tracker r2160 - in branches/indexer-split: . src src/libtracker-common src/tracker-fts tests tests/tracker-fts
- From: jamiemcc svn gnome org
- To: svn-commits-list gnome org
- Subject: tracker r2160 - in branches/indexer-split: . src src/libtracker-common src/tracker-fts tests tests/tracker-fts
- Date: Wed, 27 Aug 2008 01:53:51 +0000 (UTC)
Author: jamiemcc
Date: Wed Aug 27 01:53:50 2008
New Revision: 2160
URL: http://svn.gnome.org/viewvc/tracker?rev=2160&view=rev
Log:
2008-08-26 Jamie McCracken <jamiemcc at gnome org>
* src/libtracker-common/tracker-parser.c
* src/libtracker-common/tracker-parser.h
Made new API for tracker-parser more compatible
with sqlite FTS3.
* tracker-fts
* tracker-fts/tracker-fts.c
* tracker-fts/tracker-fts-hash.c
* tracker-fts/Makefile.am
* tracker-fts/tracker-fts.h
* tracker-fts/tracker-fts-hash.h
Created forked version of sqlite fts3 module called tracker-fts
Integrated our parser and tweaked it to work fully with FTS3 api
* tests/Makefile.am
* configure.ac
* src/Makefile.am
* tests/tracker-fts
* tests/tracker-fts/tracker-fts-test.c
* tests/tracker-fts/Makefile.am
Added test files for tracker-fts module
Added:
branches/indexer-split/src/tracker-fts/
branches/indexer-split/src/tracker-fts/Makefile.am
branches/indexer-split/src/tracker-fts/tracker-fts-hash.c
branches/indexer-split/src/tracker-fts/tracker-fts-hash.h
branches/indexer-split/src/tracker-fts/tracker-fts.c
branches/indexer-split/src/tracker-fts/tracker-fts.h
branches/indexer-split/tests/tracker-fts/
branches/indexer-split/tests/tracker-fts/Makefile.am
branches/indexer-split/tests/tracker-fts/tracker-fts-test.c
Modified:
branches/indexer-split/ChangeLog
branches/indexer-split/configure.ac
branches/indexer-split/src/Makefile.am
branches/indexer-split/src/libtracker-common/tracker-parser.c
branches/indexer-split/src/libtracker-common/tracker-parser.h
branches/indexer-split/tests/Makefile.am
Modified: branches/indexer-split/configure.ac
==============================================================================
--- branches/indexer-split/configure.ac (original)
+++ branches/indexer-split/configure.ac Wed Aug 27 01:53:50 2008
@@ -978,6 +978,7 @@
src/tracker-applet/tracker-applet.desktop.in
src/trackerd/Makefile
src/tracker-extract/Makefile
+ src/tracker-fts/Makefile
src/tracker-indexer/Makefile
src/tracker-indexer/modules/Makefile
src/tracker-preferences/Makefile
@@ -996,6 +997,7 @@
tests/trackerd/Makefile
tests/trackerd/xesam/Makefile
tests/tracker-indexer/Makefile
+ tests/tracker-fts/Makefile
thumbnailers/application/Makefile
thumbnailers/image/Makefile
thumbnailers/Makefile
Modified: branches/indexer-split/src/Makefile.am
==============================================================================
--- branches/indexer-split/src/Makefile.am (original)
+++ branches/indexer-split/src/Makefile.am Wed Aug 27 01:53:50 2008
@@ -29,6 +29,7 @@
libtracker-common \
libtracker-db \
libtracker \
+ tracker-fts \
trackerd \
tracker-utils \
tracker-extract \
Modified: branches/indexer-split/src/libtracker-common/tracker-parser.c
==============================================================================
--- branches/indexer-split/src/libtracker-common/tracker-parser.c (original)
+++ branches/indexer-split/src/libtracker-common/tracker-parser.c Wed Aug 27 01:53:50 2008
@@ -45,6 +45,8 @@
#define IS_HYPHEN(c) ((c) == 0x002D)
#define IS_UNDERSCORE(c) ((c) == 0x005F)
#define IS_NEWLINE(c) ((c) == 0x000D)
+#define IS_O(c) ((c) == 0x006F)
+#define IS_R(c) ((c) == 0x0072)
typedef enum {
TRACKER_PARSER_WORD_ASCII_HIGHER,
@@ -374,18 +376,17 @@
return p;
}
-static gchar *
+static gboolean
pango_next (TrackerParser *parser,
- guint skip_words,
- guint *byte_offset_start,
- guint *byte_offset_end,
+ gint *byte_offset_start,
+ gint *byte_offset_end,
gboolean *is_new_paragraph)
+
{
/* CJK text does not need stemming or other treatment */
gint word_start = -1;
gint old_word_start = -1;
- guint words_parsed = 0;
- guint32 i;
+ guint i;
*is_new_paragraph = FALSE;
@@ -399,12 +400,7 @@
gchar *start_word, *end_word;
old_word_start = word_start;
- words_parsed++;
-
- if (words_parsed <= skip_words) {
- continue;
- }
-
+
start_word = g_utf8_offset_to_pointer (parser->txt, word_start);
end_word = g_utf8_offset_to_pointer (parser->txt, i);
@@ -429,11 +425,15 @@
*is_new_paragraph = TRUE;
}
+ parser->word_length = strlen (index_word);
+ parser->word = index_word;
+
*byte_offset_start = (start_word - parser->txt);
*byte_offset_end = *byte_offset_start + (end_word - start_word);
parser->attr_pos = i;
- return index_word;
+
+ return TRUE;
}
@@ -443,38 +443,35 @@
parser->attr_pos = i;
- return NULL;
+ return FALSE;
}
-static gchar *
-parser_next (TrackerParser *parser,
- guint skip_words,
- guint *byte_offset_start,
- guint *byte_offset_end,
+static gboolean
+parser_next (TrackerParser *parser,
+ gint *byte_offset_start,
+ gint *byte_offset_end,
gboolean *is_new_paragraph)
{
TrackerParserWordType word_type;
gunichar word[64];
gboolean is_valid;
- gboolean filter_numbers = TRUE;
- guint length;
+ guint length;
gint char_count = 0;
glong bytes;
const gchar *p;
const gchar *start;
const gchar *end;
- guint words_skipped = 0;
gboolean do_strip = FALSE;
- *byte_offset_start = 1;
- *byte_offset_end = 2;
+ *byte_offset_start = 0;
+ *byte_offset_end = 0;
*is_new_paragraph = FALSE;
- g_return_val_if_fail (parser, NULL);
+ g_return_val_if_fail (parser, FALSE);
if (!parser->cursor) {
- return NULL;
+ return FALSE;
}
word_type = TRACKER_PARSER_WORD_IGNORE;
@@ -505,19 +502,19 @@
continue;
} else {
/* word break */
+
+ /* check if word is reserved */
+ if (is_valid && parser->parse_reserved_words) {
+ if (length == 2 && word[0] == 'o' && word[1] == 'r') {
+ break;
+ }
+ }
if (!is_valid ||
length < parser->min_word_length ||
- word_type == TRACKER_PARSER_WORD_NUM ||
- words_skipped < skip_words) {
+ word_type == TRACKER_PARSER_WORD_NUM) {
*is_new_paragraph = FALSE;
-
- if (is_valid && length >= parser->min_word_length &&
- word_type != TRACKER_PARSER_WORD_NUM &&
- words_skipped < skip_words) {
- words_skipped++;
- }
-
+
word_type = TRACKER_PARSER_WORD_IGNORE;
is_valid = TRUE;
length = 0;
@@ -538,20 +535,21 @@
}
if (!start) {
- start = g_utf8_offset_to_pointer (parser->cursor, char_count);
+ start = g_utf8_offset_to_pointer (parser->cursor, char_count-1);
/* Valid words must start with an alpha or
* underscore if we are filtering.
*/
- if (filter_numbers) {
- if (type == TRACKER_PARSER_WORD_NUM) {
- is_valid = FALSE;
- continue;
- } else {
- if (type == TRACKER_PARSER_WORD_HYPHEN) {
- is_valid = FALSE;
- continue;
- }
+
+ if (type == TRACKER_PARSER_WORD_NUM) {
+ is_valid = FALSE;
+ start = NULL;
+ continue;
+ } else {
+ if (type == TRACKER_PARSER_WORD_HYPHEN) {
+ is_valid = parser->parse_reserved_words;
+ start = NULL;
+ continue;
}
}
}
@@ -611,14 +609,12 @@
word[length -1] = c;
}
+ parser->cursor = NULL;
+
if (!is_valid) {
- parser->cursor = NULL;
- return NULL;
+ return FALSE;
}
- end = g_utf8_offset_to_pointer (parser->cursor, char_count);
-
- parser->cursor = end;
-
+
if (word_type == TRACKER_PARSER_WORD_ALPHA_NUM || word_type == TRACKER_PARSER_WORD_ALPHA) {
gchar *utf8;
gchar *processed_word;
@@ -628,20 +624,26 @@
utf8 = g_ucs4_to_utf8 (word, length, NULL, &bytes, NULL);
if (!utf8) {
- return NULL;
+ return FALSE;
}
- *byte_offset_start = start - parser->txt;
- *byte_offset_end = end - parser->txt;
+
+ *byte_offset_start = start-parser->txt;
+ *byte_offset_end = *byte_offset_start + bytes;
+
+ parser->cursor = g_utf8_next_char (parser->txt + *byte_offset_end);
processed_word = tracker_parser_process_word (parser, utf8, bytes, do_strip);
g_free (utf8);
+
+ parser->word_length = strlen (processed_word);
+ parser->word = processed_word;
- return processed_word;
+ return TRUE;
}
- return NULL;
+ return FALSE;
}
@@ -690,7 +692,7 @@
parser->max_word_length = max_word_length;
parser->min_word_length = min_word_length;
-
+ parser->word_length = 0;
parser->attrs = NULL;
return parser;
@@ -706,6 +708,8 @@
}
g_free (parser->attrs);
+
+ g_free (parser->word);
g_free (parser);
}
@@ -716,12 +720,14 @@
gint txt_size,
gboolean delimit_words,
gboolean enable_stemmer,
- gboolean enable_stop_words)
+ gboolean enable_stop_words,
+ gboolean parse_reserved_words)
{
g_return_if_fail (parser != NULL);
g_return_if_fail (txt != NULL);
g_free (parser->attrs);
+ parser->attrs = NULL;
parser->enable_stemmer = enable_stemmer;
parser->enable_stop_words = enable_stop_words;
@@ -729,6 +735,10 @@
parser->encoding = get_encoding (txt);
parser->txt_size = txt_size;
parser->txt = txt;
+ parser->parse_reserved_words = parse_reserved_words;
+
+ g_free (parser->word);
+ parser->word = NULL;
parser->word_position = 0;
@@ -736,6 +746,10 @@
if (parser->encoding == TRACKER_PARSER_ENCODING_CJK) {
PangoLogAttr *attrs;
+
+ if (parser->txt_size == -1) {
+ parser->txt_size = strlen (parser->txt);
+ }
parser->attr_length = g_utf8_strlen (parser->txt, parser->txt_size) + 1;
@@ -777,7 +791,7 @@
bytes = length;
}
- if (do_strip && get_encoding (word) == TRACKER_PARSER_ENCODING_LATIN) {
+ if (do_strip) {
stripped_word = strip_word (word, bytes, &len);
} else {
stripped_word = NULL;
@@ -806,7 +820,7 @@
gchar *result;
result = g_strdup (stem_word);
- g_free (str);
+ g_free (str);
return result;
}
@@ -827,34 +841,45 @@
return result;
}
-gchar *
+const gchar *
tracker_parser_next (TrackerParser *parser,
- guint *position,
- guint *byte_offset_start,
- guint *byte_offset_end,
+ gint *position,
+ gint *byte_offset_start,
+ gint *byte_offset_end,
gboolean *new_paragraph,
- gboolean *stop_word)
+ gboolean *stop_word,
+ gint *word_length)
{
- gchar *str;
- guint byte_start, byte_end;
+ const gchar *str;
+ gint byte_start = 0, byte_end = 0;
gboolean new_para;
+
+ str = NULL;
+
+ g_free (parser->word);
+ parser->word = NULL;
if (parser->encoding == TRACKER_PARSER_ENCODING_CJK) {
- str = pango_next (parser, 0, &byte_start, &byte_end, &new_para);
+ if (pango_next (parser, &byte_start, &byte_end, &new_para)) {
+ str = parser->word;
+ }
parser->word_position++;
*stop_word = FALSE;
} else {
- str = parser_next (parser, 0, &byte_start, &byte_end, &new_para);
- parser->word_position++;
+ if (parser_next (parser, &byte_start, &byte_end, &new_para)) {
+ str = parser->word;
+ }
if (parser->enable_stop_words && is_stop_word (parser->language, str)) {
*stop_word = TRUE;
} else {
+ parser->word_position++;
*stop_word = FALSE;
}
}
+ *word_length = parser->word_length;
*position = parser->word_position;
*byte_offset_start = byte_start;
*byte_offset_end = byte_end;
@@ -863,28 +888,6 @@
return str;
}
-void
-tracker_parser_set_posititon (TrackerParser *parser,
- guint position)
-{
- gchar *s;
- guint byte_start, byte_end;
- gboolean para;
-
- g_return_if_fail (parser != NULL);
-
- parser->word_position = 0;
- parser->cursor = parser->txt;
- parser->attr_pos = 0;
-
- if (parser->encoding == TRACKER_PARSER_ENCODING_CJK) {
- s = pango_next (parser, position, &byte_start, &byte_end, ¶);
- g_free (s);
- } else {
- s = parser_next (parser, position, &byte_start, &byte_end, ¶);
- g_free (s);
- }
-}
gchar *
tracker_parser_text_to_string (const gchar *txt,
Modified: branches/indexer-split/src/libtracker-common/tracker-parser.h
==============================================================================
--- branches/indexer-split/src/libtracker-common/tracker-parser.h (original)
+++ branches/indexer-split/src/libtracker-common/tracker-parser.h Wed Aug 27 01:53:50 2008
@@ -47,11 +47,14 @@
guint max_word_length;
guint min_word_length;
gboolean delimit_words;
+ gboolean parse_reserved_words;
/* Private members */
- guint word_position;
- TrackerParserEncoding encoding;
- const gchar *cursor;
+ gchar *word;
+ gint word_length;
+ guint word_position;
+ TrackerParserEncoding encoding;
+ const gchar *cursor;
/* Pango members for CJK text parsing */
PangoLogAttr *attrs;
@@ -67,15 +70,17 @@
gint txt_size,
gboolean delimit_words,
gboolean enable_stemmer,
- gboolean enable_stop_words);
-gchar * tracker_parser_next (TrackerParser *parser,
- guint *position,
- guint *byte_offset_start,
- guint *byte_offset_end,
+ gboolean enable_stop_words,
+ gboolean parse_reserved_words);
+const gchar * tracker_parser_next (TrackerParser *parser,
+ gint *position,
+ gint *byte_offset_start,
+ gint *byte_offset_end,
gboolean *new_paragraph,
- gboolean *stop_word);
+ gboolean *stop_word,
+ gint *word_length);
void tracker_parser_set_posititon (TrackerParser *parser,
- guint position);
+ gint position);
gboolean tracker_parser_is_stop_word (TrackerParser *parser,
const gchar *word);
gchar * tracker_parser_process_word (TrackerParser *parser,
Added: branches/indexer-split/src/tracker-fts/Makefile.am
==============================================================================
--- (empty file)
+++ branches/indexer-split/src/tracker-fts/Makefile.am Wed Aug 27 01:53:50 2008
@@ -0,0 +1,26 @@
+module_flags = -module -avoid-version -no-undefined
+tracker_fts_modulesdir = $(libdir)
+
+INCLUDES = \
+ -g \
+ -I$(top_srcdir)/src \
+ $(GLIB2_CFLAGS) \
+ $(SQLITE3_CFLAGS) \
+ $(PANGO_CFLAGS)
+
+tracker_fts_modules_LTLIBRARIES = tracker-fts.la
+
+
+tracker_fts_la_SOURCES = \
+ tracker-fts.c \
+ tracker-fts-hash.c
+
+
+tracker_fts_la_LDFLAGS = $(module_flags)
+
+tracker_fts_la_LIBADD = \
+ $(top_builddir)/src/libtracker-common/libtracker-common.la \
+ $(SQLITE3_LIBS) \
+ $(GTHREAD_LIBS) \
+ $(PANGO_LIBS) \
+ $(GLIB2_LIBS)
Added: branches/indexer-split/src/tracker-fts/tracker-fts-hash.c
==============================================================================
--- (empty file)
+++ branches/indexer-split/src/tracker-fts/tracker-fts-hash.c Wed Aug 27 01:53:50 2008
@@ -0,0 +1,373 @@
+/*
+** 2001 September 22
+**
+** 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 the implementation of generic hash-tables used in SQLite.
+** We've modified it slightly to serve as a standalone hash table
+** implementation for the full-text indexing module.
+*/
+
+/*
+** 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).
+*/
+
+#include <assert.h>
+#include <stdlib.h>
+#include <string.h>
+
+#include <sqlite3.h>
+#include "tracker-fts-hash.h"
+
+/*
+** Malloc and Free functions
+*/
+static void *fts3HashMalloc(int n){
+ void *p = sqlite3_malloc(n);
+ if( p ){
+ memset(p, 0, n);
+ }
+ return p;
+}
+static void fts3HashFree(void *p){
+ sqlite3_free(p);
+}
+
+/* Turn bulk memory into a hash table object by initializing the
+** fields of the Hash structure.
+**
+** "pNew" is a pointer to the hash table that is to be initialized.
+** keyClass is one of the constants
+** FTS3_HASH_BINARY or FTS3_HASH_STRING. The value of keyClass
+** determines what kind of key the hash table will use. "copyKey" is
+** true if the hash table should make its own private copy of keys and
+** false if it should just use the supplied pointer.
+*/
+void sqlite3Fts3HashInit(fts3Hash *pNew, int keyClass, int copyKey){
+ assert( pNew!=0 );
+ assert( keyClass>=FTS3_HASH_STRING && keyClass<=FTS3_HASH_BINARY );
+ pNew->keyClass = keyClass;
+ pNew->copyKey = copyKey;
+ pNew->first = 0;
+ pNew->count = 0;
+ pNew->htsize = 0;
+ pNew->ht = 0;
+}
+
+/* Remove all entries from a hash table. Reclaim all memory.
+** Call this routine to delete a hash table or to reset a hash table
+** to the empty state.
+*/
+void sqlite3Fts3HashClear(fts3Hash *pH){
+ fts3HashElem *elem; /* For looping over all elements of the table */
+
+ assert( pH!=0 );
+ elem = pH->first;
+ pH->first = 0;
+ fts3HashFree(pH->ht);
+ pH->ht = 0;
+ pH->htsize = 0;
+ while( elem ){
+ fts3HashElem *next_elem = elem->next;
+ if( pH->copyKey && elem->pKey ){
+ fts3HashFree(elem->pKey);
+ }
+ fts3HashFree(elem);
+ elem = next_elem;
+ }
+ pH->count = 0;
+}
+
+/*
+** Hash and comparison functions when the mode is FTS3_HASH_STRING
+*/
+static int fts3StrHash(const void *pKey, int nKey){
+ const char *z = (const char *)pKey;
+ int h = 0;
+ if( nKey<=0 ) nKey = (int) strlen(z);
+ while( nKey > 0 ){
+ h = (h<<3) ^ h ^ *z++;
+ nKey--;
+ }
+ return h & 0x7fffffff;
+}
+static int fts3StrCompare(const void *pKey1, int n1, const void *pKey2, int n2){
+ if( n1!=n2 ) return 1;
+ return strncmp((const char*)pKey1,(const char*)pKey2,n1);
+}
+
+/*
+** Hash and comparison functions when the mode is FTS3_HASH_BINARY
+*/
+static int fts3BinHash(const void *pKey, int nKey){
+ int h = 0;
+ const char *z = (const char *)pKey;
+ while( nKey-- > 0 ){
+ h = (h<<3) ^ h ^ *(z++);
+ }
+ return h & 0x7fffffff;
+}
+static int fts3BinCompare(const void *pKey1, int n1, const void *pKey2, int n2){
+ if( n1!=n2 ) return 1;
+ return memcmp(pKey1,pKey2,n1);
+}
+
+/*
+** Return a pointer to the appropriate hash function given the key class.
+**
+** The C syntax in this function definition may be unfamilar to some
+** programmers, so we provide the following additional explanation:
+**
+** The name of the function is "ftsHashFunction". The function takes a
+** single parameter "keyClass". The return value of ftsHashFunction()
+** is a pointer to another function. Specifically, the return value
+** of ftsHashFunction() is a pointer to a function that takes two parameters
+** with types "const void*" and "int" and returns an "int".
+*/
+static int (*ftsHashFunction(int keyClass))(const void*,int){
+ if( keyClass==FTS3_HASH_STRING ){
+ return &fts3StrHash;
+ }else{
+ assert( keyClass==FTS3_HASH_BINARY );
+ return &fts3BinHash;
+ }
+}
+
+/*
+** Return a pointer to the appropriate hash function given the key class.
+**
+** For help in interpreted the obscure C code in the function definition,
+** see the header comment on the previous function.
+*/
+static int (*ftsCompareFunction(int keyClass))(const void*,int,const void*,int){
+ if( keyClass==FTS3_HASH_STRING ){
+ return &fts3StrCompare;
+ }else{
+ assert( keyClass==FTS3_HASH_BINARY );
+ return &fts3BinCompare;
+ }
+}
+
+/* Link an element into the hash table
+*/
+static void fts3HashInsertElement(
+ fts3Hash *pH, /* The complete hash table */
+ struct _fts3ht *pEntry, /* The entry into which pNew is inserted */
+ fts3HashElem *pNew /* The element to be inserted */
+){
+ fts3HashElem *pHead; /* First element already in pEntry */
+ pHead = pEntry->chain;
+ if( pHead ){
+ pNew->next = pHead;
+ pNew->prev = pHead->prev;
+ if( pHead->prev ){ pHead->prev->next = pNew; }
+ else { pH->first = pNew; }
+ pHead->prev = pNew;
+ }else{
+ pNew->next = pH->first;
+ if( pH->first ){ pH->first->prev = pNew; }
+ pNew->prev = 0;
+ pH->first = pNew;
+ }
+ pEntry->count++;
+ pEntry->chain = pNew;
+}
+
+
+/* Resize the hash table so that it cantains "new_size" buckets.
+** "new_size" must be a power of 2. The hash table might fail
+** to resize if sqliteMalloc() fails.
+*/
+static void fts3Rehash(fts3Hash *pH, int new_size){
+ struct _fts3ht *new_ht; /* The new hash table */
+ fts3HashElem *elem, *next_elem; /* For looping over existing elements */
+ int (*xHash)(const void*,int); /* The hash function */
+
+ assert( (new_size & (new_size-1))==0 );
+ new_ht = (struct _fts3ht *)fts3HashMalloc( new_size*sizeof(struct _fts3ht) );
+ if( new_ht==0 ) return;
+ fts3HashFree(pH->ht);
+ pH->ht = new_ht;
+ pH->htsize = new_size;
+ xHash = ftsHashFunction(pH->keyClass);
+ for(elem=pH->first, pH->first=0; elem; elem = next_elem){
+ int h = (*xHash)(elem->pKey, elem->nKey) & (new_size-1);
+ next_elem = elem->next;
+ fts3HashInsertElement(pH, &new_ht[h], elem);
+ }
+}
+
+/* This function (for internal use only) locates an element in an
+** hash table that matches the given key. The hash for this key has
+** already been computed and is passed as the 4th parameter.
+*/
+static fts3HashElem *fts3FindElementByHash(
+ const fts3Hash *pH, /* The pH to be searched */
+ const void *pKey, /* The key we are searching for */
+ int nKey,
+ int h /* The hash for this key. */
+){
+ fts3HashElem *elem; /* Used to loop thru the element list */
+ int count; /* Number of elements left to test */
+ int (*xCompare)(const void*,int,const void*,int); /* comparison function */
+
+ if( pH->ht ){
+ struct _fts3ht *pEntry = &pH->ht[h];
+ elem = pEntry->chain;
+ count = pEntry->count;
+ xCompare = ftsCompareFunction(pH->keyClass);
+ while( count-- && elem ){
+ if( (*xCompare)(elem->pKey,elem->nKey,pKey,nKey)==0 ){
+ return elem;
+ }
+ elem = elem->next;
+ }
+ }
+ return 0;
+}
+
+/* Remove a single entry from the hash table given a pointer to that
+** element and a hash on the element's key.
+*/
+static void fts3RemoveElementByHash(
+ fts3Hash *pH, /* The pH containing "elem" */
+ fts3HashElem* elem, /* The element to be removed from the pH */
+ int h /* Hash value for the element */
+){
+ struct _fts3ht *pEntry;
+ if( elem->prev ){
+ elem->prev->next = elem->next;
+ }else{
+ pH->first = elem->next;
+ }
+ if( elem->next ){
+ elem->next->prev = elem->prev;
+ }
+ pEntry = &pH->ht[h];
+ if( pEntry->chain==elem ){
+ pEntry->chain = elem->next;
+ }
+ pEntry->count--;
+ if( pEntry->count<=0 ){
+ pEntry->chain = 0;
+ }
+ if( pH->copyKey && elem->pKey ){
+ fts3HashFree(elem->pKey);
+ }
+ fts3HashFree( elem );
+ pH->count--;
+ if( pH->count<=0 ){
+ assert( pH->first==0 );
+ assert( pH->count==0 );
+ fts3HashClear(pH);
+ }
+}
+
+/* Attempt to locate an element of the hash table pH with a key
+** that matches pKey,nKey. Return the data for this element if it is
+** found, or NULL if there is no match.
+*/
+void *sqlite3Fts3HashFind(const fts3Hash *pH, const void *pKey, int nKey){
+ int h; /* A hash on key */
+ fts3HashElem *elem; /* The element that matches key */
+ int (*xHash)(const void*,int); /* The hash function */
+
+ if( pH==0 || pH->ht==0 ) return 0;
+ xHash = ftsHashFunction(pH->keyClass);
+ assert( xHash!=0 );
+ h = (*xHash)(pKey,nKey);
+ assert( (pH->htsize & (pH->htsize-1))==0 );
+ elem = fts3FindElementByHash(pH,pKey,nKey, h & (pH->htsize-1));
+ return elem ? elem->data : 0;
+}
+
+/* Insert an element into the hash table pH. The key is pKey,nKey
+** and the data is "data".
+**
+** If no element exists with a matching key, then a new
+** element is created. A copy of the key is made if the copyKey
+** flag is set. NULL is returned.
+**
+** If another element already exists with the same key, then the
+** new data replaces the old data and the old data is returned.
+** The key is not copied in this instance. If a malloc fails, then
+** the new data is returned and the hash table is unchanged.
+**
+** If the "data" parameter to this function is NULL, then the
+** element corresponding to "key" is removed from the hash table.
