blob: 0405fb7b1e56f9aefccebcc274131b973a16b768 [file] [log] [blame]
/* fts2 has a design flaw which can lead to database corruption (see
** below). It is recommended not to use it any longer, instead use
** fts3 (or higher). If you believe that your use of fts2 is safe,
** add -DSQLITE_ENABLE_BROKEN_FTS2=1 to your CFLAGS.
*/
#if (!defined(SQLITE_CORE) || defined(SQLITE_ENABLE_FTS2)) \
&& !defined(SQLITE_ENABLE_BROKEN_FTS2)
#error fts2 has a design flaw and has been deprecated.
#endif
/* The flaw is that fts2 uses the content table's unaliased rowid as
** the unique docid. fts2 embeds the rowid in the index it builds,
** and expects the rowid to not change. The SQLite VACUUM operation
** will renumber such rowids, thereby breaking fts2. If you are using
** fts2 in a system which has disabled VACUUM, then you can continue
** to use it safely. Note that PRAGMA auto_vacuum does NOT disable
** VACUUM, though systems using auto_vacuum are unlikely to invoke
** VACUUM.
**
** Unlike fts1, which is safe across VACUUM if you never delete
** documents, fts2 has a second exposure to this flaw, in the segments
** table. So fts2 should be considered unsafe across VACUUM in all
** cases.
*/
/*
** 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 FTS2 module is being built as an extension
** (in which case SQLITE_CORE is not defined), or
**
** * The FTS2 module is being built into the core of
** SQLite (in which case SQLITE_ENABLE_FTS2 is defined).
*/
/* TODO(shess) Consider exporting this comment to an HTML file or the
** wiki.
*/
/* 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 identical to how sqlite encodes varints (see util.c).
**
**
**** Document lists ****
** A doclist (document list) holds a docid-sorted list of hits for a
** given term. Doclists hold docids, and can optionally associate
** token positions and offsets with docids.
**
** A DL_POSITIONS_OFFSETS doclist is stored like this:
**
** array {
** varint docid;
** array { (position list for column 0)
** varint position; (delta from previous position plus POS_BASE)
** varint startOffset; (delta from previous startOffset)
** varint endOffset; (delta from startOffset)
** }
** array {
** varint POS_COLUMN; (marks start of position list for new column)
** varint column; (index of new column)
** array {
** varint position; (delta from previous position plus POS_BASE)
** varint startOffset;(delta from previous startOffset)
** varint endOffset; (delta from startOffset)
** }
** }
** 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, while an "offset" is a byte offset,
** both based at 0. 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.
**
** A DL_POSITIONS doclist omits the startOffset and endOffset
** information. A DL_DOCIDS doclist omits both the position and
** offset information, becoming an array of varint-encoded docids.
**
** On-disk data is stored as type DL_DEFAULT, so we don't serialize
** the type. Due to how deletion is implemented in the segmentation
** system, on-disk doclists MUST store at least positions.
**
**
**** 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 groups into levels and
** merged in matches. 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.
*/
#if !defined(SQLITE_CORE) || defined(SQLITE_ENABLE_FTS2)
#if defined(SQLITE_ENABLE_FTS2) && !defined(SQLITE_CORE)
# define SQLITE_CORE 1
#endif
#include <assert.h>
#include <stdlib.h>
#include <stdio.h>
#include <string.h>
#include "fts2.h"
#include "fts2_hash.h"
#include "fts2_tokenizer.h"
#include "sqlite3.h"
#include "sqlite3ext.h"
SQLITE_EXTENSION_INIT1
/* TODO(shess) MAN, this thing needs some refactoring. At minimum, it
** would be nice to order the file better, perhaps something along the
** lines of:
**
** - utility functions
** - table setup functions
** - table update functions
** - table query functions
**
** Put the query functions last because they're likely to reference
** typedefs or functions from the table update section.
*/
#if 0
# define TRACE(A) printf A; fflush(stdout)
#else
# define TRACE(A)
#endif
/* It is not safe to call isspace(), tolower(), or isalnum() on
** hi-bit-set characters. This is the same solution used in the
** tokenizer.
*/
/* TODO(shess) The snippet-generation code should be using the
** tokenizer-generated tokens rather than doing its own local
** tokenization.
*/
/* TODO(shess) Is __isascii() a portable version of (c&0x80)==0? */
static int safe_isspace(char c){
return c==' ' || c=='\t' || c=='\n' || c=='\r' || c=='\v' || c=='\f';
}
static int safe_tolower(char c){
return (c>='A' && c<='Z') ? (c - 'A' + 'a') : c;
}
static int safe_isalnum(char c){
return (c>='0' && c<='9') || (c>='A' && c<='Z') || (c>='a' && c<='z');
}
typedef enum DocListType {
DL_DOCIDS, /* docids only */
DL_POSITIONS, /* docids + positions */
DL_POSITIONS_OFFSETS /* docids + positions + offsets */
} DocListType;
/*
** By default, only positions and not offsets are stored in the doclists.
** To change this so that offsets are stored too, compile with
**
** -DDL_DEFAULT=DL_POSITIONS_OFFSETS
**
** If DL_DEFAULT is set to DL_DOCIDS, your table can only be inserted
** into (no deletes or updates).
*/
#ifndef DL_DEFAULT
# define DL_DEFAULT DL_POSITIONS
#endif
enum {
POS_END = 0, /* end of this position list */
POS_COLUMN, /* followed by new column number */
POS_BASE
};
/* MERGE_COUNT controls how often we merge segments (see comment at
** top of file).
*/
#define MERGE_COUNT 16
/* utility functions */
/* CLEAR() and SCRAMBLE() abstract memset() on a pointer to a single
** record to prevent errors of the form:
**
** my_function(SomeType *b){
** memset(b, '\0', sizeof(b)); // sizeof(b)!=sizeof(*b)
** }
*/
/* TODO(shess) Obvious candidates for a header file. */
#define CLEAR(b) memset(b, '\0', sizeof(*(b)))
#ifndef NDEBUG
# define SCRAMBLE(b) memset(b, 0x55, sizeof(*(b)))
#else
# define SCRAMBLE(b)
#endif
/* We may need up to VARINT_MAX bytes to store an encoded 64-bit integer. */
#define VARINT_MAX 10
/* Write a 64-bit variable-length integer to memory starting at p[0].
* The length of data written will be between 1 and VARINT_MAX bytes.
* The number of bytes written is returned. */
static int putVarint(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 <= 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. */
static int getVarint(const char *p, sqlite_int64 *v){
const unsigned char *q = (const unsigned char *) p;
sqlite_uint64 x = 0, y = 1;
while( (*q & 0x80) == 0x80 ){
x += y * (*q++ & 0x7f);
y <<= 7;
if( q - (unsigned char *)p >= VARINT_MAX ){ /* bad data */
assert( 0 );
return 0;
}
}
x += y * (*q++);
*v = (sqlite_int64) x;
return (int) (q - (unsigned char *)p);
}
static int getVarint32(const char *p, int *pi){
sqlite_int64 i;
int ret = getVarint(p, &i);
*pi = (int) i;
assert( *pi==i );
return ret;
}
/*******************************************************************/
/* DataBuffer is used to collect data into a buffer in piecemeal
** fashion. It implements the usual distinction between amount of
** data currently stored (nData) and buffer capacity (nCapacity).
**
** dataBufferInit - create a buffer with given initial capacity.
** dataBufferReset - forget buffer's data, retaining capacity.
** dataBufferDestroy - free buffer's data.
** dataBufferSwap - swap contents of two buffers.
** dataBufferExpand - expand capacity without adding data.
** dataBufferAppend - append data.
** dataBufferAppend2 - append two pieces of data at once.
** dataBufferReplace - replace buffer's data.
*/
typedef struct DataBuffer {
char *pData; /* Pointer to malloc'ed buffer. */
int nCapacity; /* Size of pData buffer. */
int nData; /* End of data loaded into pData. */
} DataBuffer;
static void dataBufferInit(DataBuffer *pBuffer, int nCapacity){
assert( nCapacity>=0 );
pBuffer->nData = 0;
pBuffer->nCapacity = nCapacity;
pBuffer->pData = nCapacity==0 ? NULL : sqlite3_malloc(nCapacity);
}
static void dataBufferReset(DataBuffer *pBuffer){
pBuffer->nData = 0;
}
static void dataBufferDestroy(DataBuffer *pBuffer){
if( pBuffer->pData!=NULL ) sqlite3_free(pBuffer->pData);
SCRAMBLE(pBuffer);
}
static void dataBufferSwap(DataBuffer *pBuffer1, DataBuffer *pBuffer2){
DataBuffer tmp = *pBuffer1;
*pBuffer1 = *pBuffer2;
*pBuffer2 = tmp;
}
static void dataBufferExpand(DataBuffer *pBuffer, int nAddCapacity){
assert( nAddCapacity>0 );
/* TODO(shess) Consider expanding more aggressively. Note that the
** underlying malloc implementation may take care of such things for
** us already.
*/
if( pBuffer->nData+nAddCapacity>pBuffer->nCapacity ){
pBuffer->nCapacity = pBuffer->nData+nAddCapacity;
pBuffer->pData = sqlite3_realloc(pBuffer->pData, pBuffer->nCapacity);
}
}
static void dataBufferAppend(DataBuffer *pBuffer,
const char *pSource, int nSource){
assert( nSource>0 && pSource!=NULL );
dataBufferExpand(pBuffer, nSource);
memcpy(pBuffer->pData+pBuffer->nData, pSource, nSource);
pBuffer->nData += nSource;
}
static void dataBufferAppend2(DataBuffer *pBuffer,
const char *pSource1, int nSource1,
const char *pSource2, int nSource2){
assert( nSource1>0 && pSource1!=NULL );
assert( nSource2>0 && pSource2!=NULL );
dataBufferExpand(pBuffer, nSource1+nSource2);
memcpy(pBuffer->pData+pBuffer->nData, pSource1, nSource1);
memcpy(pBuffer->pData+pBuffer->nData+nSource1, pSource2, nSource2);
pBuffer->nData += nSource1+nSource2;
}
static void dataBufferReplace(DataBuffer *pBuffer,
const char *pSource, int nSource){
dataBufferReset(pBuffer);
dataBufferAppend(pBuffer, pSource, nSource);
}
/* StringBuffer is a null-terminated version of DataBuffer. */
typedef struct StringBuffer {
DataBuffer b; /* Includes null terminator. */
} StringBuffer;
static void initStringBuffer(StringBuffer *sb){
dataBufferInit(&sb->b, 100);
dataBufferReplace(&sb->b, "", 1);
}
static int stringBufferLength(StringBuffer *sb){
return sb->b.nData-1;
}
static char *stringBufferData(StringBuffer *sb){
return sb->b.pData;
}
static void stringBufferDestroy(StringBuffer *sb){
dataBufferDestroy(&sb->b);
}
static void nappend(StringBuffer *sb, const char *zFrom, int nFrom){
assert( sb->b.nData>0 );
if( nFrom>0 ){
sb->b.nData--;
dataBufferAppend2(&sb->b, zFrom, nFrom, "", 1);
}
}
static void append(StringBuffer *sb, const char *zFrom){
nappend(sb, zFrom, strlen(zFrom));
}
/* Append a list of strings separated by commas. */
static void appendList(StringBuffer *sb, int nString, char **azString){
int i;
for(i=0; i<nString; ++i){
if( i>0 ) append(sb, ", ");
append(sb, azString[i]);
}
}
static int endsInWhiteSpace(StringBuffer *p){
return stringBufferLength(p)>0 &&
safe_isspace(stringBufferData(p)[stringBufferLength(p)-1]);
