double quote characters escaped. For example:
**
** fts3QuoteId("un \"zip\"") -> "un \"\"zip\"\""
**
** The pointer returned points to memory obtained from sqlite3_malloc(). It
** is the callers responsibility to call sqlite3_free() to release this
** memory.
*/
static char *fts3QuoteId(char const *zInput){
int nRet;
char *zRet;
nRet = 2 + strlen(zInput)*2 + 1;
zRet = sqlite3_malloc(nRet);
if( zRet ){
int i;
char *z = zRet;
*(z++) = '"';
for(i=0; zInput[i]; i++){
if( zInput[i]=='"' ) *(z++) = '"';
*(z++) = zInput[i];
}
*(z++) = '"';
*(z++) = '\0';
}
return zRet;
}
/*
** Return a list of comma separated SQL expressions that could be used
** in a SELECT statement such as the following:
**
** SELECT <list of expressions> FROM %_content AS x ...
**
** to return the docid, followed by each column of text data in order
** from left to write. If parameter zFunc is not NULL, then instead of
** being returned directly each column of text data is passed to an SQL
** function named zFunc first. For example, if zFunc is "unzip" and the
** table has the three user-defined columns "a", "b", and "c", the following
** string is returned:
**
** "docid, unzip(x.'a'), unzip(x.'b'), unzip(x.'c')"
**
** The pointer returned points to a buffer allocated by sqlite3_malloc(). It
** is the responsibility of the caller to eventually free it.
**
** If *pRc is not SQLITE_OK when this function is called, it is a no-op (and
** a NULL pointer is returned). Otherwise, if an OOM error is encountered
** by this function, NULL is returned and *pRc is set to SQLITE_NOMEM. If
** no error occurs, *pRc is left unmodified.
*/
static char *fts3ReadExprList(Fts3Table *p, const char *zFunc, int *pRc){
char *zRet = 0;
char *zFree = 0;
char *zFunction;
int i;
if( !zFunc ){
zFunction = "";
}else{
zFree = zFunction = fts3QuoteId(zFunc);
}
fts3Appendf(pRc, &zRet, "docid");
for(i=0; i<p->nColumn; i++){
fts3Appendf(pRc, &zRet, ",%s(x.'c%d%q')", zFunction, i, p->azColumn[i]);
}
sqlite3_free(zFree);
return zRet;
}
/*
** Return a list of N comma separated question marks, where N is the number
** of columns in the %_content table (one for the docid plus one for each
** user-defined text column).
**
** If argument zFunc is not NULL, then all but the first question mark
** is preceded by zFunc and an open bracket, and followed by a closed
** bracket. For example, if zFunc is "zip" and the FTS3 table has three
** user-defined text columns, the following string is returned:
**
** "?, zip(?), zip(?), zip(?)"
**
** The pointer returned points to a buffer allocated by sqlite3_malloc(). It
** is the responsibility of the caller to eventually free it.
**
** If *pRc is not SQLITE_OK when this function is called, it is a no-op (and
** a NULL pointer is returned). Otherwise, if an OOM error is encountered
** by this function, NULL is returned and *pRc is set to SQLITE_NOMEM. If
** no error occurs, *pRc is left unmodified.
*/
static char *fts3WriteExprList(Fts3Table *p, const char *zFunc, int *pRc){
char *zRet = 0;
char *zFree = 0;
char *zFunction;
int i;
if( !zFunc ){
zFunction = "";
}else{
zFree = zFunction = fts3QuoteId(zFunc);
}
fts3Appendf(pRc, &zRet, "?");
for(i=0; i<p->nColumn; i++){
fts3Appendf(pRc, &zRet, ",%s(?)", zFunction);
}
sqlite3_free(zFree);
return zRet;
}
/*
** This function is the implementation of both the xConnect and xCreate
** methods of the FTS3 virtual table.
**
** The argv[] array contains the following:
**
** argv[0] -> module name ("fts3" or "fts4")
** argv[1] -> database name
** argv[2] -> table name
** argv[...] -> "column name" and other module argument fields.
*/
static int fts3InitVtab(
int isCreate, /* True for xCreate, false for xConnect */
sqlite3 *db, /* The SQLite database connection */
void *pAux, /* Hash table containing tokenizers */
int argc, /* Number of elements in argv array */
const char * const *argv, /* xCreate/xConnect argument array */
sqlite3_vtab **ppVTab, /* Write the resulting vtab structure here */
char **pzErr /* Write any error message here */
){
Fts3Hash *pHash = (Fts3Hash *)pAux;
Fts3Table *p = 0; /* Pointer to allocated vtab */
int rc = SQLITE_OK; /* Return code */
int i; /* Iterator variable */
int nByte; /* Size of allocation used for *p */
int iCol; /* Column index */
int nString = 0; /* Bytes required to hold all column names */
int nCol = 0; /* Number of columns in the FTS table */
char *zCsr; /* Space for holding column names */
int nDb; /* Bytes required to hold database name */
int nName; /* Bytes required to hold table name */
int isFts4 = (argv[0][3]=='4'); /* True for FTS4, false for FTS3 */
int bNoDocsize = 0; /* True to omit %_docsize table */
const char **aCol; /* Array of column names */
sqlite3_tokenizer *pTokenizer = 0; /* Tokenizer for this table */
char *zCompress = 0;
char *zUncompress = 0;
assert( strlen(argv[0])==4 );
assert( (sqlite3_strnicmp(argv[0], "fts4", 4)==0 && isFts4)
|| (sqlite3_strnicmp(argv[0], "fts3", 4)==0 && !isFts4)
);
nDb = (int)strlen(argv[1]) + 1;
nName = (int)strlen(argv[2]) + 1;
aCol = (const char **)sqlite3_malloc(sizeof(const char *) * (argc-2) );
if( !aCol ) return SQLITE_NOMEM;
memset((void *)aCol, 0, sizeof(const char *) * (argc-2));
/* Loop through all of the arguments passed by the user to the FTS3/4
** module (i.e. all the column names and special arguments). This loop
** does the following:
**
** + Figures out the number of columns the FTSX table will have, and
** the number of bytes of space that must be allocated to store copies
** of the column names.
**
** + If there is a tokenizer specification included in the arguments,
** initializes the tokenizer pTokenizer.
*/
for(i=3; rc==SQLITE_OK && i<argc; i++){
char const *z = argv[i];
int nKey;
char *zVal;
/* Check if this is a tokenizer specification */
if( !pTokenizer
&& strlen(z)>8
&& 0==sqlite3_strnicmp(z, "tokenize", 8)
&& 0==sqlite3Fts3IsIdChar(z[8])
){
rc = sqlite3Fts3InitTokenizer(pHash, &z[9], &pTokenizer, pzErr);
}
/* Check if it is an FTS4 special argument. */
else if( isFts4 && fts3IsSpecialColumn(z, &nKey, &zVal) ){
if( !zVal ){
rc = SQLITE_NOMEM;
goto fts3_init_out;
}
if( nKey==9 && 0==sqlite3_strnicmp(z, "matchinfo", 9) ){
if( strlen(zVal)==4 && 0==sqlite3_strnicmp(zVal, "fts3", 4) ){
bNoDocsize = 1;
}else{
*pzErr = sqlite3_mprintf("unrecognized matchinfo: %s", zVal);
rc = SQLITE_ERROR;
}
}else if( nKey==8 && 0==sqlite3_strnicmp(z, "compress", 8) ){
zCompress = zVal;
zVal = 0;
}else if( nKey==10 && 0==sqlite3_strnicmp(z, "uncompress", 10) ){
zUncompress = zVal;
zVal = 0;
}else{
*pzErr = sqlite3_mprintf("unrecognized parameter: %s", z);
rc = SQLITE_ERROR;
}
sqlite3_free(zVal);
}
/* Otherwise, the argument is a column name. */
else {
nString += (int)(strlen(z) + 1);
aCol[nCol++] = z;
}
}
if( rc!=SQLITE_OK ) goto fts3_init_out;
if( nCol==0 ){
assert( nString==0 );
aCol[0] = "content";
nString = 8;
nCol = 1;
}
if( pTokenizer==0 ){
rc = sqlite3Fts3InitTokenizer(pHash, "simple", &pTokenizer, pzErr);
if( rc!=SQLITE_OK ) goto fts3_init_out;
}
assert( pTokenizer );
/* Allocate and populate the Fts3Table structure. */
nByte = sizeof(Fts3Table) + /* Fts3Table */
nCol * sizeof(char *) + /* azColumn */
nName + /* zName */
nDb + /* zDb */
nString; /* Space for azColumn strings */
p = (Fts3Table*)sqlite3_malloc(nByte);
if( p==0 ){
rc = SQLITE_NOMEM;
goto fts3_init_out;
}
memset(p, 0, nByte);
p->db = db;
p->nColumn = nCol;
p->nPendingData = 0;
p->azColumn = (char **)&p[1];
p->pTokenizer = pTokenizer;
p->nNodeSize = 1000;
p->nMaxPendingData = FTS3_MAX_PENDING_DATA;
p->bHasDocsize = (isFts4 && bNoDocsize==0);
p->bHasStat = isFts4;
fts3HashInit(&p->pendingTerms, FTS3_HASH_STRING, 1);
/* Fill in the zName and zDb fields of the vtab structure. */
zCsr = (char *)&p->azColumn[nCol];
p->zName = zCsr;
memcpy(zCsr, argv[2], nName);
zCsr += nName;
p->zDb = zCsr;
memcpy(zCsr, argv[1], nDb);
zCsr += nDb;
/* Fill in the azColumn array */
for(iCol=0; iCol<nCol; iCol++){
char *z;
int n;
z = (char *)sqlite3Fts3NextToken(aCol[iCol], &n);
memcpy(zCsr, z, n);
zCsr[n] = '\0';
sqlite3Fts3Dequote(zCsr);
p->azColumn[iCol] = zCsr;
zCsr += n+1;
assert( zCsr <= &((char *)p)[nByte] );
}
if( (zCompress==0)!=(zUncompress==0) ){
char const *zMiss = (zCompress==0 ? "compress" : "uncompress");
rc = SQLITE_ERROR;
*pzErr = sqlite3_mprintf("missing %s parameter in fts4 constructor", zMiss);
}
p->zReadExprlist = fts3ReadExprList(p, zUncompress, &rc);
p->zWriteExprlist = fts3WriteExprList(p, zCompress, &rc);
if( rc!=SQLITE_OK ) goto fts3_init_out;
/* If this is an xCreate call, create the underlying tables in the
** database. TODO: For xConnect(), it could verify that said tables exist.
*/
if( isCreate ){
rc = fts3CreateTables(p);