+*/
+void *sqlite3Fts3HashInsert(
+ fts3Hash *pH, /* The hash table to insert into */
+ const void *pKey, /* The key */
+ int nKey, /* Number of bytes in the key */
+ void *data /* The data */
+){
+ int hraw; /* Raw hash value of the key */
+ int h; /* the hash of the key modulo hash table size */
+ fts3HashElem *elem; /* Used to loop thru the element list */
+ fts3HashElem *new_elem; /* New element added to the pH */
+ int (*xHash)(const void*,int); /* The hash function */
+
+ assert( pH!=0 );
+ xHash = ftsHashFunction(pH->keyClass);
+ assert( xHash!=0 );
+ hraw = (*xHash)(pKey, nKey);
+ assert( (pH->htsize & (pH->htsize-1))==0 );
+ h = hraw & (pH->htsize-1);
+ elem = fts3FindElementByHash(pH,pKey,nKey,h);
+ if( elem ){
+ void *old_data = elem->data;
+ if( data==0 ){
+ fts3RemoveElementByHash(pH,elem,h);
+ }else{
+ elem->data = data;
+ }
+ return old_data;
+ }
+ if( data==0 ) return 0;
+ new_elem = (fts3HashElem*)fts3HashMalloc( sizeof(fts3HashElem) );
+ if( new_elem==0 ) return data;
+ if( pH->copyKey && pKey!=0 ){
+ new_elem->pKey = fts3HashMalloc( nKey );
+ if( new_elem->pKey==0 ){
+ fts3HashFree(new_elem);
+ return data;
+ }
+ memcpy((void*)new_elem->pKey, pKey, nKey);
+ }else{
+ new_elem->pKey = (void*)pKey;
+ }
+ new_elem->nKey = nKey;
+ pH->count++;
+ if( pH->htsize==0 ){
+ fts3Rehash(pH,8);
+ if( pH->htsize==0 ){
+ pH->count = 0;
+ fts3HashFree(new_elem);
+ return data;
+ }
+ }
+ if( pH->count > pH->htsize ){
+ fts3Rehash(pH,pH->htsize*2);
+ }
+ assert( pH->htsize>0 );
+ assert( (pH->htsize & (pH->htsize-1))==0 );
+ h = hraw & (pH->htsize-1);
+ fts3HashInsertElement(pH, &pH->ht[h], new_elem);
+ new_elem->data = data;
+ return 0;
+}
+
+
Added: branches/indexer-split/src/tracker-fts/tracker-fts-hash.h
==============================================================================
--- (empty file)
+++ branches/indexer-split/src/tracker-fts/tracker-fts-hash.h Wed Aug 27 01:53:50 2008
@@ -0,0 +1,110 @@
+/*
+** 2001 September 22
+**
+** 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 the header file for the generic hash-table implemenation
+** used in SQLite. We've modified it slightly to serve as a standalone
+** hash table implementation for the full-text indexing module.
+**
+*/
+#ifndef _FTS3_HASH_H_
+#define _FTS3_HASH_H_
+
+/* Forward declarations of structures. */
+typedef struct fts3Hash fts3Hash;
+typedef struct fts3HashElem fts3HashElem;
+
+/* A complete hash table is an instance of the following structure.
+** The internals of this structure are intended to be opaque -- client
+** code should not attempt to access or modify the fields of this structure
+** directly. Change this structure only by using the routines below.
+** However, many of the "procedures" and "functions" for modifying and
+** accessing this structure are really macros, so we can't really make
+** this structure opaque.
+*/
+struct fts3Hash {
+ char keyClass; /* HASH_INT, _POINTER, _STRING, _BINARY */
+ char copyKey; /* True if copy of key made on insert */
+ int count; /* Number of entries in this table */
+ fts3HashElem *first; /* The first element of the array */
+ int htsize; /* Number of buckets in the hash table */
+ struct _fts3ht { /* the hash table */
+ int count; /* Number of entries with this hash */
+ fts3HashElem *chain; /* Pointer to first entry with this hash */
+ } *ht;
+};
+
+/* Each element in the hash table is an instance of the following
+** structure. All elements are stored on a single doubly-linked list.
+**
+** Again, this structure is intended to be opaque, but it can't really
+** be opaque because it is used by macros.
+*/
+struct fts3HashElem {
+ fts3HashElem *next, *prev; /* Next and previous elements in the table */
+ void *data; /* Data associated with this element */
+ void *pKey; int nKey; /* Key associated with this element */
+};
+
+/*
+** There are 2 different modes of operation for a hash table:
+**
+** FTS3_HASH_STRING pKey points to a string that is nKey bytes long
+** (including the null-terminator, if any). Case
+** is respected in comparisons.
+**
+** FTS3_HASH_BINARY pKey points to binary data nKey bytes long.
+** memcmp() is used to compare keys.
+**
+** A copy of the key is made if the copyKey parameter to fts3HashInit is 1.
+*/
+#define FTS3_HASH_STRING 1
+#define FTS3_HASH_BINARY 2
+
+/*
+** Access routines. To delete, insert a NULL pointer.
+*/
+void sqlite3Fts3HashInit(fts3Hash*, int keytype, int copyKey);
+void *sqlite3Fts3HashInsert(fts3Hash*, const void *pKey, int nKey, void *pData);
+void *sqlite3Fts3HashFind(const fts3Hash*, const void *pKey, int nKey);
+void sqlite3Fts3HashClear(fts3Hash*);
+
+/*
+** Shorthand for the functions above
+*/
+#define fts3HashInit sqlite3Fts3HashInit
+#define fts3HashInsert sqlite3Fts3HashInsert
+#define fts3HashFind sqlite3Fts3HashFind
+#define fts3HashClear sqlite3Fts3HashClear
+
+/*
+** Macros for looping over all elements of a hash table. The idiom is
+** like this:
+**
+** fts3Hash h;
+** fts3HashElem *p;
+** ...
+** for(p=fts3HashFirst(&h); p; p=fts3HashNext(p)){
+** SomeStructure *pData = fts3HashData(p);
+** // do something with pData
+** }
+*/
+#define fts3HashFirst(H) ((H)->first)
+#define fts3HashNext(E) ((E)->next)
+#define fts3HashData(E) ((E)->data)
+#define fts3HashKey(E) ((E)->pKey)
+#define fts3HashKeysize(E) ((E)->nKey)
+
+/*
+** Number of entries in a hash table
+*/
+#define fts3HashCount(H) ((H)->count)
+
+#endif /* _FTS3_HASH_H_ */
Added: branches/indexer-split/src/tracker-fts/tracker-fts.c
==============================================================================
--- (empty file)
+++ branches/indexer-split/src/tracker-fts/tracker-fts.c Wed Aug 27 01:53:50 2008
@@ -0,0 +1,7230 @@
+/*
+** 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.
+*/
+
+
+// gcc -shared -o tracker-fts *.c
+//gcc -Wall -fPIC -c *.c
+
+//gcc -shared -Wl,-soname,libtracker-fts.so.1 -o libtracker-fts.so.1.0 *.o
+
+/*
+** 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).
+*/
+
+/* 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.
+*/
+
+#include <assert.h>
+#include <stdlib.h>
+#include <stdio.h>
+#include <string.h>
+#include <ctype.h>
+#include <sqlite3ext.h>
+
+#include <libtracker-common/tracker-config.h>
+#include <libtracker-common/tracker-language.h>
+#include <libtracker-common/tracker-parser.h>
+
+
+#include "tracker-fts.h"
+#include "tracker-fts-hash.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 FTSTRACE(A) printf A; fflush(stdout)
+#else
+# define FTSTRACE(A)
+#endif
+
+/*
+** Default span for NEAR operators.
+*/
+#define SQLITE_FTS3_DEFAULT_NEAR_PARAM 10
+
+/* 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&0x80)==0 ? isspace(c) : 0;
+}
+static int safe_tolower(char c){
+ return (c&0x80)==0 ? tolower(c) : c;
+}
+static int safe_isalnum(char c){
+ return (c&0x80)==0 ? isalnum(c) : 0;
+}
+
+int sqlite3_extension_init(
+ sqlite3 *db,
+ char **pzErrMsg,
+ const sqlite3_api_routines *pApi
+);
+
+
+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 fts3PutVarint(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 fts3GetVarint(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 fts3GetVarint32(const char *p, int *pi){
+ sqlite_int64 i;
+ int ret = fts3GetVarint(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 = fts3GetVarint(pReader->pData, &iDummy);
+ assert( !dlrAtEnd(pReader) );
+ return pReader->pData+n;
+}
+static int dlrPosDataLen(DLReader *pReader){
+ sqlite_int64 iDummy;
+ int n = fts3GetVarint(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 = fts3GetVarint(pReader->pData, &iDocidDelta);
+ pReader->iDocid += iDocidDelta;
+ if( pReader->iType>=DL_POSITIONS ){
+ assert( n<pReader->nData );
+ while( 1 ){
+ n += fts3GetVarint32(pReader->pData+n, &iDummy);
+ assert( n<=pReader->nData );
+ if( iDummy==POS_END ) break;
+ if( iDummy==POS_COLUMN ){
+ n += fts3GetVarint32(pReader->pData+n, &iDummy);
+ assert( n<pReader->nData );
+ }else if( pReader->iType==DL_POSITIONS_OFFSETS ){
+ n += fts3GetVarint32(pReader->pData+n, &iDummy);
+ n += fts3GetVarint32(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 = fts3GetVarint(pData, &iDocidDelta);
+ iPrevDocid += iDocidDelta;
+ if( iType>DL_DOCIDS ){
+ int iDummy;
+ while( 1 ){
+ n += fts3GetVarint32(pData+n, &iDummy);
+ if( iDummy==POS_END ) break;
+ if( iDummy==POS_COLUMN ){
+ n += fts3GetVarint32(pData+n, &iDummy);
+ }else if( iType>DL_POSITIONS ){
+ n += fts3GetVarint32(pData+n, &iDummy);
+ n += fts3GetVarint32(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 = fts3GetVarint(pData, &iDocid);
+ assert( nFirstOld<nData || (nFirstOld==nData && pWriter->iType==DL_DOCIDS) );
+ nFirstNew = fts3PutVarint(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 = fts3PutVarint(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 = fts3GetVarint32(pReader->pData, &i);
+ if( i==POS_COLUMN ){
+ n += fts3GetVarint32(pReader->pData+n, &pReader->iColumn);
+ pReader->iPosition = 0;
+ pReader->iStartOffset = 0;
+ n += fts3GetVarint32(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 += fts3GetVarint32(pReader->pData+n, &i);
+ pReader->iStartOffset += i;
+ n += fts3GetVarint32(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 += fts3PutVarint(c+n, POS_COLUMN);
+ n += fts3PutVarint(c+n, iColumn);
+ pWriter->iColumn = iColumn;
+ pWriter->iPos = 0;
+ pWriter->iOffset = 0;
+ }
+ assert( iPos>=pWriter->iPos );
+ n += fts3PutVarint(c+n, POS_BASE+(iPos-pWriter->iPos));
+ pWriter->iPos = iPos;
+ if( pWriter->dlw->iType==DL_POSITIONS_OFFSETS ){
+ assert( iStartOffset>=pWriter->iOffset );
+ n += fts3PutVarint(c+n, iStartOffset-pWriter->iOffset);
+ pWriter->iOffset = iStartOffset;
+ assert( iEndOffset>=iStartOffset );
+ n += fts3PutVarint(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 = fts3PutVarint(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 = fts3PutVarint(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 = fts3PutVarint(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);
+}
+
+/*
+** This function is used as part of the implementation of phrase and
+** NEAR matching.
+**
+** pLeft and pRight are DLReaders positioned to the same docid in
+** lists of type DL_POSITION. This function writes an entry to the
+** DLWriter pOut for each position in pRight that is less than
+** (nNear+1) greater (but not equal to or smaller) than a position
+** in pLeft. For example, if nNear is 0, and the positions contained
+** by pLeft and pRight are:
+**
+** pLeft: 5 10 15 20
+** pRight: 6 9 17 21
+**
+** then the docid is added to pOut. If pOut is of type DL_POSITIONS,
+** then a positionids "6" and "21" are also added to pOut.
+**
+** If boolean argument isSaveLeft is true, then positionids are copied
+** from pLeft instead of pRight. In the example above, the positions "5"
+** and "20" would be added instead of "6" and "21".
+*/
+static void posListPhraseMerge(
+ DLReader *pLeft,
+ DLReader *pRight,
+ int nNear,
+ int isSaveLeft,
+ 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)>=plrPosition(&right) ){
+ plrStep(&right);
+ }else{
+ if( (plrPosition(&right)-plrPosition(&left))<=(nNear+1) ){
+ if( !match ){
+ plwInit(&writer, pOut, dlrDocid(pLeft));
+ match = 1;
+ }
+ if( !isSaveLeft ){
+ plwAdd(&writer, plrColumn(&right), plrPosition(&right), 0, 0);
+ }else{
+ plwAdd(&writer, plrColumn(&left), plrPosition(&left), 0, 0);
+ }
+ plrStep(&right);
+ }else{
+ plrStep(&left);
+ }
+ }
+ }
+
+ if( match ){
+ plwTerminate(&writer);
+ plwDestroy(&writer);
+ }
+
+ plrDestroy(&left);
+ plrDestroy(&right);
+}
+
+/*
+** Compare the values pointed to by the PLReaders passed as arguments.
+** Return -1 if the value pointed to by pLeft is considered less than
+** the value pointed to by pRight, +1 if it is considered greater
+** than it, or 0 if it is equal. i.e.
+**
+** (*pLeft - *pRight)
+**
+** A PLReader that is in the EOF condition is considered greater than
+** any other. If neither argument is in EOF state, the return value of
+** plrColumn() is used. If the plrColumn() values are equal, the
+** comparison is on the basis of plrPosition().
+*/
+static int plrCompare(PLReader *pLeft, PLReader *pRight){
+ assert(!plrAtEnd(pLeft) || !plrAtEnd(pRight));
+
+ if( plrAtEnd(pRight) || plrAtEnd(pLeft) ){
+ return (plrAtEnd(pRight) ? -1 : 1);
+ }
+ if( plrColumn(pLeft)!=plrColumn(pRight) ){
+ return ((plrColumn(pLeft)<plrColumn(pRight)) ? -1 : 1);
+ }
+ if( plrPosition(pLeft)!=plrPosition(pRight) ){
+ return ((plrPosition(pLeft)<plrPosition(pRight)) ? -1 : 1);
+ }
+ return 0;
+}
+
+/* We have two doclists with positions: pLeft and pRight. Depending
+** on the value of the nNear parameter, perform either a phrase
+** intersection (if nNear==0) or a NEAR intersection (if nNear>0)
+** and write the results into pOut.
+**
+** A phrase intersection means that two documents only match
+** if pLeft.iPos+1==pRight.iPos.
+**
+** A NEAR intersection means that two documents only match if
+** (abs(pLeft.iPos-pRight.iPos)<nNear).
+**
+** If a NEAR intersection is requested, then the nPhrase argument should
+** be passed the number of tokens in the two operands to the NEAR operator
+** combined. For example:
+**
+** Query syntax nPhrase
+** ------------------------------------
+** "A B C" NEAR "D E" 5
+** A NEAR B 2
+**
+** 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,
+ int nNear, /* 0 for a phrase merge, non-zero for a NEAR merge */
+ int nPhrase, /* Number of tokens in left+right operands to NEAR */
+ DocListType iType, /* Type of doclist to write to pOut */
+ 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{
+ if( nNear==0 ){
+ posListPhraseMerge(&left, &right, 0, 0, &writer);
+ }else{
+ /* This case occurs when two terms (simple terms or phrases) are
+ * connected by a NEAR operator, span (nNear+1). i.e.
+ *
+ * '"terrible company" NEAR widget'
+ */
+ DataBuffer one = {0, 0, 0};
+ DataBuffer two = {0, 0, 0};
+
+ DLWriter dlwriter2;
+ DLReader dr1 = {0, 0, 0, 0, 0};
+ DLReader dr2 = {0, 0, 0, 0, 0};
+
+ dlwInit(&dlwriter2, iType, &one);
+ posListPhraseMerge(&right, &left, nNear-3+nPhrase, 1, &dlwriter2);
+ dlwInit(&dlwriter2, iType, &two);
+ posListPhraseMerge(&left, &right, nNear-1, 0, &dlwriter2);
+
+ if( one.nData) dlrInit(&dr1, iType, one.pData, one.nData);
+ if( two.nData) dlrInit(&dr2, iType, two.pData, two.nData);
+
+ if( !dlrAtEnd(&dr1) || !dlrAtEnd(&dr2) ){
+ PLReader pr1 = {0};
+ PLReader pr2 = {0};
+
+ PLWriter plwriter;
+ plwInit(&plwriter, &writer, dlrDocid(dlrAtEnd(&dr1)?&dr2:&dr1));
+
+ if( one.nData ) plrInit(&pr1, &dr1);
+ if( two.nData ) plrInit(&pr2, &dr2);
+ while( !plrAtEnd(&pr1) || !plrAtEnd(&pr2) ){
+ int iCompare = plrCompare(&pr1, &pr2);
+ switch( iCompare ){
+ case -1:
+ plwCopy(&plwriter, &pr1);
+ plrStep(&pr1);
+ break;
+ case 1:
+ plwCopy(&plwriter, &pr2);
+ plrStep(&pr2);
+ break;
+ case 0:
+ plwCopy(&plwriter, &pr1);
+ plrStep(&pr1);
+ plrStep(&pr2);
+ break;
+ }
+ }
+ plwTerminate(&plwriter);
+ }
+ dataBufferDestroy(&one);
+ dataBufferDestroy(&two);
+ }
+ 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;
+ FTSTRACE(("FTS3 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;
+ FTSTRACE(("FTS3 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. Each word which may match against
+** document content is a term. Operators, like NEAR or OR, are
+** not terms. Query terms are organized as a flat list stored
+** in the Query.pTerms array.
+**
+** If the QueryTerm.nPhrase variable is non-zero, then the QueryTerm
+** is the first in a contiguous string of terms that are either part
+** of the same phrase, or connected by the NEAR operator.
+**
+** If the QueryTerm.nNear variable is non-zero, then the token is followed
+** by a NEAR operator with span set to (nNear-1). For example, the
+** following query:
+**
+** The QueryTerm.iPhrase variable stores the index of the token within
+** its phrase, indexed starting at 1, or 1 if the token is not part
+** of any phrase.
+**
+** For example, the data structure used to represent the following query:
+**
+** ... MATCH 'sqlite NEAR/5 google NEAR/2 "search engine"'
+**
+** is:
+**
+** {nPhrase=4, iPhrase=1, nNear=6, pTerm="sqlite"},
+** {nPhrase=0, iPhrase=1, nNear=3, pTerm="google"},
+** {nPhrase=0, iPhrase=1, nNear=0, pTerm="search"},
+** {nPhrase=0, iPhrase=2, nNear=0, pTerm="engine"},
+**
+** compiling the FTS3 syntax to Query structures is done by the parseQuery()
+** function.
+*/
+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 nNear; /* term followed by a NEAR operator with span=(nNear-1) */
+ 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 nextIsNear; /* 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 */
+ int iToken; /* The index of the matching document token */
+ 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_DOCID, /* lookup by docid */
+ 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 */ NULL, /* generated in contentSelectStatement() */
+ /* CONTENT_UPDATE */ NULL, /* generated in contentUpdateStatement() */
+ /* CONTENT_DELETE */ "delete from %_content where docid = ?",
+ /* CONTENT_EXISTS */ "select docid from %_content limit 1",
+
+ /* BLOCK_INSERT */
+ "insert into %_segments (blockid, block) values (null, ?)",
+ /* BLOCK_SELECT */ "select block from %_segments where blockid = ?",
+ /* BLOCK_DELETE */ "delete from %_segments where blockid 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 */
+ TrackerParser *parser; /* tokenizer for inserts and queries */
+ int max_words;
+
+ /* 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 blockid between ? and ? order by blockid"
+
+ /* 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;
+ fts3Hash 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 fts3Module; /* forward declaration */
+
+/* Return a dynamically generated statement of the form
+ * insert into %_content (docid, ...) values (?, ...)
+ */
+static const char *contentInsertStatement(fulltext_vtab *v){
+ StringBuffer sb;
+ int i;
+
+ initStringBuffer(&sb);
+ append(&sb, "insert into %_content (docid, ");
+ 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
+ * select <content columns> from %_content where docid = ?
+ */
+static const char *contentSelectStatement(fulltext_vtab *v){
+ StringBuffer sb;
+ initStringBuffer(&sb);
+ append(&sb, "SELECT ");
+ appendList(&sb, v->nColumn, v->azContentColumn);
+ append(&sb, " FROM %_content WHERE docid = ?");
+ return stringBufferData(&sb);
+}
+
+/* Return a dynamically generated statement of the form
+ * update %_content set [col_0] = ?, [col_1] = ?, ...
+ * where docid = ?
+ */
+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 docid = ?");
+ 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_SELECT_STMT:
+ zStmt = contentSelectStatement(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 (docid, ...) values ([docid], [pValues])
+** If the docid contains SQL NULL, then a unique docid will be
+** generated.
+*/
+static int content_insert(fulltext_vtab *v, sqlite3_value *docid,
+ 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, docid);
+ 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 docid = [iDocid] */
+static int content_update(fulltext_vtab *v, sqlite3_value **pValues,
+ sqlite_int64 iDocid){
+ 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, iDocid);
+ 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 docid = [iDocid]
+ * 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 iDocid,
+ 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, iDocid);
+ 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 docid = [iDocid ] */
+static int content_delete(fulltext_vtab *v, sqlite_int64 iDocid){
+ 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, iDocid);
+ 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 blockid 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;
+
+ /* blockid column is an alias for rowid. */
+ *piBlockid = sqlite3_last_insert_rowid(v->db);
+ return SQLITE_OK;
+}
+
+/* delete from %_segments
+** where blockid 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;
+
+ FTSTRACE(("FTS3 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->parser!=NULL ){
+ tracker_parser_free (v->parser);
+ v->parser = 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
+** ftsIdChar(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,
+** isFtsIdChar[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 isFtsIdChar[] = {
+/* 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 ftsIdChar(C) (((c=C)&0x80)!=0 || (c>0x1f && isFtsIdChar[c-0x20]))
+
+
+/*
+** Return the length of the token that begins at z[0].
+** Store the token type in *tokenType before returning.
+*/
+static int ftsGetToken(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( !ftsIdChar(*z) ){
+ break;
+ }
+ for(i=1; ftsIdChar(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 FtsToken {
+ const char *z; /* Pointer to token text. Not '\000' terminated */
+ short int n; /* Length of the token text in bytes. */
+} FtsToken;
+
+/*
+** 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;
+ FtsToken *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 = ftsGetToken(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 = ftsGetToken(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 fts3(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.
+ */
+ pSpec->azTokenizer = tokenizeString(zTokenizer, &n);
+ tokenListToIdList(pSpec->azTokenizer);
+
+ return SQLITE_OK;
+}
+
+/*
+** Generate a CREATE TABLE statement that describes the schema of
+** the virtual table. Return a pointer to this schema string.
+**
+** Space is obtained from sqlite3_mprintf() and should be freed
+** using sqlite3_free().
+*/
+static char *fulltextSchema(
+ int nColumn, /* Number of columns */
+ const char *const* azColumn, /* List of columns */
+ const char *zTableName /* Name of the table */
+){
+ int i;
+ char *zSchema, *zNext;
+ const char *zSep = "(";
+ zSchema = sqlite3_mprintf("CREATE TABLE x");
+ for(i=0; i<nColumn; i++){
+ zNext = sqlite3_mprintf("%s%s%Q", zSchema, zSep, azColumn[i]);
+ sqlite3_free(zSchema);
+ zSchema = zNext;
+ zSep = ",";
+ }
+ zNext = sqlite3_mprintf("%s,%Q HIDDEN", zSchema, zTableName);
+ sqlite3_free(zSchema);
+ zSchema = zNext;
+ zNext = sqlite3_mprintf("%s,docid HIDDEN)", zSchema);
+ sqlite3_free(zSchema);
+ return zNext;
+}
+
+/*
+** Build a new sqlite3_vtab structure that will describe the
+** fulltext index defined by spec.
+*/
+static int constructVtab(
+ sqlite3 *db, /* The SQLite database connection */
+ TableSpec *spec, /* Parsed spec information from parseSpec() */
+ sqlite3_vtab **ppVTab, /* Write the resulting vtab structure here */
+ char **pzErr /* Write any error message here */
+){
+ int rc;
+ fulltext_vtab *v = 0;
+ char *schema;
+
+
+ v = (fulltext_vtab *) sqlite3_malloc(sizeof(fulltext_vtab));
+ if( v==0 ) return SQLITE_NOMEM;
+ CLEAR(v);
+ /* sqlite will initialize v->base */
+ v->db = db;
+ v->zDb = spec->zDb; /* Freed when azColumn is freed */
+ v->zName = spec->zName; /* Freed when azColumn is freed */
+ v->nColumn = spec->nColumn;
+ v->azContentColumn = spec->azContentColumn;
+ spec->azContentColumn = 0;
+ v->azColumn = spec->azColumn;
+ spec->azColumn = 0;
+
+/* comment out tokenizer stuff
+ if( spec->azTokenizer==0 ){
+ return SQLITE_NOMEM;
+ }
+
+ zTok = spec->azTokenizer[0];
+ if( !zTok ){
+ zTok = "simple";
+ }
+ nTok = strlen(zTok)+1;
+
+ m = (sqlite3_tokenizer_module *)sqlite3Fts3HashFind(pHash, zTok, nTok);
+ if( !m ){
+ *pzErr = sqlite3_mprintf("unknown tokenizer: %s", spec->azTokenizer[0]);
+ rc = SQLITE_ERROR;
+ goto err;
+ }
+
+ for(n=0; spec->azTokenizer[n]; n++){}
+ if( n ){
+ rc = m->xCreate(n-1, (const char*const*)&spec->azTokenizer[1],
+ &v->pTokenizer);
+ }else{
+ rc = m->xCreate(0, 0, &v->pTokenizer);
+ }
+ if( rc!=SQLITE_OK ) goto err;
+ */
+
+
+ /* set up our parser */
+
+ TrackerConfig *config = tracker_config_new ();
+
+ TrackerLanguage *language = tracker_language_new (config);
+
+ int min_len = tracker_config_get_min_word_length (config);
+ int max_len = tracker_config_get_max_word_length (config);
+ v->max_words = tracker_config_get_max_words_to_index (config);
+
+ v->parser = tracker_parser_new (language, max_len, min_len);
+
+ g_object_unref (language);
+
+
+ /* TODO: verify the existence of backing tables foo_content, foo_term */
+
+ schema = fulltextSchema(v->nColumn, (const char*const*)v->azColumn,
+ spec->zName);
+ rc = sqlite3_declare_vtab(db, schema);
+ sqlite3_free(schema);
+ if( rc!=SQLITE_OK ) goto err;
+
+ memset(v->pFulltextStatements, 0, sizeof(v->pFulltextStatements));
+
+ /* Indicate that the buffer is not live. */
+ v->nPendingData = -1;
+
+ *ppVTab = &v->base;
+ FTSTRACE(("FTS3 Connect %p\n", v));
+
+ return rc;
+
+err:
+ fulltext_vtab_destroy(v);
+ return rc;
+}
+
+static int fulltextConnect(
+ sqlite3 *db,
+ void *pAux,
+ int argc, const char *const*argv,
+ sqlite3_vtab **ppVTab,
+ char **pzErr
+){
+ TableSpec spec;
+ int rc = parseSpec(&spec, argc, argv, pzErr);
+ if( rc!=SQLITE_OK ) return rc;
+
+ rc = constructVtab(db, &spec, ppVTab, pzErr);
+ clearTableSpec(&spec);
+ return rc;
+}
+
+/* The %_content table holds the text of each document, with
+** the docid column exposed as the SQLite rowid for the table.
+*/
+/* TODO(shess) This comment needs elaboration to match the updated
+** code. Work it into the top-of-file comment at that time.