}
/* If the StringBuffer ends in something other than white space, add a
** single space character to the end.
*/
static void appendWhiteSpace(StringBuffer *p){
if( stringBufferLength(p)==0 ) return;
if( !endsInWhiteSpace(p) ) append(p, " ");
}
/* Remove white space from the end of the StringBuffer */
static void trimWhiteSpace(StringBuffer *p){
while( endsInWhiteSpace(p) ){
p->b.pData[--p->b.nData-1] = '\0';
}
}
/*******************************************************************/
/* DLReader is used to read document elements from a doclist. The
** current docid is cached, so dlrDocid() is fast. DLReader does not
** own the doclist buffer.
**
** dlrAtEnd - true if there's no more data to read.
** dlrDocid - docid of current document.
** dlrDocData - doclist data for current document (including docid).
** dlrDocDataBytes - length of same.
** dlrAllDataBytes - length of all remaining data.
** dlrPosData - position data for current document.
** dlrPosDataLen - length of pos data for current document (incl POS_END).
** dlrStep - step to current document.
** dlrInit - initial for doclist of given type against given data.
** dlrDestroy - clean up.
**
** Expected usage is something like:
**
** DLReader reader;
** dlrInit(&reader, pData, nData);
** while( !dlrAtEnd(&reader) ){
** // calls to dlrDocid() and kin.
** dlrStep(&reader);
** }
** dlrDestroy(&reader);
*/
typedef struct DLReader {
DocListType iType;
const char *pData;
int nData;
sqlite_int64 iDocid;
int nElement;
} DLReader;
static int dlrAtEnd(DLReader *pReader){
assert( pReader->nData>=0 );
return pReader->nData==0;
}
static sqlite_int64 dlrDocid(DLReader *pReader){
assert( !dlrAtEnd(pReader) );
return pReader->iDocid;
}
static const char *dlrDocData(DLReader *pReader){
assert( !dlrAtEnd(pReader) );
return pReader->pData;
}
static int dlrDocDataBytes(DLReader *pReader){
assert( !dlrAtEnd(pReader) );
return pReader->nElement;
}
static int dlrAllDataBytes(DLReader *pReader){
assert( !dlrAtEnd(pReader) );
return pReader->nData;
}
/* TODO(shess) Consider adding a field to track iDocid varint length
** to make these two functions faster. This might matter (a tiny bit)
** for queries.
*/
static const char *dlrPosData(DLReader *pReader){
sqlite_int64 iDummy;
int n = getVarint(pReader->pData, &iDummy);
assert( !dlrAtEnd(pReader) );
return pReader->pData+n;
}
static int dlrPosDataLen(DLReader *pReader){
sqlite_int64 iDummy;
int n = getVarint(pReader->pData, &iDummy);
assert( !dlrAtEnd(pReader) );
return pReader->nElement-n;
}
static void dlrStep(DLReader *pReader){
assert( !dlrAtEnd(pReader) );
/* Skip past current doclist element. */
assert( pReader->nElement<=pReader->nData );
pReader->pData += pReader->nElement;
pReader->nData -= pReader->nElement;
/* If there is more data, read the next doclist element. */
if( pReader->nData!=0 ){
sqlite_int64 iDocidDelta;
int iDummy, n = getVarint(pReader->pData, &iDocidDelta);
pReader->iDocid += iDocidDelta;
if( pReader->iType>=DL_POSITIONS ){
assert( n<pReader->nData );
while( 1 ){
n += getVarint32(pReader->pData+n, &iDummy);
assert( n<=pReader->nData );
if( iDummy==POS_END ) break;
if( iDummy==POS_COLUMN ){
n += getVarint32(pReader->pData+n, &iDummy);
assert( n<pReader->nData );
}else if( pReader->iType==DL_POSITIONS_OFFSETS ){
n += getVarint32(pReader->pData+n, &iDummy);
n += getVarint32(pReader->pData+n, &iDummy);
assert( n<pReader->nData );
}
}
}
pReader->nElement = n;
assert( pReader->nElement<=pReader->nData );
}
}
static void dlrInit(DLReader *pReader, DocListType iType,
const char *pData, int nData){
assert( pData!=NULL && nData!=0 );
pReader->iType = iType;
pReader->pData = pData;
pReader->nData = nData;
pReader->nElement = 0;
pReader->iDocid = 0;
/* Load the first element's data. There must be a first element. */
dlrStep(pReader);
}
static void dlrDestroy(DLReader *pReader){
SCRAMBLE(pReader);
}
#ifndef NDEBUG
/* Verify that the doclist can be validly decoded. Also returns the
** last docid found because it is convenient in other assertions for
** DLWriter.
*/
static void docListValidate(DocListType iType, const char *pData, int nData,
sqlite_int64 *pLastDocid){
sqlite_int64 iPrevDocid = 0;
assert( nData>0 );
assert( pData!=0 );
assert( pData+nData>pData );
while( nData!=0 ){
sqlite_int64 iDocidDelta;
int n = getVarint(pData, &iDocidDelta);
iPrevDocid += iDocidDelta;
if( iType>DL_DOCIDS ){
int iDummy;
while( 1 ){
n += getVarint32(pData+n, &iDummy);
if( iDummy==POS_END ) break;
if( iDummy==POS_COLUMN ){
n += getVarint32(pData+n, &iDummy);
}else if( iType>DL_POSITIONS ){
n += getVarint32(pData+n, &iDummy);
n += getVarint32(pData+n, &iDummy);
}
assert( n<=nData );
}
}
assert( n<=nData );
pData += n;
nData -= n;
}
if( pLastDocid ) *pLastDocid = iPrevDocid;
}
#define ASSERT_VALID_DOCLIST(i, p, n, o) docListValidate(i, p, n, o)
#else
#define ASSERT_VALID_DOCLIST(i, p, n, o) assert( 1 )
#endif
/*******************************************************************/
/* DLWriter is used to write doclist data to a DataBuffer. DLWriter
** always appends to the buffer and does not own it.
**
** dlwInit - initialize to write a given type doclistto a buffer.
** dlwDestroy - clear the writer's memory. Does not free buffer.
** dlwAppend - append raw doclist data to buffer.
** dlwCopy - copy next doclist from reader to writer.
** dlwAdd - construct doclist element and append to buffer.
** Only apply dlwAdd() to DL_DOCIDS doclists (else use PLWriter).
*/
typedef struct DLWriter {
DocListType iType;
DataBuffer *b;
sqlite_int64 iPrevDocid;
#ifndef NDEBUG
int has_iPrevDocid;
#endif
} DLWriter;
static void dlwInit(DLWriter *pWriter, DocListType iType, DataBuffer *b){
pWriter->b = b;
pWriter->iType = iType;
pWriter->iPrevDocid = 0;
#ifndef NDEBUG
pWriter->has_iPrevDocid = 0;
#endif
}
static void dlwDestroy(DLWriter *pWriter){
SCRAMBLE(pWriter);
}
/* iFirstDocid is the first docid in the doclist in pData. It is
** needed because pData may point within a larger doclist, in which
** case the first item would be delta-encoded.
**
** iLastDocid is the final docid in the doclist in pData. It is
** needed to create the new iPrevDocid for future delta-encoding. The
** code could decode the passed doclist to recreate iLastDocid, but
** the only current user (docListMerge) already has decoded this
** information.
*/
/* TODO(shess) This has become just a helper for docListMerge.
** Consider a refactor to make this cleaner.
*/
static void dlwAppend(DLWriter *pWriter,
const char *pData, int nData,
sqlite_int64 iFirstDocid, sqlite_int64 iLastDocid){
sqlite_int64 iDocid = 0;
char c[VARINT_MAX];
int nFirstOld, nFirstNew; /* Old and new varint len of first docid. */
#ifndef NDEBUG
sqlite_int64 iLastDocidDelta;
#endif
/* Recode the initial docid as delta from iPrevDocid. */
nFirstOld = getVarint(pData, &iDocid);
assert( nFirstOld<nData || (nFirstOld==nData && pWriter->iType==DL_DOCIDS) );
nFirstNew = putVarint(c, iFirstDocid-pWriter->iPrevDocid);
/* Verify that the incoming doclist is valid AND that it ends with
** the expected docid. This is essential because we'll trust this
** docid in future delta-encoding.
*/
ASSERT_VALID_DOCLIST(pWriter->iType, pData, nData, &iLastDocidDelta);
assert( iLastDocid==iFirstDocid-iDocid+iLastDocidDelta );
/* Append recoded initial docid and everything else. Rest of docids
** should have been delta-encoded from previous initial docid.
*/
if( nFirstOld<nData ){
dataBufferAppend2(pWriter->b, c, nFirstNew,
pData+nFirstOld, nData-nFirstOld);
}else{
dataBufferAppend(pWriter->b, c, nFirstNew);
}
pWriter->iPrevDocid = iLastDocid;
}
static void dlwCopy(DLWriter *pWriter, DLReader *pReader){
dlwAppend(pWriter, dlrDocData(pReader), dlrDocDataBytes(pReader),
dlrDocid(pReader), dlrDocid(pReader));
}
static void dlwAdd(DLWriter *pWriter, sqlite_int64 iDocid){
char c[VARINT_MAX];
int n = putVarint(c, iDocid-pWriter->iPrevDocid);
/* Docids must ascend. */
assert( !pWriter->has_iPrevDocid || iDocid>pWriter->iPrevDocid );
assert( pWriter->iType==DL_DOCIDS );
dataBufferAppend(pWriter->b, c, n);
pWriter->iPrevDocid = iDocid;
#ifndef NDEBUG
pWriter->has_iPrevDocid = 1;