}
/* Figure out the page-size for the database. This is required in order to
** estimate the cost of loading large doclists from the database (see
** function sqlite3Fts3SegReaderCost() for details).
*/
fts3DatabasePageSize(&rc, p);
/* Declare the table schema to SQLite. */
fts3DeclareVtab(&rc, p);
fts3_init_out:
sqlite3_free(zCompress);
sqlite3_free(zUncompress);
sqlite3_free((void *)aCol);
if( rc!=SQLITE_OK ){
if( p ){
fts3DisconnectMethod((sqlite3_vtab *)p);
}else if( pTokenizer ){
pTokenizer->pModule->xDestroy(pTokenizer);
}
}else{
*ppVTab = &p->base;
}
return rc;
}
/*
** The xConnect() and xCreate() methods for the virtual table. All the
** work is done in function fts3InitVtab().
*/
static int fts3ConnectMethod(
sqlite3 *db, /* Database connection */
void *pAux, /* Pointer to tokenizer hash table */
int argc, /* Number of elements in argv array */
const char * const *argv, /* xCreate/xConnect argument array */
sqlite3_vtab **ppVtab, /* OUT: New sqlite3_vtab object */
char **pzErr /* OUT: sqlite3_malloc'd error message */
){
return fts3InitVtab(0, db, pAux, argc, argv, ppVtab, pzErr);
}
static int fts3CreateMethod(
sqlite3 *db, /* Database connection */
void *pAux, /* Pointer to tokenizer hash table */
int argc, /* Number of elements in argv array */
const char * const *argv, /* xCreate/xConnect argument array */
sqlite3_vtab **ppVtab, /* OUT: New sqlite3_vtab object */
char **pzErr /* OUT: sqlite3_malloc'd error message */
){
return fts3InitVtab(1, db, pAux, argc, argv, ppVtab, pzErr);
}
/*
** Implementation of the xBestIndex method for FTS3 tables. There
** are three possible strategies, in order of preference:
**
** 1. Direct lookup by rowid or docid.
** 2. Full-text search using a MATCH operator on a non-docid column.
** 3. Linear scan of %_content table.
*/
static int fts3BestIndexMethod(sqlite3_vtab *pVTab, sqlite3_index_info *pInfo){
Fts3Table *p = (Fts3Table *)pVTab;
int i; /* Iterator variable */
int iCons = -1; /* Index of constraint to use */
/* By default use a full table scan. This is an expensive option,
** so search through the constraints to see if a more efficient
** strategy is possible.
*/
pInfo->idxNum = FTS3_FULLSCAN_SEARCH;
pInfo->estimatedCost = 500000;
for(i=0; i<pInfo->nConstraint; i++){
struct sqlite3_index_constraint *pCons = &pInfo->aConstraint[i];
if( pCons->usable==0 ) continue;
/* A direct lookup on the rowid or docid column. Assign a cost of 1.0. */
if( pCons->op==SQLITE_INDEX_CONSTRAINT_EQ
&& (pCons->iColumn<0 || pCons->iColumn==p->nColumn+1 )
){
pInfo->idxNum = FTS3_DOCID_SEARCH;
pInfo->estimatedCost = 1.0;
iCons = i;
}
/* A MATCH constraint. Use a full-text search.
**
** If there is more than one MATCH constraint available, use the first
** one encountered. If there is both a MATCH constraint and a direct
** rowid/docid lookup, prefer the MATCH strategy. This is done even
** though the rowid/docid lookup is faster than a MATCH query, selecting
** it would lead to an "unable to use function MATCH in the requested
** context" error.
*/
if( pCons->op==SQLITE_INDEX_CONSTRAINT_MATCH
&& pCons->iColumn>=0 && pCons->iColumn<=p->nColumn
){
pInfo->idxNum = FTS3_FULLTEXT_SEARCH + pCons->iColumn;
pInfo->estimatedCost = 2.0;
iCons = i;
break;
}
}
if( iCons>=0 ){
pInfo->aConstraintUsage[iCons].argvIndex = 1;
pInfo->aConstraintUsage[iCons].omit = 1;
}
return SQLITE_OK;
}
/*
** Implementation of xOpen method.
*/
static int fts3OpenMethod(sqlite3_vtab *pVTab, sqlite3_vtab_cursor **ppCsr){
sqlite3_vtab_cursor *pCsr; /* Allocated cursor */
UNUSED_PARAMETER(pVTab);
/* Allocate a buffer large enough for an Fts3Cursor structure. If the
** allocation succeeds, zero it and return SQLITE_OK. Otherwise,
** if the allocation fails, return SQLITE_NOMEM.
*/
*ppCsr = pCsr = (sqlite3_vtab_cursor *)sqlite3_malloc(sizeof(Fts3Cursor));
if( !pCsr ){
return SQLITE_NOMEM;
}
memset(pCsr, 0, sizeof(Fts3Cursor));
return SQLITE_OK;
}
/*
** Close the cursor. For additional information see the documentation
** on the xClose method of the virtual table interface.
*/
static int fts3CloseMethod(sqlite3_vtab_cursor *pCursor){
Fts3Cursor *pCsr = (Fts3Cursor *)pCursor;
assert( ((Fts3Table *)pCsr->base.pVtab)->pSegments==0 );
sqlite3_finalize(pCsr->pStmt);
sqlite3Fts3ExprFree(pCsr->pExpr);
sqlite3Fts3FreeDeferredTokens(pCsr);
sqlite3_free(pCsr->aDoclist);
sqlite3_free(pCsr->aMatchinfo);
sqlite3_free(pCsr);
return SQLITE_OK;
}
/*
** Position the pCsr->pStmt statement so that it is on the row
** of the %_content table that contains the last match. Return
** SQLITE_OK on success.
*/
static int fts3CursorSeek(sqlite3_context *pContext, Fts3Cursor *pCsr){
if( pCsr->isRequireSeek ){
pCsr->isRequireSeek = 0;
sqlite3_bind_int64(pCsr->pStmt, 1, pCsr->iPrevId);
if( SQLITE_ROW==sqlite3_step(pCsr->pStmt) ){
return SQLITE_OK;
}else{
int rc = sqlite3_reset(pCsr->pStmt);
if( rc==SQLITE_OK ){
/* If no row was found and no error has occured, then the %_content
** table is missing a row that is present in the full-text index.
** The data structures are corrupt.
*/
rc = SQLITE_CORRUPT;
}
pCsr->isEof = 1;
if( pContext ){
sqlite3_result_error_code(pContext, rc);
}
return rc;
}
}else{
return SQLITE_OK;
}
}
/*
** This function is used to process a single interior node when searching
** a b-tree for a term or term prefix. The node data is passed to this
** function via the zNode/nNode parameters. The term to search for is
** passed in zTerm/nTerm.
**
** If piFirst is not NULL, then this function sets *piFirst to the blockid
** of the child node that heads the sub-tree that may contain the term.
**
** If piLast is not NULL, then *piLast is set to the right-most child node
** that heads a sub-tree that may contain a term for which zTerm/nTerm is
** a prefix.
**
** If an OOM error occurs, SQLITE_NOMEM is returned. Otherwise, SQLITE_OK.
*/
static int fts3ScanInteriorNode(
const char *zTerm, /* Term to select leaves for */
int nTerm, /* Size of term zTerm in bytes */
const char *zNode, /* Buffer containing segment interior node */
int nNode, /* Size of buffer at zNode */
sqlite3_int64 *piFirst, /* OUT: Selected child node */
sqlite3_int64 *piLast /* OUT: Selected child node */
){
int rc = SQLITE_OK; /* Return code */
const char *zCsr = zNode; /* Cursor to iterate through node */
const char *zEnd = &zCsr[nNode];/* End of interior node buffer */
char *zBuffer = 0; /* Buffer to load terms into */
int nAlloc = 0; /* Size of allocated buffer */
int isFirstTerm = 1; /* True when processing first term on page */
sqlite3_int64 iChild; /* Block id of child node to descend to */
/* Skip over the 'height' varint that occurs at the start of every
** interior node. Then load the blockid of the left-child of the b-tree
** node into variable iChild.
**
** Even if the data structure on disk is corrupted, this (reading two
** varints from the buffer) does not risk an overread. If zNode is a
** root node, then the buffer comes from a SELECT statement. SQLite does
** not make this guarantee explicitly, but in practice there are always
** either more than 20 bytes of allocated space following the nNode bytes of
** contents, or two zero bytes. Or, if the node is read from the %_segments
** table, then there are always 20 bytes of zeroed padding following the
** nNode bytes of content (see sqlite3Fts3ReadBlock() for details).
*/
zCsr += sqlite3Fts3GetVarint(zCsr, &iChild);
zCsr += sqlite3Fts3GetVarint(zCsr, &iChild);
if( zCsr>zEnd ){
return SQLITE_CORRUPT;
}
while( zCsr<zEnd && (piFirst || piLast) ){
int cmp; /* memcmp() result */
int nSuffix; /* Size of term suffix */
int nPrefix = 0; /* Size of term prefix */
int nBuffer; /* Total term size */
/* Load the next term on the node into zBuffer. Use realloc() to expand
** the size of zBuffer if required. */
if( !isFirstTerm ){
zCsr += sqlite3Fts3GetVarint32(zCsr, &nPrefix);
}
isFirstTerm = 0;
zCsr += sqlite3Fts3GetVarint32(zCsr, &nSuffix);
/* NOTE(shess): Previous code checked for negative nPrefix and
** nSuffix and suffix overrunning zEnd. Additionally corrupt if
** the prefix is longer than the previous term, or if the suffix
** causes overflow.
*/
if( nPrefix<0 || nSuffix<0 /* || nPrefix>nBuffer */
|| &zCsr[nSuffix]<zCsr || &zCsr[nSuffix]>zEnd ){
rc = SQLITE_CORRUPT;
goto finish_scan;
}
if( nPrefix+nSuffix>nAlloc ){
char *zNew;
nAlloc = (nPrefix+nSuffix) * 2;
zNew = (char *)sqlite3_realloc(zBuffer, nAlloc);
if( !zNew ){
rc = SQLITE_NOMEM;
goto finish_scan;
}
zBuffer = zNew;
}
memcpy(&zBuffer[nPrefix], zCsr, nSuffix);
nBuffer = nPrefix + nSuffix;
zCsr += nSuffix;
/* Compare the term we are searching for with the term just loaded from
** the interior node. If the specified term is greater than or equal
** to the term from the interior node, then all terms on the sub-tree
** headed by node iChild are smaller than zTerm. No need to search
** iChild.
**
** If the interior node term is larger than the specified term, then
** the tree headed by iChild may contain the specified term.
*/
cmp = memcmp(zTerm, zBuffer, (nBuffer>nTerm ? nTerm : nBuffer));
if( piFirst && (cmp<0 || (cmp==0 && nBuffer>nTerm)) ){
*piFirst = iChild;
piFirst = 0;
}
if( piLast && cmp<0 ){
*piLast = iChild;
piLast = 0;
}
iChild++;
};
if( piFirst ) *piFirst = iChild;
if( piLast ) *piLast = iChild;
finish_scan:
sqlite3_free(zBuffer);
return rc;