+*/
+static int fulltextCreate(sqlite3 *db, void *pAux,
+ int argc, const char * const *argv,
+ sqlite3_vtab **ppVTab, char **pzErr){
+ int rc;
+ TableSpec spec;
+ StringBuffer schema;
+ FTSTRACE(("FTS3 Create\n"));
+
+ rc = parseSpec(&spec, argc, argv, pzErr);
+ if( rc!=SQLITE_OK ) return rc;
+
+ initStringBuffer(&schema);
+ append(&schema, "CREATE TABLE %_content(");
+ append(&schema, " docid INTEGER PRIMARY KEY,");
+ appendList(&schema, spec.nColumn, spec.azContentColumn);
+ append(&schema, ")");
+ rc = sql_exec(db, spec.zDb, spec.zName, stringBufferData(&schema));
+ stringBufferDestroy(&schema);
+ if( rc!=SQLITE_OK ) goto out;
+
+ rc = sql_exec(db, spec.zDb, spec.zName,
+ "create table %_segments("
+ " blockid INTEGER PRIMARY KEY,"
+ " block blob"
+ ");"
+ );
+ if( rc!=SQLITE_OK ) goto out;
+
+ rc = sql_exec(db, spec.zDb, spec.zName,
+ "create table %_segdir("
+ " level integer,"
+ " idx integer,"
+ " start_block integer,"
+ " leaves_end_block integer,"
+ " end_block integer,"
+ " root blob,"
+ " primary key(level, idx)"
+ ");");
+ if( rc!=SQLITE_OK ) goto out;
+
+ rc = constructVtab(db, &spec, ppVTab, pzErr);
+
+out:
+ clearTableSpec(&spec);
+ return rc;
+}
+
+/* Decide how to handle an SQL query. */
+static int fulltextBestIndex(sqlite3_vtab *pVTab, sqlite3_index_info *pInfo){
+ fulltext_vtab *v = (fulltext_vtab *)pVTab;
+ int i;
+ FTSTRACE(("FTS3 BestIndex\n"));
+
+ for(i=0; i<pInfo->nConstraint; ++i){
+ const struct sqlite3_index_constraint *pConstraint;
+ pConstraint = &pInfo->aConstraint[i];
+ if( pConstraint->usable ) {
+ if( (pConstraint->iColumn==-1 || pConstraint->iColumn==v->nColumn+1) &&
+ pConstraint->op==SQLITE_INDEX_CONSTRAINT_EQ ){
+ pInfo->idxNum = QUERY_DOCID; /* lookup by docid */
+ FTSTRACE(("FTS3 QUERY_DOCID\n"));
+ } else if( pConstraint->iColumn>=0 && pConstraint->iColumn<=v->nColumn &&
+ pConstraint->op==SQLITE_INDEX_CONSTRAINT_MATCH ){
+ /* full-text search */
+ pInfo->idxNum = QUERY_FULLTEXT + pConstraint->iColumn;
+ FTSTRACE(("FTS3 QUERY_FULLTEXT %d\n", pConstraint->iColumn));
+ } else continue;
+
+ pInfo->aConstraintUsage[i].argvIndex = 1;
+ pInfo->aConstraintUsage[i].omit = 1;
+
+ /* An arbitrary value for now.
+ * TODO: Perhaps docid matches should be considered cheaper than
+ * full-text searches. */
+ pInfo->estimatedCost = 1.0;
+
+ return SQLITE_OK;
+ }
+ }
+ pInfo->idxNum = QUERY_GENERIC;
+ return SQLITE_OK;
+}
+
+static int fulltextDisconnect(sqlite3_vtab *pVTab){
+ FTSTRACE(("FTS3 Disconnect %p\n", pVTab));
+ fulltext_vtab_destroy((fulltext_vtab *)pVTab);
+ return SQLITE_OK;
+}
+
+static int fulltextDestroy(sqlite3_vtab *pVTab){
+ fulltext_vtab *v = (fulltext_vtab *)pVTab;
+ int rc;
+
+ FTSTRACE(("FTS3 Destroy %p\n", pVTab));
+ rc = sql_exec(v->db, v->zDb, v->zName,
+ "drop table if exists %_content;"
+ "drop table if exists %_segments;"
+ "drop table if exists %_segdir;"
+ );
+ if( rc!=SQLITE_OK ) return rc;
+
+ fulltext_vtab_destroy((fulltext_vtab *)pVTab);
+ return SQLITE_OK;
+}
+
+static int fulltextOpen(sqlite3_vtab *pVTab, sqlite3_vtab_cursor **ppCursor){
+ fulltext_cursor *c;
+
+ c = (fulltext_cursor *) sqlite3_malloc(sizeof(fulltext_cursor));
+ if( c ){
+ memset(c, 0, sizeof(fulltext_cursor));
+ /* sqlite will initialize c->base */
+ *ppCursor = &c->base;
+ FTSTRACE(("FTS3 Open %p: %p\n", pVTab, c));
+ return SQLITE_OK;
+ }else{
+ return SQLITE_NOMEM;
+ }
+}
+
+
+/* Free all of the dynamically allocated memory held by *q
+*/
+static void queryClear(Query *q){
+ int i;
+ for(i = 0; i < q->nTerms; ++i){
+ sqlite3_free(q->pTerms[i].pTerm);
+ }
+ sqlite3_free(q->pTerms);
+ CLEAR(q);
+}
+
+/* Free all of the dynamically allocated memory held by the
+** Snippet
+*/
+static void snippetClear(Snippet *p){
+ sqlite3_free(p->aMatch);
+ sqlite3_free(p->zOffset);
+ sqlite3_free(p->zSnippet);
+ CLEAR(p);
+}
+/*
+** Append a single entry to the p->aMatch[] log.
+*/
+static void snippetAppendMatch(
+ Snippet *p, /* Append the entry to this snippet */
+ int iCol, int iTerm, /* The column and query term */
+ int iToken, /* Matching token in document */
+ int iStart, int nByte /* Offset and size of the match */
+){
+ int i;
+ struct snippetMatch *pMatch;
+ if( p->nMatch+1>=p->nAlloc ){
+ p->nAlloc = p->nAlloc*2 + 10;
+ p->aMatch = sqlite3_realloc(p->aMatch, p->nAlloc*sizeof(p->aMatch[0]) );
+ if( p->aMatch==0 ){
+ p->nMatch = 0;
+ p->nAlloc = 0;
+ return;
+ }
+ }
+ i = p->nMatch++;
+ pMatch = &p->aMatch[i];
+ pMatch->iCol = iCol;
+ pMatch->iTerm = iTerm;
+ pMatch->iToken = iToken;
+ pMatch->iStart = iStart;
+ pMatch->nByte = nByte;
+}
+
+/*
+** Sizing information for the circular buffer used in snippetOffsetsOfColumn()
+*/
+#define FTS3_ROTOR_SZ (32)
+#define FTS3_ROTOR_MASK (FTS3_ROTOR_SZ-1)
+
+/*
+** Add entries to pSnippet->aMatch[] for every match that occurs against
+** document zDoc[0..nDoc-1] which is stored in column iColumn.
+*/
+static void snippetOffsetsOfColumn(
+ Query *pQuery,
+ Snippet *pSnippet,
+ int iColumn,
+ const char *zDoc,
+ int nDoc
+){
+
+ fulltext_vtab *pVtab; /* The full text index */
+ int nColumn; /* Number of columns in the index */
+ const QueryTerm *aTerm; /* Query string terms */
+ int nTerm; /* Number of query string terms */
+ int i, j; /* Loop counters */
+ unsigned int match, prevMatch; /* Phrase search bitmasks */
+ const char *zToken; /* Next token from the tokenizer */
+ int nToken; /* Size of zToken */
+ int iBegin, iEnd, iPos; /* Offsets of beginning and end */
+ gboolean new_paragraph, stop_word;
+
+ /* The following variables keep a circular buffer of the last
+ ** few tokens */
+ unsigned int iRotor = 0; /* Index of current token */
+ int iRotorBegin[FTS3_ROTOR_SZ]; /* Beginning offset of token */
+ int iRotorLen[FTS3_ROTOR_SZ]; /* Length of token */
+
+ pVtab = pQuery->pFts;
+ nColumn = pVtab->nColumn;
+
+ tracker_parser_reset (pVtab->parser, zDoc, nDoc, FALSE, TRUE, TRUE, FALSE);
+
+ aTerm = pQuery->pTerms;
+ nTerm = pQuery->nTerms;
+
+ if( nTerm>=FTS3_ROTOR_SZ ){
+ nTerm = FTS3_ROTOR_SZ - 1;
+ }
+
+ prevMatch = 0;
+
+ while(1){
+// rc = pTModule->xNext(pTCursor, &zToken, &nToken, &iBegin, &iEnd, &iPos);
+
+ zToken = tracker_parser_next (pVtab->parser,
+ &iPos,
+ &iBegin,
+ &iEnd,
+ &new_paragraph,
+ &stop_word,
+ &nToken);
+
+ if (!zToken) break;
+
+ if (stop_word) {
+ continue;
+ }
+
+ iRotorBegin[iRotor&FTS3_ROTOR_MASK] = iBegin;
+ iRotorLen[iRotor&FTS3_ROTOR_MASK] = iEnd-iBegin;
+ match = 0;
+ for(i=0; i<nTerm; i++){
+ int iCol;
+ iCol = aTerm[i].iColumn;
+ if( iCol>=0 && iCol<nColumn && iCol!=iColumn ) continue;
+ if( aTerm[i].nTerm>nToken ) continue;
+ if( !aTerm[i].isPrefix && aTerm[i].nTerm<nToken ) continue;
+ assert( aTerm[i].nTerm<=nToken );
+ if( memcmp(aTerm[i].pTerm, zToken, aTerm[i].nTerm) ) continue;
+ if( aTerm[i].iPhrase>1 && (prevMatch & (1<<i))==0 ) continue;
+ match |= 1<<i;
+ if( i==nTerm-1 || aTerm[i+1].iPhrase==1 ){
+ for(j=aTerm[i].iPhrase-1; j>=0; j--){
+ int k = (iRotor-j) & FTS3_ROTOR_MASK;
+ snippetAppendMatch(pSnippet, iColumn, i-j, iPos-j,
+ iRotorBegin[k], iRotorLen[k]);
+ }
+ }
+ }
+ prevMatch = match<<1;
+ iRotor++;
+ }
+// pTModule->xClose(pTCursor);
+}
+
+/*
+** Remove entries from the pSnippet structure to account for the NEAR
+** operator. When this is called, pSnippet contains the list of token
+** offsets produced by treating all NEAR operators as AND operators.
+** This function removes any entries that should not be present after
+** accounting for the NEAR restriction. For example, if the queried
+** document is:
+**
+** "A B C D E A"
+**
+** and the query is:
+**
+** A NEAR/0 E
+**
+** then when this function is called the Snippet contains token offsets
+** 0, 4 and 5. This function removes the "0" entry (because the first A
+** is not near enough to an E).
+*/
+static void trimSnippetOffsetsForNear(Query *pQuery, Snippet *pSnippet){
+ int ii;
+ int iDir = 1;
+
+ while(iDir>-2) {
+ assert( iDir==1 || iDir==-1 );
+ for(ii=0; ii<pSnippet->nMatch; ii++){
+ int jj;
+ int nNear;
+ struct snippetMatch *pMatch = &pSnippet->aMatch[ii];
+ QueryTerm *pQueryTerm = &pQuery->pTerms[pMatch->iTerm];
+
+ if( (pMatch->iTerm+iDir)<0
+ || (pMatch->iTerm+iDir)>=pQuery->nTerms
+ ){
+ continue;
+ }
+
+ nNear = pQueryTerm->nNear;
+ if( iDir<0 ){
+ nNear = pQueryTerm[-1].nNear;
+ }
+
+ if( pMatch->iTerm>=0 && nNear ){
+ int isOk = 0;
+ int iNextTerm = pMatch->iTerm+iDir;
+ int iPrevTerm = iNextTerm;
+
+ int iEndToken;
+ int iStartToken;
+
+ if( iDir<0 ){
+ int nPhrase = 1;
+ iStartToken = pMatch->iToken;
+ while( (pMatch->iTerm+nPhrase)<pQuery->nTerms
+ && pQuery->pTerms[pMatch->iTerm+nPhrase].iPhrase>1
+ ){
+ nPhrase++;
+ }
+ iEndToken = iStartToken + nPhrase - 1;
+ }else{
+ iEndToken = pMatch->iToken;
+ iStartToken = pMatch->iToken+1-pQueryTerm->iPhrase;
+ }
+
+ while( pQuery->pTerms[iNextTerm].iPhrase>1 ){
+ iNextTerm--;
+ }
+ while( (iPrevTerm+1)<pQuery->nTerms &&
+ pQuery->pTerms[iPrevTerm+1].iPhrase>1
+ ){
+ iPrevTerm++;
+ }
+
+ for(jj=0; isOk==0 && jj<pSnippet->nMatch; jj++){
+ struct snippetMatch *p = &pSnippet->aMatch[jj];
+ if( p->iCol==pMatch->iCol && ((
+ p->iTerm==iNextTerm &&
+ p->iToken>iEndToken &&
+ p->iToken<=iEndToken+nNear
+ ) || (
+ p->iTerm==iPrevTerm &&
+ p->iToken<iStartToken &&
+ p->iToken>=iStartToken-nNear
+ ))){
+ isOk = 1;
+ }
+ }
+ if( !isOk ){
+ for(jj=1-pQueryTerm->iPhrase; jj<=0; jj++){
+ pMatch[jj].iTerm = -1;
+ }
+ ii = -1;
+ iDir = 1;
+ }
+ }
+ }
+ iDir -= 2;
+ }
+}
+
+/*
+** Compute all offsets for the current row of the query.
+** If the offsets have already been computed, this routine is a no-op.
+*/
+static void snippetAllOffsets(fulltext_cursor *p){
+ int nColumn;
+ int iColumn, i;
+ int iFirst, iLast;
+ fulltext_vtab *pFts;
+
+ if( p->snippet.nMatch ) return;
+ if( p->q.nTerms==0 ) return;
+ pFts = p->q.pFts;
+ nColumn = pFts->nColumn;
+ iColumn = (p->iCursorType - QUERY_FULLTEXT);
+ if( iColumn<0 || iColumn>=nColumn ){
+ iFirst = 0;
+ iLast = nColumn-1;
+ }else{
+ iFirst = iColumn;
+ iLast = iColumn;
+ }
+ for(i=iFirst; i<=iLast; i++){
+ const char *zDoc;
+ int nDoc;
+ zDoc = (const char*)sqlite3_column_text(p->pStmt, i+1);
+ nDoc = sqlite3_column_bytes(p->pStmt, i+1);
+ snippetOffsetsOfColumn(&p->q, &p->snippet, i, zDoc, nDoc);
+ }
+
+ trimSnippetOffsetsForNear(&p->q, &p->snippet);
+}
+
+/*
+** Convert the information in the aMatch[] array of the snippet
+** into the string zOffset[0..nOffset-1].
+*/
+static void snippetOffsetText(Snippet *p){
+ int i;
+ int cnt = 0;
+ StringBuffer sb;
+ char zBuf[200];
+ if( p->zOffset ) return;
+ initStringBuffer(&sb);
+ for(i=0; i<p->nMatch; i++){
+ struct snippetMatch *pMatch = &p->aMatch[i];
+ if( pMatch->iTerm>=0 ){
+ /* If snippetMatch.iTerm is less than 0, then the match was
+ ** discarded as part of processing the NEAR operator (see the
+ ** trimSnippetOffsetsForNear() function for details). Ignore
+ ** it in this case
+ */
+ zBuf[0] = ' ';
+ sqlite3_snprintf(sizeof(zBuf)-1, &zBuf[cnt>0], "%d %d %d %d",
+ pMatch->iCol, pMatch->iTerm, pMatch->iStart, pMatch->nByte);
+ append(&sb, zBuf);
+ cnt++;
+ }
+ }
+ p->zOffset = stringBufferData(&sb);
+ p->nOffset = stringBufferLength(&sb);
+}
+
+/*
+** zDoc[0..nDoc-1] is phrase of text. aMatch[0..nMatch-1] are a set
+** of matching words some of which might be in zDoc. zDoc is column
+** number iCol.
+**
+** iBreak is suggested spot in zDoc where we could begin or end an
+** excerpt. Return a value similar to iBreak but possibly adjusted
+** to be a little left or right so that the break point is better.
+*/
+static int wordBoundary(
+ int iBreak, /* The suggested break point */
+ const char *zDoc, /* Document text */
+ int nDoc, /* Number of bytes in zDoc[] */
+ struct snippetMatch *aMatch, /* Matching words */
+ int nMatch, /* Number of entries in aMatch[] */
+ int iCol /* The column number for zDoc[] */
+){
+ int i;
+ if( iBreak<=10 ){
+ return 0;
+ }
+ if( iBreak>=nDoc-10 ){
+ return nDoc;
+ }
+ for(i=0; i<nMatch && aMatch[i].iCol<iCol; i++){}
+ while( i<nMatch && aMatch[i].iStart+aMatch[i].nByte<iBreak ){ i++; }
+ if( i<nMatch ){
+ if( aMatch[i].iStart<iBreak+10 ){
+ return aMatch[i].iStart;
+ }
+ if( i>0 && aMatch[i-1].iStart+aMatch[i-1].nByte>=iBreak ){
+ return aMatch[i-1].iStart;
+ }
+ }
+ for(i=1; i<=10; i++){
+ if( safe_isspace(zDoc[iBreak-i]) ){
+ return iBreak - i + 1;
+ }
+ if( safe_isspace(zDoc[iBreak+i]) ){
+ return iBreak + i + 1;
+ }
+ }
+ return iBreak;
+}
+
+
+
+/*
+** Allowed values for Snippet.aMatch[].snStatus
+*/
+#define SNIPPET_IGNORE 0 /* It is ok to omit this match from the snippet */
+#define SNIPPET_DESIRED 1 /* We want to include this match in the snippet */
+
+/*
+** Generate the text of a snippet.
+*/
+static void snippetText(
+ fulltext_cursor *pCursor, /* The cursor we need the snippet for */
+ const char *zStartMark, /* Markup to appear before each match */
+ const char *zEndMark, /* Markup to appear after each match */
+ const char *zEllipsis /* Ellipsis mark */
+){
+ int i, j;
+ struct snippetMatch *aMatch;
+ int nMatch;
+ int nDesired;
+ StringBuffer sb;
+ int tailCol;
+ int tailOffset;
+ int iCol;
+ int nDoc;
+ const char *zDoc;
+ int iStart, iEnd;
+ int tailEllipsis = 0;
+ int iMatch;
+
+
+ sqlite3_free(pCursor->snippet.zSnippet);
+ pCursor->snippet.zSnippet = 0;
+ aMatch = pCursor->snippet.aMatch;
+ nMatch = pCursor->snippet.nMatch;
+ initStringBuffer(&sb);
+
+ for(i=0; i<nMatch; i++){
+ aMatch[i].snStatus = SNIPPET_IGNORE;
+ }
+ nDesired = 0;
+ for(i=0; i<pCursor->q.nTerms; i++){
+ for(j=0; j<nMatch; j++){
+ if( aMatch[j].iTerm==i ){
+ aMatch[j].snStatus = SNIPPET_DESIRED;
+ nDesired++;
+ break;
+ }
+ }
+ }
+
+ iMatch = 0;
+ tailCol = -1;
+ tailOffset = 0;
+ for(i=0; i<nMatch && nDesired>0; i++){
+ if( aMatch[i].snStatus!=SNIPPET_DESIRED ) continue;
+ nDesired--;
+ iCol = aMatch[i].iCol;
+ zDoc = (const char*)sqlite3_column_text(pCursor->pStmt, iCol+1);
+ nDoc = sqlite3_column_bytes(pCursor->pStmt, iCol+1);
+ iStart = aMatch[i].iStart - 40;
+ iStart = wordBoundary(iStart, zDoc, nDoc, aMatch, nMatch, iCol);
+ if( iStart<=10 ){
+ iStart = 0;
+ }
+ if( iCol==tailCol && iStart<=tailOffset+20 ){
+ iStart = tailOffset;
+ }
+ if( (iCol!=tailCol && tailCol>=0) || iStart!=tailOffset ){
+ trimWhiteSpace(&sb);
+ appendWhiteSpace(&sb);
+ append(&sb, zEllipsis);
+ appendWhiteSpace(&sb);
+ }
+ iEnd = aMatch[i].iStart + aMatch[i].nByte + 40;
+ iEnd = wordBoundary(iEnd, zDoc, nDoc, aMatch, nMatch, iCol);
+ if( iEnd>=nDoc-10 ){
+ iEnd = nDoc;
+ tailEllipsis = 0;
+ }else{
+ tailEllipsis = 1;
+ }
+ while( iMatch<nMatch && aMatch[iMatch].iCol<iCol ){ iMatch++; }
+ while( iStart<iEnd ){
+ while( iMatch<nMatch && aMatch[iMatch].iStart<iStart
+ && aMatch[iMatch].iCol<=iCol ){
+ iMatch++;
+ }
+ if( iMatch<nMatch && aMatch[iMatch].iStart<iEnd
+ && aMatch[iMatch].iCol==iCol ){
+ nappend(&sb, &zDoc[iStart], aMatch[iMatch].iStart - iStart);
+ iStart = aMatch[iMatch].iStart;
+ append(&sb, zStartMark);
+ nappend(&sb, &zDoc[iStart], aMatch[iMatch].nByte);
+ append(&sb, zEndMark);
+ iStart += aMatch[iMatch].nByte;
+ for(j=iMatch+1; j<nMatch; j++){
+ if( aMatch[j].iTerm==aMatch[iMatch].iTerm
+ && aMatch[j].snStatus==SNIPPET_DESIRED ){
+ nDesired--;
+ aMatch[j].snStatus = SNIPPET_IGNORE;
+ }
+ }
+ }else{
+ nappend(&sb, &zDoc[iStart], iEnd - iStart);
+ iStart = iEnd;
+ }
+ }
+ tailCol = iCol;
+ tailOffset = iEnd;
+ }
+ trimWhiteSpace(&sb);
+ if( tailEllipsis ){
+ appendWhiteSpace(&sb);
+ append(&sb, zEllipsis);
+ }
+ pCursor->snippet.zSnippet = stringBufferData(&sb);
+ pCursor->snippet.nSnippet = stringBufferLength(&sb);
+}
+
+
+/*
+** Close the cursor. For additional information see the documentation
+** on the xClose method of the virtual table interface.
+*/
+static int fulltextClose(sqlite3_vtab_cursor *pCursor){
+ fulltext_cursor *c = (fulltext_cursor *) pCursor;
+ FTSTRACE(("FTS3 Close %p\n", c));
+ sqlite3_finalize(c->pStmt);
+ queryClear(&c->q);
+ snippetClear(&c->snippet);
+ if( c->result.nData!=0 ) dlrDestroy(&c->reader);
+ dataBufferDestroy(&c->result);
+ sqlite3_free(c);
+ return SQLITE_OK;
+}
+
+static int fulltextNext(sqlite3_vtab_cursor *pCursor){
+ fulltext_cursor *c = (fulltext_cursor *) pCursor;
+ int rc;
+
+ FTSTRACE(("FTS3 Next %p\n", pCursor));
+ snippetClear(&c->snippet);
+ if( c->iCursorType < QUERY_FULLTEXT ){
+ /* TODO(shess) Handle SQLITE_SCHEMA AND SQLITE_BUSY. */
+ rc = sqlite3_step(c->pStmt);
+ switch( rc ){
+ case SQLITE_ROW:
+ c->eof = 0;
+ return SQLITE_OK;
+ case SQLITE_DONE:
+ c->eof = 1;
+ return SQLITE_OK;
+ default:
+ c->eof = 1;
+ return rc;
+ }
+ } else { /* full-text query */
+ rc = sqlite3_reset(c->pStmt);
+ if( rc!=SQLITE_OK ) return rc;
+
+ if( c->result.nData==0 || dlrAtEnd(&c->reader) ){
+ c->eof = 1;
+ return SQLITE_OK;
+ }
+ rc = sqlite3_bind_int64(c->pStmt, 1, dlrDocid(&c->reader));
+ dlrStep(&c->reader);
+ if( rc!=SQLITE_OK ) return rc;
+ /* TODO(shess) Handle SQLITE_SCHEMA AND SQLITE_BUSY. */
+ rc = sqlite3_step(c->pStmt);
+ if( rc==SQLITE_ROW ){ /* the case we expect */
+ c->eof = 0;
+ return SQLITE_OK;
+ }
+ /* an error occurred; abort */
+ return rc==SQLITE_DONE ? SQLITE_ERROR : rc;
+ }
+}
+
+
+/* TODO(shess) If we pushed LeafReader to the top of the file, or to
+** another file, term_select() could be pushed above
+** docListOfTerm().
+*/
+static int termSelect(fulltext_vtab *v, int iColumn,
+ const char *pTerm, int nTerm, int isPrefix,
+ DocListType iType, DataBuffer *out);
+
+/* Return a DocList corresponding to the query term *pTerm. If *pTerm
+** is the first term of a phrase query, go ahead and evaluate the phrase
+** query and return the doclist for the entire phrase query.
+**
+** The resulting DL_DOCIDS doclist is stored in pResult, which is
+** overwritten.
+*/
+static int docListOfTerm(
+ fulltext_vtab *v, /* The full text index */
+ int iColumn, /* column to restrict to. No restriction if >=nColumn */
+ QueryTerm *pQTerm, /* Term we are looking for, or 1st term of a phrase */
+ DataBuffer *pResult /* Write the result here */
+){
+ DataBuffer left, right, new;
+ int i, rc;
+
+ /* No phrase search if no position info. */
+ assert( pQTerm->nPhrase==0 || DL_DEFAULT!=DL_DOCIDS );
+
+ /* This code should never be called with buffered updates. */
+ assert( v->nPendingData<0 );
+
+ dataBufferInit(&left, 0);
+ rc = termSelect(v, iColumn, pQTerm->pTerm, pQTerm->nTerm, pQTerm->isPrefix,
+ (0<pQTerm->nPhrase ? DL_POSITIONS : DL_DOCIDS), &left);
+ if( rc ) return rc;
+ for(i=1; i<=pQTerm->nPhrase && left.nData>0; i++){
+ /* If this token is connected to the next by a NEAR operator, and
+ ** the next token is the start of a phrase, then set nPhraseRight
+ ** to the number of tokens in the phrase. Otherwise leave it at 1.
+ */
+ int nPhraseRight = 1;
+ while( (i+nPhraseRight)<=pQTerm->nPhrase
+ && pQTerm[i+nPhraseRight].nNear==0
+ ){
+ nPhraseRight++;
+ }
+
+ dataBufferInit(&right, 0);
+ rc = termSelect(v, iColumn, pQTerm[i].pTerm, pQTerm[i].nTerm,
+ pQTerm[i].isPrefix, DL_POSITIONS, &right);
+ if( rc ){
+ dataBufferDestroy(&left);
+ return rc;
+ }
+ dataBufferInit(&new, 0);
+ docListPhraseMerge(left.pData, left.nData, right.pData, right.nData,
+ pQTerm[i-1].nNear, pQTerm[i-1].iPhrase + nPhraseRight,
+ ((i<pQTerm->nPhrase) ? DL_POSITIONS : DL_DOCIDS),
+ &new);
+ dataBufferDestroy(&left);
+ dataBufferDestroy(&right);
+ left = new;
+ }
+ *pResult = left;
+ return SQLITE_OK;
+}
+
+/* Add a new term pTerm[0..nTerm-1] to the query *q.