#endif
}
/*******************************************************************/
/* PLReader is used to read data from a document's position list. As
** the caller steps through the list, data is cached so that varints
** only need to be decoded once.
**
** plrInit, plrDestroy - create/destroy a reader.
** plrColumn, plrPosition, plrStartOffset, plrEndOffset - accessors
** plrAtEnd - at end of stream, only call plrDestroy once true.
** plrStep - step to the next element.
*/
typedef struct PLReader {
/* These refer to the next position's data. nData will reach 0 when
** reading the last position, so plrStep() signals EOF by setting
** pData to NULL.
*/
const char *pData;
int nData;
DocListType iType;
int iColumn; /* the last column read */
int iPosition; /* the last position read */
int iStartOffset; /* the last start offset read */
int iEndOffset; /* the last end offset read */
} PLReader;
static int plrAtEnd(PLReader *pReader){
return pReader->pData==NULL;
}
static int plrColumn(PLReader *pReader){
assert( !plrAtEnd(pReader) );
return pReader->iColumn;
}
static int plrPosition(PLReader *pReader){
assert( !plrAtEnd(pReader) );
return pReader->iPosition;
}
static int plrStartOffset(PLReader *pReader){
assert( !plrAtEnd(pReader) );
return pReader->iStartOffset;
}
static int plrEndOffset(PLReader *pReader){
assert( !plrAtEnd(pReader) );
return pReader->iEndOffset;
}
static void plrStep(PLReader *pReader){
int i, n;
assert( !plrAtEnd(pReader) );
if( pReader->nData==0 ){
pReader->pData = NULL;
return;
}
n = getVarint32(pReader->pData, &i);
if( i==POS_COLUMN ){
n += getVarint32(pReader->pData+n, &pReader->iColumn);
pReader->iPosition = 0;
pReader->iStartOffset = 0;
n += getVarint32(pReader->pData+n, &i);
}
/* Should never see adjacent column changes. */
assert( i!=POS_COLUMN );
if( i==POS_END ){
pReader->nData = 0;
pReader->pData = NULL;
return;
}
pReader->iPosition += i-POS_BASE;
if( pReader->iType==DL_POSITIONS_OFFSETS ){
n += getVarint32(pReader->pData+n, &i);
pReader->iStartOffset += i;
n += getVarint32(pReader->pData+n, &i);
pReader->iEndOffset = pReader->iStartOffset+i;
}
assert( n<=pReader->nData );
pReader->pData += n;
pReader->nData -= n;
}
static void plrInit(PLReader *pReader, DLReader *pDLReader){
pReader->pData = dlrPosData(pDLReader);
pReader->nData = dlrPosDataLen(pDLReader);
pReader->iType = pDLReader->iType;
pReader->iColumn = 0;
pReader->iPosition = 0;
pReader->iStartOffset = 0;
pReader->iEndOffset = 0;
plrStep(pReader);
}
static void plrDestroy(PLReader *pReader){
SCRAMBLE(pReader);
}
/*******************************************************************/
/* PLWriter is used in constructing a document's position list. As a
** convenience, if iType is DL_DOCIDS, PLWriter becomes a no-op.
** PLWriter writes to the associated DLWriter's buffer.
**
** plwInit - init for writing a document's poslist.
** plwDestroy - clear a writer.
** plwAdd - append position and offset information.
** plwCopy - copy next position's data from reader to writer.
** plwTerminate - add any necessary doclist terminator.
**
** Calling plwAdd() after plwTerminate() may result in a corrupt
** doclist.
*/
/* TODO(shess) Until we've written the second item, we can cache the
** first item's information. Then we'd have three states:
**
** - initialized with docid, no positions.
** - docid and one position.
** - docid and multiple positions.
**
** Only the last state needs to actually write to dlw->b, which would
** be an improvement in the DLCollector case.
*/
typedef struct PLWriter {
DLWriter *dlw;
int iColumn; /* the last column written */
int iPos; /* the last position written */
int iOffset; /* the last start offset written */
} PLWriter;
/* TODO(shess) In the case where the parent is reading these values
** from a PLReader, we could optimize to a copy if that PLReader has
** the same type as pWriter.
*/
static void plwAdd(PLWriter *pWriter, int iColumn, int iPos,
int iStartOffset, int iEndOffset){
/* Worst-case space for POS_COLUMN, iColumn, iPosDelta,
** iStartOffsetDelta, and iEndOffsetDelta.
*/
char c[5*VARINT_MAX];
int n = 0;
/* Ban plwAdd() after plwTerminate(). */
assert( pWriter->iPos!=-1 );
if( pWriter->dlw->iType==DL_DOCIDS ) return;
if( iColumn!=pWriter->iColumn ){
n += putVarint(c+n, POS_COLUMN);
n += putVarint(c+n, iColumn);
pWriter->iColumn = iColumn;
pWriter->iPos = 0;
pWriter->iOffset = 0;
}
assert( iPos>=pWriter->iPos );
n += putVarint(c+n, POS_BASE+(iPos-pWriter->iPos));
pWriter->iPos = iPos;
if( pWriter->dlw->iType==DL_POSITIONS_OFFSETS ){
assert( iStartOffset>=pWriter->iOffset );
n += putVarint(c+n, iStartOffset-pWriter->iOffset);
pWriter->iOffset = iStartOffset;
assert( iEndOffset>=iStartOffset );
n += putVarint(c+n, iEndOffset-iStartOffset);
}
dataBufferAppend(pWriter->dlw->b, c, n);
}
static void plwCopy(PLWriter *pWriter, PLReader *pReader){
plwAdd(pWriter, plrColumn(pReader), plrPosition(pReader),
plrStartOffset(pReader), plrEndOffset(pReader));
}
static void plwInit(PLWriter *pWriter, DLWriter *dlw, sqlite_int64 iDocid){
char c[VARINT_MAX];
int n;
pWriter->dlw = dlw;
/* Docids must ascend. */
assert( !pWriter->dlw->has_iPrevDocid || iDocid>pWriter->dlw->iPrevDocid );
n = putVarint(c, iDocid-pWriter->dlw->iPrevDocid);
dataBufferAppend(pWriter->dlw->b, c, n);
pWriter->dlw->iPrevDocid = iDocid;
#ifndef NDEBUG
pWriter->dlw->has_iPrevDocid = 1;
#endif
pWriter->iColumn = 0;
pWriter->iPos = 0;
pWriter->iOffset = 0;
}
/* TODO(shess) Should plwDestroy() also terminate the doclist? But
** then plwDestroy() would no longer be just a destructor, it would
** also be doing work, which isn't consistent with the overall idiom.
** Another option would be for plwAdd() to always append any necessary
** terminator, so that the output is always correct. But that would
** add incremental work to the common case with the only benefit being
** API elegance. Punt for now.
*/
static void plwTerminate(PLWriter *pWriter){
if( pWriter->dlw->iType>DL_DOCIDS ){
char c[VARINT_MAX];
int n = putVarint(c, POS_END);
dataBufferAppend(pWriter->dlw->b, c, n);
}
#ifndef NDEBUG
/* Mark as terminated for assert in plwAdd(). */
pWriter->iPos = -1;
#endif
}
static void plwDestroy(PLWriter *pWriter){
SCRAMBLE(pWriter);
}
/*******************************************************************/
/* DLCollector wraps PLWriter and DLWriter to provide a
** dynamically-allocated doclist area to use during tokenization.
**
** dlcNew - malloc up and initialize a collector.
** dlcDelete - destroy a collector and all contained items.
** dlcAddPos - append position and offset information.
** dlcAddDoclist - add the collected doclist to the given buffer.
** dlcNext - terminate the current document and open another.
*/
typedef struct DLCollector {
DataBuffer b;
DLWriter dlw;
PLWriter plw;
} DLCollector;
/* TODO(shess) This could also be done by calling plwTerminate() and
** dataBufferAppend(). I tried that, expecting nominal performance
** differences, but it seemed to pretty reliably be worth 1% to code
** it this way. I suspect it is the incremental malloc overhead (some
** percentage of the plwTerminate() calls will cause a realloc), so
** this might be worth revisiting if the DataBuffer implementation
** changes.
*/
static void dlcAddDoclist(DLCollector *pCollector, DataBuffer *b){
if( pCollector->dlw.iType>DL_DOCIDS ){
char c[VARINT_MAX];
int n = putVarint(c, POS_END);
dataBufferAppend2(b, pCollector->b.pData, pCollector->b.nData, c, n);
}else{
dataBufferAppend(b, pCollector->b.pData, pCollector->b.nData);
}
}
static void dlcNext(DLCollector *pCollector, sqlite_int64 iDocid){
plwTerminate(&pCollector->plw);
plwDestroy(&pCollector->plw);
plwInit(&pCollector->plw, &pCollector->dlw, iDocid);
}
static void dlcAddPos(DLCollector *pCollector, int iColumn, int iPos,
int iStartOffset, int iEndOffset){
plwAdd(&pCollector->plw, iColumn, iPos, iStartOffset, iEndOffset);
}
static DLCollector *dlcNew(sqlite_int64 iDocid, DocListType iType){
DLCollector *pCollector = sqlite3_malloc(sizeof(DLCollector));
dataBufferInit(&pCollector->b, 0);
dlwInit(&pCollector->dlw, iType, &pCollector->b);
plwInit(&pCollector->plw, &pCollector->dlw, iDocid);
return pCollector;
}
static void dlcDelete(DLCollector *pCollector){
plwDestroy(&pCollector->plw);
dlwDestroy(&pCollector->dlw);
dataBufferDestroy(&pCollector->b);
SCRAMBLE(pCollector);
sqlite3_free(pCollector);
}
/* Copy the doclist data of iType in pData/nData into *out, trimming
** unnecessary data as we go. Only columns matching iColumn are
** copied, all columns copied if iColumn is -1. Elements with no
** matching columns are dropped. The output is an iOutType doclist.
*/
/* NOTE(shess) This code is only valid after all doclists are merged.
** If this is run before merges, then doclist items which represent
** deletion will be trimmed, and will thus not effect a deletion
** during the merge.
*/
static void docListTrim(DocListType iType, const char *pData, int nData,
int iColumn, DocListType iOutType, DataBuffer *out){
DLReader dlReader;
DLWriter dlWriter;
assert( iOutType<=iType );
dlrInit(&dlReader, iType, pData, nData);
dlwInit(&dlWriter, iOutType, out);
while( !dlrAtEnd(&dlReader) ){
PLReader plReader;
PLWriter plWriter;
int match = 0;
plrInit(&plReader, &dlReader);
while( !plrAtEnd(&plReader) ){
if( iColumn==-1 || plrColumn(&plReader)==iColumn ){
if( !match ){
plwInit(&plWriter, &dlWriter, dlrDocid(&dlReader));
match = 1;
}
plwAdd(&plWriter, plrColumn(&plReader), plrPosition(&plReader),
plrStartOffset(&plReader), plrEndOffset(&plReader));
}
plrStep(&plReader);
}
if( match ){
plwTerminate(&plWriter);
plwDestroy(&plWriter);
}
plrDestroy(&plReader);
dlrStep(&dlReader);
}
dlwDestroy(&dlWriter);
dlrDestroy(&dlReader);
}
/* Used by docListMerge() to keep doclists in the ascending order by
** docid, then ascending order by age (so the newest comes first).
*/
typedef struct OrderedDLReader {
DLReader *pReader;
/* TODO(shess) If we assume that docListMerge pReaders is ordered by
** age (which we do), then we could use pReader comparisons to break
** ties.
*/
int idx;
} OrderedDLReader;
/* Order eof to end, then by docid asc, idx desc. */
static int orderedDLReaderCmp(OrderedDLReader *r1, OrderedDLReader *r2){
if( dlrAtEnd(r1->pReader) ){
if( dlrAtEnd(r2->pReader) ) return 0; /* Both atEnd(). */
return 1; /* Only r1 atEnd(). */
}
if( dlrAtEnd(r2->pReader) ) return -1; /* Only r2 atEnd(). */
if( dlrDocid(r1->pReader)<dlrDocid(r2->pReader) ) return -1;
if( dlrDocid(r1->pReader)>dlrDocid(r2->pReader) ) return 1;
/* Descending on idx. */
return r2->idx-r1->idx;
}
/* Bubble p[0] to appropriate place in p[1..n-1]. Assumes that
** p[1..n-1] is already sorted.
*/
/* TODO(shess) Is this frequent enough to warrant a binary search?