}
/*
** The buffer pointed to by argument zNode (size nNode bytes) contains an
** interior node of a b-tree segment. The zTerm buffer (size nTerm bytes)
** contains a term. This function searches the sub-tree headed by the zNode
** node for the range of leaf nodes that may contain the specified term
** or terms for which the specified term is a prefix.
**
** If piLeaf is not NULL, then *piLeaf is set to the blockid of the
** left-most leaf node in the tree that may contain the specified term.
** If piLeaf2 is not NULL, then *piLeaf2 is set to the blockid of the
** right-most leaf node that may contain a term for which the specified
** term is a prefix.
**
** It is possible that the range of returned leaf nodes does not contain
** the specified term or any terms for which it is a prefix. However, if the
** segment does contain any such terms, they are stored within the identified
** range. Because this function only inspects interior segment nodes (and
** never loads leaf nodes into memory), it is not possible to be sure.
**
** If an error occurs, an error code other than SQLITE_OK is returned.
*/
static int fts3SelectLeaf(
Fts3Table *p, /* Virtual table handle */
const char *zTerm, /* Term to select leaves for */
int nTerm, /* Size of term zTerm in bytes */
const char *zNode, /* Buffer containing segment interior node */
int nNode, /* Size of buffer at zNode */
sqlite3_int64 *piLeaf, /* Selected leaf node */
sqlite3_int64 *piLeaf2 /* Selected leaf node */
){
int rc; /* Return code */
int iHeight; /* Height of this node in tree */
assert( piLeaf || piLeaf2 );
sqlite3Fts3GetVarint32(zNode, &iHeight);
rc = fts3ScanInteriorNode(zTerm, nTerm, zNode, nNode, piLeaf, piLeaf2);
assert( !piLeaf2 || !piLeaf || rc!=SQLITE_OK || (*piLeaf<=*piLeaf2) );
if( rc==SQLITE_OK && iHeight>1 ){
char *zBlob = 0; /* Blob read from %_segments table */
int nBlob; /* Size of zBlob in bytes */
if( piLeaf && piLeaf2 && (*piLeaf!=*piLeaf2) ){
rc = sqlite3Fts3ReadBlock(p, *piLeaf, &zBlob, &nBlob);
if( rc==SQLITE_OK ){
rc = fts3SelectLeaf(p, zTerm, nTerm, zBlob, nBlob, piLeaf, 0);
}
sqlite3_free(zBlob);
piLeaf = 0;
zBlob = 0;
}
if( rc==SQLITE_OK ){
rc = sqlite3Fts3ReadBlock(p, piLeaf ? *piLeaf : *piLeaf2, &zBlob, &nBlob);
}
if( rc==SQLITE_OK ){
rc = fts3SelectLeaf(p, zTerm, nTerm, zBlob, nBlob, piLeaf, piLeaf2);
}
sqlite3_free(zBlob);
}
return rc;
}
/*
** This function is used to create delta-encoded serialized lists of FTS3
** varints. Each call to this function appends a single varint to a list.
*/
static void fts3PutDeltaVarint(
char **pp, /* IN/OUT: Output pointer */
sqlite3_int64 *piPrev, /* IN/OUT: Previous value written to list */
sqlite3_int64 iVal /* Write this value to the list */
){
assert( iVal-*piPrev > 0 || (*piPrev==0 && iVal==0) );
*pp += sqlite3Fts3PutVarint(*pp, iVal-*piPrev);
*piPrev = iVal;
}
/*
** When this function is called, *ppPoslist is assumed to point to the
** start of a position-list. After it returns, *ppPoslist points to the
** first byte after the position-list.
**
** A position list is list of positions (delta encoded) and columns for
** a single document record of a doclist. So, in other words, this
** routine advances *ppPoslist so that it points to the next docid in
** the doclist, or to the first byte past the end of the doclist.
**
** If pp is not NULL, then the contents of the position list are copied
** to *pp. *pp is set to point to the first byte past the last byte copied
** before this function returns.
*/
static void fts3PoslistCopy(char **pp, char **ppPoslist){
char *pEnd = *ppPoslist;
char c = 0;
/* The end of a position list is marked by a zero encoded as an FTS3
** varint. A single POS_END (0) byte. Except, if the 0 byte is preceded by
** a byte with the 0x80 bit set, then it is not a varint 0, but the tail
** of some other, multi-byte, value.
**
** The following while-loop moves pEnd to point to the first byte that is not
** immediately preceded by a byte with the 0x80 bit set. Then increments
** pEnd once more so that it points to the byte immediately following the
** last byte in the position-list.
*/
while( *pEnd | c ){
c = *pEnd++ & 0x80;
testcase( c!=0 && (*pEnd)==0 );
}
pEnd++; /* Advance past the POS_END terminator byte */
if( pp ){
int n = (int)(pEnd - *ppPoslist);
char *p = *pp;
memcpy(p, *ppPoslist, n);
p += n;
*pp = p;
}
*ppPoslist = pEnd;
}
/*
** When this function is called, *ppPoslist is assumed to point to the
** start of a column-list. After it returns, *ppPoslist points to the
** to the terminator (POS_COLUMN or POS_END) byte of the column-list.
**
** A column-list is list of delta-encoded positions for a single column
** within a single document within a doclist.
**
** The column-list is terminated either by a POS_COLUMN varint (1) or
** a POS_END varint (0). This routine leaves *ppPoslist pointing to
** the POS_COLUMN or POS_END that terminates the column-list.
**
** If pp is not NULL, then the contents of the column-list are copied
** to *pp. *pp is set to point to the first byte past the last byte copied
** before this function returns. The POS_COLUMN or POS_END terminator
** is not copied into *pp.
*/
static void fts3ColumnlistCopy(char **pp, char **ppPoslist){
char *pEnd = *ppPoslist;
char c = 0;
/* A column-list is terminated by either a 0x01 or 0x00 byte that is
** not part of a multi-byte varint.
*/
while( 0xFE & (*pEnd | c) ){
c = *pEnd++ & 0x80;
testcase( c!=0 && ((*pEnd)&0xfe)==0 );
}
if( pp ){
int n = (int)(pEnd - *ppPoslist);
char *p = *pp;
memcpy(p, *ppPoslist, n);
p += n;
*pp = p;
}
*ppPoslist = pEnd;
}
/*
** Value used to signify the end of an position-list. This is safe because
** it is not possible to have a document with 2^31 terms.
*/
#define POSITION_LIST_END 0x7fffffff
/*
** This function is used to help parse position-lists. When this function is
** called, *pp may point to the start of the next varint in the position-list
** being parsed, or it may point to 1 byte past the end of the position-list
** (in which case **pp will be a terminator bytes POS_END (0) or
** (1)).
**
** If *pp points past the end of the current position-list, set *pi to
** POSITION_LIST_END and return. Otherwise, read the next varint from *pp,
** increment the current value of *pi by the value read, and set *pp to
** point to the next value before returning.
**
** Before calling this routine *pi must be initialized to the value of
** the previous position, or zero if we are reading the first position
** in the position-list. Because positions are delta-encoded, the value
** of the previous position is needed in order to compute the value of
** the next position.
*/
static void fts3ReadNextPos(
char **pp, /* IN/OUT: Pointer into position-list buffer */
sqlite3_int64 *pi /* IN/OUT: Value read from position-list */
){
if( (**pp)&0xFE ){
fts3GetDeltaVarint(pp, pi);
*pi -= 2;
}else{
*pi = POSITION_LIST_END;
}
}
/*
** If parameter iCol is not 0, write an POS_COLUMN (1) byte followed by
** the value of iCol encoded as a varint to *pp. This will start a new
** column list.
**
** Set *pp to point to the byte just after the last byte written before
** returning (do not modify it if iCol==0). Return the total number of bytes
** written (0 if iCol==0).
*/
static int fts3PutColNumber(char **pp, int iCol){
int n = 0; /* Number of bytes written */
if( iCol ){
char *p = *pp; /* Output pointer */
n = 1 + sqlite3Fts3PutVarint(&p[1], iCol);
*p = 0x01;
*pp = &p[n];
}
return n;
}
/*
** Compute the union of two position lists. The output written
** into *pp contains all positions of both *pp1 and *pp2 in sorted
** order and with any duplicates removed. All pointers are
** updated appropriately. The caller is responsible for insuring
** that there is enough space in *pp to hold the complete output.
*/
static void fts3PoslistMerge(
char **pp, /* Output buffer */
char **pp1, /* Left input list */
char **pp2 /* Right input list */
){
char *p = *pp;
char *p1 = *pp1;
char *p2 = *pp2;
while( *p1 || *p2 ){
int iCol1; /* The current column index in pp1 */
int iCol2; /* The current column index in pp2 */
if( *p1==POS_COLUMN ) sqlite3Fts3GetVarint32(&p1[1], &iCol1);
else if( *p1==POS_END ) iCol1 = POSITION_LIST_END;
else iCol1 = 0;
if( *p2==POS_COLUMN ) sqlite3Fts3GetVarint32(&p2[1], &iCol2);
else if( *p2==POS_END ) iCol2 = POSITION_LIST_END;
else iCol2 = 0;
if( iCol1==iCol2 ){
sqlite3_int64 i1 = 0; /* Last position from pp1 */
sqlite3_int64 i2 = 0; /* Last position from pp2 */
sqlite3_int64 iPrev = 0;
int n = fts3PutColNumber(&p, iCol1);
p1 += n;
p2 += n;
/* At this point, both p1 and p2 point to the start of column-lists
** for the same column (the column with index iCol1 and iCol2).
** A column-list is a list of non-negative delta-encoded varints, each
** incremented by 2 before being stored. Each list is terminated by a
** POS_END (0) or POS_COLUMN (1). The following block merges the two lists
** and writes the results to buffer p. p is left pointing to the byte
** after the list written. No terminator (POS_END or POS_COLUMN) is
** written to the output.
*/
fts3GetDeltaVarint(&p1, &i1);
fts3GetDeltaVarint(&p2, &i2);
do {
fts3PutDeltaVarint(&p, &iPrev, (i1<i2) ? i1 : i2);
iPrev -= 2;
if( i1==i2 ){
fts3ReadNextPos(&p1, &i1);
fts3ReadNextPos(&p2, &i2);
}else if( i1<i2 ){
fts3ReadNextPos(&p1, &i1);
}else{
fts3ReadNextPos(&p2, &i2);
}
}while( i1!=POSITION_LIST_END || i2!=POSITION_LIST_END );
}else if( iCol1<iCol2 ){
p1 += fts3PutColNumber(&p, iCol1);
fts3ColumnlistCopy(&p, &p1);
}else{
p2 += fts3PutColNumber(&p, iCol2);
fts3ColumnlistCopy(&p, &p2);
}
}
*p++ = POS_END;
*pp = p;
*pp1 = p1 + 1;
*pp2 = p2 + 1;