+*/
+static void queryAdd(Query *q, const char *pTerm, int nTerm){
+ QueryTerm *t;
+ ++q->nTerms;
+ q->pTerms = sqlite3_realloc(q->pTerms, q->nTerms * sizeof(q->pTerms[0]));
+ if( q->pTerms==0 ){
+ q->nTerms = 0;
+ return;
+ }
+ t = &q->pTerms[q->nTerms - 1];
+ CLEAR(t);
+ t->pTerm = sqlite3_malloc(nTerm+1);
+ memcpy(t->pTerm, pTerm, nTerm);
+ t->pTerm[nTerm] = 0;
+ t->nTerm = nTerm;
+ t->isOr = q->nextIsOr;
+ t->isPrefix = 0;
+ q->nextIsOr = 0;
+ t->iColumn = q->nextColumn;
+ q->nextColumn = q->dfltColumn;
+}
+
+/*
+** Check to see if the string zToken[0...nToken-1] matches any
+** column name in the virtual table. If it does,
+** return the zero-indexed column number. If not, return -1.
+*/
+static int checkColumnSpecifier(
+ fulltext_vtab *pVtab, /* The virtual table */
+ const char *zToken, /* Text of the token */
+ int nToken /* Number of characters in the token */
+){
+ int i;
+ for(i=0; i<pVtab->nColumn; i++){
+ if( memcmp(pVtab->azColumn[i], zToken, nToken)==0
+ && pVtab->azColumn[i][nToken]==0 ){
+ return i;
+ }
+ }
+ return -1;
+}
+
+/*
+** Parse the text at pSegment[0..nSegment-1]. Add additional terms
+** to the query being assemblied in pQuery.
+**
+** inPhrase is true if pSegment[0..nSegement-1] is contained within
+** double-quotes. If inPhrase is true, then the first term
+** is marked with the number of terms in the phrase less one and
+** OR and "-" syntax is ignored. If inPhrase is false, then every
+** term found is marked with nPhrase=0 and OR and "-" syntax is significant.
+*/
+static int tokenizeSegment(
+ TrackerParser *parser, /* The tokenizer to use */
+ const char *pSegment, int nSegment, /* Query expression being parsed */
+ int inPhrase, /* True if within "..." */
+ Query *pQuery /* Append results here */
+){
+ int firstIndex = pQuery->nTerms;
+ int iCol;
+ int nTerm = 1;
+
+ tracker_parser_reset (parser, pSegment, nSegment, FALSE, TRUE, TRUE, TRUE);
+
+ while( 1 ){
+ const char *pToken;
+ int nToken, iBegin, iEnd, iPos, stop_word, new_paragraph;
+
+
+ pToken = tracker_parser_next (parser, &iPos,
+ &iBegin,
+ &iEnd,
+ &new_paragraph,
+ &stop_word,
+ &nToken);
+ if (!pToken) {
+ break;
+ }
+
+// printf("token being indexed is %s, pos is %d, begin is %d, end is %d and length is %d\n", pToken, iPos, iBegin, iEnd, nToken);
+
+ if( !inPhrase &&
+ pSegment[iEnd]==':') {
+
+ int len = iEnd - iBegin;
+ char *field = g_strndup (pSegment + iBegin, len);
+
+ // printf ("field is %s\n", field);
+
+ if ((iCol = checkColumnSpecifier(pQuery->pFts, field, len))>=0 ){
+ pQuery->nextColumn = iCol;
+ g_free (field);
+ continue;
+ }
+ }
+ if( !inPhrase && pQuery->nTerms>0 && nToken==2
+ && pToken[0] == 'o' && pToken[1] == 'r'
+ ){
+ pQuery->nextIsOr = 1;
+ continue;
+ }
+ if( !inPhrase && pQuery->nTerms>0 && !pQuery->nextIsOr && nToken==4
+ && pToken[0]=='n'
+ && pToken[1]=='e'
+ && pToken[2]=='a'
+ && pToken[3]=='r'
+ ){
+ QueryTerm *pTerm = &pQuery->pTerms[pQuery->nTerms-1];
+ if( (iBegin+6)<nSegment
+ && pSegment[iBegin+4] == '/'
+ && pSegment[iBegin+5]>='0' && pSegment[iBegin+5]<='9'
+ ){
+ pTerm->nNear = (pSegment[iBegin+5] - '0');
+ nToken += 2;
+ if( pSegment[iBegin+6]>='0' && pSegment[iBegin+6]<=9 ){
+ pTerm->nNear = pTerm->nNear * 10 + (pSegment[iBegin+6] - '0');
+ iEnd++;
+ }
+ pToken = tracker_parser_next (parser, &iPos,
+ &iBegin,
+ &iEnd,
+ &new_paragraph,
+ &stop_word,
+ &nToken);
+ if (!pToken) {
+ break;
+ }
+
+
+ } else {
+ pTerm->nNear = SQLITE_FTS3_DEFAULT_NEAR_PARAM;
+ }
+ pTerm->nNear++;
+ continue;
+ }
+
+ if (stop_word != 0) {
+ continue;
+ }
+
+ queryAdd(pQuery, pToken, nToken);
+ if( !inPhrase && iBegin>0) {
+
+ // printf("first char is %c, prev char is %c\n", pSegment[iBegin], pSegment[iBegin-1]);
+
+ if (pSegment[iBegin-1]=='-' ){
+ pQuery->pTerms[pQuery->nTerms-1].isNot = 1;
+ }
+ }
+ if( iEnd<nSegment && pSegment[iEnd]=='*' ){
+ pQuery->pTerms[pQuery->nTerms-1].isPrefix = 1;
+ }
+ pQuery->pTerms[pQuery->nTerms-1].iPhrase = nTerm;
+ if( inPhrase ){
+ nTerm++;
+ }
+ }
+
+ if( inPhrase && pQuery->nTerms>firstIndex ){
+ pQuery->pTerms[firstIndex].nPhrase = pQuery->nTerms - firstIndex - 1;
+ }
+
+ return SQLITE_OK;
+}
+
+/* Parse a query string, yielding a Query object pQuery.
+**
+** The calling function will need to queryClear() to clean up
+** the dynamically allocated memory held by pQuery.
+*/
+static int parseQuery(
+ fulltext_vtab *v, /* The fulltext index */
+ const char *zInput, /* Input text of the query string */
+ int nInput, /* Size of the input text */
+ int dfltColumn, /* Default column of the index to match against */
+ Query *pQuery /* Write the parse results here. */
+){
+ int iInput, inPhrase = 0;
+ int ii;
+ QueryTerm *aTerm;
+
+ if( zInput==0 ) nInput = 0;
+ if( nInput<0 ) nInput = strlen(zInput);
+ pQuery->nTerms = 0;
+ pQuery->pTerms = NULL;
+ pQuery->nextIsOr = 0;
+ pQuery->nextColumn = dfltColumn;
+ pQuery->dfltColumn = dfltColumn;
+ pQuery->pFts = v;
+
+ for(iInput=0; iInput<nInput; ++iInput){
+ int i;
+ for(i=iInput; i<nInput && zInput[i]!='"'; ++i){}
+ if( i>iInput ){
+ tokenizeSegment(v->parser, zInput+iInput, i-iInput, inPhrase,
+ pQuery);
+ }
+ iInput = i;
+ if( i<nInput ){
+ assert( zInput[i]=='"' );
+ inPhrase = !inPhrase;
+ }
+ }
+
+ if( inPhrase ){
+ /* unmatched quote */
+ queryClear(pQuery);
+ return SQLITE_ERROR;
+ }
+
+ /* Modify the values of the QueryTerm.nPhrase variables to account for
+ ** the NEAR operator. For the purposes of QueryTerm.nPhrase, phrases
+ ** and tokens connected by the NEAR operator are handled as a single
+ ** phrase. See comments above the QueryTerm structure for details.
+ */
+ aTerm = pQuery->pTerms;
+ for(ii=0; ii<pQuery->nTerms; ii++){
+ if( aTerm[ii].nNear || aTerm[ii].nPhrase ){
+ while (aTerm[ii+aTerm[ii].nPhrase].nNear) {
+ aTerm[ii].nPhrase += (1 + aTerm[ii+aTerm[ii].nPhrase+1].nPhrase);
+ }
+ }
+ }
+
+ return SQLITE_OK;
+}
+
+/* TODO(shess) Refactor the code to remove this forward decl. */
+static int flushPendingTerms(fulltext_vtab *v);
+
+/* Perform a full-text query using the search expression in
+** zInput[0..nInput-1]. Return a list of matching documents
+** in pResult.
+**
+** Queries must match column iColumn. Or if iColumn>=nColumn
+** they are allowed to match against any column.
+*/
+static int fulltextQuery(
+ fulltext_vtab *v, /* The full text index */
+ int iColumn, /* Match against this column by default */
+ const char *zInput, /* The query string */
+ int nInput, /* Number of bytes in zInput[] */
+ DataBuffer *pResult, /* Write the result doclist here */
+ Query *pQuery /* Put parsed query string here */
+){
+ int i, iNext, rc;
+ DataBuffer left, right, or, new;
+ int nNot = 0;
+ QueryTerm *aTerm;
+
+ /* TODO(shess) Instead of flushing pendingTerms, we could query for
+ ** the relevant term and merge the doclist into what we receive from
+ ** the database. Wait and see if this is a common issue, first.
+ **
+ ** A good reason not to flush is to not generate update-related
+ ** error codes from here.
+ */
+
+ /* Flush any buffered updates before executing the query. */
+ rc = flushPendingTerms(v);
+ if( rc!=SQLITE_OK ) return rc;
+
+ /* TODO(shess) I think that the queryClear() calls below are not
+ ** necessary, because fulltextClose() already clears the query.
+ */
+ rc = parseQuery(v, zInput, nInput, iColumn, pQuery);
+ if( rc!=SQLITE_OK ) return rc;
+
+ /* Empty or NULL queries return no results. */
+ if( pQuery->nTerms==0 ){
+ dataBufferInit(pResult, 0);
+ return SQLITE_OK;
+ }
+
+ /* Merge AND terms. */
+ /* TODO(shess) I think we can early-exit if( i>nNot && left.nData==0 ). */
+ aTerm = pQuery->pTerms;
+ for(i = 0; i<pQuery->nTerms; i=iNext){
+ if( aTerm[i].isNot ){
+ /* Handle all NOT terms in a separate pass */
+ nNot++;
+ iNext = i + aTerm[i].nPhrase+1;
+ continue;
+ }
+ iNext = i + aTerm[i].nPhrase + 1;
+ rc = docListOfTerm(v, aTerm[i].iColumn, &aTerm[i], &right);
+ if( rc ){
+ if( i!=nNot ) dataBufferDestroy(&left);
+ queryClear(pQuery);
+ return rc;
+ }
+ while( iNext<pQuery->nTerms && aTerm[iNext].isOr ){
+ rc = docListOfTerm(v, aTerm[iNext].iColumn, &aTerm[iNext], &or);
+ iNext += aTerm[iNext].nPhrase + 1;
+ if( rc ){
+ if( i!=nNot ) dataBufferDestroy(&left);
+ dataBufferDestroy(&right);
+ queryClear(pQuery);
+ return rc;
+ }
+ dataBufferInit(&new, 0);
+ docListOrMerge(right.pData, right.nData, or.pData, or.nData, &new);
+ dataBufferDestroy(&right);
+ dataBufferDestroy(&or);
+ right = new;
+ }
+ if( i==nNot ){ /* first term processed. */
+ left = right;
+ }else{
+ dataBufferInit(&new, 0);
+ docListAndMerge(left.pData, left.nData, right.pData, right.nData, &new);
+ dataBufferDestroy(&right);
+ dataBufferDestroy(&left);
+ left = new;
+ }
+ }
+
+ if( nNot==pQuery->nTerms ){
+ /* We do not yet know how to handle a query of only NOT terms */
+ return SQLITE_ERROR;
+ }
+
+ /* Do the EXCEPT terms */
+ for(i=0; i<pQuery->nTerms; i += aTerm[i].nPhrase + 1){
+ if( !aTerm[i].isNot ) continue;
+ rc = docListOfTerm(v, aTerm[i].iColumn, &aTerm[i], &right);
+ if( rc ){
+ queryClear(pQuery);
+ dataBufferDestroy(&left);
+ return rc;
+ }
+ dataBufferInit(&new, 0);
+ docListExceptMerge(left.pData, left.nData, right.pData, right.nData, &new);
+ dataBufferDestroy(&right);
+ dataBufferDestroy(&left);
+ left = new;
+ }
+
+ *pResult = left;
+ return rc;
+}
+
+/*
+** This is the xFilter interface for the virtual table. See
+** the virtual table xFilter method documentation for additional
+** information.
+**
+** If idxNum==QUERY_GENERIC then do a full table scan against
+** the %_content table.
+**
+** If idxNum==QUERY_DOCID then do a docid lookup for a single entry
+** in the %_content table.
+**
+** If idxNum>=QUERY_FULLTEXT then use the full text index. The
+** column on the left-hand side of the MATCH operator is column
+** number idxNum-QUERY_FULLTEXT, 0 indexed. argv[0] is the right-hand
+** side of the MATCH operator.
+*/
+/* TODO(shess) Upgrade the cursor initialization and destruction to
+** account for fulltextFilter() being called multiple times on the
+** same cursor. The current solution is very fragile. Apply fix to
+** fts3 as appropriate.
+*/
+static int fulltextFilter(
+ sqlite3_vtab_cursor *pCursor, /* The cursor used for this query */
+ int idxNum, const char *idxStr, /* Which indexing scheme to use */
+ int argc, sqlite3_value **argv /* Arguments for the indexing scheme */
+){
+ fulltext_cursor *c = (fulltext_cursor *) pCursor;
+ fulltext_vtab *v = cursor_vtab(c);
+ int rc;
+ StringBuffer sb;
+
+ FTSTRACE(("FTS3 Filter %p\n",pCursor));
+
+ initStringBuffer(&sb);
+ append(&sb, "SELECT docid, ");
+ appendList(&sb, v->nColumn, v->azContentColumn);
+ append(&sb, " FROM %_content");
+ if( idxNum!=QUERY_GENERIC ) append(&sb, " WHERE docid = ?");
+ sqlite3_finalize(c->pStmt);
+ rc = sql_prepare(v->db, v->zDb, v->zName, &c->pStmt, stringBufferData(&sb));
+ stringBufferDestroy(&sb);
+ if( rc!=SQLITE_OK ) return rc;
+
+ c->iCursorType = idxNum;
+ switch( idxNum ){
+ case QUERY_GENERIC:
+ break;
+
+ case QUERY_DOCID:
+ rc = sqlite3_bind_int64(c->pStmt, 1, sqlite3_value_int64(argv[0]));
+ if( rc!=SQLITE_OK ) return rc;
+ break;
+
+ default: /* full-text search */
+ {
+ const char *zQuery = (const char *)sqlite3_value_text(argv[0]);
+ assert( idxNum<=QUERY_FULLTEXT+v->nColumn);
+ assert( argc==1 );
+ queryClear(&c->q);
+ if( c->result.nData!=0 ){
+ /* This case happens if the same cursor is used repeatedly. */
+ dlrDestroy(&c->reader);
+ dataBufferReset(&c->result);
+ }else{
+ dataBufferInit(&c->result, 0);
+ }
+ rc = fulltextQuery(v, idxNum-QUERY_FULLTEXT, zQuery, -1, &c->result, &c->q);
+ if( rc!=SQLITE_OK ) return rc;
+ if( c->result.nData!=0 ){
+ dlrInit(&c->reader, DL_DOCIDS, c->result.pData, c->result.nData);
+ }
+ break;
+ }
+ }
+
+ return fulltextNext(pCursor);
+}
+
+/* This is the xEof method of the virtual table. The SQLite core
+** calls this routine to find out if it has reached the end of
+** a query's results set.
+*/
+static int fulltextEof(sqlite3_vtab_cursor *pCursor){
+ fulltext_cursor *c = (fulltext_cursor *) pCursor;
+ return c->eof;
+}
+
+/* This is the xColumn method of the virtual table. The SQLite
+** core calls this method during a query when it needs the value
+** of a column from the virtual table. This method needs to use
+** one of the sqlite3_result_*() routines to store the requested
+** value back in the pContext.
+*/
+static int fulltextColumn(sqlite3_vtab_cursor *pCursor,
+ sqlite3_context *pContext, int idxCol){
+ fulltext_cursor *c = (fulltext_cursor *) pCursor;
+ fulltext_vtab *v = cursor_vtab(c);
+
+ if( idxCol<v->nColumn ){
+ sqlite3_value *pVal = sqlite3_column_value(c->pStmt, idxCol+1);
+ sqlite3_result_value(pContext, pVal);
+ }else if( idxCol==v->nColumn ){
+ /* The extra column whose name is the same as the table.
+ ** Return a blob which is a pointer to the cursor
+ */
+ sqlite3_result_blob(pContext, &c, sizeof(c), SQLITE_TRANSIENT);
+ }else if( idxCol==v->nColumn+1 ){
+ /* The docid column, which is an alias for rowid. */
+ sqlite3_value *pVal = sqlite3_column_value(c->pStmt, 0);
+ sqlite3_result_value(pContext, pVal);
+ }
+ return SQLITE_OK;
+}
+
+/* This is the xRowid method. The SQLite core calls this routine to
+** retrieve the rowid for the current row of the result set. fts3
+** exposes %_content.docid as the rowid for the virtual table. The
+** rowid should be written to *pRowid.
+*/
+static int fulltextRowid(sqlite3_vtab_cursor *pCursor, sqlite_int64 *pRowid){
+ fulltext_cursor *c = (fulltext_cursor *) pCursor;
+
+ *pRowid = sqlite3_column_int64(c->pStmt, 0);
+ return SQLITE_OK;
+}
+
+/* Add all terms in [zText] to pendingTerms table. If [iColumn] > 0,
+** we also store positions and offsets in the hash table using that
+** column number.
+*/
+static int buildTerms(fulltext_vtab *v, sqlite_int64 iDocid,
+ const char *zText, int iColumn){
+ const char *pToken;
+ int nTokenBytes;
+ int iStartOffset, iEndOffset, iPosition, stop_word, new_paragraph;
+ int rc;
+ TrackerParser *parser = v->parser;
+
+ if (!zText) return SQLITE_OK;
+
+ tracker_parser_reset (parser, zText, strlen (zText), FALSE, TRUE, TRUE, FALSE);
+
+ while( 1 ){
+
+ pToken = tracker_parser_next (parser, &iPosition,
+ &iStartOffset,
+ &iEndOffset,
+ &new_paragraph,
+ &stop_word,
+ &nTokenBytes);
+ if (!pToken) {
+ break;
+ }
+
+ // printf("token being indexed is %s, begin is %d, end is %d and length is %d\n", pToken, iStartOffset, iEndOffset, nTokenBytes);
+
+ if (stop_word) {
+ continue;
+ }
+
+
+
+
+ DLCollector *p;
+ int nData; /* Size of doclist before our update. */
+
+ /* Positions can't be negative; we use -1 as a terminator
+ * internally. Token can't be NULL or empty. */
+ if( iPosition<0 || pToken == NULL || nTokenBytes == 0 ){
+ rc = SQLITE_ERROR;
+ break;
+ }
+
+ p = fts3HashFind(&v->pendingTerms, pToken, nTokenBytes);
+ if( p==NULL ){
+ nData = 0;
+ p = dlcNew(iDocid, DL_DEFAULT);
+ fts3HashInsert(&v->pendingTerms, pToken, nTokenBytes, p);
+
+ /* Overhead for our hash table entry, the key, and the value. */
+ v->nPendingData += sizeof(struct fts3HashElem)+sizeof(*p)+nTokenBytes;
+ }else{
+ nData = p->b.nData;
+ if( p->dlw.iPrevDocid!=iDocid ) dlcNext(p, iDocid);
+ }
+ if( iColumn>=0 ){
+ dlcAddPos(p, iColumn, iPosition, iStartOffset, iEndOffset);
+ }
+
+ /* Accumulate data added by dlcNew or dlcNext, and dlcAddPos. */
+ v->nPendingData += p->b.nData-nData;
+ }
+
+ /* TODO(shess) Check return? Should this be able to cause errors at
+ ** this point? Actually, same question about sqlite3_finalize(),
+ ** though one could argue that failure there means that the data is
+ ** not durable. *ponder*
+ */
+
+ return SQLITE_OK;
+
+}
+
+/* Add doclists for all terms in [pValues] to pendingTerms table. */
+static int insertTerms(fulltext_vtab *v, sqlite_int64 iDocid,
+ sqlite3_value **pValues){
+ int i;
+ for(i = 0; i < v->nColumn ; ++i){
+ char *zText = (char*)sqlite3_value_text(pValues[i]);
+ int rc = buildTerms(v, iDocid, zText, i);
+ if( rc!=SQLITE_OK ) return rc;
+ }
+ return SQLITE_OK;
+}
+
+/* Add empty doclists for all terms in the given row's content to
+** pendingTerms.
+*/
+static int deleteTerms(fulltext_vtab *v, sqlite_int64 iDocid){
+ const char **pValues;
+ int i, rc;
+
+ /* TODO(shess) Should we allow such tables at all? */
+ if( DL_DEFAULT==DL_DOCIDS ) return SQLITE_ERROR;
+
+ rc = content_select(v, iDocid, &pValues);
+ if( rc!=SQLITE_OK ) return rc;
+
+ for(i = 0 ; i < v->nColumn; ++i) {
+ rc = buildTerms(v, iDocid, pValues[i], -1);
+ if( rc!=SQLITE_OK ) break;
+ }
+
+ freeStringArray(v->nColumn, pValues);
+ return SQLITE_OK;
+}
+
+/* TODO(shess) Refactor the code to remove this forward decl. */
+static int initPendingTerms(fulltext_vtab *v, sqlite_int64 iDocid);
+
+/* Insert a row into the %_content table; set *piDocid to be the ID of the
+** new row. Add doclists for terms to pendingTerms.
+*/
+static int index_insert(fulltext_vtab *v, sqlite3_value *pRequestDocid,
+ sqlite3_value **pValues, sqlite_int64 *piDocid){
+ int rc;
+
+ rc = content_insert(v, pRequestDocid, pValues); /* execute an SQL INSERT */
+ if( rc!=SQLITE_OK ) return rc;
+
+ /* docid column is an alias for rowid. */
+ *piDocid = sqlite3_last_insert_rowid(v->db);
+ rc = initPendingTerms(v, *piDocid);
+ if( rc!=SQLITE_OK ) return rc;
+
+ return insertTerms(v, *piDocid, pValues);
+}
+
+/* Delete a row from the %_content table; add empty doclists for terms
+** to pendingTerms.
+*/
+static int index_delete(fulltext_vtab *v, sqlite_int64 iRow){
+ int rc = initPendingTerms(v, iRow);
+ if( rc!=SQLITE_OK ) return rc;
+
+ rc = deleteTerms(v, iRow);
+ if( rc!=SQLITE_OK ) return rc;
+
+ return content_delete(v, iRow); /* execute an SQL DELETE */
+}
+
+/* Update a row in the %_content table; add delete doclists to
+** pendingTerms for old terms not in the new data, add insert doclists
+** to pendingTerms for terms in the new data.
+*/
+static int index_update(fulltext_vtab *v, sqlite_int64 iRow,
+ sqlite3_value **pValues){
+ int rc = initPendingTerms(v, iRow);
+ if( rc!=SQLITE_OK ) return rc;
+
+ /* Generate an empty doclist for each term that previously appeared in this
+ * row. */
+ rc = deleteTerms(v, iRow);
+ if( rc!=SQLITE_OK ) return rc;
+
+ rc = content_update(v, pValues, iRow); /* execute an SQL UPDATE */
+ if( rc!=SQLITE_OK ) return rc;
+
+ /* Now add positions for terms which appear in the updated row. */
+ return insertTerms(v, iRow, pValues);
+}
+
+/*******************************************************************/
+/* InteriorWriter is used to collect terms and block references into
+** interior nodes in %_segments. See commentary at top of file for
+** format.
+*/
+
+/* How large interior nodes can grow. */
+#define INTERIOR_MAX 2048
+
+/* Minimum number of terms per interior node (except the root). This
+** prevents large terms from making the tree too skinny - must be >0
+** so that the tree always makes progress. Note that the min tree
+** fanout will be INTERIOR_MIN_TERMS+1.
+*/
+#define INTERIOR_MIN_TERMS 7
+#if INTERIOR_MIN_TERMS<1
+# error INTERIOR_MIN_TERMS must be greater than 0.
+#endif
+
+/* ROOT_MAX controls how much data is stored inline in the segment
+** directory.
+*/
+/* TODO(shess) Push ROOT_MAX down to whoever is writing things. It's
+** only here so that interiorWriterRootInfo() and leafWriterRootInfo()
+** can both see it, but if the caller passed it in, we wouldn't even
+** need a define.
+*/
+#define ROOT_MAX 1024
+#if ROOT_MAX<VARINT_MAX*2
+# error ROOT_MAX must have enough space for a header.
+#endif
+
+/* InteriorBlock stores a linked-list of interior blocks while a lower
+** layer is being constructed.
+*/
+typedef struct InteriorBlock {
+ DataBuffer term; /* Leftmost term in block's subtree. */
+ DataBuffer data; /* Accumulated data for the block. */
+ struct InteriorBlock *next;
+} InteriorBlock;
+
+static InteriorBlock *interiorBlockNew(int iHeight, sqlite_int64 iChildBlock,
+ const char *pTerm, int nTerm){
+ InteriorBlock *block = sqlite3_malloc(sizeof(InteriorBlock));
+ char c[VARINT_MAX+VARINT_MAX];
+ int n;
+
+ if( block ){
+ memset(block, 0, sizeof(*block));
+ dataBufferInit(&block->term, 0);
+ dataBufferReplace(&block->term, pTerm, nTerm);
+
+ n = fts3PutVarint(c, iHeight);
+ n += fts3PutVarint(c+n, iChildBlock);
+ dataBufferInit(&block->data, INTERIOR_MAX);
+ dataBufferReplace(&block->data, c, n);
+ }
+ return block;
+}
+
+#ifndef NDEBUG
+/* Verify that the data is readable as an interior node. */
+static void interiorBlockValidate(InteriorBlock *pBlock){
+ const char *pData = pBlock->data.pData;
+ int nData = pBlock->data.nData;
+ int n, iDummy;
+ sqlite_int64 iBlockid;
+
+ assert( nData>0 );
+ assert( pData!=0 );
+ assert( pData+nData>pData );
+
+ /* Must lead with height of node as a varint(n), n>0 */
+ n = fts3GetVarint32(pData, &iDummy);
+ assert( n>0 );
+ assert( iDummy>0 );
+ assert( n<nData );
+ pData += n;
+ nData -= n;
+
+ /* Must contain iBlockid. */
+ n = fts3GetVarint(pData, &iBlockid);
+ assert( n>0 );
+ assert( n<=nData );
+ pData += n;
+ nData -= n;
+
+ /* Zero or more terms of positive length */
+ if( nData!=0 ){
+ /* First term is not delta-encoded. */
+ n = fts3GetVarint32(pData, &iDummy);
+ assert( n>0 );
+ assert( iDummy>0 );
+ assert( n+iDummy>0);
+ assert( n+iDummy<=nData );
+ pData += n+iDummy;
+ nData -= n+iDummy;
+
+ /* Following terms delta-encoded. */
+ while( nData!=0 ){
+ /* Length of shared prefix. */
+ n = fts3GetVarint32(pData, &iDummy);
+ assert( n>0 );
+ assert( iDummy>=0 );
+ assert( n<nData );
+ pData += n;
+ nData -= n;
+
+ /* Length and data of distinct suffix. */
+ n = fts3GetVarint32(pData, &iDummy);
+ assert( n>0 );
+ assert( iDummy>0 );
+ assert( n+iDummy>0);
+ assert( n+iDummy<=nData );
+ pData += n+iDummy;
+ nData -= n+iDummy;
+ }
+ }
+}
+#define ASSERT_VALID_INTERIOR_BLOCK(x) interiorBlockValidate(x)
+#else
+#define ASSERT_VALID_INTERIOR_BLOCK(x) assert( 1 )
+#endif
+
+typedef struct InteriorWriter {
+ int iHeight; /* from 0 at leaves. */
+ InteriorBlock *first, *last;
+ struct InteriorWriter *parentWriter;
+
+ DataBuffer term; /* Last term written to block "last". */
+ sqlite_int64 iOpeningChildBlock; /* First child block in block "last". */
+#ifndef NDEBUG
+ sqlite_int64 iLastChildBlock; /* for consistency checks. */
+#endif
+} InteriorWriter;
+
+/* Initialize an interior node where pTerm[nTerm] marks the leftmost
+** term in the tree. iChildBlock is the leftmost child block at the
+** next level down the tree.