** Before implementing that, instrument the code to check. In most
** current usage, I expect that p[0] will be less than p[1] a very
** high proportion of the time.
*/
static void orderedDLReaderReorder(OrderedDLReader *p, int n){
while( n>1 && orderedDLReaderCmp(p, p+1)>0 ){
OrderedDLReader tmp = p[0];
p[0] = p[1];
p[1] = tmp;
n--;
p++;
}
}
/* Given an array of doclist readers, merge their doclist elements
** into out in sorted order (by docid), dropping elements from older
** readers when there is a duplicate docid. pReaders is assumed to be
** ordered by age, oldest first.
*/
/* TODO(shess) nReaders must be <= MERGE_COUNT. This should probably
** be fixed.
*/
static void docListMerge(DataBuffer *out,
DLReader *pReaders, int nReaders){
OrderedDLReader readers[MERGE_COUNT];
DLWriter writer;
int i, n;
const char *pStart = 0;
int nStart = 0;
sqlite_int64 iFirstDocid = 0, iLastDocid = 0;
assert( nReaders>0 );
if( nReaders==1 ){
dataBufferAppend(out, dlrDocData(pReaders), dlrAllDataBytes(pReaders));
return;
}
assert( nReaders<=MERGE_COUNT );
n = 0;
for(i=0; i<nReaders; i++){
assert( pReaders[i].iType==pReaders[0].iType );
readers[i].pReader = pReaders+i;
readers[i].idx = i;
n += dlrAllDataBytes(&pReaders[i]);
}
/* Conservatively size output to sum of inputs. Output should end
** up strictly smaller than input.
*/
dataBufferExpand(out, n);
/* Get the readers into sorted order. */
while( i-->0 ){
orderedDLReaderReorder(readers+i, nReaders-i);
}
dlwInit(&writer, pReaders[0].iType, out);
while( !dlrAtEnd(readers[0].pReader) ){
sqlite_int64 iDocid = dlrDocid(readers[0].pReader);
/* If this is a continuation of the current buffer to copy, extend
** that buffer. memcpy() seems to be more efficient if it has a
** lots of data to copy.
*/
if( dlrDocData(readers[0].pReader)==pStart+nStart ){
nStart += dlrDocDataBytes(readers[0].pReader);
}else{
if( pStart!=0 ){
dlwAppend(&writer, pStart, nStart, iFirstDocid, iLastDocid);
}
pStart = dlrDocData(readers[0].pReader);
nStart = dlrDocDataBytes(readers[0].pReader);
iFirstDocid = iDocid;
}
iLastDocid = iDocid;
dlrStep(readers[0].pReader);
/* Drop all of the older elements with the same docid. */
for(i=1; i<nReaders &&
!dlrAtEnd(readers[i].pReader) &&
dlrDocid(readers[i].pReader)==iDocid; i++){
dlrStep(readers[i].pReader);
}
/* Get the readers back into order. */
while( i-->0 ){
orderedDLReaderReorder(readers+i, nReaders-i);
}
}
/* Copy over any remaining elements. */
if( nStart>0 ) dlwAppend(&writer, pStart, nStart, iFirstDocid, iLastDocid);
dlwDestroy(&writer);
}
/* Helper function for posListUnion(). Compares the current position
** between left and right, returning as standard C idiom of <0 if
** left<right, >0 if left>right, and 0 if left==right. "End" always
** compares greater.
*/
static int posListCmp(PLReader *pLeft, PLReader *pRight){
assert( pLeft->iType==pRight->iType );
if( pLeft->iType==DL_DOCIDS ) return 0;
if( plrAtEnd(pLeft) ) return plrAtEnd(pRight) ? 0 : 1;
if( plrAtEnd(pRight) ) return -1;
if( plrColumn(pLeft)<plrColumn(pRight) ) return -1;
if( plrColumn(pLeft)>plrColumn(pRight) ) return 1;
if( plrPosition(pLeft)<plrPosition(pRight) ) return -1;
if( plrPosition(pLeft)>plrPosition(pRight) ) return 1;
if( pLeft->iType==DL_POSITIONS ) return 0;
if( plrStartOffset(pLeft)<plrStartOffset(pRight) ) return -1;
if( plrStartOffset(pLeft)>plrStartOffset(pRight) ) return 1;
if( plrEndOffset(pLeft)<plrEndOffset(pRight) ) return -1;
if( plrEndOffset(pLeft)>plrEndOffset(pRight) ) return 1;
return 0;
}
/* Write the union of position lists in pLeft and pRight to pOut.
** "Union" in this case meaning "All unique position tuples". Should
** work with any doclist type, though both inputs and the output
** should be the same type.
*/
static void posListUnion(DLReader *pLeft, DLReader *pRight, DLWriter *pOut){
PLReader left, right;
PLWriter writer;
assert( dlrDocid(pLeft)==dlrDocid(pRight) );
assert( pLeft->iType==pRight->iType );
assert( pLeft->iType==pOut->iType );
plrInit(&left, pLeft);
plrInit(&right, pRight);
plwInit(&writer, pOut, dlrDocid(pLeft));
while( !plrAtEnd(&left) || !plrAtEnd(&right) ){
int c = posListCmp(&left, &right);
if( c<0 ){
plwCopy(&writer, &left);
plrStep(&left);
}else if( c>0 ){
plwCopy(&writer, &right);
plrStep(&right);
}else{
plwCopy(&writer, &left);
plrStep(&left);
plrStep(&right);
}
}
plwTerminate(&writer);
plwDestroy(&writer);
plrDestroy(&left);
plrDestroy(&right);
}
/* Write the union of doclists in pLeft and pRight to pOut. For
** docids in common between the inputs, the union of the position
** lists is written. Inputs and outputs are always type DL_DEFAULT.
*/
static void docListUnion(
const char *pLeft, int nLeft,
const char *pRight, int nRight,
DataBuffer *pOut /* Write the combined doclist here */
){
DLReader left, right;
DLWriter writer;
if( nLeft==0 ){
if( nRight!=0) dataBufferAppend(pOut, pRight, nRight);
return;
}
if( nRight==0 ){
dataBufferAppend(pOut, pLeft, nLeft);
return;
}
dlrInit(&left, DL_DEFAULT, pLeft, nLeft);
dlrInit(&right, DL_DEFAULT, pRight, nRight);
dlwInit(&writer, DL_DEFAULT, pOut);
while( !dlrAtEnd(&left) || !dlrAtEnd(&right) ){
if( dlrAtEnd(&right) ){
dlwCopy(&writer, &left);
dlrStep(&left);
}else if( dlrAtEnd(&left) ){
dlwCopy(&writer, &right);
dlrStep(&right);
}else if( dlrDocid(&left)<dlrDocid(&right) ){
dlwCopy(&writer, &left);
dlrStep(&left);
}else if( dlrDocid(&left)>dlrDocid(&right) ){
dlwCopy(&writer, &right);
dlrStep(&right);
}else{
posListUnion(&left, &right, &writer);
dlrStep(&left);
dlrStep(&right);
}
}
dlrDestroy(&left);
dlrDestroy(&right);
dlwDestroy(&writer);
}
/* pLeft and pRight are DLReaders positioned to the same docid.
**
** If there are no instances in pLeft or pRight where the position
** of pLeft is one less than the position of pRight, then this
** routine adds nothing to pOut.
**
** If there are one or more instances where positions from pLeft
** are exactly one less than positions from pRight, then add a new
** document record to pOut. If pOut wants to hold positions, then
** include the positions from pRight that are one more than a
** position in pLeft. In other words: pRight.iPos==pLeft.iPos+1.
*/
static void posListPhraseMerge(DLReader *pLeft, DLReader *pRight,
DLWriter *pOut){
PLReader left, right;
PLWriter writer;
int match = 0;
assert( dlrDocid(pLeft)==dlrDocid(pRight) );
assert( pOut->iType!=DL_POSITIONS_OFFSETS );
plrInit(&left, pLeft);
plrInit(&right, pRight);
while( !plrAtEnd(&left) && !plrAtEnd(&right) ){
if( plrColumn(&left)<plrColumn(&right) ){
plrStep(&left);
}else if( plrColumn(&left)>plrColumn(&right) ){
plrStep(&right);
}else if( plrPosition(&left)+1<plrPosition(&right) ){
plrStep(&left);
}else if( plrPosition(&left)+1>plrPosition(&right) ){
plrStep(&right);
}else{
if( !match ){
plwInit(&writer, pOut, dlrDocid(pLeft));
match = 1;
}
plwAdd(&writer, plrColumn(&right), plrPosition(&right), 0, 0);
plrStep(&left);
plrStep(&right);
}
}
if( match ){
plwTerminate(&writer);
plwDestroy(&writer);
}
plrDestroy(&left);
plrDestroy(&right);
}
/* We have two doclists with positions: pLeft and pRight.
** Write the phrase intersection of these two doclists into pOut.
**
** A phrase intersection means that two documents only match
** if pLeft.iPos+1==pRight.iPos.
**
** iType controls the type of data written to pOut. If iType is
** DL_POSITIONS, the positions are those from pRight.
*/
static void docListPhraseMerge(
const char *pLeft, int nLeft,
const char *pRight, int nRight,
DocListType iType,
DataBuffer *pOut /* Write the combined doclist here */
){
DLReader left, right;
DLWriter writer;
if( nLeft==0 || nRight==0 ) return;
assert( iType!=DL_POSITIONS_OFFSETS );
dlrInit(&left, DL_POSITIONS, pLeft, nLeft);
dlrInit(&right, DL_POSITIONS, pRight, nRight);
dlwInit(&writer, iType, pOut);
while( !dlrAtEnd(&left) && !dlrAtEnd(&right) ){
if( dlrDocid(&left)<dlrDocid(&right) ){
dlrStep(&left);
}else if( dlrDocid(&right)<dlrDocid(&left) ){
dlrStep(&right);
}else{
posListPhraseMerge(&left, &right, &writer);
dlrStep(&left);
dlrStep(&right);
}
}
dlrDestroy(&left);
dlrDestroy(&right);
dlwDestroy(&writer);
}
/* We have two DL_DOCIDS doclists: pLeft and pRight.