}
/*
** nToken==1 searches for adjacent positions.
**
** This function is used to merge two position lists into one. When it is
** called, *pp1 and *pp2 must both point to position lists. A position-list is
** the part of a doclist that follows each document id. For example, if a row
** contains:
**
** 'a b c'|'x y z'|'a b b a'
**
** Then the position list for this row for token 'b' would consist of:
**
** 0x02 0x01 0x02 0x03 0x03 0x00
**
** When this function returns, both *pp1 and *pp2 are left pointing to the
** byte following the 0x00 terminator of their respective position lists.
**
** If isSaveLeft is 0, an entry is added to the output position list for
** each position in *pp2 for which there exists one or more positions in
** *pp1 so that (pos(*pp2)>pos(*pp1) && pos(*pp2)-pos(*pp1)<=nToken). i.e.
** when the *pp1 token appears before the *pp2 token, but not more than nToken
** slots before it.
*/
static int fts3PoslistPhraseMerge(
char **pp, /* IN/OUT: Preallocated output buffer */
int nToken, /* Maximum difference in token positions */
int isSaveLeft, /* Save the left position */
int isExact, /* If *pp1 is exactly nTokens before *pp2 */
char **pp1, /* IN/OUT: Left input list */
char **pp2 /* IN/OUT: Right input list */
){
char *p = (pp ? *pp : 0);
char *p1 = *pp1;
char *p2 = *pp2;
int iCol1 = 0;
int iCol2 = 0;
/* Never set both isSaveLeft and isExact for the same invocation. */
assert( isSaveLeft==0 || isExact==0 );
assert( *p1!=0 && *p2!=0 );
if( *p1==POS_COLUMN ){
p1++;
p1 += sqlite3Fts3GetVarint32(p1, &iCol1);
}
if( *p2==POS_COLUMN ){
p2++;
p2 += sqlite3Fts3GetVarint32(p2, &iCol2);
}
while( 1 ){
if( iCol1==iCol2 ){
char *pSave = p;
sqlite3_int64 iPrev = 0;
sqlite3_int64 iPos1 = 0;
sqlite3_int64 iPos2 = 0;
if( pp && iCol1 ){
*p++ = POS_COLUMN;
p += sqlite3Fts3PutVarint(p, iCol1);
}
assert( *p1!=POS_END && *p1!=POS_COLUMN );
assert( *p2!=POS_END && *p2!=POS_COLUMN );
fts3GetDeltaVarint(&p1, &iPos1); iPos1 -= 2;
fts3GetDeltaVarint(&p2, &iPos2); iPos2 -= 2;
while( 1 ){
if( iPos2==iPos1+nToken
|| (isExact==0 && iPos2>iPos1 && iPos2<=iPos1+nToken)
){
sqlite3_int64 iSave;
if( !pp ){
fts3PoslistCopy(0, &p2);
fts3PoslistCopy(0, &p1);
*pp1 = p1;
*pp2 = p2;
return 1;
}
iSave = isSaveLeft ? iPos1 : iPos2;
fts3PutDeltaVarint(&p, &iPrev, iSave+2); iPrev -= 2;
pSave = 0;
}
if( (!isSaveLeft && iPos2<=(iPos1+nToken)) || iPos2<=iPos1 ){
if( (*p2&0xFE)==0 ) break;
fts3GetDeltaVarint(&p2, &iPos2); iPos2 -= 2;
}else{
if( (*p1&0xFE)==0 ) break;
fts3GetDeltaVarint(&p1, &iPos1); iPos1 -= 2;
}
}
if( pSave ){
assert( pp && p );
p = pSave;
}
fts3ColumnlistCopy(0, &p1);
fts3ColumnlistCopy(0, &p2);
assert( (*p1&0xFE)==0 && (*p2&0xFE)==0 );
if( 0==*p1 || 0==*p2 ) break;
p1++;
p1 += sqlite3Fts3GetVarint32(p1, &iCol1);
p2++;
p2 += sqlite3Fts3GetVarint32(p2, &iCol2);
}
/* Advance pointer p1 or p2 (whichever corresponds to the smaller of
** iCol1 and iCol2) so that it points to either the 0x00 that marks the
** end of the position list, or the 0x01 that precedes the next
** column-number in the position list.
*/
else if( iCol1<iCol2 ){
fts3ColumnlistCopy(0, &p1);
if( 0==*p1 ) break;
p1++;
p1 += sqlite3Fts3GetVarint32(p1, &iCol1);
}else{
fts3ColumnlistCopy(0, &p2);
if( 0==*p2 ) break;
p2++;
p2 += sqlite3Fts3GetVarint32(p2, &iCol2);
}
}
fts3PoslistCopy(0, &p2);
fts3PoslistCopy(0, &p1);
*pp1 = p1;
*pp2 = p2;
if( !pp || *pp==p ){
return 0;
}
*p++ = 0x00;
*pp = p;
return 1;
}
/*
** Merge two position-lists as required by the NEAR operator.
*/
static int fts3PoslistNearMerge(
char **pp, /* Output buffer */
char *aTmp, /* Temporary buffer space */
int nRight, /* Maximum difference in token positions */
int nLeft, /* Maximum difference in token positions */
char **pp1, /* IN/OUT: Left input list */
char **pp2 /* IN/OUT: Right input list */
){
char *p1 = *pp1;
char *p2 = *pp2;
if( !pp ){
if( fts3PoslistPhraseMerge(0, nRight, 0, 0, pp1, pp2) ) return 1;
*pp1 = p1;
*pp2 = p2;
return fts3PoslistPhraseMerge(0, nLeft, 0, 0, pp2, pp1);
}else{
char *pTmp1 = aTmp;
char *pTmp2;
char *aTmp2;
int res = 1;
fts3PoslistPhraseMerge(&pTmp1, nRight, 0, 0, pp1, pp2);
aTmp2 = pTmp2 = pTmp1;
*pp1 = p1;
*pp2 = p2;
fts3PoslistPhraseMerge(&pTmp2, nLeft, 1, 0, pp2, pp1);
if( pTmp1!=aTmp && pTmp2!=aTmp2 ){
fts3PoslistMerge(pp, &aTmp, &aTmp2);
}else if( pTmp1!=aTmp ){
fts3PoslistCopy(pp, &aTmp);
}else if( pTmp2!=aTmp2 ){
fts3PoslistCopy(pp, &aTmp2);
}else{
res = 0;
}
return res;
}
}
/*
** Values that may be used as the first parameter to fts3DoclistMerge().
*/
#define MERGE_NOT 2 /* D + D -> D */
#define MERGE_AND 3 /* D + D -> D */
#define MERGE_OR 4 /* D + D -> D */
#define MERGE_POS_OR 5 /* P + P -> P */
#define MERGE_PHRASE 6 /* P + P -> D */
#define MERGE_POS_PHRASE 7 /* P + P -> P */
#define MERGE_NEAR 8 /* P + P -> D */
#define MERGE_POS_NEAR 9 /* P + P -> P */
/*
** Merge the two doclists passed in buffer a1 (size n1 bytes) and a2
** (size n2 bytes). The output is written to pre-allocated buffer aBuffer,
** which is guaranteed to be large enough to hold the results. The number
** of bytes written to aBuffer is stored in *pnBuffer before returning.
**
** If successful, SQLITE_OK is returned. Otherwise, if a malloc error
** occurs while allocating a temporary buffer as part of the merge operation,
** SQLITE_NOMEM is returned.
*/
static int fts3DoclistMerge(
int mergetype, /* One of the MERGE_XXX constants */
int nParam1, /* Used by MERGE_NEAR and MERGE_POS_NEAR */
int nParam2, /* Used by MERGE_NEAR and MERGE_POS_NEAR */
char *aBuffer, /* Pre-allocated output buffer */
int *pnBuffer, /* OUT: Bytes written to aBuffer */
char *a1, /* Buffer containing first doclist */
int n1, /* Size of buffer a1 */
char *a2, /* Buffer containing second doclist */
int n2, /* Size of buffer a2 */
int *pnDoc /* OUT: Number of docids in output */
){
sqlite3_int64 i1 = 0;
sqlite3_int64 i2 = 0;
sqlite3_int64 iPrev = 0;
char *p = aBuffer;
char *p1 = a1;
char *p2 = a2;
char *pEnd1 = &a1[n1];
char *pEnd2 = &a2[n2];
int nDoc = 0;
assert( mergetype==MERGE_OR || mergetype==MERGE_POS_OR
|| mergetype==MERGE_AND || mergetype==MERGE_NOT
|| mergetype==MERGE_PHRASE || mergetype==MERGE_POS_PHRASE
|| mergetype==MERGE_NEAR || mergetype==MERGE_POS_NEAR
);
if( !aBuffer ){
*pnBuffer = 0;
return SQLITE_NOMEM;
}
/* Read the first docid from each doclist */
fts3GetDeltaVarint2(&p1, pEnd1, &i1);
fts3GetDeltaVarint2(&p2, pEnd2, &i2);
switch( mergetype ){
case MERGE_OR:
case MERGE_POS_OR:
while( p1 || p2 ){
if( p2 && p1 && i1==i2 ){
fts3PutDeltaVarint(&p, &iPrev, i1);
if( mergetype==MERGE_POS_OR ) fts3PoslistMerge(&p, &p1, &p2);
fts3GetDeltaVarint2(&p1, pEnd1, &i1);
fts3GetDeltaVarint2(&p2, pEnd2, &i2);
}else if( !p2 || (p1 && i1<i2) ){
fts3PutDeltaVarint(&p, &iPrev, i1);
if( mergetype==MERGE_POS_OR ) fts3PoslistCopy(&p, &p1);
fts3GetDeltaVarint2(&p1, pEnd1, &i1);
}else{
fts3PutDeltaVarint(&p, &iPrev, i2);
if( mergetype==MERGE_POS_OR ) fts3PoslistCopy(&p, &p2);
fts3GetDeltaVarint2(&p2, pEnd2, &i2);
}
}
break;
case MERGE_AND:
while( p1 && p2 ){
if( i1==i2 ){
fts3PutDeltaVarint(&p, &iPrev, i1);
fts3GetDeltaVarint2(&p1, pEnd1, &i1);
fts3GetDeltaVarint2(&p2, pEnd2, &i2);
nDoc++;
}else if( i1<i2 ){
fts3GetDeltaVarint2(&p1, pEnd1, &i1);
}else{
fts3GetDeltaVarint2(&p2, pEnd2, &i2);
}
}
break;
case MERGE_NOT:
while( p1 ){
if( p2 && i1==i2 ){
fts3GetDeltaVarint2(&p1, pEnd1, &i1);
fts3GetDeltaVarint2(&p2, pEnd2, &i2);
}else if( !p2 || i1<i2 ){
fts3PutDeltaVarint(&p, &iPrev, i1);
fts3GetDeltaVarint2(&p1, pEnd1, &i1);
}else{
fts3GetDeltaVarint2(&p2, pEnd2, &i2);
}
}
break;
case MERGE_POS_PHRASE:
case MERGE_PHRASE: {
char **ppPos = (mergetype==MERGE_PHRASE ? 0 : &p);
while( p1 && p2 ){
if( i1==i2 ){
char *pSave = p;
sqlite3_int64 iPrevSave = iPrev;
fts3PutDeltaVarint(&p, &iPrev, i1);
if( 0==fts3PoslistPhraseMerge(ppPos, nParam1, 0, 1, &p1, &p2) ){
p = pSave;
iPrev = iPrevSave;
}else{
nDoc++;
}
fts3GetDeltaVarint2(&p1, pEnd1, &i1);
fts3GetDeltaVarint2(&p2, pEnd2, &i2);
}else if( i1<i2 ){
fts3PoslistCopy(0, &p1);
fts3GetDeltaVarint2(&p1, pEnd1, &i1);
}else{
fts3PoslistCopy(0, &p2);
fts3GetDeltaVarint2(&p2, pEnd2, &i2);
}
}
break;
}
default: assert( mergetype==MERGE_POS_NEAR || mergetype==MERGE_NEAR ); {
char *aTmp = 0;
char **ppPos = 0;
if( mergetype==MERGE_POS_NEAR ){
ppPos = &p;
aTmp = sqlite3_malloc(2*(n1+n2+1));
if( !aTmp ){
return SQLITE_NOMEM;
}
}
while( p1 && p2 ){
if( i1==i2 ){
char *pSave = p;
sqlite3_int64 iPrevSave = iPrev;
fts3PutDeltaVarint(&p, &iPrev, i1);
if( !fts3PoslistNearMerge(ppPos, aTmp, nParam1, nParam2, &p1, &p2) ){
iPrev = iPrevSave;
p = pSave;
}
fts3GetDeltaVarint2(&p1, pEnd1, &i1);
fts3GetDeltaVarint2(&p2, pEnd2, &i2);
}else if( i1<i2 ){
fts3PoslistCopy(0, &p1);
fts3GetDeltaVarint2(&p1, pEnd1, &i1);
}else{
fts3PoslistCopy(0, &p2);
fts3GetDeltaVarint2(&p2, pEnd2, &i2);
}
}
sqlite3_free(aTmp);
break;
}
}
if( pnDoc ) *pnDoc = nDoc;
*pnBuffer = (int)(p-aBuffer);
return SQLITE_OK;
}
/*
** A pointer to an instance of this structure is used as the context
** argument to sqlite3Fts3SegReaderIterate()
*/
typedef struct TermSelect TermSelect;
struct TermSelect {
int isReqPos;
char *aaOutput[16]; /* Malloc'd output buffer */
int anOutput[16]; /* Size of output in bytes */
};