+*/
+static void interiorWriterInit(int iHeight, const char *pTerm, int nTerm,
+ sqlite_int64 iChildBlock,
+ InteriorWriter *pWriter){
+ InteriorBlock *block;
+ assert( iHeight>0 );
+ CLEAR(pWriter);
+
+ pWriter->iHeight = iHeight;
+ pWriter->iOpeningChildBlock = iChildBlock;
+#ifndef NDEBUG
+ pWriter->iLastChildBlock = iChildBlock;
+#endif
+ block = interiorBlockNew(iHeight, iChildBlock, pTerm, nTerm);
+ pWriter->last = pWriter->first = block;
+ ASSERT_VALID_INTERIOR_BLOCK(pWriter->last);
+ dataBufferInit(&pWriter->term, 0);
+}
+
+/* Append the child node rooted at iChildBlock to the interior node,
+** with pTerm[nTerm] as the leftmost term in iChildBlock's subtree.
+*/
+static void interiorWriterAppend(InteriorWriter *pWriter,
+ const char *pTerm, int nTerm,
+ sqlite_int64 iChildBlock){
+ char c[VARINT_MAX+VARINT_MAX];
+ int n, nPrefix = 0;
+
+ ASSERT_VALID_INTERIOR_BLOCK(pWriter->last);
+
+ /* The first term written into an interior node is actually
+ ** associated with the second child added (the first child was added
+ ** in interiorWriterInit, or in the if clause at the bottom of this
+ ** function). That term gets encoded straight up, with nPrefix left
+ ** at 0.
+ */
+ if( pWriter->term.nData==0 ){
+ n = fts3PutVarint(c, nTerm);
+ }else{
+ while( nPrefix<pWriter->term.nData &&
+ pTerm[nPrefix]==pWriter->term.pData[nPrefix] ){
+ nPrefix++;
+ }
+
+ n = fts3PutVarint(c, nPrefix);
+ n += fts3PutVarint(c+n, nTerm-nPrefix);
+ }
+
+#ifndef NDEBUG
+ pWriter->iLastChildBlock++;
+#endif
+ assert( pWriter->iLastChildBlock==iChildBlock );
+
+ /* Overflow to a new block if the new term makes the current block
+ ** too big, and the current block already has enough terms.
+ */
+ if( pWriter->last->data.nData+n+nTerm-nPrefix>INTERIOR_MAX &&
+ iChildBlock-pWriter->iOpeningChildBlock>INTERIOR_MIN_TERMS ){
+ pWriter->last->next = interiorBlockNew(pWriter->iHeight, iChildBlock,
+ pTerm, nTerm);
+ pWriter->last = pWriter->last->next;
+ pWriter->iOpeningChildBlock = iChildBlock;
+ dataBufferReset(&pWriter->term);
+ }else{
+ dataBufferAppend2(&pWriter->last->data, c, n,
+ pTerm+nPrefix, nTerm-nPrefix);
+ dataBufferReplace(&pWriter->term, pTerm, nTerm);
+ }
+ ASSERT_VALID_INTERIOR_BLOCK(pWriter->last);
+}
+
+/* Free the space used by pWriter, including the linked-list of
+** InteriorBlocks, and parentWriter, if present.
+*/
+static int interiorWriterDestroy(InteriorWriter *pWriter){
+ InteriorBlock *block = pWriter->first;
+
+ while( block!=NULL ){
+ InteriorBlock *b = block;
+ block = block->next;
+ dataBufferDestroy(&b->term);
+ dataBufferDestroy(&b->data);
+ sqlite3_free(b);
+ }
+ if( pWriter->parentWriter!=NULL ){
+ interiorWriterDestroy(pWriter->parentWriter);
+ sqlite3_free(pWriter->parentWriter);
+ }
+ dataBufferDestroy(&pWriter->term);
+ SCRAMBLE(pWriter);
+ return SQLITE_OK;
+}
+
+/* If pWriter can fit entirely in ROOT_MAX, return it as the root info
+** directly, leaving *piEndBlockid unchanged. Otherwise, flush
+** pWriter to %_segments, building a new layer of interior nodes, and
+** recursively ask for their root into.
+*/
+static int interiorWriterRootInfo(fulltext_vtab *v, InteriorWriter *pWriter,
+ char **ppRootInfo, int *pnRootInfo,
+ sqlite_int64 *piEndBlockid){
+ InteriorBlock *block = pWriter->first;
+ sqlite_int64 iBlockid = 0;
+ int rc;
+
+ /* If we can fit the segment inline */
+ if( block==pWriter->last && block->data.nData<ROOT_MAX ){
+ *ppRootInfo = block->data.pData;
+ *pnRootInfo = block->data.nData;
+ return SQLITE_OK;
+ }
+
+ /* Flush the first block to %_segments, and create a new level of
+ ** interior node.
+ */
+ ASSERT_VALID_INTERIOR_BLOCK(block);
+ rc = block_insert(v, block->data.pData, block->data.nData, &iBlockid);
+ if( rc!=SQLITE_OK ) return rc;
+ *piEndBlockid = iBlockid;
+
+ pWriter->parentWriter = sqlite3_malloc(sizeof(*pWriter->parentWriter));
+ interiorWriterInit(pWriter->iHeight+1,
+ block->term.pData, block->term.nData,
+ iBlockid, pWriter->parentWriter);
+
+ /* Flush additional blocks and append to the higher interior
+ ** node.
+ */
+ for(block=block->next; block!=NULL; block=block->next){
+ ASSERT_VALID_INTERIOR_BLOCK(block);
+ rc = block_insert(v, block->data.pData, block->data.nData, &iBlockid);
+ if( rc!=SQLITE_OK ) return rc;
+ *piEndBlockid = iBlockid;
+
+ interiorWriterAppend(pWriter->parentWriter,
+ block->term.pData, block->term.nData, iBlockid);
+ }
+
+ /* Parent node gets the chance to be the root. */
+ return interiorWriterRootInfo(v, pWriter->parentWriter,
+ ppRootInfo, pnRootInfo, piEndBlockid);
+}
+
+/****************************************************************/
+/* InteriorReader is used to read off the data from an interior node
+** (see comment at top of file for the format).
+*/
+typedef struct InteriorReader {
+ const char *pData;
+ int nData;
+
+ DataBuffer term; /* previous term, for decoding term delta. */
+
+ sqlite_int64 iBlockid;
+} InteriorReader;
+
+static void interiorReaderDestroy(InteriorReader *pReader){
+ dataBufferDestroy(&pReader->term);
+ SCRAMBLE(pReader);
+}
+
+/* TODO(shess) The assertions are great, but what if we're in NDEBUG
+** and the blob is empty or otherwise contains suspect data?
+*/
+static void interiorReaderInit(const char *pData, int nData,
+ InteriorReader *pReader){
+ int n, nTerm;
+
+ /* Require at least the leading flag byte */
+ assert( nData>0 );
+ assert( pData[0]!='\0' );
+
+ CLEAR(pReader);
+
+ /* Decode the base blockid, and set the cursor to the first term. */
+ n = fts3GetVarint(pData+1, &pReader->iBlockid);
+ assert( 1+n<=nData );
+ pReader->pData = pData+1+n;
+ pReader->nData = nData-(1+n);
+
+ /* A single-child interior node (such as when a leaf node was too
+ ** large for the segment directory) won't have any terms.
+ ** Otherwise, decode the first term.
+ */
+ if( pReader->nData==0 ){
+ dataBufferInit(&pReader->term, 0);
+ }else{
+ n = fts3GetVarint32(pReader->pData, &nTerm);
+ dataBufferInit(&pReader->term, nTerm);
+ dataBufferReplace(&pReader->term, pReader->pData+n, nTerm);
+ assert( n+nTerm<=pReader->nData );
+ pReader->pData += n+nTerm;
+ pReader->nData -= n+nTerm;
+ }
+}
+
+static int interiorReaderAtEnd(InteriorReader *pReader){
+ return pReader->term.nData==0;
+}
+
+static sqlite_int64 interiorReaderCurrentBlockid(InteriorReader *pReader){
+ return pReader->iBlockid;
+}
+
+static int interiorReaderTermBytes(InteriorReader *pReader){
+ assert( !interiorReaderAtEnd(pReader) );
+ return pReader->term.nData;
+}
+static const char *interiorReaderTerm(InteriorReader *pReader){
+ assert( !interiorReaderAtEnd(pReader) );
+ return pReader->term.pData;
+}
+
+/* Step forward to the next term in the node. */
+static void interiorReaderStep(InteriorReader *pReader){
+ assert( !interiorReaderAtEnd(pReader) );
+
+ /* If the last term has been read, signal eof, else construct the
+ ** next term.
+ */
+ if( pReader->nData==0 ){
+ dataBufferReset(&pReader->term);
+ }else{
+ int n, nPrefix, nSuffix;
+
+ n = fts3GetVarint32(pReader->pData, &nPrefix);
+ n += fts3GetVarint32(pReader->pData+n, &nSuffix);
+
+ /* Truncate the current term and append suffix data. */
+ pReader->term.nData = nPrefix;
+ dataBufferAppend(&pReader->term, pReader->pData+n, nSuffix);
+
+ assert( n+nSuffix<=pReader->nData );
+ pReader->pData += n+nSuffix;
+ pReader->nData -= n+nSuffix;
+ }
+ pReader->iBlockid++;
+}
+
+/* Compare the current term to pTerm[nTerm], returning strcmp-style
+** results. If isPrefix, equality means equal through nTerm bytes.
+*/
+static int interiorReaderTermCmp(InteriorReader *pReader,
+ const char *pTerm, int nTerm, int isPrefix){
+ const char *pReaderTerm = interiorReaderTerm(pReader);
+ int nReaderTerm = interiorReaderTermBytes(pReader);
+ int c, n = nReaderTerm<nTerm ? nReaderTerm : nTerm;
+
+ if( n==0 ){
+ if( nReaderTerm>0 ) return -1;
+ if( nTerm>0 ) return 1;
+ return 0;
+ }
+
+ c = memcmp(pReaderTerm, pTerm, n);
+ if( c!=0 ) return c;
+ if( isPrefix && n==nTerm ) return 0;
+ return nReaderTerm - nTerm;
+}
+
+/****************************************************************/
+/* LeafWriter is used to collect terms and associated doclist data
+** into leaf blocks in %_segments (see top of file for format info).
+** Expected usage is:
+**
+** LeafWriter writer;
+** leafWriterInit(0, 0, &writer);
+** while( sorted_terms_left_to_process ){
+** // data is doclist data for that term.
+** rc = leafWriterStep(v, &writer, pTerm, nTerm, pData, nData);
+** if( rc!=SQLITE_OK ) goto err;
+** }
+** rc = leafWriterFinalize(v, &writer);
+**err:
+** leafWriterDestroy(&writer);
+** return rc;
+**
+** leafWriterStep() may write a collected leaf out to %_segments.
+** leafWriterFinalize() finishes writing any buffered data and stores
+** a root node in %_segdir. leafWriterDestroy() frees all buffers and
+** InteriorWriters allocated as part of writing this segment.
+**
+** TODO(shess) Document leafWriterStepMerge().
+*/
+
+/* Put terms with data this big in their own block. */
+#define STANDALONE_MIN 1024
+
+/* Keep leaf blocks below this size. */
+#define LEAF_MAX 2048
+
+typedef struct LeafWriter {
+ int iLevel;
+ int idx;
+ sqlite_int64 iStartBlockid; /* needed to create the root info */
+ sqlite_int64 iEndBlockid; /* when we're done writing. */
+
+ DataBuffer term; /* previous encoded term */
+ DataBuffer data; /* encoding buffer */
+
+ /* bytes of first term in the current node which distinguishes that
+ ** term from the last term of the previous node.
+ */
+ int nTermDistinct;
+
+ InteriorWriter parentWriter; /* if we overflow */
+ int has_parent;
+} LeafWriter;
+
+static void leafWriterInit(int iLevel, int idx, LeafWriter *pWriter){
+ CLEAR(pWriter);
+ pWriter->iLevel = iLevel;
+ pWriter->idx = idx;
+
+ dataBufferInit(&pWriter->term, 32);
+
+ /* Start out with a reasonably sized block, though it can grow. */
+ dataBufferInit(&pWriter->data, LEAF_MAX);
+}
+
+#ifndef NDEBUG
+/* Verify that the data is readable as a leaf node. */
+static void leafNodeValidate(const char *pData, int nData){
+ int n, iDummy;
+
+ if( nData==0 ) return;
+ assert( nData>0 );
+ assert( pData!=0 );
+ assert( pData+nData>pData );
+
+ /* Must lead with a varint(0) */
+ n = fts3GetVarint32(pData, &iDummy);
+ assert( iDummy==0 );
+ assert( n>0 );
+ assert( n<nData );
+ pData += n;
+ nData -= n;
+
+ /* Leading term length and data must fit in buffer. */
+ n = fts3GetVarint32(pData, &iDummy);
+ assert( n>0 );
+ assert( iDummy>0 );
+ assert( n+iDummy>0 );
+ assert( n+iDummy<nData );
+ pData += n+iDummy;
+ nData -= n+iDummy;
+
+ /* Leading term's doclist length and data must fit. */
+ n = fts3GetVarint32(pData, &iDummy);
+ assert( n>0 );
+ assert( iDummy>0 );
+ assert( n+iDummy>0 );
+ assert( n+iDummy<=nData );
+ ASSERT_VALID_DOCLIST(DL_DEFAULT, pData+n, iDummy, NULL);
+ pData += n+iDummy;
+ nData -= n+iDummy;
+
+ /* Verify that trailing terms and doclists also are readable. */
+ while( nData!=0 ){
+ n = fts3GetVarint32(pData, &iDummy);
+ assert( n>0 );
+ assert( iDummy>=0 );
+ assert( n<nData );
+ pData += n;
+ nData -= n;
+ n = fts3GetVarint32(pData, &iDummy);
+ assert( n>0 );
+ assert( iDummy>0 );
+ assert( n+iDummy>0 );
+ assert( n+iDummy<nData );
+ pData += n+iDummy;
+ nData -= n+iDummy;
+
+ n = fts3GetVarint32(pData, &iDummy);
+ assert( n>0 );
+ assert( iDummy>0 );
+ assert( n+iDummy>0 );
+ assert( n+iDummy<=nData );
+ ASSERT_VALID_DOCLIST(DL_DEFAULT, pData+n, iDummy, NULL);
+ pData += n+iDummy;
+ nData -= n+iDummy;
+ }
+}
+#define ASSERT_VALID_LEAF_NODE(p, n) leafNodeValidate(p, n)
+#else
+#define ASSERT_VALID_LEAF_NODE(p, n) assert( 1 )
+#endif
+
+/* Flush the current leaf node to %_segments, and adding the resulting
+** blockid and the starting term to the interior node which will
+** contain it.
+*/
+static int leafWriterInternalFlush(fulltext_vtab *v, LeafWriter *pWriter,
+ int iData, int nData){
+ sqlite_int64 iBlockid = 0;
+ const char *pStartingTerm;
+ int nStartingTerm, rc, n;
+
+ /* Must have the leading varint(0) flag, plus at least some
+ ** valid-looking data.
+ */
+ assert( nData>2 );
+ assert( iData>=0 );
+ assert( iData+nData<=pWriter->data.nData );
+ ASSERT_VALID_LEAF_NODE(pWriter->data.pData+iData, nData);
+
+ rc = block_insert(v, pWriter->data.pData+iData, nData, &iBlockid);
+ if( rc!=SQLITE_OK ) return rc;
+ assert( iBlockid!=0 );
+
+ /* Reconstruct the first term in the leaf for purposes of building
+ ** the interior node.
+ */
+ n = fts3GetVarint32(pWriter->data.pData+iData+1, &nStartingTerm);
+ pStartingTerm = pWriter->data.pData+iData+1+n;
+ assert( pWriter->data.nData>iData+1+n+nStartingTerm );
+ assert( pWriter->nTermDistinct>0 );
+ assert( pWriter->nTermDistinct<=nStartingTerm );
+ nStartingTerm = pWriter->nTermDistinct;
+
+ if( pWriter->has_parent ){
+ interiorWriterAppend(&pWriter->parentWriter,
+ pStartingTerm, nStartingTerm, iBlockid);
+ }else{
+ interiorWriterInit(1, pStartingTerm, nStartingTerm, iBlockid,
+ &pWriter->parentWriter);
+ pWriter->has_parent = 1;
+ }
+
+ /* Track the span of this segment's leaf nodes. */
+ if( pWriter->iEndBlockid==0 ){
+ pWriter->iEndBlockid = pWriter->iStartBlockid = iBlockid;
+ }else{
+ pWriter->iEndBlockid++;
+ assert( iBlockid==pWriter->iEndBlockid );
+ }
+
+ return SQLITE_OK;
+}
+static int leafWriterFlush(fulltext_vtab *v, LeafWriter *pWriter){
+ int rc = leafWriterInternalFlush(v, pWriter, 0, pWriter->data.nData);
+ if( rc!=SQLITE_OK ) return rc;
+
+ /* Re-initialize the output buffer. */
+ dataBufferReset(&pWriter->data);
+
+ return SQLITE_OK;
+}
+
+/* Fetch the root info for the segment. If the entire leaf fits
+** within ROOT_MAX, then it will be returned directly, otherwise it
+** will be flushed and the root info will be returned from the
+** interior node. *piEndBlockid is set to the blockid of the last
+** interior or leaf node written to disk (0 if none are written at
+** all).
+*/
+static int leafWriterRootInfo(fulltext_vtab *v, LeafWriter *pWriter,
+ char **ppRootInfo, int *pnRootInfo,
+ sqlite_int64 *piEndBlockid){
+ /* we can fit the segment entirely inline */
+ if( !pWriter->has_parent && pWriter->data.nData<ROOT_MAX ){
+ *ppRootInfo = pWriter->data.pData;
+ *pnRootInfo = pWriter->data.nData;
+ *piEndBlockid = 0;
+ return SQLITE_OK;
+ }
+
+ /* Flush remaining leaf data. */
+ if( pWriter->data.nData>0 ){
+ int rc = leafWriterFlush(v, pWriter);
+ if( rc!=SQLITE_OK ) return rc;
+ }
+
+ /* We must have flushed a leaf at some point. */
+ assert( pWriter->has_parent );
+
+ /* Tenatively set the end leaf blockid as the end blockid. If the
+ ** interior node can be returned inline, this will be the final
+ ** blockid, otherwise it will be overwritten by
+ ** interiorWriterRootInfo().
+ */
+ *piEndBlockid = pWriter->iEndBlockid;
+
+ return interiorWriterRootInfo(v, &pWriter->parentWriter,
+ ppRootInfo, pnRootInfo, piEndBlockid);
+}
+
+/* Collect the rootInfo data and store it into the segment directory.
+** This has the effect of flushing the segment's leaf data to
+** %_segments, and also flushing any interior nodes to %_segments.
+*/
+static int leafWriterFinalize(fulltext_vtab *v, LeafWriter *pWriter){
+ sqlite_int64 iEndBlockid;
+ char *pRootInfo;
+ int rc, nRootInfo;
+
+ rc = leafWriterRootInfo(v, pWriter, &pRootInfo, &nRootInfo, &iEndBlockid);
+ if( rc!=SQLITE_OK ) return rc;
+
+ /* Don't bother storing an entirely empty segment. */
+ if( iEndBlockid==0 && nRootInfo==0 ) return SQLITE_OK;
+
+ return segdir_set(v, pWriter->iLevel, pWriter->idx,
+ pWriter->iStartBlockid, pWriter->iEndBlockid,
+ iEndBlockid, pRootInfo, nRootInfo);
+}
+
+static void leafWriterDestroy(LeafWriter *pWriter){
+ if( pWriter->has_parent ) interiorWriterDestroy(&pWriter->parentWriter);
+ dataBufferDestroy(&pWriter->term);
+ dataBufferDestroy(&pWriter->data);
+}
+
+/* Encode a term into the leafWriter, delta-encoding as appropriate.
+** Returns the length of the new term which distinguishes it from the
+** previous term, which can be used to set nTermDistinct when a node
+** boundary is crossed.
+*/
+static int leafWriterEncodeTerm(LeafWriter *pWriter,
+ const char *pTerm, int nTerm){
+ char c[VARINT_MAX+VARINT_MAX];
+ int n, nPrefix = 0;
+
+ assert( nTerm>0 );
+ while( nPrefix<pWriter->term.nData &&
+ pTerm[nPrefix]==pWriter->term.pData[nPrefix] ){
+ nPrefix++;
+ /* Failing this implies that the terms weren't in order. */
+ assert( nPrefix<nTerm );
+ }
+
+ if( pWriter->data.nData==0 ){
+ /* Encode the node header and leading term as:
+ ** varint(0)
+ ** varint(nTerm)
+ ** char pTerm[nTerm]
+ */
+ n = fts3PutVarint(c, '\0');
+ n += fts3PutVarint(c+n, nTerm);
+ dataBufferAppend2(&pWriter->data, c, n, pTerm, nTerm);
+ }else{
+ /* Delta-encode the term as:
+ ** varint(nPrefix)
+ ** varint(nSuffix)
+ ** char pTermSuffix[nSuffix]
+ */
+ n = fts3PutVarint(c, nPrefix);
+ n += fts3PutVarint(c+n, nTerm-nPrefix);
+ dataBufferAppend2(&pWriter->data, c, n, pTerm+nPrefix, nTerm-nPrefix);
+ }
+ dataBufferReplace(&pWriter->term, pTerm, nTerm);
+
+ return nPrefix+1;
+}
+
+/* Used to avoid a memmove when a large amount of doclist data is in
+** the buffer. This constructs a node and term header before
+** iDoclistData and flushes the resulting complete node using
+** leafWriterInternalFlush().
+*/
+static int leafWriterInlineFlush(fulltext_vtab *v, LeafWriter *pWriter,
+ const char *pTerm, int nTerm,
+ int iDoclistData){
+ char c[VARINT_MAX+VARINT_MAX];
+ int iData, n = fts3PutVarint(c, 0);
+ n += fts3PutVarint(c+n, nTerm);
+
+ /* There should always be room for the header. Even if pTerm shared
+ ** a substantial prefix with the previous term, the entire prefix
+ ** could be constructed from earlier data in the doclist, so there
+ ** should be room.
+ */
+ assert( iDoclistData>=n+nTerm );
+
+ iData = iDoclistData-(n+nTerm);
+ memcpy(pWriter->data.pData+iData, c, n);
+ memcpy(pWriter->data.pData+iData+n, pTerm, nTerm);
+
+ return leafWriterInternalFlush(v, pWriter, iData, pWriter->data.nData-iData);
+}
+
+/* Push pTerm[nTerm] along with the doclist data to the leaf layer of
+** %_segments.
+*/
+static int leafWriterStepMerge(fulltext_vtab *v, LeafWriter *pWriter,
+ const char *pTerm, int nTerm,
+ DLReader *pReaders, int nReaders){
+ char c[VARINT_MAX+VARINT_MAX];
+ int iTermData = pWriter->data.nData, iDoclistData;
+ int i, nData, n, nActualData, nActual, rc, nTermDistinct;
+
+ ASSERT_VALID_LEAF_NODE(pWriter->data.pData, pWriter->data.nData);
+ nTermDistinct = leafWriterEncodeTerm(pWriter, pTerm, nTerm);
+
+ /* Remember nTermDistinct if opening a new node. */
+ if( iTermData==0 ) pWriter->nTermDistinct = nTermDistinct;
+
+ iDoclistData = pWriter->data.nData;
+
+ /* Estimate the length of the merged doclist so we can leave space
+ ** to encode it.
+ */
+ for(i=0, nData=0; i<nReaders; i++){
+ nData += dlrAllDataBytes(&pReaders[i]);
+ }
+ n = fts3PutVarint(c, nData);
+ dataBufferAppend(&pWriter->data, c, n);
+
+ docListMerge(&pWriter->data, pReaders, nReaders);
+ ASSERT_VALID_DOCLIST(DL_DEFAULT,
+ pWriter->data.pData+iDoclistData+n,
+ pWriter->data.nData-iDoclistData-n, NULL);
+
+ /* The actual amount of doclist data at this point could be smaller
+ ** than the length we encoded. Additionally, the space required to
+ ** encode this length could be smaller. For small doclists, this is
+ ** not a big deal, we can just use memmove() to adjust things.
+ */
+ nActualData = pWriter->data.nData-(iDoclistData+n);
+ nActual = fts3PutVarint(c, nActualData);
+ assert( nActualData<=nData );
+ assert( nActual<=n );
+
+ /* If the new doclist is big enough for force a standalone leaf
+ ** node, we can immediately flush it inline without doing the
+ ** memmove().
+ */
+ /* TODO(shess) This test matches leafWriterStep(), which does this
+ ** test before it knows the cost to varint-encode the term and
+ ** doclist lengths. At some point, change to
+ ** pWriter->data.nData-iTermData>STANDALONE_MIN.
+ */
+ if( nTerm+nActualData>STANDALONE_MIN ){
+ /* Push leaf node from before this term. */
+ if( iTermData>0 ){
+ rc = leafWriterInternalFlush(v, pWriter, 0, iTermData);
+ if( rc!=SQLITE_OK ) return rc;
+
+ pWriter->nTermDistinct = nTermDistinct;
+ }
+
+ /* Fix the encoded doclist length. */
+ iDoclistData += n - nActual;
+ memcpy(pWriter->data.pData+iDoclistData, c, nActual);
+
+ /* Push the standalone leaf node. */
+ rc = leafWriterInlineFlush(v, pWriter, pTerm, nTerm, iDoclistData);
+ if( rc!=SQLITE_OK ) return rc;
+
+ /* Leave the node empty. */
+ dataBufferReset(&pWriter->data);
+
+ return rc;
+ }
+
+ /* At this point, we know that the doclist was small, so do the
+ ** memmove if indicated.
+ */
+ if( nActual<n ){
+ memmove(pWriter->data.pData+iDoclistData+nActual,
+ pWriter->data.pData+iDoclistData+n,
+ pWriter->data.nData-(iDoclistData+n));
+ pWriter->data.nData -= n-nActual;
+ }
+
+ /* Replace written length with actual length. */
+ memcpy(pWriter->data.pData+iDoclistData, c, nActual);
+
+ /* If the node is too large, break things up. */
+ /* TODO(shess) This test matches leafWriterStep(), which does this
+ ** test before it knows the cost to varint-encode the term and
+ ** doclist lengths. At some point, change to
+ ** pWriter->data.nData>LEAF_MAX.
+ */
+ if( iTermData+nTerm+nActualData>LEAF_MAX ){
+ /* Flush out the leading data as a node */
+ rc = leafWriterInternalFlush(v, pWriter, 0, iTermData);
+ if( rc!=SQLITE_OK ) return rc;
+
+ pWriter->nTermDistinct = nTermDistinct;
+
+ /* Rebuild header using the current term */
+ n = fts3PutVarint(pWriter->data.pData, 0);
+ n += fts3PutVarint(pWriter->data.pData+n, nTerm);
+ memcpy(pWriter->data.pData+n, pTerm, nTerm);
+ n += nTerm;
+
+ /* There should always be room, because the previous encoding
+ ** included all data necessary to construct the term.