** Write the intersection of these two doclists into pOut as a
** DL_DOCIDS doclist.
*/
static void docListAndMerge(
const char *pLeft, int nLeft,
const char *pRight, int nRight,
DataBuffer *pOut /* Write the combined doclist here */
){
DLReader left, right;
DLWriter writer;
if( nLeft==0 || nRight==0 ) return;
dlrInit(&left, DL_DOCIDS, pLeft, nLeft);
dlrInit(&right, DL_DOCIDS, pRight, nRight);
dlwInit(&writer, DL_DOCIDS, pOut);
while( !dlrAtEnd(&left) && !dlrAtEnd(&right) ){
if( dlrDocid(&left)<dlrDocid(&right) ){
dlrStep(&left);
}else if( dlrDocid(&right)<dlrDocid(&left) ){
dlrStep(&right);
}else{
dlwAdd(&writer, dlrDocid(&left));
dlrStep(&left);
dlrStep(&right);
}
}
dlrDestroy(&left);
dlrDestroy(&right);
dlwDestroy(&writer);
}
/* We have two DL_DOCIDS doclists: pLeft and pRight.
** Write the union of these two doclists into pOut as a
** DL_DOCIDS doclist.
*/
static void docListOrMerge(
const char *pLeft, int nLeft,
const char *pRight, int nRight,
DataBuffer *pOut /* Write the combined doclist here */
){
DLReader left, right;
DLWriter writer;
if( nLeft==0 ){
if( nRight!=0 ) dataBufferAppend(pOut, pRight, nRight);
return;
}
if( nRight==0 ){
dataBufferAppend(pOut, pLeft, nLeft);
return;
}
dlrInit(&left, DL_DOCIDS, pLeft, nLeft);
dlrInit(&right, DL_DOCIDS, pRight, nRight);
dlwInit(&writer, DL_DOCIDS, pOut);
while( !dlrAtEnd(&left) || !dlrAtEnd(&right) ){
if( dlrAtEnd(&right) ){
dlwAdd(&writer, dlrDocid(&left));
dlrStep(&left);
}else if( dlrAtEnd(&left) ){
dlwAdd(&writer, dlrDocid(&right));
dlrStep(&right);
}else if( dlrDocid(&left)<dlrDocid(&right) ){
dlwAdd(&writer, dlrDocid(&left));
dlrStep(&left);
}else if( dlrDocid(&right)<dlrDocid(&left) ){
dlwAdd(&writer, dlrDocid(&right));
dlrStep(&right);
}else{
dlwAdd(&writer, dlrDocid(&left));
dlrStep(&left);
dlrStep(&right);
}
}
dlrDestroy(&left);
dlrDestroy(&right);
dlwDestroy(&writer);
}
/* We have two DL_DOCIDS doclists: pLeft and pRight.
** Write into pOut as DL_DOCIDS doclist containing all documents that
** occur in pLeft but not in pRight.
*/
static void docListExceptMerge(
const char *pLeft, int nLeft,
const char *pRight, int nRight,
DataBuffer *pOut /* Write the combined doclist here */
){
DLReader left, right;
DLWriter writer;
if( nLeft==0 ) return;
if( nRight==0 ){
dataBufferAppend(pOut, pLeft, nLeft);
return;
}
dlrInit(&left, DL_DOCIDS, pLeft, nLeft);
dlrInit(&right, DL_DOCIDS, pRight, nRight);
dlwInit(&writer, DL_DOCIDS, pOut);
while( !dlrAtEnd(&left) ){
while( !dlrAtEnd(&right) && dlrDocid(&right)<dlrDocid(&left) ){
dlrStep(&right);
}
if( dlrAtEnd(&right) || dlrDocid(&left)<dlrDocid(&right) ){
dlwAdd(&writer, dlrDocid(&left));
}
dlrStep(&left);
}
dlrDestroy(&left);
dlrDestroy(&right);
dlwDestroy(&writer);
}
static char *string_dup_n(const char *s, int n){
char *str = sqlite3_malloc(n + 1);
memcpy(str, s, n);
str[n] = '\0';
return str;
}
/* Duplicate a string; the caller must free() the returned string.
* (We don't use strdup() since it is not part of the standard C library and
* may not be available everywhere.) */
static char *string_dup(const char *s){
return string_dup_n(s, strlen(s));
}
/* Format a string, replacing each occurrence of the % character with
* zDb.zName. This may be more convenient than sqlite_mprintf()
* when one string is used repeatedly in a format string.
* The caller must free() the returned string. */
static char *string_format(const char *zFormat,
const char *zDb, const char *zName){
const char *p;
size_t len = 0;
size_t nDb = strlen(zDb);
size_t nName = strlen(zName);
size_t nFullTableName = nDb+1+nName;
char *result;
char *r;
/* first compute length needed */
for(p = zFormat ; *p ; ++p){
len += (*p=='%' ? nFullTableName : 1);
}
len += 1; /* for null terminator */
r = result = sqlite3_malloc(len);
for(p = zFormat; *p; ++p){
if( *p=='%' ){
memcpy(r, zDb, nDb);
r += nDb;
*r++ = '.';
memcpy(r, zName, nName);
r += nName;
} else {
*r++ = *p;
}
}
*r++ = '\0';
assert( r == result + len );
return result;
}
static int sql_exec(sqlite3 *db, const char *zDb, const char *zName,
const char *zFormat){
char *zCommand = string_format(zFormat, zDb, zName);
int rc;
TRACE(("FTS2 sql: %s\n", zCommand));
rc = sqlite3_exec(db, zCommand, NULL, 0, NULL);
sqlite3_free(zCommand);
return rc;
}
static int sql_prepare(sqlite3 *db, const char *zDb, const char *zName,
sqlite3_stmt **ppStmt, const char *zFormat){
char *zCommand = string_format(zFormat, zDb, zName);
int rc;
TRACE(("FTS2 prepare: %s\n", zCommand));
rc = sqlite3_prepare_v2(db, zCommand, -1, ppStmt, NULL);
sqlite3_free(zCommand);
return rc;
}
/* end utility functions */
/* Forward reference */
typedef struct fulltext_vtab fulltext_vtab;
/* A single term in a query is represented by an instances of
** the following structure.
*/
typedef struct QueryTerm {
short int nPhrase; /* How many following terms are part of the same phrase */
short int iPhrase; /* This is the i-th term of a phrase. */
short int iColumn; /* Column of the index that must match this term */
signed char isOr; /* this term is preceded by "OR" */
signed char isNot; /* this term is preceded by "-" */
signed char isPrefix; /* this term is followed by "*" */
char *pTerm; /* text of the term. '\000' terminated. malloced */
int nTerm; /* Number of bytes in pTerm[] */
} QueryTerm;
/* A query string is parsed into a Query structure.
*
* We could, in theory, allow query strings to be complicated
* nested expressions with precedence determined by parentheses.
* But none of the major search engines do this. (Perhaps the
* feeling is that an parenthesized expression is two complex of
* an idea for the average user to grasp.) Taking our lead from
* the major search engines, we will allow queries to be a list
* of terms (with an implied AND operator) or phrases in double-quotes,
* with a single optional "-" before each non-phrase term to designate
* negation and an optional OR connector.
*
* OR binds more tightly than the implied AND, which is what the
* major search engines seem to do. So, for example:
*
* [one two OR three] ==> one AND (two OR three)
* [one OR two three] ==> (one OR two) AND three
*
* A "-" before a term matches all entries that lack that term.
* The "-" must occur immediately before the term with in intervening
* space. This is how the search engines do it.
*
* A NOT term cannot be the right-hand operand of an OR. If this
* occurs in the query string, the NOT is ignored:
*
* [one OR -two] ==> one OR two
*
*/
typedef struct Query {
fulltext_vtab *pFts; /* The full text index */
int nTerms; /* Number of terms in the query */
QueryTerm *pTerms; /* Array of terms. Space obtained from malloc() */
int nextIsOr; /* Set the isOr flag on the next inserted term */
int nextColumn; /* Next word parsed must be in this column */
int dfltColumn; /* The default column */
} Query;
/*
** An instance of the following structure keeps track of generated
** matching-word offset information and snippets.
*/
typedef struct Snippet {
int nMatch; /* Total number of matches */
int nAlloc; /* Space allocated for aMatch[] */
struct snippetMatch { /* One entry for each matching term */
char snStatus; /* Status flag for use while constructing snippets */
short int iCol; /* The column that contains the match */
short int iTerm; /* The index in Query.pTerms[] of the matching term */
short int nByte; /* Number of bytes in the term */
int iStart; /* The offset to the first character of the term */
} *aMatch; /* Points to space obtained from malloc */
char *zOffset; /* Text rendering of aMatch[] */
int nOffset; /* strlen(zOffset) */
char *zSnippet; /* Snippet text */
int nSnippet; /* strlen(zSnippet) */
} Snippet;
typedef enum QueryType {
QUERY_GENERIC, /* table scan */
QUERY_ROWID, /* lookup by rowid */
QUERY_FULLTEXT /* QUERY_FULLTEXT + [i] is a full-text search for column i*/
} QueryType;
typedef enum fulltext_statement {
CONTENT_INSERT_STMT,
CONTENT_SELECT_STMT,
CONTENT_UPDATE_STMT,
CONTENT_DELETE_STMT,
CONTENT_EXISTS_STMT,
BLOCK_INSERT_STMT,
BLOCK_SELECT_STMT,
BLOCK_DELETE_STMT,
BLOCK_DELETE_ALL_STMT,
SEGDIR_MAX_INDEX_STMT,
SEGDIR_SET_STMT,
SEGDIR_SELECT_LEVEL_STMT,
SEGDIR_SPAN_STMT,
SEGDIR_DELETE_STMT,
SEGDIR_SELECT_SEGMENT_STMT,
SEGDIR_SELECT_ALL_STMT,
SEGDIR_DELETE_ALL_STMT,
SEGDIR_COUNT_STMT,
MAX_STMT /* Always at end! */
} fulltext_statement;