/*
** Merge all doclists in the TermSelect.aaOutput[] array into a single
** doclist stored in TermSelect.aaOutput[0]. If successful, delete all
** other doclists (except the aaOutput[0] one) and return SQLITE_OK.
**
** If an OOM error occurs, return SQLITE_NOMEM. In this case it is
** the responsibility of the caller to free any doclists left in the
** TermSelect.aaOutput[] array.
*/
static int fts3TermSelectMerge(TermSelect *pTS){
int mergetype = (pTS->isReqPos ? MERGE_POS_OR : MERGE_OR);
char *aOut = 0;
int nOut = 0;
int i;
/* Loop through the doclists in the aaOutput[] array. Merge them all
** into a single doclist.
*/
for(i=0; i<SizeofArray(pTS->aaOutput); i++){
if( pTS->aaOutput[i] ){
if( !aOut ){
aOut = pTS->aaOutput[i];
nOut = pTS->anOutput[i];
pTS->aaOutput[i] = 0;
}else{
int nNew = nOut + pTS->anOutput[i];
char *aNew = sqlite3_malloc(nNew);
if( !aNew ){
sqlite3_free(aOut);
return SQLITE_NOMEM;
}
fts3DoclistMerge(mergetype, 0, 0,
aNew, &nNew, pTS->aaOutput[i], pTS->anOutput[i], aOut, nOut, 0
);
sqlite3_free(pTS->aaOutput[i]);
sqlite3_free(aOut);
pTS->aaOutput[i] = 0;
aOut = aNew;
nOut = nNew;
}
}
}
pTS->aaOutput[0] = aOut;
pTS->anOutput[0] = nOut;
return SQLITE_OK;
}
/*
** This function is used as the sqlite3Fts3SegReaderIterate() callback when
** querying the full-text index for a doclist associated with a term or
** term-prefix.
*/
static int fts3TermSelectCb(
Fts3Table *p, /* Virtual table object */
void *pContext, /* Pointer to TermSelect structure */
char *zTerm,
int nTerm,
char *aDoclist,
int nDoclist
){
TermSelect *pTS = (TermSelect *)pContext;
UNUSED_PARAMETER(p);
UNUSED_PARAMETER(zTerm);
UNUSED_PARAMETER(nTerm);
if( pTS->aaOutput[0]==0 ){
/* If this is the first term selected, copy the doclist to the output
** buffer using memcpy(). TODO: Add a way to transfer control of the
** aDoclist buffer from the caller so as to avoid the memcpy().
*/
pTS->aaOutput[0] = sqlite3_malloc(nDoclist);
pTS->anOutput[0] = nDoclist;
if( pTS->aaOutput[0] ){
memcpy(pTS->aaOutput[0], aDoclist, nDoclist);
}else{
return SQLITE_NOMEM;
}
}else{
int mergetype = (pTS->isReqPos ? MERGE_POS_OR : MERGE_OR);
char *aMerge = aDoclist;
int nMerge = nDoclist;
int iOut;
for(iOut=0; iOut<SizeofArray(pTS->aaOutput); iOut++){
char *aNew;
int nNew;
if( pTS->aaOutput[iOut]==0 ){
assert( iOut>0 );
pTS->aaOutput[iOut] = aMerge;
pTS->anOutput[iOut] = nMerge;
break;
}
nNew = nMerge + pTS->anOutput[iOut];
aNew = sqlite3_malloc(nNew);
if( !aNew ){
if( aMerge!=aDoclist ){
sqlite3_free(aMerge);
}
return SQLITE_NOMEM;
}
fts3DoclistMerge(mergetype, 0, 0, aNew, &nNew,
pTS->aaOutput[iOut], pTS->anOutput[iOut], aMerge, nMerge, 0
);
if( iOut>0 ) sqlite3_free(aMerge);
sqlite3_free(pTS->aaOutput[iOut]);
pTS->aaOutput[iOut] = 0;
aMerge = aNew;
nMerge = nNew;
if( (iOut+1)==SizeofArray(pTS->aaOutput) ){
pTS->aaOutput[iOut] = aMerge;
pTS->anOutput[iOut] = nMerge;
}
}
}
return SQLITE_OK;
}
static int fts3DeferredTermSelect(
Fts3DeferredToken *pToken, /* Phrase token */
int isTermPos, /* True to include positions */
int *pnOut, /* OUT: Size of list */
char **ppOut /* OUT: Body of list */
){
char *aSource;
int nSource;
aSource = sqlite3Fts3DeferredDoclist(pToken, &nSource);
if( !aSource ){
*pnOut = 0;
*ppOut = 0;
}else if( isTermPos ){
*ppOut = sqlite3_malloc(nSource);
if( !*ppOut ) return SQLITE_NOMEM;
memcpy(*ppOut, aSource, nSource);
*pnOut = nSource;
}else{
sqlite3_int64 docid;
*pnOut = sqlite3Fts3GetVarint(aSource, &docid);
*ppOut = sqlite3_malloc(*pnOut);
if( !*ppOut ) return SQLITE_NOMEM;
sqlite3Fts3PutVarint(*ppOut, docid);
}
return SQLITE_OK;
}
int sqlite3Fts3SegReaderCursor(
Fts3Table *p, /* FTS3 table handle */
int iLevel, /* Level of segments to scan */
const char *zTerm, /* Term to query for */
int nTerm, /* Size of zTerm in bytes */
int isPrefix, /* True for a prefix search */
int isScan, /* True to scan from zTerm to EOF */
Fts3SegReaderCursor *pCsr /* Cursor object to populate */
){
int rc = SQLITE_OK;
int rc2;
int iAge = 0;
sqlite3_stmt *pStmt = 0;
Fts3SegReader *pPending = 0;
assert( iLevel==FTS3_SEGCURSOR_ALL
|| iLevel==FTS3_SEGCURSOR_PENDING
|| iLevel>=0
);
assert( FTS3_SEGCURSOR_PENDING<0 );
assert( FTS3_SEGCURSOR_ALL<0 );
assert( iLevel==FTS3_SEGCURSOR_ALL || (zTerm==0 && isPrefix==1) );
assert( isPrefix==0 || isScan==0 );
memset(pCsr, 0, sizeof(Fts3SegReaderCursor));
/* If iLevel is less than 0, include a seg-reader for the pending-terms. */
assert( isScan==0 || fts3HashCount(&p->pendingTerms)==0 );
if( iLevel<0 && isScan==0 ){
rc = sqlite3Fts3SegReaderPending(p, zTerm, nTerm, isPrefix, &pPending);
if( rc==SQLITE_OK && pPending ){
int nByte = (sizeof(Fts3SegReader *) * 16);
pCsr->apSegment = (Fts3SegReader **)sqlite3_malloc(nByte);
if( pCsr->apSegment==0 ){
rc = SQLITE_NOMEM;
}else{
pCsr->apSegment[0] = pPending;
pCsr->nSegment = 1;
pPending = 0;
}
}
}
if( iLevel!=FTS3_SEGCURSOR_PENDING ){
if( rc==SQLITE_OK ){
rc = sqlite3Fts3AllSegdirs(p, iLevel, &pStmt);
}
while( rc==SQLITE_OK && SQLITE_ROW==(rc = sqlite3_step(pStmt)) ){
/* Read the values returned by the SELECT into local variables. */
sqlite3_int64 iStartBlock = sqlite3_column_int64(pStmt, 1);
sqlite3_int64 iLeavesEndBlock = sqlite3_column_int64(pStmt, 2);
sqlite3_int64 iEndBlock = sqlite3_column_int64(pStmt, 3);
int nRoot = sqlite3_column_bytes(pStmt, 4);
char const *zRoot = sqlite3_column_blob(pStmt, 4);
/* If nSegment is a multiple of 16 the array needs to be extended. */
if( (pCsr->nSegment%16)==0 ){
Fts3SegReader **apNew;
int nByte = (pCsr->nSegment + 16)*sizeof(Fts3SegReader*);
apNew = (Fts3SegReader **)sqlite3_realloc(pCsr->apSegment, nByte);
if( !apNew ){
rc = SQLITE_NOMEM;
goto finished;
}
pCsr->apSegment = apNew;
}
/* If zTerm is not NULL, and this segment is not stored entirely on its
** root node, the range of leaves scanned can be reduced. Do this. */
if( iStartBlock && zTerm ){
sqlite3_int64 *pi = (isPrefix ? &iLeavesEndBlock : 0);
rc = fts3SelectLeaf(p, zTerm, nTerm, zRoot, nRoot, &iStartBlock, pi);
if( rc!=SQLITE_OK ) goto finished;
if( isPrefix==0 && isScan==0 ) iLeavesEndBlock = iStartBlock;
}
rc = sqlite3Fts3SegReaderNew(iAge, iStartBlock, iLeavesEndBlock,
iEndBlock, zRoot, nRoot, &pCsr->apSegment[pCsr->nSegment]
);
if( rc!=SQLITE_OK ) goto finished;
pCsr->nSegment++;
iAge++;
}
}
finished:
rc2 = sqlite3_reset(pStmt);
if( rc==SQLITE_DONE ) rc = rc2;
sqlite3Fts3SegReaderFree(pPending);
return rc;
}
static int fts3TermSegReaderCursor(
Fts3Cursor *pCsr, /* Virtual table cursor handle */
const char *zTerm, /* Term to query for */
int nTerm, /* Size of zTerm in bytes */
int isPrefix, /* True for a prefix search */
Fts3SegReaderCursor **ppSegcsr /* OUT: Allocated seg-reader cursor */
){
Fts3SegReaderCursor *pSegcsr; /* Object to allocate and return */
int rc = SQLITE_NOMEM; /* Return code */
pSegcsr = sqlite3_malloc(sizeof(Fts3SegReaderCursor));
if( pSegcsr ){
Fts3Table *p = (Fts3Table *)pCsr->base.pVtab;
int i;
int nCost = 0;
rc = sqlite3Fts3SegReaderCursor(
p, FTS3_SEGCURSOR_ALL, zTerm, nTerm, isPrefix, 0, pSegcsr);
for(i=0; rc==SQLITE_OK && i<pSegcsr->nSegment; i++){
rc = sqlite3Fts3SegReaderCost(pCsr, pSegcsr->apSegment[i], &nCost);
}
pSegcsr->nCost = nCost;
}
*ppSegcsr = pSegcsr;
return rc;
}
static void fts3SegReaderCursorFree(Fts3SegReaderCursor *pSegcsr){
sqlite3Fts3SegReaderFinish(pSegcsr);
sqlite3_free(pSegcsr);