+ */
+ assert( n<iDoclistData );
+ /* So long as STANDALONE_MIN is half or less of LEAF_MAX, the
+ ** following memcpy() is safe (as opposed to needing a memmove).
+ */
+ assert( 2*STANDALONE_MIN<=LEAF_MAX );
+ assert( n+pWriter->data.nData-iDoclistData<iDoclistData );
+ memcpy(pWriter->data.pData+n,
+ pWriter->data.pData+iDoclistData,
+ pWriter->data.nData-iDoclistData);
+ pWriter->data.nData -= iDoclistData-n;
+ }
+ ASSERT_VALID_LEAF_NODE(pWriter->data.pData, pWriter->data.nData);
+
+ return SQLITE_OK;
+}
+
+/* Push pTerm[nTerm] along with the doclist data to the leaf layer of
+** %_segments.
+*/
+/* TODO(shess) Revise writeZeroSegment() so that doclists are
+** constructed directly in pWriter->data.
+*/
+static int leafWriterStep(fulltext_vtab *v, LeafWriter *pWriter,
+ const char *pTerm, int nTerm,
+ const char *pData, int nData){
+ int rc;
+ DLReader reader;
+
+ dlrInit(&reader, DL_DEFAULT, pData, nData);
+ rc = leafWriterStepMerge(v, pWriter, pTerm, nTerm, &reader, 1);
+ dlrDestroy(&reader);
+
+ return rc;
+}
+
+
+/****************************************************************/
+/* LeafReader is used to iterate over an individual leaf node. */
+typedef struct LeafReader {
+ DataBuffer term; /* copy of current term. */
+
+ const char *pData; /* data for current term. */
+ int nData;
+} LeafReader;
+
+static void leafReaderDestroy(LeafReader *pReader){
+ dataBufferDestroy(&pReader->term);
+ SCRAMBLE(pReader);
+}
+
+static int leafReaderAtEnd(LeafReader *pReader){
+ return pReader->nData<=0;
+}
+
+/* Access the current term. */
+static int leafReaderTermBytes(LeafReader *pReader){
+ return pReader->term.nData;
+}
+static const char *leafReaderTerm(LeafReader *pReader){
+ assert( pReader->term.nData>0 );
+ return pReader->term.pData;
+}
+
+/* Access the doclist data for the current term. */
+static int leafReaderDataBytes(LeafReader *pReader){
+ int nData;
+ assert( pReader->term.nData>0 );
+ fts3GetVarint32(pReader->pData, &nData);
+ return nData;
+}
+static const char *leafReaderData(LeafReader *pReader){
+ int n, nData;
+ assert( pReader->term.nData>0 );
+ n = fts3GetVarint32(pReader->pData, &nData);
+ return pReader->pData+n;
+}
+
+static void leafReaderInit(const char *pData, int nData,
+ LeafReader *pReader){
+ int nTerm, n;
+
+ assert( nData>0 );
+ assert( pData[0]=='\0' );
+
+ CLEAR(pReader);
+
+ /* Read the first term, skipping the header byte. */
+ n = fts3GetVarint32(pData+1, &nTerm);
+ dataBufferInit(&pReader->term, nTerm);
+ dataBufferReplace(&pReader->term, pData+1+n, nTerm);
+
+ /* Position after the first term. */
+ assert( 1+n+nTerm<nData );
+ pReader->pData = pData+1+n+nTerm;
+ pReader->nData = nData-1-n-nTerm;
+}
+
+/* Step the reader forward to the next term. */
+static void leafReaderStep(LeafReader *pReader){
+ int n, nData, nPrefix, nSuffix;
+ assert( !leafReaderAtEnd(pReader) );
+
+ /* Skip previous entry's data block. */
+ n = fts3GetVarint32(pReader->pData, &nData);
+ assert( n+nData<=pReader->nData );
+ pReader->pData += n+nData;
+ pReader->nData -= n+nData;
+
+ if( !leafReaderAtEnd(pReader) ){
+ /* Construct the new term using a prefix from the old term plus a
+ ** suffix from the leaf data.
+ */
+ n = fts3GetVarint32(pReader->pData, &nPrefix);
+ n += fts3GetVarint32(pReader->pData+n, &nSuffix);
+ assert( n+nSuffix<pReader->nData );
+ pReader->term.nData = nPrefix;
+ dataBufferAppend(&pReader->term, pReader->pData+n, nSuffix);
+
+ pReader->pData += n+nSuffix;
+ pReader->nData -= n+nSuffix;
+ }
+}
+
+/* strcmp-style comparison of pReader's current term against pTerm.
+** If isPrefix, equality means equal through nTerm bytes.
+*/
+static int leafReaderTermCmp(LeafReader *pReader,
+ const char *pTerm, int nTerm, int isPrefix){
+ int c, n = pReader->term.nData<nTerm ? pReader->term.nData : nTerm;
+ if( n==0 ){
+ if( pReader->term.nData>0 ) return -1;
+ if(nTerm>0 ) return 1;
+ return 0;
+ }
+
+ c = memcmp(pReader->term.pData, pTerm, n);
+ if( c!=0 ) return c;
+ if( isPrefix && n==nTerm ) return 0;
+ return pReader->term.nData - nTerm;
+}
+
+
+/****************************************************************/
+/* LeavesReader wraps LeafReader to allow iterating over the entire
+** leaf layer of the tree.
+*/
+typedef struct LeavesReader {
+ int idx; /* Index within the segment. */
+
+ sqlite3_stmt *pStmt; /* Statement we're streaming leaves from. */
+ int eof; /* we've seen SQLITE_DONE from pStmt. */
+
+ LeafReader leafReader; /* reader for the current leaf. */
+ DataBuffer rootData; /* root data for inline. */
+} LeavesReader;
+
+/* Access the current term. */
+static int leavesReaderTermBytes(LeavesReader *pReader){
+ assert( !pReader->eof );
+ return leafReaderTermBytes(&pReader->leafReader);
+}
+static const char *leavesReaderTerm(LeavesReader *pReader){
+ assert( !pReader->eof );
+ return leafReaderTerm(&pReader->leafReader);
+}
+
+/* Access the doclist data for the current term. */
+static int leavesReaderDataBytes(LeavesReader *pReader){
+ assert( !pReader->eof );
+ return leafReaderDataBytes(&pReader->leafReader);
+}
+static const char *leavesReaderData(LeavesReader *pReader){
+ assert( !pReader->eof );
+ return leafReaderData(&pReader->leafReader);
+}
+
+static int leavesReaderAtEnd(LeavesReader *pReader){
+ return pReader->eof;
+}
+
+/* loadSegmentLeaves() may not read all the way to SQLITE_DONE, thus
+** leaving the statement handle open, which locks the table.
+*/
+/* TODO(shess) This "solution" is not satisfactory. Really, there
+** should be check-in function for all statement handles which
+** arranges to call sqlite3_reset(). This most likely will require
+** modification to control flow all over the place, though, so for now
+** just punt.
+**
+** Note the the current system assumes that segment merges will run to
+** completion, which is why this particular probably hasn't arisen in
+** this case. Probably a brittle assumption.
+*/
+static int leavesReaderReset(LeavesReader *pReader){
+ return sqlite3_reset(pReader->pStmt);
+}
+
+static void leavesReaderDestroy(LeavesReader *pReader){
+ /* If idx is -1, that means we're using a non-cached statement
+ ** handle in the optimize() case, so we need to release it.
+ */
+ if( pReader->pStmt!=NULL && pReader->idx==-1 ){
+ sqlite3_finalize(pReader->pStmt);
+ }
+ leafReaderDestroy(&pReader->leafReader);
+ dataBufferDestroy(&pReader->rootData);
+ SCRAMBLE(pReader);
+}
+
+/* Initialize pReader with the given root data (if iStartBlockid==0
+** the leaf data was entirely contained in the root), or from the
+** stream of blocks between iStartBlockid and iEndBlockid, inclusive.
+*/
+static int leavesReaderInit(fulltext_vtab *v,
+ int idx,
+ sqlite_int64 iStartBlockid,
+ sqlite_int64 iEndBlockid,
+ const char *pRootData, int nRootData,
+ LeavesReader *pReader){
+ CLEAR(pReader);
+ pReader->idx = idx;
+
+ dataBufferInit(&pReader->rootData, 0);
+ if( iStartBlockid==0 ){
+ /* Entire leaf level fit in root data. */
+ dataBufferReplace(&pReader->rootData, pRootData, nRootData);
+ leafReaderInit(pReader->rootData.pData, pReader->rootData.nData,
+ &pReader->leafReader);
+ }else{
+ sqlite3_stmt *s;
+ int rc = sql_get_leaf_statement(v, idx, &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;
+
+ rc = sqlite3_step(s);
+ if( rc==SQLITE_DONE ){
+ pReader->eof = 1;
+ return SQLITE_OK;
+ }
+ if( rc!=SQLITE_ROW ) return rc;
+
+ pReader->pStmt = s;
+ leafReaderInit(sqlite3_column_blob(pReader->pStmt, 0),
+ sqlite3_column_bytes(pReader->pStmt, 0),
+ &pReader->leafReader);
+ }
+ return SQLITE_OK;
+}
+
+/* Step the current leaf forward to the next term. If we reach the
+** end of the current leaf, step forward to the next leaf block.
+*/
+static int leavesReaderStep(fulltext_vtab *v, LeavesReader *pReader){
+ assert( !leavesReaderAtEnd(pReader) );
+ leafReaderStep(&pReader->leafReader);
+
+ if( leafReaderAtEnd(&pReader->leafReader) ){
+ int rc;
+ if( pReader->rootData.pData ){
+ pReader->eof = 1;
+ return SQLITE_OK;
+ }
+ rc = sqlite3_step(pReader->pStmt);
+ if( rc!=SQLITE_ROW ){
+ pReader->eof = 1;
+ return rc==SQLITE_DONE ? SQLITE_OK : rc;
+ }
+ leafReaderDestroy(&pReader->leafReader);
+ leafReaderInit(sqlite3_column_blob(pReader->pStmt, 0),
+ sqlite3_column_bytes(pReader->pStmt, 0),
+ &pReader->leafReader);
+ }
+ return SQLITE_OK;
+}
+
+/* Order LeavesReaders by their term, ignoring idx. Readers at eof
+** always sort to the end.
+*/
+static int leavesReaderTermCmp(LeavesReader *lr1, LeavesReader *lr2){
+ if( leavesReaderAtEnd(lr1) ){
+ if( leavesReaderAtEnd(lr2) ) return 0;
+ return 1;
+ }
+ if( leavesReaderAtEnd(lr2) ) return -1;
+
+ return leafReaderTermCmp(&lr1->leafReader,
+ leavesReaderTerm(lr2), leavesReaderTermBytes(lr2),
+ 0);
+}
+
+/* Similar to leavesReaderTermCmp(), with additional ordering by idx
+** so that older segments sort before newer segments.
+*/
+static int leavesReaderCmp(LeavesReader *lr1, LeavesReader *lr2){
+ int c = leavesReaderTermCmp(lr1, lr2);
+ if( c!=0 ) return c;
+ return lr1->idx-lr2->idx;
+}
+
+/* Assume that pLr[1]..pLr[nLr] are sorted. Bubble pLr[0] into its
+** sorted position.
+*/
+static void leavesReaderReorder(LeavesReader *pLr, int nLr){
+ while( nLr>1 && leavesReaderCmp(pLr, pLr+1)>0 ){
+ LeavesReader tmp = pLr[0];
+ pLr[0] = pLr[1];
+ pLr[1] = tmp;
+ nLr--;
+ pLr++;
+ }
+}
+
+/* Initializes pReaders with the segments from level iLevel, returning
+** the number of segments in *piReaders. Leaves pReaders in sorted
+** order.
+*/
+static int leavesReadersInit(fulltext_vtab *v, int iLevel,
+ LeavesReader *pReaders, int *piReaders){
+ sqlite3_stmt *s;
+ int i, rc = sql_get_statement(v, SEGDIR_SELECT_LEVEL_STMT, &s);
+ if( rc!=SQLITE_OK ) return rc;
+
+ rc = sqlite3_bind_int(s, 1, iLevel);
+ if( rc!=SQLITE_OK ) return rc;
+
+ i = 0;
+ while( (rc = sqlite3_step(s))==SQLITE_ROW ){
+ sqlite_int64 iStart = sqlite3_column_int64(s, 0);
+ sqlite_int64 iEnd = sqlite3_column_int64(s, 1);
+ const char *pRootData = sqlite3_column_blob(s, 2);
+ int nRootData = sqlite3_column_bytes(s, 2);
+
+ assert( i<MERGE_COUNT );
+ rc = leavesReaderInit(v, i, iStart, iEnd, pRootData, nRootData,
+ &pReaders[i]);
+ if( rc!=SQLITE_OK ) break;
+
+ i++;
+ }
+ if( rc!=SQLITE_DONE ){
+ while( i-->0 ){
+ leavesReaderDestroy(&pReaders[i]);
+ }
+ return rc;
+ }
+
+ *piReaders = i;
+
+ /* Leave our results sorted by term, then age. */
+ while( i-- ){
+ leavesReaderReorder(pReaders+i, *piReaders-i);
+ }
+ return SQLITE_OK;
+}
+
+/* Merge doclists from pReaders[nReaders] into a single doclist, which
+** is written to pWriter. Assumes pReaders is ordered oldest to
+** newest.
+*/
+/* TODO(shess) Consider putting this inline in segmentMerge(). */
+static int leavesReadersMerge(fulltext_vtab *v,
+ LeavesReader *pReaders, int nReaders,
+ LeafWriter *pWriter){
+ DLReader dlReaders[MERGE_COUNT];
+ const char *pTerm = leavesReaderTerm(pReaders);
+ int i, nTerm = leavesReaderTermBytes(pReaders);
+
+ assert( nReaders<=MERGE_COUNT );
+
+ for(i=0; i<nReaders; i++){
+ dlrInit(&dlReaders[i], DL_DEFAULT,
+ leavesReaderData(pReaders+i),
+ leavesReaderDataBytes(pReaders+i));
+ }
+
+ return leafWriterStepMerge(v, pWriter, pTerm, nTerm, dlReaders, nReaders);
+}
+
+/* Forward ref due to mutual recursion with segdirNextIndex(). */
+static int segmentMerge(fulltext_vtab *v, int iLevel);
+
+/* Put the next available index at iLevel into *pidx. If iLevel
+** already has MERGE_COUNT segments, they are merged to a higher
+** level to make room.
+*/
+static int segdirNextIndex(fulltext_vtab *v, int iLevel, int *pidx){
+ int rc = segdir_max_index(v, iLevel, pidx);
+ if( rc==SQLITE_DONE ){ /* No segments at iLevel. */
+ *pidx = 0;
+ }else if( rc==SQLITE_ROW ){
+ if( *pidx==(MERGE_COUNT-1) ){
+ rc = segmentMerge(v, iLevel);
+ if( rc!=SQLITE_OK ) return rc;
+ *pidx = 0;
+ }else{
+ (*pidx)++;
+ }
+ }else{
+ return rc;
+ }
+ return SQLITE_OK;
+}
+
+/* Merge MERGE_COUNT segments at iLevel into a new segment at
+** iLevel+1. If iLevel+1 is already full of segments, those will be
+** merged to make room.
+*/
+static int segmentMerge(fulltext_vtab *v, int iLevel){
+ LeafWriter writer;
+ LeavesReader lrs[MERGE_COUNT];
+ int i, rc, idx = 0;
+
+ /* Determine the next available segment index at the next level,
+ ** merging as necessary.
+ */
+ rc = segdirNextIndex(v, iLevel+1, &idx);
+ if( rc!=SQLITE_OK ) return rc;
+
+ /* TODO(shess) This assumes that we'll always see exactly
+ ** MERGE_COUNT segments to merge at a given level. That will be
+ ** broken if we allow the developer to request preemptive or
+ ** deferred merging.
+ */
+ memset(&lrs, '\0', sizeof(lrs));
+ rc = leavesReadersInit(v, iLevel, lrs, &i);
+ if( rc!=SQLITE_OK ) return rc;
+ assert( i==MERGE_COUNT );
+
+ leafWriterInit(iLevel+1, idx, &writer);
+
+ /* Since leavesReaderReorder() pushes readers at eof to the end,
+ ** when the first reader is empty, all will be empty.
+ */
+ while( !leavesReaderAtEnd(lrs) ){
+ /* Figure out how many readers share their next term. */
+ for(i=1; i<MERGE_COUNT && !leavesReaderAtEnd(lrs+i); i++){
+ if( 0!=leavesReaderTermCmp(lrs, lrs+i) ) break;
+ }
+
+ rc = leavesReadersMerge(v, lrs, i, &writer);
+ if( rc!=SQLITE_OK ) goto err;
+
+ /* Step forward those that were merged. */
+ while( i-->0 ){
+ rc = leavesReaderStep(v, lrs+i);
+ if( rc!=SQLITE_OK ) goto err;
+
+ /* Reorder by term, then by age. */
+ leavesReaderReorder(lrs+i, MERGE_COUNT-i);
+ }
+ }
+
+ for(i=0; i<MERGE_COUNT; i++){
+ leavesReaderDestroy(&lrs[i]);
+ }
+
+ rc = leafWriterFinalize(v, &writer);
+ leafWriterDestroy(&writer);
+ if( rc!=SQLITE_OK ) return rc;
+
+ /* Delete the merged segment data. */
+ return segdir_delete(v, iLevel);
+
+ err:
+ for(i=0; i<MERGE_COUNT; i++){
+ leavesReaderDestroy(&lrs[i]);
+ }
+ leafWriterDestroy(&writer);
+ return rc;
+}
+
+/* Accumulate the union of *acc and *pData into *acc. */
+static void docListAccumulateUnion(DataBuffer *acc,
+ const char *pData, int nData) {
+ DataBuffer tmp = *acc;
+ dataBufferInit(acc, tmp.nData+nData);
+ docListUnion(tmp.pData, tmp.nData, pData, nData, acc);
+ dataBufferDestroy(&tmp);
+}
+
+/* TODO(shess) It might be interesting to explore different merge
+** strategies, here. For instance, since this is a sorted merge, we
+** could easily merge many doclists in parallel. With some
+** comprehension of the storage format, we could merge all of the
+** doclists within a leaf node directly from the leaf node's storage.
+** It may be worthwhile to merge smaller doclists before larger
+** doclists, since they can be traversed more quickly - but the
+** results may have less overlap, making them more expensive in a
+** different way.
+*/
+
+/* Scan pReader for pTerm/nTerm, and merge the term's doclist over
+** *out (any doclists with duplicate docids overwrite those in *out).
+** Internal function for loadSegmentLeaf().
+*/
+static int loadSegmentLeavesInt(fulltext_vtab *v, LeavesReader *pReader,
+ const char *pTerm, int nTerm, int isPrefix,
+ DataBuffer *out){
+ /* doclist data is accumulated into pBuffers similar to how one does
+ ** increment in binary arithmetic. If index 0 is empty, the data is
+ ** stored there. If there is data there, it is merged and the
+ ** results carried into position 1, with further merge-and-carry
+ ** until an empty position is found.
+ */
+ DataBuffer *pBuffers = NULL;
+ int nBuffers = 0, nMaxBuffers = 0, rc;
+
+ assert( nTerm>0 );
+
+ for(rc=SQLITE_OK; rc==SQLITE_OK && !leavesReaderAtEnd(pReader);
+ rc=leavesReaderStep(v, pReader)){
+ /* TODO(shess) Really want leavesReaderTermCmp(), but that name is
+ ** already taken to compare the terms of two LeavesReaders. Think
+ ** on a better name. [Meanwhile, break encapsulation rather than
+ ** use a confusing name.]
+ */
+ int c = leafReaderTermCmp(&pReader->leafReader, pTerm, nTerm, isPrefix);
+ if( c>0 ) break; /* Past any possible matches. */
+ if( c==0 ){
+ const char *pData = leavesReaderData(pReader);
+ int iBuffer, nData = leavesReaderDataBytes(pReader);
+
+ /* Find the first empty buffer. */
+ for(iBuffer=0; iBuffer<nBuffers; ++iBuffer){
+ if( 0==pBuffers[iBuffer].nData ) break;
+ }
+
+ /* Out of buffers, add an empty one. */
+ if( iBuffer==nBuffers ){
+ if( nBuffers==nMaxBuffers ){
+ DataBuffer *p;
+ nMaxBuffers += 20;
+
+ /* Manual realloc so we can handle NULL appropriately. */
+ p = sqlite3_malloc(nMaxBuffers*sizeof(*pBuffers));
+ if( p==NULL ){
+ rc = SQLITE_NOMEM;
+ break;
+ }
+
+ if( nBuffers>0 ){
+ assert(pBuffers!=NULL);
+ memcpy(p, pBuffers, nBuffers*sizeof(*pBuffers));
+ sqlite3_free(pBuffers);
+ }
+ pBuffers = p;
+ }
+ dataBufferInit(&(pBuffers[nBuffers]), 0);
+ nBuffers++;
+ }
+
+ /* At this point, must have an empty at iBuffer. */
+ assert(iBuffer<nBuffers && pBuffers[iBuffer].nData==0);
+
+ /* If empty was first buffer, no need for merge logic. */
+ if( iBuffer==0 ){
+ dataBufferReplace(&(pBuffers[0]), pData, nData);
+ }else{
+ /* pAcc is the empty buffer the merged data will end up in. */
+ DataBuffer *pAcc = &(pBuffers[iBuffer]);
+ DataBuffer *p = &(pBuffers[0]);
+
+ /* Handle position 0 specially to avoid need to prime pAcc
+ ** with pData/nData.
+ */
+ dataBufferSwap(p, pAcc);
+ docListAccumulateUnion(pAcc, pData, nData);
+
+ /* Accumulate remaining doclists into pAcc. */
+ for(++p; p<pAcc; ++p){
+ docListAccumulateUnion(pAcc, p->pData, p->nData);
+
+ /* dataBufferReset() could allow a large doclist to blow up
+ ** our memory requirements.
+ */
+ if( p->nCapacity<1024 ){
+ dataBufferReset(p);
+ }else{
+ dataBufferDestroy(p);
+ dataBufferInit(p, 0);
+ }
+ }
+ }
+ }
+ }
+
+ /* Union all the doclists together into *out. */
+ /* TODO(shess) What if *out is big? Sigh. */
+ if( rc==SQLITE_OK && nBuffers>0 ){
+ int iBuffer;
+ for(iBuffer=0; iBuffer<nBuffers; ++iBuffer){
+ if( pBuffers[iBuffer].nData>0 ){
+ if( out->nData==0 ){
+ dataBufferSwap(out, &(pBuffers[iBuffer]));
+ }else{
+ docListAccumulateUnion(out, pBuffers[iBuffer].pData,
+ pBuffers[iBuffer].nData);
+ }
+ }
+ }
+ }
+
+ while( nBuffers-- ){
+ dataBufferDestroy(&(pBuffers[nBuffers]));
+ }
+ if( pBuffers!=NULL ) sqlite3_free(pBuffers);
+
+ return rc;
+}
+
+/* Call loadSegmentLeavesInt() with pData/nData as input. */
+static int loadSegmentLeaf(fulltext_vtab *v, const char *pData, int nData,
+ const char *pTerm, int nTerm, int isPrefix,
+ DataBuffer *out){
+ LeavesReader reader;
+ int rc;
+
+ assert( nData>1 );
+ assert( *pData=='\0' );
+ rc = leavesReaderInit(v, 0, 0, 0, pData, nData, &reader);
+ if( rc!=SQLITE_OK ) return rc;
+
+ rc = loadSegmentLeavesInt(v, &reader, pTerm, nTerm, isPrefix, out);
+ leavesReaderReset(&reader);
+ leavesReaderDestroy(&reader);
+ return rc;
+}
+
+/* Call loadSegmentLeavesInt() with the leaf nodes from iStartLeaf to
+** iEndLeaf (inclusive) as input, and merge the resulting doclist into
+** out.
+*/
+static int loadSegmentLeaves(fulltext_vtab *v,
+ sqlite_int64 iStartLeaf, sqlite_int64 iEndLeaf,
+ const char *pTerm, int nTerm, int isPrefix,
+ DataBuffer *out){
+ int rc;
+ LeavesReader reader;
+
+ assert( iStartLeaf<=iEndLeaf );
+ rc = leavesReaderInit(v, 0, iStartLeaf, iEndLeaf, NULL, 0, &reader);
+ if( rc!=SQLITE_OK ) return rc;
+
+ rc = loadSegmentLeavesInt(v, &reader, pTerm, nTerm, isPrefix, out);
+ leavesReaderReset(&reader);
+ leavesReaderDestroy(&reader);
+ return rc;
+}
+
+/* Taking pData/nData as an interior node, find the sequence of child
+** nodes which could include pTerm/nTerm/isPrefix. Note that the
+** interior node terms logically come between the blocks, so there is
+** one more blockid than there are terms (that block contains terms >=
+** the last interior-node term).
+*/
+/* TODO(shess) The calling code may already know that the end child is
+** not worth calculating, because the end may be in a later sibling
+** node. Consider whether breaking symmetry is worthwhile. I suspect
+** it is not worthwhile.
+*/
+static void getChildrenContaining(const char *pData, int nData,
+ const char *pTerm, int nTerm, int isPrefix,
+ sqlite_int64 *piStartChild,
+ sqlite_int64 *piEndChild){
+ InteriorReader reader;
+
+ assert( nData>1 );
+ assert( *pData!='\0' );
+ interiorReaderInit(pData, nData, &reader);
+
+ /* Scan for the first child which could contain pTerm/nTerm. */
+ while( !interiorReaderAtEnd(&reader) ){
+ if( interiorReaderTermCmp(&reader, pTerm, nTerm, 0)>0 ) break;
+ interiorReaderStep(&reader);
+ }
+ *piStartChild = interiorReaderCurrentBlockid(&reader);
+
+ /* Keep scanning to find a term greater than our term, using prefix
+ ** comparison if indicated. If isPrefix is false, this will be the
+ ** same blockid as the starting block.
+ */
+ while( !interiorReaderAtEnd(&reader) ){
+ if( interiorReaderTermCmp(&reader, pTerm, nTerm, isPrefix)>0 ) break;
+ interiorReaderStep(&reader);
+ }
+ *piEndChild = interiorReaderCurrentBlockid(&reader);
+
+ interiorReaderDestroy(&reader);
+
+ /* Children must ascend, and if !prefix, both must be the same. */
+ assert( *piEndChild>=*piStartChild );
+ assert( isPrefix || *piStartChild==*piEndChild );
+}
+
+/* Read block at iBlockid and pass it with other params to
+** getChildrenContaining().
+*/
+static int loadAndGetChildrenContaining(
+ fulltext_vtab *v,
+ sqlite_int64 iBlockid,
+ const char *pTerm, int nTerm, int isPrefix,
+ sqlite_int64 *piStartChild, sqlite_int64 *piEndChild
+){
+ sqlite3_stmt *s = NULL;
+ int rc;
+
+ assert( iBlockid!=0 );
+ assert( pTerm!=NULL );
+ assert( nTerm!=0 ); /* TODO(shess) Why not allow this? */
+ assert( piStartChild!=NULL );
+ assert( piEndChild!=NULL );
+
+ rc = sql_get_statement(v, BLOCK_SELECT_STMT, &s);
+ if( rc!=SQLITE_OK ) return rc;
+
+ rc = sqlite3_bind_int64(s, 1, iBlockid);
+ if( rc!=SQLITE_OK ) return rc;
+
+ rc = sqlite3_step(s);
+ if( rc==SQLITE_DONE ) return SQLITE_ERROR;
+ if( rc!=SQLITE_ROW ) return rc;
+
+ getChildrenContaining(sqlite3_column_blob(s, 0), sqlite3_column_bytes(s, 0),
+ pTerm, nTerm, isPrefix, piStartChild, piEndChild);
+
+ /* 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_OK;
+}
+
+/* Traverse the tree represented by pData[nData] looking for
+** pTerm[nTerm], placing its doclist into *out. This is internal to
+** loadSegment() to make error-handling cleaner.