/* These must exactly match the enum above. */
/* TODO(shess): Is there some risk that a statement will be used in two
** cursors at once, e.g. if a query joins a virtual table to itself?
** If so perhaps we should move some of these to the cursor object.
*/
static const char *const fulltext_zStatement[MAX_STMT] = {
/* CONTENT_INSERT */ NULL, /* generated in contentInsertStatement() */
/* CONTENT_SELECT */ "select * from %_content where rowid = ?",
/* CONTENT_UPDATE */ NULL, /* generated in contentUpdateStatement() */
/* CONTENT_DELETE */ "delete from %_content where rowid = ?",
/* CONTENT_EXISTS */ "select rowid from %_content limit 1",
/* BLOCK_INSERT */ "insert into %_segments values (?)",
/* BLOCK_SELECT */ "select block from %_segments where rowid = ?",
/* BLOCK_DELETE */ "delete from %_segments where rowid between ? and ?",
/* BLOCK_DELETE_ALL */ "delete from %_segments",
/* SEGDIR_MAX_INDEX */ "select max(idx) from %_segdir where level = ?",
/* SEGDIR_SET */ "insert into %_segdir values (?, ?, ?, ?, ?, ?)",
/* SEGDIR_SELECT_LEVEL */
"select start_block, leaves_end_block, root from %_segdir "
" where level = ? order by idx",
/* SEGDIR_SPAN */
"select min(start_block), max(end_block) from %_segdir "
" where level = ? and start_block <> 0",
/* SEGDIR_DELETE */ "delete from %_segdir where level = ?",
/* NOTE(shess): The first three results of the following two
** statements must match.
*/
/* SEGDIR_SELECT_SEGMENT */
"select start_block, leaves_end_block, root from %_segdir "
" where level = ? and idx = ?",
/* SEGDIR_SELECT_ALL */
"select start_block, leaves_end_block, root from %_segdir "
" order by level desc, idx asc",
/* SEGDIR_DELETE_ALL */ "delete from %_segdir",
/* SEGDIR_COUNT */ "select count(*), ifnull(max(level),0) from %_segdir",
};
/*
** A connection to a fulltext index is an instance of the following
** structure. The xCreate and xConnect methods create an instance
** of this structure and xDestroy and xDisconnect free that instance.
** All other methods receive a pointer to the structure as one of their
** arguments.
*/
struct fulltext_vtab {
sqlite3_vtab base; /* Base class used by SQLite core */
sqlite3 *db; /* The database connection */
const char *zDb; /* logical database name */
const char *zName; /* virtual table name */
int nColumn; /* number of columns in virtual table */
char **azColumn; /* column names. malloced */
char **azContentColumn; /* column names in content table; malloced */
sqlite3_tokenizer *pTokenizer; /* tokenizer for inserts and queries */
/* Precompiled statements which we keep as long as the table is
** open.
*/
sqlite3_stmt *pFulltextStatements[MAX_STMT];
/* Precompiled statements used for segment merges. We run a
** separate select across the leaf level of each tree being merged.
*/
sqlite3_stmt *pLeafSelectStmts[MERGE_COUNT];
/* The statement used to prepare pLeafSelectStmts. */
#define LEAF_SELECT \
"select block from %_segments where rowid between ? and ? order by rowid"
/* These buffer pending index updates during transactions.
** nPendingData estimates the memory size of the pending data. It
** doesn't include the hash-bucket overhead, nor any malloc
** overhead. When nPendingData exceeds kPendingThreshold, the
** buffer is flushed even before the transaction closes.
** pendingTerms stores the data, and is only valid when nPendingData
** is >=0 (nPendingData<0 means pendingTerms has not been
** initialized). iPrevDocid is the last docid written, used to make
** certain we're inserting in sorted order.
*/
int nPendingData;
#define kPendingThreshold (1*1024*1024)
sqlite_int64 iPrevDocid;
fts2Hash pendingTerms;
};
/*
** When the core wants to do a query, it create a cursor using a
** call to xOpen. This structure is an instance of a cursor. It
** is destroyed by xClose.
*/
typedef struct fulltext_cursor {
sqlite3_vtab_cursor base; /* Base class used by SQLite core */
QueryType iCursorType; /* Copy of sqlite3_index_info.idxNum */
sqlite3_stmt *pStmt; /* Prepared statement in use by the cursor */
int eof; /* True if at End Of Results */
Query q; /* Parsed query string */
Snippet snippet; /* Cached snippet for the current row */
int iColumn; /* Column being searched */
DataBuffer result; /* Doclist results from fulltextQuery */
DLReader reader; /* Result reader if result not empty */
} fulltext_cursor;
static struct fulltext_vtab *cursor_vtab(fulltext_cursor *c){
return (fulltext_vtab *) c->base.pVtab;
}
static const sqlite3_module fts2Module; /* forward declaration */
/* Return a dynamically generated statement of the form
* insert into %_content (rowid, ...) values (?, ...)
*/
static const char *contentInsertStatement(fulltext_vtab *v){
StringBuffer sb;
int i;
initStringBuffer(&sb);
append(&sb, "insert into %_content (rowid, ");
appendList(&sb, v->nColumn, v->azContentColumn);
append(&sb, ") values (?");
for(i=0; i<v->nColumn; ++i)
append(&sb, ", ?");
append(&sb, ")");
return stringBufferData(&sb);
}
/* Return a dynamically generated statement of the form
* update %_content set [col_0] = ?, [col_1] = ?, ...
* where rowid = ?
*/
static const char *contentUpdateStatement(fulltext_vtab *v){
StringBuffer sb;
int i;
initStringBuffer(&sb);
append(&sb, "update %_content set ");
for(i=0; i<v->nColumn; ++i) {
if( i>0 ){
append(&sb, ", ");
}
append(&sb, v->azContentColumn[i]);
append(&sb, " = ?");
}
append(&sb, " where rowid = ?");
return stringBufferData(&sb);
}
/* Puts a freshly-prepared statement determined by iStmt in *ppStmt.
** If the indicated statement has never been prepared, it is prepared
** and cached, otherwise the cached version is reset.
*/
static int sql_get_statement(fulltext_vtab *v, fulltext_statement iStmt,
sqlite3_stmt **ppStmt){
assert( iStmt<MAX_STMT );
if( v->pFulltextStatements[iStmt]==NULL ){
const char *zStmt;
int rc;
switch( iStmt ){
case CONTENT_INSERT_STMT:
zStmt = contentInsertStatement(v); break;
case CONTENT_UPDATE_STMT:
zStmt = contentUpdateStatement(v); break;
default:
zStmt = fulltext_zStatement[iStmt];
}
rc = sql_prepare(v->db, v->zDb, v->zName, &v->pFulltextStatements[iStmt],
zStmt);
if( zStmt != fulltext_zStatement[iStmt]) sqlite3_free((void *) zStmt);
if( rc!=SQLITE_OK ) return rc;
} else {
int rc = sqlite3_reset(v->pFulltextStatements[iStmt]);
if( rc!=SQLITE_OK ) return rc;
}
*ppStmt = v->pFulltextStatements[iStmt];
return SQLITE_OK;
}
/* Like sqlite3_step(), but convert SQLITE_DONE to SQLITE_OK and
** SQLITE_ROW to SQLITE_ERROR. Useful for statements like UPDATE,
** where we expect no results.
*/
static int sql_single_step(sqlite3_stmt *s){
int rc = sqlite3_step(s);
return (rc==SQLITE_DONE) ? SQLITE_OK : rc;
}
/* Like sql_get_statement(), but for special replicated LEAF_SELECT
** statements. idx -1 is a special case for an uncached version of
** the statement (used in the optimize implementation).
*/
/* TODO(shess) Write version for generic statements and then share
** that between the cached-statement functions.
*/
static int sql_get_leaf_statement(fulltext_vtab *v, int idx,
sqlite3_stmt **ppStmt){
assert( idx>=-1 && idx<MERGE_COUNT );
if( idx==-1 ){
return sql_prepare(v->db, v->zDb, v->zName, ppStmt, LEAF_SELECT);
}else if( v->pLeafSelectStmts[idx]==NULL ){
int rc = sql_prepare(v->db, v->zDb, v->zName, &v->pLeafSelectStmts[idx],
LEAF_SELECT);
if( rc!=SQLITE_OK ) return rc;
}else{
int rc = sqlite3_reset(v->pLeafSelectStmts[idx]);
if( rc!=SQLITE_OK ) return rc;
}
*ppStmt = v->pLeafSelectStmts[idx];
return SQLITE_OK;
}
/* insert into %_content (rowid, ...) values ([rowid], [pValues]) */
static int content_insert(fulltext_vtab *v, sqlite3_value *rowid,
sqlite3_value **pValues){
sqlite3_stmt *s;
int i;
int rc = sql_get_statement(v, CONTENT_INSERT_STMT, &s);
if( rc!=SQLITE_OK ) return rc;
rc = sqlite3_bind_value(s, 1, rowid);
if( rc!=SQLITE_OK ) return rc;
for(i=0; i<v->nColumn; ++i){
rc = sqlite3_bind_value(s, 2+i, pValues[i]);
if( rc!=SQLITE_OK ) return rc;
}
return sql_single_step(s);
}
/* update %_content set col0 = pValues[0], col1 = pValues[1], ...
* where rowid = [iRowid] */
static int content_update(fulltext_vtab *v, sqlite3_value **pValues,
sqlite_int64 iRowid){
sqlite3_stmt *s;
int i;
int rc = sql_get_statement(v, CONTENT_UPDATE_STMT, &s);
if( rc!=SQLITE_OK ) return rc;
for(i=0; i<v->nColumn; ++i){
rc = sqlite3_bind_value(s, 1+i, pValues[i]);
if( rc!=SQLITE_OK ) return rc;
}
rc = sqlite3_bind_int64(s, 1+v->nColumn, iRowid);
if( rc!=SQLITE_OK ) return rc;
return sql_single_step(s);
}
static void freeStringArray(int nString, const char **pString){
int i;
for (i=0 ; i < nString ; ++i) {
if( pString[i]!=NULL ) sqlite3_free((void *) pString[i]);
}
sqlite3_free((void *) pString);
}
/* select * from %_content where rowid = [iRow]
* The caller must delete the returned array and all strings in it.
* null fields will be NULL in the returned array.