}
/*
** This function retreives the doclist for the specified term (or term
** prefix) from the database.
**
** The returned doclist may be in one of two formats, depending on the
** value of parameter isReqPos. If isReqPos is zero, then the doclist is
** a sorted list of delta-compressed docids (a bare doclist). If isReqPos
** is non-zero, then the returned list is in the same format as is stored
** in the database without the found length specifier at the start of on-disk
** doclists.
*/
static int fts3TermSelect(
Fts3Table *p, /* Virtual table handle */
Fts3PhraseToken *pTok, /* Token to query for */
int iColumn, /* Column to query (or -ve for all columns) */
int isReqPos, /* True to include position lists in output */
int *pnOut, /* OUT: Size of buffer at *ppOut */
char **ppOut /* OUT: Malloced result buffer */
){
int rc; /* Return code */
Fts3SegReaderCursor *pSegcsr; /* Seg-reader cursor for this term */
TermSelect tsc; /* Context object for fts3TermSelectCb() */
Fts3SegFilter filter; /* Segment term filter configuration */
pSegcsr = pTok->pSegcsr;
memset(&tsc, 0, sizeof(TermSelect));
tsc.isReqPos = isReqPos;
filter.flags = FTS3_SEGMENT_IGNORE_EMPTY
| (pTok->isPrefix ? FTS3_SEGMENT_PREFIX : 0)
| (isReqPos ? FTS3_SEGMENT_REQUIRE_POS : 0)
| (iColumn<p->nColumn ? FTS3_SEGMENT_COLUMN_FILTER : 0);
filter.iCol = iColumn;
filter.zTerm = pTok->z;
filter.nTerm = pTok->n;
rc = sqlite3Fts3SegReaderStart(p, pSegcsr, &filter);
while( SQLITE_OK==rc
&& SQLITE_ROW==(rc = sqlite3Fts3SegReaderStep(p, pSegcsr))
){
rc = fts3TermSelectCb(p, (void *)&tsc,
pSegcsr->zTerm, pSegcsr->nTerm, pSegcsr->aDoclist, pSegcsr->nDoclist
);
}
if( rc==SQLITE_OK ){
rc = fts3TermSelectMerge(&tsc);
}
if( rc==SQLITE_OK ){
*ppOut = tsc.aaOutput[0];
*pnOut = tsc.anOutput[0];
}else{
int i;
for(i=0; i<SizeofArray(tsc.aaOutput); i++){
sqlite3_free(tsc.aaOutput[i]);
}
}
fts3SegReaderCursorFree(pSegcsr);
pTok->pSegcsr = 0;
return rc;
}
/*
** This function counts the total number of docids in the doclist stored
** in buffer aList[], size nList bytes.
**
** If the isPoslist argument is true, then it is assumed that the doclist
** contains a position-list following each docid. Otherwise, it is assumed
** that the doclist is simply a list of docids stored as delta encoded
** varints.
*/
static int fts3DoclistCountDocids(int isPoslist, char *aList, int nList){
int nDoc = 0; /* Return value */
if( aList ){
char *aEnd = &aList[nList]; /* Pointer to one byte after EOF */
char *p = aList; /* Cursor */
if( !isPoslist ){
/* The number of docids in the list is the same as the number of
** varints. In FTS3 a varint consists of a single byte with the 0x80
** bit cleared and zero or more bytes with the 0x80 bit set. So to
** count the varints in the buffer, just count the number of bytes
** with the 0x80 bit clear. */
while( p<aEnd ) nDoc += (((*p++)&0x80)==0);
}else{
while( p<aEnd ){
nDoc++;
while( (*p++)&0x80 ); /* Skip docid varint */
fts3PoslistCopy(0, &p); /* Skip over position list */
}
}
}
return nDoc;
}
/*
** Call sqlite3Fts3DeferToken() for each token in the expression pExpr.
*/
static int fts3DeferExpression(Fts3Cursor *pCsr, Fts3Expr *pExpr){
int rc = SQLITE_OK;
if( pExpr ){
rc = fts3DeferExpression(pCsr, pExpr->pLeft);
if( rc==SQLITE_OK ){
rc = fts3DeferExpression(pCsr, pExpr->pRight);
}
if( pExpr->eType==FTSQUERY_PHRASE ){
int iCol = pExpr->pPhrase->iColumn;
int i;
for(i=0; rc==SQLITE_OK && i<pExpr->pPhrase->nToken; i++){
Fts3PhraseToken *pToken = &pExpr->pPhrase->aToken[i];
if( pToken->pDeferred==0 ){
rc = sqlite3Fts3DeferToken(pCsr, pToken, iCol);
}
}
}
}
return rc;
}
/*
** This function removes the position information from a doclist. When
** called, buffer aList (size *pnList bytes) contains a doclist that includes
** position information. This function removes the position information so
** that aList contains only docids, and adjusts *pnList to reflect the new
** (possibly reduced) size of the doclist.
*/
static void fts3DoclistStripPositions(
char *aList, /* IN/OUT: Buffer containing doclist */
int *pnList /* IN/OUT: Size of doclist in bytes */
){
if( aList ){
char *aEnd = &aList[*pnList]; /* Pointer to one byte after EOF */
char *p = aList; /* Input cursor */
char *pOut = aList; /* Output cursor */
while( p<aEnd ){
sqlite3_int64 delta;
p += sqlite3Fts3GetVarint(p, &delta);
fts3PoslistCopy(0, &p);
pOut += sqlite3Fts3PutVarint(pOut, delta);
}
*pnList = (int)(pOut - aList);
}
}
/*
** Return a DocList corresponding to the phrase *pPhrase.
**
** If this function returns SQLITE_OK, but *pnOut is set to a negative value,
** then no tokens in the phrase were looked up in the full-text index. This
** is only possible when this function is called from within xFilter(). The
** caller should assume that all documents match the phrase. The actual
** filtering will take place in xNext().
*/
static int fts3PhraseSelect(
Fts3Cursor *pCsr, /* Virtual table cursor handle */
Fts3Phrase *pPhrase, /* Phrase to return a doclist for */
int isReqPos, /* True if output should contain positions */
char **paOut, /* OUT: Pointer to malloc'd result buffer */
int *pnOut /* OUT: Size of buffer at *paOut */
){
char *pOut = 0;
int nOut = 0;
int rc = SQLITE_OK;
int ii;
int iCol = pPhrase->iColumn;
int isTermPos = (pPhrase->nToken>1 || isReqPos);
Fts3Table *p = (Fts3Table *)pCsr->base.pVtab;
int isFirst = 1;
int iPrevTok = 0;
int nDoc = 0;
/* If this is an xFilter() evaluation, create a segment-reader for each
** phrase token. Or, if this is an xNext() or snippet/offsets/matchinfo
** evaluation, only create segment-readers if there are no Fts3DeferredToken
** objects attached to the phrase-tokens.
*/
for(ii=0; ii<pPhrase->nToken; ii++){
Fts3PhraseToken *pTok = &pPhrase->aToken[ii];
if( pTok->pSegcsr==0 ){
if( (pCsr->eEvalmode==FTS3_EVAL_FILTER)
|| (pCsr->eEvalmode==FTS3_EVAL_NEXT && pCsr->pDeferred==0)
|| (pCsr->eEvalmode==FTS3_EVAL_MATCHINFO && pTok->bFulltext)
){
rc = fts3TermSegReaderCursor(
pCsr, pTok->z, pTok->n, pTok->isPrefix, &pTok->pSegcsr
);
if( rc!=SQLITE_OK ) return rc;
}
}
}
for(ii=0; ii<pPhrase->nToken; ii++){
Fts3PhraseToken *pTok; /* Token to find doclist for */
int iTok = 0; /* The token being queried this iteration */
char *pList = 0; /* Pointer to token doclist */
int nList = 0; /* Size of buffer at pList */
/* Select a token to process. If this is an xFilter() call, then tokens
** are processed in order from least to most costly. Otherwise, tokens
** are processed in the order in which they occur in the phrase.
*/
if( pCsr->eEvalmode==FTS3_EVAL_MATCHINFO ){
assert( isReqPos );
iTok = ii;
pTok = &pPhrase->aToken[iTok];
if( pTok->bFulltext==0 ) continue;
}else if( pCsr->eEvalmode==FTS3_EVAL_NEXT || isReqPos ){
iTok = ii;
pTok = &pPhrase->aToken[iTok];
}else{
int nMinCost = 0x7FFFFFFF;
int jj;
/* Find the remaining token with the lowest cost. */
for(jj=0; jj<pPhrase->nToken; jj++){
Fts3SegReaderCursor *pSegcsr = pPhrase->aToken[jj].pSegcsr;
if( pSegcsr && pSegcsr->nCost<nMinCost ){
iTok = jj;
nMinCost = pSegcsr->nCost;
}
}
pTok = &pPhrase->aToken[iTok];