+*/
+static int loadSegmentInt(fulltext_vtab *v, const char *pData, int nData,
+ sqlite_int64 iLeavesEnd,
+ const char *pTerm, int nTerm, int isPrefix,
+ DataBuffer *out){
+ /* Special case where root is a leaf. */
+ if( *pData=='\0' ){
+ return loadSegmentLeaf(v, pData, nData, pTerm, nTerm, isPrefix, out);
+ }else{
+ int rc;
+ sqlite_int64 iStartChild, iEndChild;
+
+ /* Process pData as an interior node, then loop down the tree
+ ** until we find the set of leaf nodes to scan for the term.
+ */
+ getChildrenContaining(pData, nData, pTerm, nTerm, isPrefix,
+ &iStartChild, &iEndChild);
+ while( iStartChild>iLeavesEnd ){
+ sqlite_int64 iNextStart, iNextEnd;
+ rc = loadAndGetChildrenContaining(v, iStartChild, pTerm, nTerm, isPrefix,
+ &iNextStart, &iNextEnd);
+ if( rc!=SQLITE_OK ) return rc;
+
+ /* If we've branched, follow the end branch, too. */
+ if( iStartChild!=iEndChild ){
+ sqlite_int64 iDummy;
+ rc = loadAndGetChildrenContaining(v, iEndChild, pTerm, nTerm, isPrefix,
+ &iDummy, &iNextEnd);
+ if( rc!=SQLITE_OK ) return rc;
+ }
+
+ assert( iNextStart<=iNextEnd );
+ iStartChild = iNextStart;
+ iEndChild = iNextEnd;
+ }
+ assert( iStartChild<=iLeavesEnd );
+ assert( iEndChild<=iLeavesEnd );
+
+ /* Scan through the leaf segments for doclists. */
+ return loadSegmentLeaves(v, iStartChild, iEndChild,
+ pTerm, nTerm, isPrefix, out);
+ }
+}
+
+/* Call loadSegmentInt() to collect the doclist for pTerm/nTerm, then
+** merge its doclist over *out (any duplicate doclists read from the
+** segment rooted at pData will overwrite those in *out).
+*/
+/* TODO(shess) Consider changing this to determine the depth of the
+** leaves using either the first characters of interior nodes (when
+** ==1, we're one level above the leaves), or the first character of
+** the root (which will describe the height of the tree directly).
+** Either feels somewhat tricky to me.
+*/
+/* TODO(shess) The current merge is likely to be slow for large
+** doclists (though it should process from newest/smallest to
+** oldest/largest, so it may not be that bad). It might be useful to
+** modify things to allow for N-way merging. This could either be
+** within a segment, with pairwise merges across segments, or across
+** all segments at once.
+*/
+static int loadSegment(fulltext_vtab *v, const char *pData, int nData,
+ sqlite_int64 iLeavesEnd,
+ const char *pTerm, int nTerm, int isPrefix,
+ DataBuffer *out){
+ DataBuffer result;
+ int rc;
+
+ assert( nData>1 );
+
+ /* This code should never be called with buffered updates. */
+ assert( v->nPendingData<0 );
+
+ dataBufferInit(&result, 0);
+ rc = loadSegmentInt(v, pData, nData, iLeavesEnd,
+ pTerm, nTerm, isPrefix, &result);
+ if( rc==SQLITE_OK && result.nData>0 ){
+ if( out->nData==0 ){
+ DataBuffer tmp = *out;
+ *out = result;
+ result = tmp;
+ }else{
+ DataBuffer merged;
+ DLReader readers[2];
+
+ dlrInit(&readers[0], DL_DEFAULT, out->pData, out->nData);
+ dlrInit(&readers[1], DL_DEFAULT, result.pData, result.nData);
+ dataBufferInit(&merged, out->nData+result.nData);
+ docListMerge(&merged, readers, 2);
+ dataBufferDestroy(out);
+ *out = merged;
+ dlrDestroy(&readers[0]);
+ dlrDestroy(&readers[1]);
+ }
+ }
+ dataBufferDestroy(&result);
+ return rc;
+}
+
+/* Scan the database and merge together the posting lists for the term
+** into *out.
+*/
+static int termSelect(fulltext_vtab *v, int iColumn,
+ const char *pTerm, int nTerm, int isPrefix,
+ DocListType iType, DataBuffer *out){
+ DataBuffer doclist;
+ sqlite3_stmt *s;
+ int rc = sql_get_statement(v, SEGDIR_SELECT_ALL_STMT, &s);
+ if( rc!=SQLITE_OK ) return rc;
+
+ /* This code should never be called with buffered updates. */
+ assert( v->nPendingData<0 );
+
+ dataBufferInit(&doclist, 0);
+
+ /* Traverse the segments from oldest to newest so that newer doclist
+ ** elements for given docids overwrite older elements.
+ */
+ while( (rc = sqlite3_step(s))==SQLITE_ROW ){
+ const char *pData = sqlite3_column_blob(s, 2);
+ const int nData = sqlite3_column_bytes(s, 2);
+ const sqlite_int64 iLeavesEnd = sqlite3_column_int64(s, 1);
+ rc = loadSegment(v, pData, nData, iLeavesEnd, pTerm, nTerm, isPrefix,
+ &doclist);
+ if( rc!=SQLITE_OK ) goto err;
+ }
+ if( rc==SQLITE_DONE ){
+ if( doclist.nData!=0 ){
+ /* TODO(shess) The old term_select_all() code applied the column
+ ** restrict as we merged segments, leading to smaller buffers.
+ ** This is probably worthwhile to bring back, once the new storage
+ ** system is checked in.
+ */
+ if( iColumn==v->nColumn) iColumn = -1;
+ docListTrim(DL_DEFAULT, doclist.pData, doclist.nData,
+ iColumn, iType, out);
+ }
+ rc = SQLITE_OK;
+ }
+
+ err:
+ dataBufferDestroy(&doclist);
+ return rc;
+}
+
+/****************************************************************/
+/* Used to hold hashtable data for sorting. */
+typedef struct TermData {
+ const char *pTerm;
+ int nTerm;
+ DLCollector *pCollector;
+} TermData;
+
+/* Orders TermData elements in strcmp fashion ( <0 for less-than, 0
+** for equal, >0 for greater-than).
+*/
+static int termDataCmp(const void *av, const void *bv){
+ const TermData *a = (const TermData *)av;
+ const TermData *b = (const TermData *)bv;
+ int n = a->nTerm<b->nTerm ? a->nTerm : b->nTerm;
+ int c = memcmp(a->pTerm, b->pTerm, n);
+ if( c!=0 ) return c;
+ return a->nTerm-b->nTerm;
+}
+
+/* Order pTerms data by term, then write a new level 0 segment using
+** LeafWriter.
+*/
+static int writeZeroSegment(fulltext_vtab *v, fts3Hash *pTerms){
+ fts3HashElem *e;
+ int idx, rc, i, n;
+ TermData *pData;
+ LeafWriter writer;
+ DataBuffer dl;
+
+ /* Determine the next index at level 0, merging as necessary. */
+ rc = segdirNextIndex(v, 0, &idx);
+ if( rc!=SQLITE_OK ) return rc;
+
+ n = fts3HashCount(pTerms);
+ pData = sqlite3_malloc(n*sizeof(TermData));
+
+ for(i = 0, e = fts3HashFirst(pTerms); e; i++, e = fts3HashNext(e)){
+ assert( i<n );
+ pData[i].pTerm = fts3HashKey(e);
+ pData[i].nTerm = fts3HashKeysize(e);
+ pData[i].pCollector = fts3HashData(e);
+ }
+ assert( i==n );
+
+ /* TODO(shess) Should we allow user-defined collation sequences,
+ ** here? I think we only need that once we support prefix searches.
+ */
+ if( n>1 ) qsort(pData, n, sizeof(*pData), termDataCmp);
+
+ /* TODO(shess) Refactor so that we can write directly to the segment
+ ** DataBuffer, as happens for segment merges.
+ */
+ leafWriterInit(0, idx, &writer);
+ dataBufferInit(&dl, 0);
+ for(i=0; i<n; i++){
+ dataBufferReset(&dl);
+ dlcAddDoclist(pData[i].pCollector, &dl);
+ rc = leafWriterStep(v, &writer,
+ pData[i].pTerm, pData[i].nTerm, dl.pData, dl.nData);
+ if( rc!=SQLITE_OK ) goto err;
+ }
+ rc = leafWriterFinalize(v, &writer);
+
+ err:
+ dataBufferDestroy(&dl);
+ sqlite3_free(pData);
+ leafWriterDestroy(&writer);
+ return rc;
+}
+
+/* If pendingTerms has data, free it. */
+static int clearPendingTerms(fulltext_vtab *v){
+ if( v->nPendingData>=0 ){
+ fts3HashElem *e;
+ for(e=fts3HashFirst(&v->pendingTerms); e; e=fts3HashNext(e)){
+ dlcDelete(fts3HashData(e));
+ }
+ fts3HashClear(&v->pendingTerms);
+ v->nPendingData = -1;
+ }
+ return SQLITE_OK;
+}
+
+/* If pendingTerms has data, flush it to a level-zero segment, and
+** free it.
+*/
+static int flushPendingTerms(fulltext_vtab *v){
+ if( v->nPendingData>=0 ){
+ int rc = writeZeroSegment(v, &v->pendingTerms);
+ if( rc==SQLITE_OK ) clearPendingTerms(v);
+ return rc;
+ }
+ return SQLITE_OK;
+}
+
+/* If pendingTerms is "too big", or docid is out of order, flush it.
+** Regardless, be certain that pendingTerms is initialized for use.
+*/
+static int initPendingTerms(fulltext_vtab *v, sqlite_int64 iDocid){
+ /* TODO(shess) Explore whether partially flushing the buffer on
+ ** forced-flush would provide better performance. I suspect that if
+ ** we ordered the doclists by size and flushed the largest until the
+ ** buffer was half empty, that would let the less frequent terms
+ ** generate longer doclists.
+ */
+ if( iDocid<=v->iPrevDocid || v->nPendingData>kPendingThreshold ){
+ int rc = flushPendingTerms(v);
+ if( rc!=SQLITE_OK ) return rc;
+ }
+ if( v->nPendingData<0 ){
+ fts3HashInit(&v->pendingTerms, FTS3_HASH_STRING, 1);
+ v->nPendingData = 0;
+ }
+ v->iPrevDocid = iDocid;
+ return SQLITE_OK;
+}
+
+/* This function implements the xUpdate callback; it is the top-level entry
+ * point for inserting, deleting or updating a row in a full-text table. */
+static int fulltextUpdate(sqlite3_vtab *pVtab, int nArg, sqlite3_value **ppArg,
+ sqlite_int64 *pRowid){
+ fulltext_vtab *v = (fulltext_vtab *) pVtab;
+ int rc;
+
+ FTSTRACE(("FTS3 Update %p\n", pVtab));
+
+ if( nArg<2 ){
+ rc = index_delete(v, sqlite3_value_int64(ppArg[0]));
+ if( rc==SQLITE_OK ){
+ /* If we just deleted the last row in the table, clear out the
+ ** index data.
+ */
+ rc = content_exists(v);
+ if( rc==SQLITE_ROW ){
+ rc = SQLITE_OK;
+ }else if( rc==SQLITE_DONE ){
+ /* Clear the pending terms so we don't flush a useless level-0
+ ** segment when the transaction closes.
+ */
+ rc = clearPendingTerms(v);
+ if( rc==SQLITE_OK ){
+ rc = segdir_delete_all(v);
+ }
+ }
+ }
+ } else if( sqlite3_value_type(ppArg[0]) != SQLITE_NULL ){
+ /* An update:
+ * ppArg[0] = old rowid
+ * ppArg[1] = new rowid
+ * ppArg[2..2+v->nColumn-1] = values
+ * ppArg[2+v->nColumn] = value for magic column (we ignore this)
+ * ppArg[2+v->nColumn+1] = value for docid
+ */
+ sqlite_int64 rowid = sqlite3_value_int64(ppArg[0]);
+ if( sqlite3_value_type(ppArg[1]) != SQLITE_INTEGER ||
+ sqlite3_value_int64(ppArg[1]) != rowid ){
+ rc = SQLITE_ERROR; /* we don't allow changing the rowid */
+ }else if( sqlite3_value_type(ppArg[2+v->nColumn+1]) != SQLITE_INTEGER ||
+ sqlite3_value_int64(ppArg[2+v->nColumn+1]) != rowid ){
+ rc = SQLITE_ERROR; /* we don't allow changing the docid */
+ }else{
+ assert( nArg==2+v->nColumn+2);
+ rc = index_update(v, rowid, &ppArg[2]);
+ }
+ } else {
+ /* An insert:
+ * ppArg[1] = requested rowid
+ * ppArg[2..2+v->nColumn-1] = values
+ * ppArg[2+v->nColumn] = value for magic column (we ignore this)
+ * ppArg[2+v->nColumn+1] = value for docid
+ */
+ sqlite3_value *pRequestDocid = ppArg[2+v->nColumn+1];
+ assert( nArg==2+v->nColumn+2);
+ if( SQLITE_NULL != sqlite3_value_type(pRequestDocid) &&
+ SQLITE_NULL != sqlite3_value_type(ppArg[1]) ){
+ /* TODO(shess) Consider allowing this to work if the values are
+ ** identical. I'm inclined to discourage that usage, though,
+ ** given that both rowid and docid are special columns. Better
+ ** would be to define one or the other as the default winner,
+ ** but should it be fts3-centric (docid) or SQLite-centric
+ ** (rowid)?
+ */
+ rc = SQLITE_ERROR;
+ }else{
+ if( SQLITE_NULL == sqlite3_value_type(pRequestDocid) ){
+ pRequestDocid = ppArg[1];
+ }
+ rc = index_insert(v, pRequestDocid, &ppArg[2], pRowid);
+ }
+ }
+
+ return rc;
+}
+
+static int fulltextSync(sqlite3_vtab *pVtab){
+ FTSTRACE(("FTS3 xSync()\n"));
+ return flushPendingTerms((fulltext_vtab *)pVtab);
+}
+
+static int fulltextBegin(sqlite3_vtab *pVtab){
+ fulltext_vtab *v = (fulltext_vtab *) pVtab;
+ FTSTRACE(("FTS3 xBegin()\n"));
+
+ /* Any buffered updates should have been cleared by the previous
+ ** transaction.
+ */
+ assert( v->nPendingData<0 );
+ return clearPendingTerms(v);
+}
+
+static int fulltextCommit(sqlite3_vtab *pVtab){
+ fulltext_vtab *v = (fulltext_vtab *) pVtab;
+ FTSTRACE(("FTS3 xCommit()\n"));
+
+ /* Buffered updates should have been cleared by fulltextSync(). */
+ assert( v->nPendingData<0 );
+ return clearPendingTerms(v);
+}
+
+static int fulltextRollback(sqlite3_vtab *pVtab){
+ FTSTRACE(("FTS3 xRollback()\n"));
+ return clearPendingTerms((fulltext_vtab *)pVtab);
+}
+
+/*
+** Implementation of the snippet() function for FTS3
+*/
+static void snippetFunc(
+ sqlite3_context *pContext,
+ int argc,
+ sqlite3_value **argv
+){
+ fulltext_cursor *pCursor;
+ if( argc<1 ) return;
+ if( sqlite3_value_type(argv[0])!=SQLITE_BLOB ||
+ sqlite3_value_bytes(argv[0])!=sizeof(pCursor) ){
+ sqlite3_result_error(pContext, "illegal first argument to html_snippet",-1);
+ }else{
+ const char *zStart = "<b>";
+ const char *zEnd = "</b>";
+ const char *zEllipsis = "<b>...</b>";
+ memcpy(&pCursor, sqlite3_value_blob(argv[0]), sizeof(pCursor));
+ if( argc>=2 ){
+ zStart = (const char*)sqlite3_value_text(argv[1]);
+ if( argc>=3 ){
+ zEnd = (const char*)sqlite3_value_text(argv[2]);
+ if( argc>=4 ){
+ zEllipsis = (const char*)sqlite3_value_text(argv[3]);
+ }
+ }
+ }
+ snippetAllOffsets(pCursor);
+ snippetText(pCursor, zStart, zEnd, zEllipsis);
+ sqlite3_result_text(pContext, pCursor->snippet.zSnippet,
+ pCursor->snippet.nSnippet, SQLITE_STATIC);
+ }
+}
+
+/*
+** Implementation of the offsets() function for FTS3
+*/
+static void snippetOffsetsFunc(
+ sqlite3_context *pContext,
+ int argc,
+ sqlite3_value **argv
+){
+ fulltext_cursor *pCursor;
+ if( argc<1 ) return;
+ if( sqlite3_value_type(argv[0])!=SQLITE_BLOB ||
+ sqlite3_value_bytes(argv[0])!=sizeof(pCursor) ){
+ sqlite3_result_error(pContext, "illegal first argument to offsets",-1);
+ }else{
+ memcpy(&pCursor, sqlite3_value_blob(argv[0]), sizeof(pCursor));
+ snippetAllOffsets(pCursor);
+ snippetOffsetText(&pCursor->snippet);
+ sqlite3_result_text(pContext,
+ pCursor->snippet.zOffset, pCursor->snippet.nOffset,
+ SQLITE_STATIC);
+ }
+}
+
+/* OptLeavesReader is nearly identical to LeavesReader, except that
+** where LeavesReader is geared towards the merging of complete
+** segment levels (with exactly MERGE_COUNT segments), OptLeavesReader
+** is geared towards implementation of the optimize() function, and
+** can merge all segments simultaneously. This version may be
+** somewhat less efficient than LeavesReader because it merges into an
+** accumulator rather than doing an N-way merge, but since segment
+** size grows exponentially (so segment count logrithmically) this is
+** probably not an immediate problem.
+*/
+/* TODO(shess): Prove that assertion, or extend the merge code to
+** merge tree fashion (like the prefix-searching code does).
+*/
+/* TODO(shess): OptLeavesReader and LeavesReader could probably be
+** merged with little or no loss of performance for LeavesReader. The
+** merged code would need to handle >MERGE_COUNT segments, and would
+** also need to be able to optionally optimize away deletes.
+*/
+typedef struct OptLeavesReader {
+ /* Segment number, to order readers by age. */
+ int segment;
+ LeavesReader reader;
+} OptLeavesReader;
+
+static int optLeavesReaderAtEnd(OptLeavesReader *pReader){
+ return leavesReaderAtEnd(&pReader->reader);
+}
+static int optLeavesReaderTermBytes(OptLeavesReader *pReader){
+ return leavesReaderTermBytes(&pReader->reader);
+}
+static const char *optLeavesReaderData(OptLeavesReader *pReader){
+ return leavesReaderData(&pReader->reader);
+}
+static int optLeavesReaderDataBytes(OptLeavesReader *pReader){
+ return leavesReaderDataBytes(&pReader->reader);
+}
+static const char *optLeavesReaderTerm(OptLeavesReader *pReader){
+ return leavesReaderTerm(&pReader->reader);
+}
+static int optLeavesReaderStep(fulltext_vtab *v, OptLeavesReader *pReader){
+ return leavesReaderStep(v, &pReader->reader);
+}
+static int optLeavesReaderTermCmp(OptLeavesReader *lr1, OptLeavesReader *lr2){
+ return leavesReaderTermCmp(&lr1->reader, &lr2->reader);
+}
+/* Order by term ascending, segment ascending (oldest to newest), with
+** exhausted readers to the end.
+*/
+static int optLeavesReaderCmp(OptLeavesReader *lr1, OptLeavesReader *lr2){
+ int c = optLeavesReaderTermCmp(lr1, lr2);
+ if( c!=0 ) return c;
+ return lr1->segment-lr2->segment;
+}
+/* Bubble pLr[0] to appropriate place in pLr[1..nLr-1]. Assumes that
+** pLr[1..nLr-1] is already sorted.
+*/
+static void optLeavesReaderReorder(OptLeavesReader *pLr, int nLr){
+ while( nLr>1 && optLeavesReaderCmp(pLr, pLr+1)>0 ){
+ OptLeavesReader tmp = pLr[0];
+ pLr[0] = pLr[1];
+ pLr[1] = tmp;
+ nLr--;
+ pLr++;
+ }
+}
+
+/* optimize() helper function. Put the readers in order and iterate
+** through them, merging doclists for matching terms into pWriter.
+** Returns SQLITE_OK on success, or the SQLite error code which
+** prevented success.
+*/
+static int optimizeInternal(fulltext_vtab *v,
+ OptLeavesReader *readers, int nReaders,
+ LeafWriter *pWriter){
+ int i, rc = SQLITE_OK;
+ DataBuffer doclist, merged, tmp;
+
+ /* Order the readers. */
+ i = nReaders;
+ while( i-- > 0 ){
+ optLeavesReaderReorder(&readers[i], nReaders-i);
+ }
+
+ dataBufferInit(&doclist, LEAF_MAX);
+ dataBufferInit(&merged, LEAF_MAX);
+
+ /* Exhausted readers bubble to the end, so when the first reader is
+ ** at eof, all are at eof.
+ */
+ while( !optLeavesReaderAtEnd(&readers[0]) ){
+
+ /* Figure out how many readers share the next term. */
+ for(i=1; i<nReaders && !optLeavesReaderAtEnd(&readers[i]); i++){
+ if( 0!=optLeavesReaderTermCmp(&readers[0], &readers[i]) ) break;
+ }
+
+ /* Special-case for no merge. */
+ if( i==1 ){
+ /* Trim deletions from the doclist. */
+ dataBufferReset(&merged);
+ docListTrim(DL_DEFAULT,
+ optLeavesReaderData(&readers[0]),
+ optLeavesReaderDataBytes(&readers[0]),
+ -1, DL_DEFAULT, &merged);
+ }else{
+ DLReader dlReaders[MERGE_COUNT];
+ int iReader, nReaders;
+
+ /* Prime the pipeline with the first reader's doclist. After
+ ** one pass index 0 will reference the accumulated doclist.
+ */
+ dlrInit(&dlReaders[0], DL_DEFAULT,
+ optLeavesReaderData(&readers[0]),
+ optLeavesReaderDataBytes(&readers[0]));
+ iReader = 1;
+
+ assert( iReader<i ); /* Must execute the loop at least once. */
+ while( iReader<i ){
+ /* Merge 16 inputs per pass. */
+ for( nReaders=1; iReader<i && nReaders<MERGE_COUNT;
+ iReader++, nReaders++ ){
+ dlrInit(&dlReaders[nReaders], DL_DEFAULT,
+ optLeavesReaderData(&readers[iReader]),
+ optLeavesReaderDataBytes(&readers[iReader]));
+ }
+
+ /* Merge doclists and swap result into accumulator. */
+ dataBufferReset(&merged);
+ docListMerge(&merged, dlReaders, nReaders);
+ tmp = merged;
+ merged = doclist;
+ doclist = tmp;
+
+ while( nReaders-- > 0 ){
+ dlrDestroy(&dlReaders[nReaders]);
+ }
+
+ /* Accumulated doclist to reader 0 for next pass. */
+ dlrInit(&dlReaders[0], DL_DEFAULT, doclist.pData, doclist.nData);
+ }
+
+ /* Destroy reader that was left in the pipeline. */
+ dlrDestroy(&dlReaders[0]);
+
+ /* Trim deletions from the doclist. */
+ dataBufferReset(&merged);
+ docListTrim(DL_DEFAULT, doclist.pData, doclist.nData,
+ -1, DL_DEFAULT, &merged);
+ }
+
+ /* Only pass doclists with hits (skip if all hits deleted). */
+ if( merged.nData>0 ){
+ rc = leafWriterStep(v, pWriter,
+ optLeavesReaderTerm(&readers[0]),
+ optLeavesReaderTermBytes(&readers[0]),
+ merged.pData, merged.nData);
+ if( rc!=SQLITE_OK ) goto err;
+ }
+
+ /* Step merged readers to next term and reorder. */
+ while( i-- > 0 ){
+ rc = optLeavesReaderStep(v, &readers[i]);
+ if( rc!=SQLITE_OK ) goto err;
+
+ optLeavesReaderReorder(&readers[i], nReaders-i);
+ }
+ }
+
+ err:
+ dataBufferDestroy(&doclist);
+ dataBufferDestroy(&merged);
+ return rc;
+}
+
+/* Implement optimize() function for FTS3. optimize(t) merges all
+** segments in the fts index into a single segment. 't' is the magic
+** table-named column.
+*/
+static void optimizeFunc(sqlite3_context *pContext,
+ int argc, sqlite3_value **argv){
+ fulltext_cursor *pCursor;
+ if( argc>1 ){
+ sqlite3_result_error(pContext, "excess arguments to optimize()",-1);
+ }else if( sqlite3_value_type(argv[0])!=SQLITE_BLOB ||
+ sqlite3_value_bytes(argv[0])!=sizeof(pCursor) ){
+ sqlite3_result_error(pContext, "illegal first argument to optimize",-1);
+ }else{
+ fulltext_vtab *v;
+ int i, rc, iMaxLevel;
+ OptLeavesReader *readers;
+ int nReaders;
+ LeafWriter writer;
+ sqlite3_stmt *s;
+
+ memcpy(&pCursor, sqlite3_value_blob(argv[0]), sizeof(pCursor));
+ v = cursor_vtab(pCursor);
+
+ /* Flush any buffered updates before optimizing. */
+ rc = flushPendingTerms(v);
+ if( rc!=SQLITE_OK ) goto err;
+
+ rc = segdir_count(v, &nReaders, &iMaxLevel);
+ if( rc!=SQLITE_OK ) goto err;
+ if( nReaders==0 || nReaders==1 ){
+ sqlite3_result_text(pContext, "Index already optimal", -1,
+ SQLITE_STATIC);
+ return;
+ }
+
+ rc = sql_get_statement(v, SEGDIR_SELECT_ALL_STMT, &s);
+ if( rc!=SQLITE_OK ) goto err;
+
+ readers = sqlite3_malloc(nReaders*sizeof(readers[0]));
+ if( readers==NULL ) goto err;
+
+ /* Note that there will already be a segment at this position
+ ** until we call segdir_delete() on iMaxLevel.
+ */
+ leafWriterInit(iMaxLevel, 0, &writer);
+
+ i = 0;
+ while( (rc = sqlite3_step(s))==SQLITE_ROW ){
+ sqlite_int64 iStart = sqlite3_column_int64(s, 0);
+ sqlite_int64 iEnd = sqlite3_column_int64(s, 1);
+ const char *pRootData = sqlite3_column_blob(s, 2);
+ int nRootData = sqlite3_column_bytes(s, 2);
+
+ assert( i<nReaders );
+ rc = leavesReaderInit(v, -1, iStart, iEnd, pRootData, nRootData,
+ &readers[i].reader);
+ if( rc!=SQLITE_OK ) break;
+
+ readers[i].segment = i;
+ i++;
+ }
+
+ /* If we managed to succesfully read them all, optimize them. */
+ if( rc==SQLITE_DONE ){
+ assert( i==nReaders );
+ rc = optimizeInternal(v, readers, nReaders, &writer);
+ }
+
+ while( i-- > 0 ){
+ leavesReaderDestroy(&readers[i].reader);
+ }
+ sqlite3_free(readers);
+
+ /* If we've successfully gotten to here, delete the old segments
+ ** and flush the interior structure of the new segment.