*
* TODO: Perhaps we should return pointer/length strings here for consistency
* with other code which uses pointer/length. */
static int content_select(fulltext_vtab *v, sqlite_int64 iRow,
const char ***pValues){
sqlite3_stmt *s;
const char **values;
int i;
int rc;
*pValues = NULL;
rc = sql_get_statement(v, CONTENT_SELECT_STMT, &s);
if( rc!=SQLITE_OK ) return rc;
rc = sqlite3_bind_int64(s, 1, iRow);
if( rc!=SQLITE_OK ) return rc;
rc = sqlite3_step(s);
if( rc!=SQLITE_ROW ) return rc;
values = (const char **) sqlite3_malloc(v->nColumn * sizeof(const char *));
for(i=0; i<v->nColumn; ++i){
if( sqlite3_column_type(s, i)==SQLITE_NULL ){
values[i] = NULL;
}else{
values[i] = string_dup((char*)sqlite3_column_text(s, i));
}
}
/* We expect only one row. We must execute another sqlite3_step()
* to complete the iteration; otherwise the table will remain locked. */
rc = sqlite3_step(s);
if( rc==SQLITE_DONE ){
*pValues = values;
return SQLITE_OK;
}
freeStringArray(v->nColumn, values);
return rc;
}
/* delete from %_content where rowid = [iRow ] */
static int content_delete(fulltext_vtab *v, sqlite_int64 iRow){
sqlite3_stmt *s;
int rc = sql_get_statement(v, CONTENT_DELETE_STMT, &s);
if( rc!=SQLITE_OK ) return rc;
rc = sqlite3_bind_int64(s, 1, iRow);
if( rc!=SQLITE_OK ) return rc;
return sql_single_step(s);
}
/* Returns SQLITE_ROW if any rows exist in %_content, SQLITE_DONE if
** no rows exist, and any error in case of failure.
*/
static int content_exists(fulltext_vtab *v){
sqlite3_stmt *s;
int rc = sql_get_statement(v, CONTENT_EXISTS_STMT, &s);
if( rc!=SQLITE_OK ) return rc;
rc = sqlite3_step(s);
if( rc!=SQLITE_ROW ) return rc;
/* We expect only one row. We must execute another sqlite3_step()
* to complete the iteration; otherwise the table will remain locked. */
rc = sqlite3_step(s);
if( rc==SQLITE_DONE ) return SQLITE_ROW;
if( rc==SQLITE_ROW ) return SQLITE_ERROR;
return rc;
}
/* insert into %_segments values ([pData])
** returns assigned rowid in *piBlockid
*/
static int block_insert(fulltext_vtab *v, const char *pData, int nData,
sqlite_int64 *piBlockid){
sqlite3_stmt *s;
int rc = sql_get_statement(v, BLOCK_INSERT_STMT, &s);
if( rc!=SQLITE_OK ) return rc;
rc = sqlite3_bind_blob(s, 1, pData, nData, SQLITE_STATIC);
if( rc!=SQLITE_OK ) return rc;
rc = sqlite3_step(s);
if( rc==SQLITE_ROW ) return SQLITE_ERROR;
if( rc!=SQLITE_DONE ) return rc;
*piBlockid = sqlite3_last_insert_rowid(v->db);
return SQLITE_OK;
}
/* delete from %_segments
** where rowid between [iStartBlockid] and [iEndBlockid]
**
** Deletes the range of blocks, inclusive, used to delete the blocks
** which form a segment.
*/
static int block_delete(fulltext_vtab *v,
sqlite_int64 iStartBlockid, sqlite_int64 iEndBlockid){
sqlite3_stmt *s;
int rc = sql_get_statement(v, BLOCK_DELETE_STMT, &s);
if( rc!=SQLITE_OK ) return rc;
rc = sqlite3_bind_int64(s, 1, iStartBlockid);
if( rc!=SQLITE_OK ) return rc;
rc = sqlite3_bind_int64(s, 2, iEndBlockid);
if( rc!=SQLITE_OK ) return rc;
return sql_single_step(s);
}
/* Returns SQLITE_ROW with *pidx set to the maximum segment idx found
** at iLevel. Returns SQLITE_DONE if there are no segments at
** iLevel. Otherwise returns an error.
*/
static int segdir_max_index(fulltext_vtab *v, int iLevel, int *pidx){
sqlite3_stmt *s;
int rc = sql_get_statement(v, SEGDIR_MAX_INDEX_STMT, &s);
if( rc!=SQLITE_OK ) return rc;
rc = sqlite3_bind_int(s, 1, iLevel);
if( rc!=SQLITE_OK ) return rc;
rc = sqlite3_step(s);
/* Should always get at least one row due to how max() works. */
if( rc==SQLITE_DONE ) return SQLITE_DONE;
if( rc!=SQLITE_ROW ) return rc;
/* NULL means that there were no inputs to max(). */
if( SQLITE_NULL==sqlite3_column_type(s, 0) ){
rc = sqlite3_step(s);
if( rc==SQLITE_ROW ) return SQLITE_ERROR;
return rc;
}
*pidx = sqlite3_column_int(s, 0);
/* We expect only one row. We must execute another sqlite3_step()
* to complete the iteration; otherwise the table will remain locked. */
rc = sqlite3_step(s);
if( rc==SQLITE_ROW ) return SQLITE_ERROR;
if( rc!=SQLITE_DONE ) return rc;
return SQLITE_ROW;
}
/* insert into %_segdir values (
** [iLevel], [idx],
** [iStartBlockid], [iLeavesEndBlockid], [iEndBlockid],
** [pRootData]
** )
*/
static int segdir_set(fulltext_vtab *v, int iLevel, int idx,
sqlite_int64 iStartBlockid,
sqlite_int64 iLeavesEndBlockid,
sqlite_int64 iEndBlockid,
const char *pRootData, int nRootData){
sqlite3_stmt *s;
int rc = sql_get_statement(v, SEGDIR_SET_STMT, &s);
if( rc!=SQLITE_OK ) return rc;
rc = sqlite3_bind_int(s, 1, iLevel);
if( rc!=SQLITE_OK ) return rc;
rc = sqlite3_bind_int(s, 2, idx);
if( rc!=SQLITE_OK ) return rc;
rc = sqlite3_bind_int64(s, 3, iStartBlockid);
if( rc!=SQLITE_OK ) return rc;
rc = sqlite3_bind_int64(s, 4, iLeavesEndBlockid);
if( rc!=SQLITE_OK ) return rc;
rc = sqlite3_bind_int64(s, 5, iEndBlockid);
if( rc!=SQLITE_OK ) return rc;
rc = sqlite3_bind_blob(s, 6, pRootData, nRootData, SQLITE_STATIC);
if( rc!=SQLITE_OK ) return rc;
return sql_single_step(s);
}
/* Queries %_segdir for the block span of the segments in level
** iLevel. Returns SQLITE_DONE if there are no blocks for iLevel,
** SQLITE_ROW if there are blocks, else an error.
*/
static int segdir_span(fulltext_vtab *v, int iLevel,
sqlite_int64 *piStartBlockid,
sqlite_int64 *piEndBlockid){
sqlite3_stmt *s;
int rc = sql_get_statement(v, SEGDIR_SPAN_STMT, &s);
if( rc!=SQLITE_OK ) return rc;
rc = sqlite3_bind_int(s, 1, iLevel);
if( rc!=SQLITE_OK ) return rc;
rc = sqlite3_step(s);
if( rc==SQLITE_DONE ) return SQLITE_DONE; /* Should never happen */
if( rc!=SQLITE_ROW ) return rc;
/* This happens if all segments at this level are entirely inline. */
if( SQLITE_NULL==sqlite3_column_type(s, 0) ){
/* We expect only one row. We must execute another sqlite3_step()
* to complete the iteration; otherwise the table will remain locked. */
int rc2 = sqlite3_step(s);
if( rc2==SQLITE_ROW ) return SQLITE_ERROR;
return rc2;
}
*piStartBlockid = sqlite3_column_int64(s, 0);
*piEndBlockid = sqlite3_column_int64(s, 1);
/* We expect only one row. We must execute another sqlite3_step()
* to complete the iteration; otherwise the table will remain locked. */
rc = sqlite3_step(s);
if( rc==SQLITE_ROW ) return SQLITE_ERROR;
if( rc!=SQLITE_DONE ) return rc;
return SQLITE_ROW;
}
/* Delete the segment blocks and segment directory records for all
** segments at iLevel.
*/
static int segdir_delete(fulltext_vtab *v, int iLevel){
sqlite3_stmt *s;
sqlite_int64 iStartBlockid, iEndBlockid;
int rc = segdir_span(v, iLevel, &iStartBlockid, &iEndBlockid);
if( rc!=SQLITE_ROW && rc!=SQLITE_DONE ) return rc;
if( rc==SQLITE_ROW ){
rc = block_delete(v, iStartBlockid, iEndBlockid);
if( rc!=SQLITE_OK ) return rc;
}
/* Delete the segment directory itself. */
rc = sql_get_statement(v, SEGDIR_DELETE_STMT, &s);
if( rc!=SQLITE_OK ) return rc;
rc = sqlite3_bind_int64(s, 1, iLevel);
if( rc!=SQLITE_OK ) return rc;
return sql_single_step(s);
}
/* Delete entire fts index, SQLITE_OK on success, relevant error on
** failure.
*/
static int segdir_delete_all(fulltext_vtab *v){
sqlite3_stmt *s;
int rc = sql_get_statement(v, SEGDIR_DELETE_ALL_STMT, &s);
if( rc!=SQLITE_OK ) return rc;
rc = sql_single_step(s);
if( rc!=SQLITE_OK ) return rc;
rc = sql_get_statement(v, BLOCK_DELETE_ALL_STMT, &s);
if( rc!=SQLITE_OK ) return rc;
return sql_single_step(s);
}
/* Returns SQLITE_OK with *pnSegments set to the number of entries in
** %_segdir and *piMaxLevel set to the highest level which has a
** segment. Otherwise returns the SQLite error which caused failure.
*/
static int segdir_count(fulltext_vtab *v, int *pnSegments, int *piMaxLevel){
sqlite3_stmt *s;
int rc = sql_get_statement(v, SEGDIR_COUNT_STMT, &s);
if( rc!=SQLITE_OK ) return rc;
rc = sqlite3_step(s);
/* TODO(shess): This case should not be possible? Should stronger
** measures be taken if it happens?
*/
if( rc==SQLITE_DONE ){
*pnSegments = 0;
*piMaxLevel = 0;
return SQLITE_OK;
}
if( rc!=SQLITE_ROW ) return rc;
*pnSegments = sqlite3_column_int(s, 0);
*piMaxLevel = sqlite3_column_int(s, 1);
/* We expect only one row. We must execute another sqlite3_step()
* to complete the iteration; otherwise the table will remain locked. */
rc = sqlite3_step(s);
if( rc==SQLITE_DONE ) return SQLITE_OK;
if( rc==SQLITE_ROW ) return SQLITE_ERROR;
return rc;
}
/* TODO(shess) clearPendingTerms() is far down the file because
** writeZeroSegment() is far down the file because LeafWriter is far
** down the file. Consider refactoring the code to move the non-vtab
** code above the vtab code so that we don't need this forward
** reference.
*/
static int clearPendingTerms(fulltext_vtab *v);