/* This branch is taken if it is determined that loading the doclist
** for the next token would require more IO than loading all documents
** currently identified by doclist pOut/nOut. No further doclists will
** be loaded from the full-text index for this phrase.
*/
if( nMinCost>nDoc && ii>0 ){
rc = fts3DeferExpression(pCsr, pCsr->pExpr);
break;
}
}
if( pCsr->eEvalmode==FTS3_EVAL_NEXT && pTok->pDeferred ){
rc = fts3DeferredTermSelect(pTok->pDeferred, isTermPos, &nList, &pList);
}else{
if( pTok->pSegcsr ){
rc = fts3TermSelect(p, pTok, iCol, isTermPos, &nList, &pList);
}
pTok->bFulltext = 1;
}
assert( rc!=SQLITE_OK || pCsr->eEvalmode || pTok->pSegcsr==0 );
if( rc!=SQLITE_OK ) break;
if( isFirst ){
pOut = pList;
nOut = nList;
if( pCsr->eEvalmode==FTS3_EVAL_FILTER && pPhrase->nToken>1 ){
nDoc = fts3DoclistCountDocids(1, pOut, nOut);
}
isFirst = 0;
iPrevTok = iTok;
}else{
/* Merge the new term list and the current output. */
char *aLeft, *aRight;
int nLeft, nRight;
int nDist;
int mt;
/* If this is the final token of the phrase, and positions were not
** requested by the caller, use MERGE_PHRASE instead of POS_PHRASE.
** This drops the position information from the output list.
*/
mt = MERGE_POS_PHRASE;
if( ii==pPhrase->nToken-1 && !isReqPos ) mt = MERGE_PHRASE;
assert( iPrevTok!=iTok );
if( iPrevTok<iTok ){
aLeft = pOut;
nLeft = nOut;
aRight = pList;
nRight = nList;
nDist = iTok-iPrevTok;
iPrevTok = iTok;
}else{
aRight = pOut;
nRight = nOut;
aLeft = pList;
nLeft = nList;
nDist = iPrevTok-iTok;
}
pOut = aRight;
fts3DoclistMerge(
mt, nDist, 0, pOut, &nOut, aLeft, nLeft, aRight, nRight, &nDoc
);
sqlite3_free(aLeft);
}
assert( nOut==0 || pOut!=0 );
}
if( rc==SQLITE_OK ){
if( ii!=pPhrase->nToken ){
assert( pCsr->eEvalmode==FTS3_EVAL_FILTER && isReqPos==0 );
fts3DoclistStripPositions(pOut, &nOut);
}
*paOut = pOut;
*pnOut = nOut;
}else{
sqlite3_free(pOut);
}
return rc;
}
/*
** This function merges two doclists according to the requirements of a
** NEAR operator.
**
** Both input doclists must include position information. The output doclist
** includes position information if the first argument to this function
** is MERGE_POS_NEAR, or does not if it is MERGE_NEAR.
*/
static int fts3NearMerge(
int mergetype, /* MERGE_POS_NEAR or MERGE_NEAR */
int nNear, /* Parameter to NEAR operator */
int nTokenLeft, /* Number of tokens in LHS phrase arg */
char *aLeft, /* Doclist for LHS (incl. positions) */
int nLeft, /* Size of LHS doclist in bytes */
int nTokenRight, /* As nTokenLeft */
char *aRight, /* As aLeft */
int nRight, /* As nRight */
char **paOut, /* OUT: Results of merge (malloced) */
int *pnOut /* OUT: Sized of output buffer */
){
char *aOut; /* Buffer to write output doclist to */
int rc; /* Return code */
assert( mergetype==MERGE_POS_NEAR || MERGE_NEAR );
aOut = sqlite3_malloc(nLeft+nRight+1);
if( aOut==0 ){
rc = SQLITE_NOMEM;
}else{
rc = fts3DoclistMerge(mergetype, nNear+nTokenRight, nNear+nTokenLeft,
aOut, pnOut, aLeft, nLeft, aRight, nRight, 0
);
if( rc!=SQLITE_OK ){
sqlite3_free(aOut);
aOut = 0;
}
}
*paOut = aOut;
return rc;
}
/*
** This function is used as part of the processing for the snippet() and
** offsets() functions.
**
** Both pLeft and pRight are expression nodes of type FTSQUERY_PHRASE. Both
** have their respective doclists (including position information) loaded
** in Fts3Expr.aDoclist/nDoclist. This function removes all entries from
** each doclist that are not within nNear tokens of a corresponding entry
** in the other doclist.
*/
int sqlite3Fts3ExprNearTrim(Fts3Expr *pLeft, Fts3Expr *pRight, int nNear){
int rc; /* Return code */
assert( pLeft->eType==FTSQUERY_PHRASE );
assert( pRight->eType==FTSQUERY_PHRASE );
assert( pLeft->isLoaded && pRight->isLoaded );
if( pLeft->aDoclist==0 || pRight->aDoclist==0 ){
sqlite3_free(pLeft->aDoclist);
sqlite3_free(pRight->aDoclist);
pRight->aDoclist = 0;
pLeft->aDoclist = 0;
rc = SQLITE_OK;
}else{
char *aOut; /* Buffer in which to assemble new doclist */
int nOut; /* Size of buffer aOut in bytes */
rc = fts3NearMerge(MERGE_POS_NEAR, nNear,
pLeft->pPhrase->nToken, pLeft->aDoclist, pLeft->nDoclist,
pRight->pPhrase->nToken, pRight->aDoclist, pRight->nDoclist,
&aOut, &nOut
);
if( rc!=SQLITE_OK ) return rc;
sqlite3_free(pRight->aDoclist);
pRight->aDoclist = aOut;
pRight->nDoclist = nOut;
rc = fts3NearMerge(MERGE_POS_NEAR, nNear,
pRight->pPhrase->nToken, pRight->aDoclist, pRight->nDoclist,
pLeft->pPhrase->nToken, pLeft->aDoclist, pLeft->nDoclist,
&aOut, &nOut
);
sqlite3_free(pLeft->aDoclist);
pLeft->aDoclist = aOut;
pLeft->nDoclist = nOut;
}
return rc;
}
/*
** Allocate an Fts3SegReaderArray for each token in the expression pExpr.
** The allocated objects are stored in the Fts3PhraseToken.pArray member
** variables of each token structure.
*/
static int fts3ExprAllocateSegReaders(
Fts3Cursor *pCsr, /* FTS3 table */
Fts3Expr *pExpr, /* Expression to create seg-readers for */
int *pnExpr /* OUT: Number of AND'd expressions */
){
int rc = SQLITE_OK; /* Return code */
assert( pCsr->eEvalmode==FTS3_EVAL_FILTER );
if( pnExpr && pExpr->eType!=FTSQUERY_AND ){
(*pnExpr)++;
pnExpr = 0;
}
if( pExpr->eType==FTSQUERY_PHRASE ){
Fts3Phrase *pPhrase = pExpr->pPhrase;
int ii;
for(ii=0; rc==SQLITE_OK && ii<pPhrase->nToken; ii++){
Fts3PhraseToken *pTok = &pPhrase->aToken[ii];
if( pTok->pSegcsr==0 ){
rc = fts3TermSegReaderCursor(
pCsr, pTok->z, pTok->n, pTok->isPrefix, &pTok->pSegcsr
);
}
}
}else{
rc = fts3ExprAllocateSegReaders(pCsr, pExpr->pLeft, pnExpr);
if( rc==SQLITE_OK ){
rc = fts3ExprAllocateSegReaders(pCsr, pExpr->pRight, pnExpr);
}
}
return rc;
}
/*
** Free the Fts3SegReaderArray objects associated with each token in the
** expression pExpr. In other words, this function frees the resources
** allocated by fts3ExprAllocateSegReaders().
*/
static void fts3ExprFreeSegReaders(Fts3Expr *pExpr){
if( pExpr ){
Fts3Phrase *pPhrase = pExpr->pPhrase;
if( pPhrase ){
int kk;
for(kk=0; kk<pPhrase->nToken; kk++){
fts3SegReaderCursorFree(pPhrase->aToken[kk].pSegcsr);
pPhrase->aToken[kk].pSegcsr = 0;
}
}
fts3ExprFreeSegReaders(pExpr->pLeft);
fts3ExprFreeSegReaders(pExpr->pRight);
}
}
/*
** Return the sum of the costs of all tokens in the expression pExpr. This
** function must be called after Fts3SegReaderArrays have been allocated
** for all tokens using fts3ExprAllocateSegReaders().
*/
static int fts3ExprCost(Fts3Expr *pExpr){
int nCost; /* Return value */
if( pExpr->eType==FTSQUERY_PHRASE ){
Fts3Phrase *pPhrase = pExpr->pPhrase;
int ii;
nCost = 0;
for(ii=0; ii<pPhrase->nToken; ii++){
Fts3SegReaderCursor *pSegcsr = pPhrase->aToken[ii].pSegcsr;
if( pSegcsr ) nCost += pSegcsr->nCost;
}
}else{
nCost = fts3ExprCost(pExpr->pLeft) + fts3ExprCost(pExpr->pRight);
}
return nCost;
}
/*
** The following is a helper function (and type) for fts3EvalExpr(). It
** must be called after Fts3SegReaders have been allocated for every token
** in the expression. See the context it is called from in fts3EvalExpr()
** for further explanation.
*/
typedef struct ExprAndCost ExprAndCost;
struct ExprAndCost {
Fts3Expr *pExpr;
int nCost;
};
static void fts3ExprAssignCosts(
Fts3Expr *pExpr, /* Expression to create seg-readers for */
ExprAndCost **ppExprCost /* OUT: Write to *ppExprCost */
){
if( pExpr->eType==FTSQUERY_AND ){
fts3ExprAssignCosts(pExpr->pLeft, ppExprCost);
fts3ExprAssignCosts(pExpr->pRight, ppExprCost);
}else{
(*ppExprCost)->pExpr = pExpr;
(*ppExprCost)->nCost = fts3ExprCost(pExpr);
(*ppExprCost)++;