+ */
+ if( rc==SQLITE_OK ){
+ for( i=0; i<=iMaxLevel; i++ ){
+ rc = segdir_delete(v, i);
+ if( rc!=SQLITE_OK ) break;
+ }
+
+ if( rc==SQLITE_OK ) rc = leafWriterFinalize(v, &writer);
+ }
+
+ leafWriterDestroy(&writer);
+
+ if( rc!=SQLITE_OK ) goto err;
+
+ sqlite3_result_text(pContext, "Index optimized", -1, SQLITE_STATIC);
+ return;
+
+ /* TODO(shess): Error-handling needs to be improved along the
+ ** lines of the dump_ functions.
+ */
+ err:
+ {
+ char buf[512];
+ sqlite3_snprintf(sizeof(buf), buf, "Error in optimize: %s",
+ sqlite3_errmsg(sqlite3_context_db_handle(pContext)));
+ sqlite3_result_error(pContext, buf, -1);
+ }
+ }
+}
+
+#ifdef SQLITE_TEST
+/* Generate an error of the form "<prefix>: <msg>". If msg is NULL,
+** pull the error from the context's db handle.
+*/
+static void generateError(sqlite3_context *pContext,
+ const char *prefix, const char *msg){
+ char buf[512];
+ if( msg==NULL ) msg = sqlite3_errmsg(sqlite3_context_db_handle(pContext));
+ sqlite3_snprintf(sizeof(buf), buf, "%s: %s", prefix, msg);
+ sqlite3_result_error(pContext, buf, -1);
+}
+
+/* Helper function to collect the set of terms in the segment into
+** pTerms. The segment is defined by the leaf nodes between
+** iStartBlockid and iEndBlockid, inclusive, or by the contents of
+** pRootData if iStartBlockid is 0 (in which case the entire segment
+** fit in a leaf).
+*/
+static int collectSegmentTerms(fulltext_vtab *v, sqlite3_stmt *s,
+ fts3Hash *pTerms){
+ const sqlite_int64 iStartBlockid = sqlite3_column_int64(s, 0);
+ const sqlite_int64 iEndBlockid = sqlite3_column_int64(s, 1);
+ const char *pRootData = sqlite3_column_blob(s, 2);
+ const int nRootData = sqlite3_column_bytes(s, 2);
+ LeavesReader reader;
+ int rc = leavesReaderInit(v, 0, iStartBlockid, iEndBlockid,
+ pRootData, nRootData, &reader);
+ if( rc!=SQLITE_OK ) return rc;
+
+ while( rc==SQLITE_OK && !leavesReaderAtEnd(&reader) ){
+ const char *pTerm = leavesReaderTerm(&reader);
+ const int nTerm = leavesReaderTermBytes(&reader);
+ void *oldValue = sqlite3Fts3HashFind(pTerms, pTerm, nTerm);
+ void *newValue = (void *)((char *)oldValue+1);
+
+ /* From the comment before sqlite3Fts3HashInsert in fts3_hash.c,
+ ** the data value passed is returned in case of malloc failure.
+ */
+ if( newValue==sqlite3Fts3HashInsert(pTerms, pTerm, nTerm, newValue) ){
+ rc = SQLITE_NOMEM;
+ }else{
+ rc = leavesReaderStep(v, &reader);
+ }
+ }
+
+ leavesReaderDestroy(&reader);
+ return rc;
+}
+
+/* Helper function to build the result string for dump_terms(). */
+static int generateTermsResult(sqlite3_context *pContext, fts3Hash *pTerms){
+ int iTerm, nTerms, nResultBytes, iByte;
+ char *result;
+ TermData *pData;
+ fts3HashElem *e;
+
+ /* Iterate pTerms to generate an array of terms in pData for
+ ** sorting.
+ */
+ nTerms = fts3HashCount(pTerms);
+ assert( nTerms>0 );
+ pData = sqlite3_malloc(nTerms*sizeof(TermData));
+ if( pData==NULL ) return SQLITE_NOMEM;
+
+ nResultBytes = 0;
+ for(iTerm = 0, e = fts3HashFirst(pTerms); e; iTerm++, e = fts3HashNext(e)){
+ nResultBytes += fts3HashKeysize(e)+1; /* Term plus trailing space */
+ assert( iTerm<nTerms );
+ pData[iTerm].pTerm = fts3HashKey(e);
+ pData[iTerm].nTerm = fts3HashKeysize(e);
+ pData[iTerm].pCollector = fts3HashData(e); /* unused */
+ }
+ assert( iTerm==nTerms );
+
+ assert( nResultBytes>0 ); /* nTerms>0, nResultsBytes must be, too. */
+ result = sqlite3_malloc(nResultBytes);
+ if( result==NULL ){
+ sqlite3_free(pData);
+ return SQLITE_NOMEM;
+ }
+
+ if( nTerms>1 ) qsort(pData, nTerms, sizeof(*pData), termDataCmp);
+
+ /* Read the terms in order to build the result. */
+ iByte = 0;
+ for(iTerm=0; iTerm<nTerms; ++iTerm){
+ memcpy(result+iByte, pData[iTerm].pTerm, pData[iTerm].nTerm);
+ iByte += pData[iTerm].nTerm;
+ result[iByte++] = ' ';
+ }
+ assert( iByte==nResultBytes );
+ assert( result[nResultBytes-1]==' ' );
+ result[nResultBytes-1] = '\0';
+
+ /* Passes away ownership of result. */
+ sqlite3_result_text(pContext, result, nResultBytes-1, sqlite3_free);
+ sqlite3_free(pData);
+ return SQLITE_OK;
+}
+
+/* Implements dump_terms() for use in inspecting the fts3 index from
+** tests. TEXT result containing the ordered list of terms joined by
+** spaces. dump_terms(t, level, idx) dumps the terms for the segment
+** specified by level, idx (in %_segdir), while dump_terms(t) dumps
+** all terms in the index. In both cases t is the fts table's magic
+** table-named column.
+*/
+static void dumpTermsFunc(
+ sqlite3_context *pContext,
+ int argc, sqlite3_value **argv
+){
+ fulltext_cursor *pCursor;
+ if( argc!=3 && argc!=1 ){
+ generateError(pContext, "dump_terms", "incorrect arguments");
+ }else if( sqlite3_value_type(argv[0])!=SQLITE_BLOB ||
+ sqlite3_value_bytes(argv[0])!=sizeof(pCursor) ){
+ generateError(pContext, "dump_terms", "illegal first argument");
+ }else{
+ fulltext_vtab *v;
+ fts3Hash terms;
+ sqlite3_stmt *s = NULL;
+ int rc;
+
+ memcpy(&pCursor, sqlite3_value_blob(argv[0]), sizeof(pCursor));
+ v = cursor_vtab(pCursor);
+
+ /* If passed only the cursor column, get all segments. Otherwise
+ ** get the segment described by the following two arguments.
+ */
+ if( argc==1 ){
+ rc = sql_get_statement(v, SEGDIR_SELECT_ALL_STMT, &s);
+ }else{
+ rc = sql_get_statement(v, SEGDIR_SELECT_SEGMENT_STMT, &s);
+ if( rc==SQLITE_OK ){
+ rc = sqlite3_bind_int(s, 1, sqlite3_value_int(argv[1]));
+ if( rc==SQLITE_OK ){
+ rc = sqlite3_bind_int(s, 2, sqlite3_value_int(argv[2]));
+ }
+ }
+ }
+
+ if( rc!=SQLITE_OK ){
+ generateError(pContext, "dump_terms", NULL);
+ return;
+ }
+
+ /* Collect the terms for each segment. */
+ sqlite3Fts3HashInit(&terms, FTS3_HASH_STRING, 1);
+ while( (rc = sqlite3_step(s))==SQLITE_ROW ){
+ rc = collectSegmentTerms(v, s, &terms);
+ if( rc!=SQLITE_OK ) break;
+ }
+
+ if( rc!=SQLITE_DONE ){
+ sqlite3_reset(s);
+ generateError(pContext, "dump_terms", NULL);
+ }else{
+ const int nTerms = fts3HashCount(&terms);
+ if( nTerms>0 ){
+ rc = generateTermsResult(pContext, &terms);
+ if( rc==SQLITE_NOMEM ){
+ generateError(pContext, "dump_terms", "out of memory");
+ }else{
+ assert( rc==SQLITE_OK );
+ }
+ }else if( argc==3 ){
+ /* The specific segment asked for could not be found. */
+ generateError(pContext, "dump_terms", "segment not found");
+ }else{
+ /* No segments found. */
+ /* TODO(shess): It should be impossible to reach this. This
+ ** case can only happen for an empty table, in which case
+ ** SQLite has no rows to call this function on.
+ */
+ sqlite3_result_null(pContext);
+ }
+ }
+ sqlite3Fts3HashClear(&terms);
+ }
+}
+
+/* Expand the DL_DEFAULT doclist in pData into a text result in
+** pContext.
+*/
+static void createDoclistResult(sqlite3_context *pContext,
+ const char *pData, int nData){
+ DataBuffer dump;
+ DLReader dlReader;
+
+ assert( pData!=NULL && nData>0 );
+
+ dataBufferInit(&dump, 0);
+ dlrInit(&dlReader, DL_DEFAULT, pData, nData);
+ for( ; !dlrAtEnd(&dlReader); dlrStep(&dlReader) ){
+ char buf[256];
+ PLReader plReader;
+
+ plrInit(&plReader, &dlReader);
+ if( DL_DEFAULT==DL_DOCIDS || plrAtEnd(&plReader) ){
+ sqlite3_snprintf(sizeof(buf), buf, "[%lld] ", dlrDocid(&dlReader));
+ dataBufferAppend(&dump, buf, strlen(buf));
+ }else{
+ int iColumn = plrColumn(&plReader);
+
+ sqlite3_snprintf(sizeof(buf), buf, "[%lld %d[",
+ dlrDocid(&dlReader), iColumn);
+ dataBufferAppend(&dump, buf, strlen(buf));
+
+ for( ; !plrAtEnd(&plReader); plrStep(&plReader) ){
+ if( plrColumn(&plReader)!=iColumn ){
+ iColumn = plrColumn(&plReader);
+ sqlite3_snprintf(sizeof(buf), buf, "] %d[", iColumn);
+ assert( dump.nData>0 );
+ dump.nData--; /* Overwrite trailing space. */
+ assert( dump.pData[dump.nData]==' ');
+ dataBufferAppend(&dump, buf, strlen(buf));
+ }
+ if( DL_DEFAULT==DL_POSITIONS_OFFSETS ){
+ sqlite3_snprintf(sizeof(buf), buf, "%d,%d,%d ",
+ plrPosition(&plReader),
+ plrStartOffset(&plReader), plrEndOffset(&plReader));
+ }else if( DL_DEFAULT==DL_POSITIONS ){
+ sqlite3_snprintf(sizeof(buf), buf, "%d ", plrPosition(&plReader));
+ }else{
+ assert( NULL=="Unhandled DL_DEFAULT value");
+ }
+ dataBufferAppend(&dump, buf, strlen(buf));
+ }
+ plrDestroy(&plReader);
+
+ assert( dump.nData>0 );
+ dump.nData--; /* Overwrite trailing space. */
+ assert( dump.pData[dump.nData]==' ');
+ dataBufferAppend(&dump, "]] ", 3);
+ }
+ }
+ dlrDestroy(&dlReader);
+
+ assert( dump.nData>0 );
+ dump.nData--; /* Overwrite trailing space. */
+ assert( dump.pData[dump.nData]==' ');
+ dump.pData[dump.nData] = '\0';
+ assert( dump.nData>0 );
+
+ /* Passes ownership of dump's buffer to pContext. */
+ sqlite3_result_text(pContext, dump.pData, dump.nData, sqlite3_free);
+ dump.pData = NULL;
+ dump.nData = dump.nCapacity = 0;
+}
+
+/* Implements dump_doclist() for use in inspecting the fts3 index from
+** tests. TEXT result containing a string representation of the
+** doclist for the indicated term. dump_doclist(t, term, level, idx)
+** dumps the doclist for term from the segment specified by level, idx
+** (in %_segdir), while dump_doclist(t, term) dumps the logical
+** doclist for the term across all segments. The per-segment doclist
+** can contain deletions, while the full-index doclist will not
+** (deletions are omitted).
+**
+** Result formats differ with the setting of DL_DEFAULTS. Examples:
+**
+** DL_DOCIDS: [1] [3] [7]
+** DL_POSITIONS: [1 0[0 4] 1[17]] [3 1[5]]
+** DL_POSITIONS_OFFSETS: [1 0[0,0,3 4,23,26] 1[17,102,105]] [3 1[5,20,23]]
+**
+** In each case the number after the outer '[' is the docid. In the
+** latter two cases, the number before the inner '[' is the column
+** associated with the values within. For DL_POSITIONS the numbers
+** within are the positions, for DL_POSITIONS_OFFSETS they are the
+** position, the start offset, and the end offset.
+*/
+static void dumpDoclistFunc(
+ sqlite3_context *pContext,
+ int argc, sqlite3_value **argv
+){
+ fulltext_cursor *pCursor;
+ if( argc!=2 && argc!=4 ){
+ generateError(pContext, "dump_doclist", "incorrect arguments");
+ }else if( sqlite3_value_type(argv[0])!=SQLITE_BLOB ||
+ sqlite3_value_bytes(argv[0])!=sizeof(pCursor) ){
+ generateError(pContext, "dump_doclist", "illegal first argument");
+ }else if( sqlite3_value_text(argv[1])==NULL ||
+ sqlite3_value_text(argv[1])[0]=='\0' ){
+ generateError(pContext, "dump_doclist", "empty second argument");
+ }else{
+ const char *pTerm = (const char *)sqlite3_value_text(argv[1]);
+ const int nTerm = strlen(pTerm);
+ fulltext_vtab *v;
+ int rc;
+ DataBuffer doclist;
+
+ memcpy(&pCursor, sqlite3_value_blob(argv[0]), sizeof(pCursor));
+ v = cursor_vtab(pCursor);
+
+ dataBufferInit(&doclist, 0);
+
+ /* termSelect() yields the same logical doclist that queries are
+ ** run against.
+ */
+ if( argc==2 ){
+ rc = termSelect(v, v->nColumn, pTerm, nTerm, 0, DL_DEFAULT, &doclist);
+ }else{
+ sqlite3_stmt *s = NULL;
+
+ /* Get our specific segment's information. */
+ rc = sql_get_statement(v, SEGDIR_SELECT_SEGMENT_STMT, &s);
+ if( rc==SQLITE_OK ){
+ rc = sqlite3_bind_int(s, 1, sqlite3_value_int(argv[2]));
+ if( rc==SQLITE_OK ){
+ rc = sqlite3_bind_int(s, 2, sqlite3_value_int(argv[3]));
+ }
+ }
+
+ if( rc==SQLITE_OK ){
+ rc = sqlite3_step(s);
+
+ if( rc==SQLITE_DONE ){
+ dataBufferDestroy(&doclist);
+ generateError(pContext, "dump_doclist", "segment not found");
+ return;
+ }
+
+ /* Found a segment, load it into doclist. */
+ if( rc==SQLITE_ROW ){
+ const sqlite_int64 iLeavesEnd = sqlite3_column_int64(s, 1);
+ const char *pData = sqlite3_column_blob(s, 2);
+ const int nData = sqlite3_column_bytes(s, 2);
+
+ /* loadSegment() is used by termSelect() to load each
+ ** segment's data.
+ */
+ rc = loadSegment(v, pData, nData, iLeavesEnd, pTerm, nTerm, 0,
+ &doclist);
+ if( rc==SQLITE_OK ){
+ rc = sqlite3_step(s);
+
+ /* Should not have more than one matching segment. */
+ if( rc!=SQLITE_DONE ){
+ sqlite3_reset(s);
+ dataBufferDestroy(&doclist);
+ generateError(pContext, "dump_doclist", "invalid segdir");
+ return;
+ }
+ rc = SQLITE_OK;
+ }
+ }
+ }
+
+ sqlite3_reset(s);
+ }
+
+ if( rc==SQLITE_OK ){
+ if( doclist.nData>0 ){
+ createDoclistResult(pContext, doclist.pData, doclist.nData);
+ }else{
+ /* TODO(shess): This can happen if the term is not present, or
+ ** if all instances of the term have been deleted and this is
+ ** an all-index dump. It may be interesting to distinguish
+ ** these cases.
+ */
+ sqlite3_result_text(pContext, "", 0, SQLITE_STATIC);
+ }
+ }else if( rc==SQLITE_NOMEM ){
+ /* Handle out-of-memory cases specially because if they are
+ ** generated in fts3 code they may not be reflected in the db
+ ** handle.
+ */
+ /* TODO(shess): Handle this more comprehensively.
+ ** sqlite3ErrStr() has what I need, but is internal.
+ */
+ generateError(pContext, "dump_doclist", "out of memory");
+ }else{
+ generateError(pContext, "dump_doclist", NULL);
+ }
+
+ dataBufferDestroy(&doclist);
+ }
+}
+#endif
+
+/*
+** This routine implements the xFindFunction method for the FTS3
+** virtual table.
+*/
+static int fulltextFindFunction(
+ sqlite3_vtab *pVtab,
+ int nArg,
+ const char *zName,
+ void (**pxFunc)(sqlite3_context*,int,sqlite3_value**),
+ void **ppArg
+){
+ if( strcmp(zName,"snippet")==0 ){
+ *pxFunc = snippetFunc;
+ return 1;
+ }else if( strcmp(zName,"offsets")==0 ){
+ *pxFunc = snippetOffsetsFunc;
+ return 1;
+ }else if( strcmp(zName,"optimize")==0 ){
+ *pxFunc = optimizeFunc;
+ return 1;
+#ifdef SQLITE_TEST
+ /* NOTE(shess): These functions are present only for testing
+ ** purposes. No particular effort is made to optimize their
+ ** execution or how they build their results.
+ */
+ }else if( strcmp(zName,"dump_terms")==0 ){
+ /* fprintf(stderr, "Found dump_terms\n"); */
+ *pxFunc = dumpTermsFunc;
+ return 1;
+ }else if( strcmp(zName,"dump_doclist")==0 ){
+ /* fprintf(stderr, "Found dump_doclist\n"); */
+ *pxFunc = dumpDoclistFunc;
+ return 1;
+#endif
+ }
+ return 0;
+}
+
+/*
+** Rename an fts3 table.
+*/
+static int fulltextRename(
+ sqlite3_vtab *pVtab,
+ const char *zName
+){
+ fulltext_vtab *p = (fulltext_vtab *)pVtab;
+ int rc = SQLITE_NOMEM;
+ char *zSql = sqlite3_mprintf(
+ "ALTER TABLE %Q.'%q_content' RENAME TO '%q_content';"
+ "ALTER TABLE %Q.'%q_segments' RENAME TO '%q_segments';"
+ "ALTER TABLE %Q.'%q_segdir' RENAME TO '%q_segdir';"
+ , p->zDb, p->zName, zName
+ , p->zDb, p->zName, zName
+ , p->zDb, p->zName, zName
+ );
+ if( zSql ){
+ rc = sqlite3_exec(p->db, zSql, 0, 0, 0);
+ sqlite3_free(zSql);
+ }
+ return rc;
+}
+
+static const sqlite3_module fts3Module = {
+ /* iVersion */ 0,
+ /* xCreate */ fulltextCreate,
+ /* xConnect */ fulltextConnect,
+ /* xBestIndex */ fulltextBestIndex,
+ /* xDisconnect */ fulltextDisconnect,
+ /* xDestroy */ fulltextDestroy,
+ /* xOpen */ fulltextOpen,
+ /* xClose */ fulltextClose,
+ /* xFilter */ fulltextFilter,
+ /* xNext */ fulltextNext,
+ /* xEof */ fulltextEof,
+ /* xColumn */ fulltextColumn,
+ /* xRowid */ fulltextRowid,
+ /* xUpdate */ fulltextUpdate,
+ /* xBegin */ fulltextBegin,
+ /* xSync */ fulltextSync,
+ /* xCommit */ fulltextCommit,
+ /* xRollback */ fulltextRollback,
+ /* xFindFunction */ fulltextFindFunction,
+ /* xRename */ fulltextRename,
+};
+
+static void hashDestroy(void *p){
+ fts3Hash *pHash = (fts3Hash *)p;
+ sqlite3Fts3HashClear(pHash);
+ sqlite3_free(pHash);
+}
+
+
+int sqlite3Fts3InitHashTable(sqlite3 *, fts3Hash *, const char *);
+
+/*
+** Initialise the fts3 extension. If this extension is built as part
+** of the sqlite library, then this function is called directly by
+** SQLite. If fts3 is built as a dynamically loadable extension, this
+** function is called by the sqlite3_extension_init() entry point.
+*/
+int sqlite3Fts3Init(sqlite3 *db){
+ int rc = SQLITE_OK;
+
+ /* Create the virtual table wrapper around the hash-table and overload
+ ** the two scalar functions. If this is successful, register the
+ ** module with sqlite.
+ */
+ if( SQLITE_OK==rc
+ && SQLITE_OK==(rc = sqlite3_overload_function(db, "snippet", -1))
+ && SQLITE_OK==(rc = sqlite3_overload_function(db, "offsets", -1))
+ && SQLITE_OK==(rc = sqlite3_overload_function(db, "optimize", -1))
+#ifdef SQLITE_TEST
+ && SQLITE_OK==(rc = sqlite3_overload_function(db, "dump_terms", -1))
+ && SQLITE_OK==(rc = sqlite3_overload_function(db, "dump_doclist", -1))
+#endif
+ ){
+ return sqlite3_create_module_v2(
+ db, "trackerfts", &fts3Module, 0, 0
+ );
+ }
+
+ /* An error has occured. Delete the hash table and return the error code. */
+ assert( rc!=SQLITE_OK );
+
+ return rc;
+}
+
+
+int sqlite3_extension_init(
+ sqlite3 *db,
+ char **pzErrMsg,
+ const sqlite3_api_routines *pApi
+){
+ SQLITE_EXTENSION_INIT2(pApi)
+ return sqlite3Fts3Init(db);
+}
+
Added: branches/indexer-split/src/tracker-fts/tracker-fts.h
==============================================================================
--- (empty file)
+++ branches/indexer-split/src/tracker-fts/tracker-fts.h Wed Aug 27 01:53:50 2008
@@ -0,0 +1,26 @@
+/*
+** 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 header file is used by programs that want to link against the
+** FTS3 library. All it does is declare the sqlite3Fts3Init() interface.
+*/
+#include <sqlite3.h>
+
+#ifdef __cplusplus
+extern "C" {
+#endif /* __cplusplus */
+
+int sqlite3Fts3Init(sqlite3 *db);
+
+#ifdef __cplusplus
+} /* extern "C" */
+#endif /* __cplusplus */
Modified: branches/indexer-split/tests/Makefile.am
==============================================================================
--- branches/indexer-split/tests/Makefile.am (original)
+++ branches/indexer-split/tests/Makefile.am Wed Aug 27 01:53:50 2008
@@ -5,5 +5,6 @@
trackerd \
scripts \
libtracker-db \
+ tracker-fts \
tracker-indexer
Added: branches/indexer-split/tests/tracker-fts/Makefile.am
==============================================================================
--- (empty file)
+++ branches/indexer-split/tests/tracker-fts/Makefile.am Wed Aug 27 01:53:50 2008
@@ -0,0 +1,27 @@
+include $(top_srcdir)/Makefile.decl
+
+noinst_PROGRAMS = $(TEST_PROGS)
+
+TEST_PROGS += tracker-fts-test
+
+INCLUDES = \
+ -DG_LOG_DOMAIN=\"Tracker\" \
+ -I$(top_srcdir)/src \
+ $(SQLITE3_CFLAGS) \
+ $(GTHREAD_CFLAGS) \
+ $(PANGO_CFLAGS) \
+ $(GLIB2_CFLAGS)
+
+
+tracker_fts_test_SOURCES = \
+ tracker-fts-test.c \
+ tracker-fts-test.h
+
+tracker_fts_test_LDADD = \
+ $(top_builddir)/src/libtracker-common/libtracker-common.la \
+ $(SQLITE3_LIBS) \
+ $(GTHREAD_LIBS) \
+ $(PANGO_LIBS) \
+ $(GLIB2_LIBS) \
+ $(top_builddir)/src/libstemmer/libstemmer.la
+
Added: branches/indexer-split/tests/tracker-fts/tracker-fts-test.c
==============================================================================
--- (empty file)
+++ branches/indexer-split/tests/tracker-fts/tracker-fts-test.c Wed Aug 27 01:53:50 2008
@@ -0,0 +1,92 @@
+#include <stdio.h>
+#include <sqlite3.h>
+#include <glib.h>
+
+static int
+callback (void *NotUsed, int argc, char **argv, char **azColName)
+{
+ int i;
+ for (i=0; i<argc; i++){
+ printf("%s = %s\n", azColName[i], argv[i] ? argv[i] : "NULL");
+ }
+
+ printf("\n");
+ return 0;
+}
+
+
+static void
+exec_sql (sqlite3 *db, const char *sql)
+{
+ char *zErrMsg = 0;
+
+
+
+ int rc = sqlite3_exec (db, sql , callback, 0, &zErrMsg);
+
+ if( rc!=SQLITE_OK ){
+ fprintf(stderr, "SQL error: %s\n", zErrMsg);
+ sqlite3_free(zErrMsg);
+ }
+
+
+
+}
+
+
+
+int
+main (int argc, char **argv)
+{
+ sqlite3 *db;
+ char *zErrMsg = 0;
+ int rc;
+ gboolean db_exists = FALSE;
+
+ g_type_init ();
+ g_thread_init (NULL);
+
+ if( argc != 2 ){
+ fprintf(stderr, "Usage: %s MATCH_TERM\n", argv[0]);
+ fprintf(stderr, "EG: %s stew\n", argv[0]);
+ exit(1);
+ }
+
+ db_exists = g_file_test ("/tmp/test.db", G_FILE_TEST_EXISTS);
+
+ rc = sqlite3_open("/tmp/test.db", &db);
+ if( rc ){
+ fprintf(stderr, "Can't open database: %s\n", sqlite3_errmsg(db));
+ sqlite3_close(db);
+ exit(1);
+ }
+
+ sqlite3_enable_load_extension(db, 1);
+
+ char *st = NULL;
+
+ sqlite3_load_extension (db, "tracker-fts.so", NULL, &st);
+
+ if (st) {
+ fprintf(stderr, "SQL error: %s\n", st);
+ sqlite3_free(st);
+ }
+
+ if (!db_exists) {
+ exec_sql (db, "create virtual table recipe using trackerfts (name, ingredients)");
+ exec_sql (db, "insert into recipe (name, ingredients) values ('broccoli stew', 'broccoli,peppers,cheese and tomatoes')");
+ exec_sql (db, "insert into recipe (name, ingredients) values ('pumpkin stew', 'pumpkin,onions,garlic and celery')");
+ exec_sql (db, "insert into recipe (name, ingredients) values ('broccoli pie', 'broccoli,cheese,onions and flour.')");
+ exec_sql (db, "insert into recipe (name, ingredients) values ('pumpkin pie', 'pumpkin,sugar,flour and butter.')");
+ }
+
+
+ char *sql = g_strdup_printf ("select rowid, name, ingredients, snippet(recipe), offsets(recipe) from recipe where recipe match '%s'", argv[1]);
+ exec_sql (db, sql);
+ g_free (sql);
+
+
+ sqlite3_close(db);
+ return 0;
+}
+
[
Date Prev][
Date Next] [
Thread Prev][
Thread Next]
[
Thread Index]
[
Date Index]
[
Author Index]