/*
** Free the memory used to contain a fulltext_vtab structure.
*/
static void fulltext_vtab_destroy(fulltext_vtab *v){
int iStmt, i;
TRACE(("FTS2 Destroy %p\n", v));
for( iStmt=0; iStmt<MAX_STMT; iStmt++ ){
if( v->pFulltextStatements[iStmt]!=NULL ){
sqlite3_finalize(v->pFulltextStatements[iStmt]);
v->pFulltextStatements[iStmt] = NULL;
}
}
for( i=0; i<MERGE_COUNT; i++ ){
if( v->pLeafSelectStmts[i]!=NULL ){
sqlite3_finalize(v->pLeafSelectStmts[i]);
v->pLeafSelectStmts[i] = NULL;
}
}
if( v->pTokenizer!=NULL ){
v->pTokenizer->pModule->xDestroy(v->pTokenizer);
v->pTokenizer = NULL;
}
clearPendingTerms(v);
sqlite3_free(v->azColumn);
for(i = 0; i < v->nColumn; ++i) {
sqlite3_free(v->azContentColumn[i]);
}
sqlite3_free(v->azContentColumn);
sqlite3_free(v);
}
/*
** Token types for parsing the arguments to xConnect or xCreate.
*/
#define TOKEN_EOF 0 /* End of file */
#define TOKEN_SPACE 1 /* Any kind of whitespace */
#define TOKEN_ID 2 /* An identifier */
#define TOKEN_STRING 3 /* A string literal */
#define TOKEN_PUNCT 4 /* A single punctuation character */
/*
** If X is a character that can be used in an identifier then
** IdChar(X) will be true. Otherwise it is false.
**
** For ASCII, any character with the high-order bit set is
** allowed in an identifier. For 7-bit characters,
** sqlite3IsIdChar[X] must be 1.
**
** Ticket #1066. the SQL standard does not allow '$' in the
** middle of identfiers. But many SQL implementations do.
** SQLite will allow '$' in identifiers for compatibility.
** But the feature is undocumented.
*/
static const char isIdChar[] = {
/* x0 x1 x2 x3 x4 x5 x6 x7 x8 x9 xA xB xC xD xE xF */
0, 0, 0, 0, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, /* 2x */
1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 0, 0, 0, 0, 0, 0, /* 3x */
0, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, /* 4x */
1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 0, 0, 0, 0, 1, /* 5x */
0, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, /* 6x */
1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 0, 0, 0, 0, 0, /* 7x */
};
#define IdChar(C) (((c=C)&0x80)!=0 || (c>0x1f && isIdChar[c-0x20]))
/*
** Return the length of the token that begins at z[0].
** Store the token type in *tokenType before returning.
*/
static int getToken(const char *z, int *tokenType){
int i, c;
switch( *z ){
case 0: {
*tokenType = TOKEN_EOF;
return 0;
}
case ' ': case '\t': case '\n': case '\f': case '\r': {
for(i=1; safe_isspace(z[i]); i++){}
*tokenType = TOKEN_SPACE;
return i;
}
case '`':
case '\'':
case '"': {
int delim = z[0];
for(i=1; (c=z[i])!=0; i++){
if( c==delim ){
if( z[i+1]==delim ){
i++;
}else{
break;
}
}
}
*tokenType = TOKEN_STRING;
return i + (c!=0);
}
case '[': {
for(i=1, c=z[0]; c!=']' && (c=z[i])!=0; i++){}
*tokenType = TOKEN_ID;
return i;
}
default: {
if( !IdChar(*z) ){
break;
}
for(i=1; IdChar(z[i]); i++){}
*tokenType = TOKEN_ID;
return i;
}
}
*tokenType = TOKEN_PUNCT;
return 1;
}
/*
** A token extracted from a string is an instance of the following
** structure.
*/
typedef struct Token {
const char *z; /* Pointer to token text. Not '\000' terminated */
short int n; /* Length of the token text in bytes. */
} Token;
/*
** Given a input string (which is really one of the argv[] parameters
** passed into xConnect or xCreate) split the string up into tokens.
** Return an array of pointers to '\000' terminated strings, one string
** for each non-whitespace token.
**
** The returned array is terminated by a single NULL pointer.
**
** Space to hold the returned array is obtained from a single
** malloc and should be freed by passing the return value to free().
** The individual strings within the token list are all a part of
** the single memory allocation and will all be freed at once.
*/
static char **tokenizeString(const char *z, int *pnToken){
int nToken = 0;
Token *aToken = sqlite3_malloc( strlen(z) * sizeof(aToken[0]) );
int n = 1;
int e, i;
int totalSize = 0;
char **azToken;
char *zCopy;
while( n>0 ){
n = getToken(z, &e);
if( e!=TOKEN_SPACE ){
aToken[nToken].z = z;
aToken[nToken].n = n;
nToken++;
totalSize += n+1;
}
z += n;
}
azToken = (char**)sqlite3_malloc( nToken*sizeof(char*) + totalSize );
zCopy = (char*)&azToken[nToken];
nToken--;
for(i=0; i<nToken; i++){
azToken[i] = zCopy;
n = aToken[i].n;
memcpy(zCopy, aToken[i].z, n);
zCopy[n] = 0;
zCopy += n+1;
}
azToken[nToken] = 0;
sqlite3_free(aToken);
*pnToken = nToken;
return azToken;
}
/*
** 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
*/
static void dequoteString(char *z){
int quote;
int i, j;
if( z==0 ) return;
quote = z[0];
switch( quote ){
case '\'': break;
case '"': break;
case '`': break; /* For MySQL compatibility */
case '[': quote = ']'; break; /* For MS SqlServer compatibility */
default: return;
}
for(i=1, j=0; z[i]; i++){
if( z[i]==quote ){
if( z[i+1]==quote ){
z[j++] = quote;
i++;
}else{
z[j++] = 0;
break;
}
}else{
z[j++] = z[i];
}
}
}
/*
** The input azIn is a NULL-terminated list of tokens. Remove the first
** token and all punctuation tokens. Remove the quotes from
** around string literal tokens.
**
** Example:
**
** input: tokenize chinese ( 'simplifed' , 'mixed' )
** output: chinese simplifed mixed
**
** Another example:
**
** input: delimiters ( '[' , ']' , '...' )
** output: [ ] ...
*/
static void tokenListToIdList(char **azIn){
int i, j;
if( azIn ){
for(i=0, j=-1; azIn[i]; i++){
if( safe_isalnum(azIn[i][0]) || azIn[i][1] ){
dequoteString(azIn[i]);
if( j>=0 ){
azIn[j] = azIn[i];
}
j++;
}
}
azIn[j] = 0;
}
}
/*
** Find the first alphanumeric token in the string zIn. Null-terminate
** this token. Remove any quotation marks. And return a pointer to
** the result.
*/
static char *firstToken(char *zIn, char **pzTail){
int n, ttype;
while(1){
n = getToken(zIn, &ttype);
if( ttype==TOKEN_SPACE ){
zIn += n;
}else if( ttype==TOKEN_EOF ){
*pzTail = zIn;
return 0;
}else{
zIn[n] = 0;
*pzTail = &zIn[1];
dequoteString(zIn);
return zIn;
}
}
/*NOTREACHED*/
}
/* Return true if...
**
** * s begins with the string t, ignoring case
** * s is longer than t
** * The first character of s beyond t is not a alphanumeric
**
** Ignore leading space in *s.
**
** To put it another way, return true if the first token of
** s[] is t[].
*/
static int startsWith(const char *s, const char *t){
while( safe_isspace(*s) ){ s++; }
while( *t ){
if( safe_tolower(*s++)!=safe_tolower(*t++) ) return 0;
}
return *s!='_' && !safe_isalnum(*s);
}
/*
** An instance of this structure defines the "spec" of a
** full text index. This structure is populated by parseSpec
** and use by fulltextConnect and fulltextCreate.
*/
typedef struct TableSpec {
const char *zDb; /* Logical database name */
const char *zName; /* Name of the full-text index */
int nColumn; /* Number of columns to be indexed */
char **azColumn; /* Original names of columns to be indexed */
char **azContentColumn; /* Column names for %_content */
char **azTokenizer; /* Name of tokenizer and its arguments */
} TableSpec;
/*
** Reclaim all of the memory used by a TableSpec
*/
static void clearTableSpec(TableSpec *p) {
sqlite3_free(p->azColumn);
sqlite3_free(p->azContentColumn);
sqlite3_free(p->azTokenizer);
}
/* Parse a CREATE VIRTUAL TABLE statement, which looks like this:
*
* CREATE VIRTUAL TABLE email
* USING fts2(subject, body, tokenize mytokenizer(myarg))
*
* We return parsed information in a TableSpec structure.
*
*/
static int parseSpec(TableSpec *pSpec, int argc, const char *const*argv,
char**pzErr){
int i, n;
char *z, *zDummy;
char **azArg;
const char *zTokenizer = 0; /* argv[] entry describing the tokenizer */
assert( argc>=3 );
/* Current interface:
** argv[0] - module name
** argv[1] - database name
** argv[2] - table name
** argv[3..] - columns, optionally followed by tokenizer specification
** and snippet delimiters specification.
*/
/* Make a copy of the complete argv[][] array in a single allocation.
** The argv[][] array is read-only and transient. We can write to the
** copy in order to modify things and the copy is persistent.
*/
CLEAR(pSpec);
for(i=n=0; i<argc; i++){
n += strlen(argv[i]) + 1;
}
azArg = sqlite3_malloc( sizeof(char*)*argc + n );
if( azArg==0 ){
return SQLITE_NOMEM;
}
z = (char*)&azArg[argc];
for(i=0; i<argc; i++){
azArg[i] = z;
strcpy(z, argv[i]);
z += strlen(z)+1;
}
/* Identify the column names and the tokenizer and delimiter arguments
** in the argv[][] array.
*/
pSpec->zDb = azArg[1];
pSpec->zName = azArg[2];
pSpec->nColumn = 0;
pSpec->azColumn = azArg;
zTokenizer = "tokenize simple";
for(i=3; i<argc; ++i){
if( startsWith(azArg[i],"tokenize") ){
zTokenizer = azArg[i];
}else{
z = azArg[pSpec->nColumn] = firstToken(azArg[i], &zDummy);
pSpec->nColumn++;
}
}
if( pSpec->nColumn==0 ){
azArg[0] = "content";
pSpec->nColumn = 1;
}
/*
** Construct the list of content column names.
**
** Each content column name will be of the form cNNAAAA
** where NN is the column number and AAAA is the sanitized
** column name. "sanitized" means that special characters are
** converted to "_". The cNN prefix guarantees that all column
** names are unique.
**
** The AAAA suffix is not strictly necessary. It is included
** for the convenience of people who might examine the generated
** %_content table and wonder what the columns are used for.
*/
pSpec->azContentColumn = sqlite3_malloc( pSpec->nColumn * sizeof(char *) );
if( pSpec->azContentColumn==0 ){
clearTableSpec(pSpec);
return SQLITE_NOMEM;
}
for(i=0; i<pSpec->nColumn; i++){
char *p;
pSpec->azContentColumn[i] = sqlite3_mprintf("c%d%s", i, azArg[i]);
for (p = pSpec->azContentColumn[i]; *p ; ++p) {
if( !safe_isalnum(*p) ) *p = '_';
}
}
/*
** Parse the tokenizer specification string.
*/