}
}
/*
** Evaluate the full-text expression pExpr against FTS3 table pTab. Store
** the resulting doclist in *paOut and *pnOut. This routine mallocs for
** the space needed to store the output. The caller is responsible for
** freeing the space when it has finished.
**
** This function is called in two distinct contexts:
**
** * From within the virtual table xFilter() method. In this case, the
** output doclist contains entries for all rows in the table, based on
** data read from the full-text index.
**
** In this case, if the query expression contains one or more tokens that
** are very common, then the returned doclist may contain a superset of
** the documents that actually match the expression.
**
** * From within the virtual table xNext() method. This call is only made
** if the call from within xFilter() found that there were very common
** tokens in the query expression and did return a superset of the
** matching documents. In this case the returned doclist contains only
** entries that correspond to the current row of the table. Instead of
** reading the data for each token from the full-text index, the data is
** already available in-memory in the Fts3PhraseToken.pDeferred structures.
** See fts3EvalDeferred() for how it gets there.
**
** In the first case above, Fts3Cursor.doDeferred==0. In the second (if it is
** required) Fts3Cursor.doDeferred==1.
**
** If the SQLite invokes the snippet(), offsets() or matchinfo() function
** as part of a SELECT on an FTS3 table, this function is called on each
** individual phrase expression in the query. If there were very common tokens
** found in the xFilter() call, then this function is called once for phrase
** for each row visited, and the returned doclist contains entries for the
** current row only. Otherwise, if there were no very common tokens, then this
** function is called once only for each phrase in the query and the returned
** doclist contains entries for all rows of the table.
**
** Fts3Cursor.doDeferred==1 when this function is called on phrases as a
** result of a snippet(), offsets() or matchinfo() invocation.
*/
static int fts3EvalExpr(
Fts3Cursor *p, /* Virtual table cursor handle */
Fts3Expr *pExpr, /* Parsed fts3 expression */
char **paOut, /* OUT: Pointer to malloc'd result buffer */
int *pnOut, /* OUT: Size of buffer at *paOut */
int isReqPos /* Require positions in output buffer */
){
int rc = SQLITE_OK; /* Return code */
/* Zero the output parameters. */
*paOut = 0;
*pnOut = 0;
if( pExpr ){
assert( pExpr->eType==FTSQUERY_NEAR || pExpr->eType==FTSQUERY_OR
|| pExpr->eType==FTSQUERY_AND || pExpr->eType==FTSQUERY_NOT
|| pExpr->eType==FTSQUERY_PHRASE
);
assert( pExpr->eType==FTSQUERY_PHRASE || isReqPos==0 );
if( pExpr->eType==FTSQUERY_PHRASE ){
rc = fts3PhraseSelect(p, pExpr->pPhrase,
isReqPos || (pExpr->pParent && pExpr->pParent->eType==FTSQUERY_NEAR),
paOut, pnOut
);
fts3ExprFreeSegReaders(pExpr);
}else if( p->eEvalmode==FTS3_EVAL_FILTER && pExpr->eType==FTSQUERY_AND ){
ExprAndCost *aExpr = 0; /* Array of AND'd expressions and costs */
int nExpr = 0; /* Size of aExpr[] */
char *aRet = 0; /* Doclist to return to caller */
int nRet = 0; /* Length of aRet[] in bytes */
int nDoc = 0x7FFFFFFF;
assert( !isReqPos );
rc = fts3ExprAllocateSegReaders(p, pExpr, &nExpr);
if( rc==SQLITE_OK ){
assert( nExpr>1 );
aExpr = sqlite3_malloc(sizeof(ExprAndCost) * nExpr);
if( !aExpr ) rc = SQLITE_NOMEM;
}
if( rc==SQLITE_OK ){
int ii; /* Used to iterate through expressions */
fts3ExprAssignCosts(pExpr, &aExpr);
aExpr -= nExpr;
for(ii=0; ii<nExpr; ii++){
char *aNew;
int nNew;
int jj;
ExprAndCost *pBest = 0;
for(jj=0; jj<nExpr; jj++){
ExprAndCost *pCand = &aExpr[jj];
if( pCand->pExpr && (pBest==0 || pCand->nCost<pBest->nCost) ){
pBest = pCand;
}
}
if( pBest->nCost>nDoc ){
rc = fts3DeferExpression(p, p->pExpr);
break;
}else{
rc = fts3EvalExpr(p, pBest->pExpr, &aNew, &nNew, 0);
if( rc!=SQLITE_OK ) break;
pBest->pExpr = 0;
if( ii==0 ){
aRet = aNew;
nRet = nNew;
nDoc = fts3DoclistCountDocids(0, aRet, nRet);
}else{
fts3DoclistMerge(
MERGE_AND, 0, 0, aRet, &nRet, aRet, nRet, aNew, nNew, &nDoc
);
sqlite3_free(aNew);
}
}
}
}
if( rc==SQLITE_OK ){
*paOut = aRet;
*pnOut = nRet;
}else{
assert( *paOut==0 );
sqlite3_free(aRet);
}
sqlite3_free(aExpr);
fts3ExprFreeSegReaders(pExpr);
}else{
char *aLeft;
char *aRight;
int nLeft;
int nRight;
assert( pExpr->eType==FTSQUERY_NEAR
|| pExpr->eType==FTSQUERY_OR
|| pExpr->eType==FTSQUERY_NOT
|| (pExpr->eType==FTSQUERY_AND && p->eEvalmode==FTS3_EVAL_NEXT)
);
if( 0==(rc = fts3EvalExpr(p, pExpr->pRight, &aRight, &nRight, isReqPos))
&& 0==(rc = fts3EvalExpr(p, pExpr->pLeft, &aLeft, &nLeft, isReqPos))
){
switch( pExpr->eType ){
case FTSQUERY_NEAR: {
Fts3Expr *pLeft;
Fts3Expr *pRight;
int mergetype = MERGE_NEAR;
if( pExpr->pParent && pExpr->pParent->eType==FTSQUERY_NEAR ){
mergetype = MERGE_POS_NEAR;
}
pLeft = pExpr->pLeft;
while( pLeft->eType==FTSQUERY_NEAR ){
pLeft=pLeft->pRight;
}
pRight = pExpr->pRight;
assert( pRight->eType==FTSQUERY_PHRASE );
assert( pLeft->eType==FTSQUERY_PHRASE );
rc = fts3NearMerge(mergetype, pExpr->nNear,
pLeft->pPhrase->nToken, aLeft, nLeft,
pRight->pPhrase->nToken, aRight, nRight,
paOut, pnOut
);
sqlite3_free(aLeft);
break;
}
case FTSQUERY_OR: {
/* Allocate a buffer for the output. The maximum size is the
** sum of the sizes of the two input buffers. The +1 term is
** so that a buffer of zero bytes is never allocated - this can
** cause fts3DoclistMerge() to incorrectly return SQLITE_NOMEM.
*/
char *aBuffer = sqlite3_malloc(nRight+nLeft+1);
rc = fts3DoclistMerge(MERGE_OR, 0, 0, aBuffer, pnOut,
aLeft, nLeft, aRight, nRight, 0
);
*paOut = aBuffer;
sqlite3_free(aLeft);
break;
}
default: {
assert( FTSQUERY_NOT==MERGE_NOT && FTSQUERY_AND==MERGE_AND );
fts3DoclistMerge(pExpr->eType, 0, 0, aLeft, pnOut,
aLeft, nLeft, aRight, nRight, 0
);
*paOut = aLeft;
break;
}
}
}
sqlite3_free(aRight);
}
}
assert( rc==SQLITE_OK || *paOut==0 );
return rc;
}
/*
** This function is called from within xNext() for each row visited by
** an FTS3 query. If evaluating the FTS3 query expression within xFilter()
** was able to determine the exact set of matching rows, this function sets
** *pbRes to true and returns SQLITE_IO immediately.
**
** Otherwise, if evaluating the query expression within xFilter() returned a
** superset of the matching documents instead of an exact set (this happens
** when the query includes very common tokens and it is deemed too expensive to
** load their doclists from disk), this function tests if the current row
** really does match the FTS3 query.
**
** If an error occurs, an SQLite error code is returned. Otherwise, SQLITE_OK
** is returned and *pbRes is set to true if the current row matches the
** FTS3 query (and should be included in the results returned to SQLite), or
** false otherwise.
*/
static int fts3EvalDeferred(
Fts3Cursor *pCsr, /* FTS3 cursor pointing at row to test */
int *pbRes /* OUT: Set to true if row is a match */
){
int rc = SQLITE_OK;
if( pCsr->pDeferred==0 ){
*pbRes = 1;
}else{
rc = fts3CursorSeek(0, pCsr);
if( rc==SQLITE_OK ){
sqlite3Fts3FreeDeferredDoclists(pCsr);
rc = sqlite3Fts3CacheDeferredDoclists(pCsr);
}
if( rc==SQLITE_OK ){
char *a = 0;
int n = 0;
rc = fts3EvalExpr(pCsr, pCsr->pExpr, &a, &n, 0);
assert( n>=0 );
*pbRes = (n>0);
sqlite3_free(a);
}
}
return rc;
}
/*
** Advance the cursor to the next row in the %_content table that
** matches the search criteria. For a MATCH search, this will be
** the next row that matches. For a full-table scan, this will be
** simply the next row in the %_content table. For a docid lookup,
** this routine simply sets the EOF flag.
**
** Return SQLITE_OK if nothing goes wrong. SQLITE_OK is returned
** even if we reach end-of-file. The fts3EofMethod() will be called
** subsequently to determine whether or not an EOF was hit.
*/
static int fts3NextMethod(sqlite3_vtab_cursor *pCursor){
int res;
int rc = SQLITE_OK; /* Return code */
Fts3Cursor *pCsr = (Fts3Cursor *)pCursor;
pCsr->eEvalmode = FTS3_EVAL_NEXT;
do {
if( pCsr->aDoclist==0 ){
if( SQLITE_ROW!=sqlite3_step(pCsr->pStmt) ){
pCsr->isEof = 1;
rc = sqlite3_reset(pCsr->pStmt);
break;
}
pCsr->iPrevId = sqlite3_column_int64(pCsr->pStmt, 0);
}else{
if( pCsr->pNextId>=&pCsr->aDoclist[pCsr->nDoclist] ){
pCsr->isEof = 1;
break;
}
sqlite3_reset(pCsr->pStmt);
fts3GetDeltaVarint(&pCsr->pNextId, &pCsr->iPrevId);
pCsr->isRequireSeek = 1;
pCsr->isMatchinfoNeeded = 1;
}
}while( SQLITE_OK==(rc = fts3EvalDeferred(pCsr, &res)) && res==0 );
return rc;
}
/*
** This is the xFilter interface for the virtual table. See
** the virtual table xFilter method documentation for additional
** information.
**
** If idxNum==FTS3_FULLSCAN_SEARCH then do a full table scan against
** the %_content table.
**
** If idxNum==FTS3_DOCID_SEARCH then do a docid lookup for a single entry
** in the %_content table.
**
** If idxNum>=FTS3_FULLTEXT_SEARCH then use the full text index. The
** column on the left-hand side of the MATCH operator is column
** number idxNum-FTS3_FULLTEXT_SEARCH, 0 indexed. argv[0] is the right-hand
** side of the MATCH operator.
*/
static int fts3FilterMethod(
sqlite3_vtab_cursor *pCursor, /* The cursor used for this query */
int idxNum, /* Strategy index */
const char *idxStr, /* Unused */
int nVal, /* Number of elements in apVal */
sqlite3_value **apVal /* Arguments for the indexing scheme */
){
const char *azSql[] = {
"SELECT %s FROM %Q.'%q_content' AS x WHERE docid = ?", /* non-full-scan */
"SELECT %s FROM %Q.'%q_content' AS x ", /* full-scan */
};
int rc; /* Return code */
char *zSql; /* SQL statement used to access %_content */
Fts3Table *p = (Fts3Table *)pCursor->pVtab;
Fts3Cursor *pCsr = (Fts3Cursor *)pCursor;
UNUSED_PARAMETER(idxStr);
UNUSED_PARAMETER(nVal);
assert( idxNum>=0 && idxNum<=(FTS3_FULLTEXT_SEARCH+p->nColumn) );
assert( nVal==0 || nVal==1 );
assert( (nVal==0)==(idxNum==FTS3_FULLSCAN_SEARCH) );
assert( p->pSegments==0 );
/* In case the cursor has been used before, clear it now. */
sqlite3_finalize(</