Added the transitive_closure, ieee754, and amatch extensions.
FossilOrigin-Name: 84018099c8715b982cd24ce9221f93c7379e8c08
diff --git a/Makefile.in b/Makefile.in
index 74bf58e..c713daa 100644
--- a/Makefile.in
+++ b/Makefile.in
@@ -388,7 +388,10 @@
# Statically linked extensions
#
TESTSRC += \
+ $(TOP)/ext/misc/amatch.c \
+ $(TOP)/ext/misc/closure.c \
$(TOP)/ext/misc/fuzzer.c \
+ $(TOP)/ext/misc/ieee754.c \
$(TOP)/ext/misc/regexp.c \
$(TOP)/ext/misc/spellfix.c \
$(TOP)/ext/misc/wholenumber.c
diff --git a/Makefile.msc b/Makefile.msc
index ea8fc10..383368c 100644
--- a/Makefile.msc
+++ b/Makefile.msc
@@ -708,7 +708,10 @@
# Statically linked extensions
#
TESTEXT = \
+ $(TOP)\ext\misc\amatch.c \
+ $(TOP)\ext\misc\closure.c \
$(TOP)\ext\misc\fuzzer.c \
+ $(TOP)\ext\misc\ieee754.c \
$(TOP)\ext\misc\regexp.c \
$(TOP)\ext\misc\spellfix.c \
$(TOP)\ext\misc\wholenumber.c
diff --git a/ext/misc/amatch.c b/ext/misc/amatch.c
new file mode 100644
index 0000000..f91c84f
--- /dev/null
+++ b/ext/misc/amatch.c
@@ -0,0 +1,1477 @@
+/*
+** 2013-03-14
+**
+** The author disclaims copyright to this source code. In place of
+** a legal notice, here is a blessing:
+**
+** May you do good and not evil.
+** May you find forgiveness for yourself and forgive others.
+** May you share freely, never taking more than you give.
+**
+*************************************************************************
+**
+** This file contains code for a demonstration virtual table that finds
+** "approximate matches" - strings from a finite set that are nearly the
+** same as a single input string. The virtual table is called "amatch".
+**
+** A amatch virtual table is created like this:
+**
+** CREATE VIRTUAL TABLE f USING approximate_match(
+** vocabulary_table=<tablename>, -- V
+** vocabulary_word=<columnname>, -- W
+** vocabulary_language=<columnname>, -- L
+** edit_distances=<edit-cost-table>
+** );
+**
+** When it is created, the new amatch table must be supplied with the
+** the name of a table V and columns V.W and V.L such that
+**
+** SELECT W FROM V WHERE L=$language
+**
+** returns the allowed vocabulary for the match. If the "vocabulary_language"
+** or L columnname is left unspecified or is an empty string, then no
+** filtering of the vocabulary by language is performed.
+**
+** For efficiency, it is essential that the vocabulary table be indexed:
+**
+** CREATE vocab_index ON V(W)
+**
+** A separate edit-cost-table provides scoring information that defines
+** what it means for one string to be "close" to another.
+**
+** The edit-cost-table must contain exactly four columns (more precisely,
+** the statement "SELECT * FROM <edit-cost-table>" must return records
+** that consist of four columns). It does not matter what the columns are
+** named.
+**
+** Each row in the edit-cost-table represents a single character
+** transformation going from user input to the vocabulary. The leftmost
+** column of the row (column 0) contains an integer identifier of the
+** language to which the transformation rule belongs (see "MULTIPLE LANGUAGES"
+** below). The second column of the row (column 1) contains the input
+** character or characters - the characters of user input. The third
+** column contains characters as they appear in the vocabulary table.
+** And the fourth column contains the integer cost of making the
+** transformation. For example:
+**
+** CREATE TABLE f_data(iLang, cFrom, cTo, Cost);
+** INSERT INTO f_data(iLang, cFrom, cTo, Cost) VALUES(0, '', 'a', 100);
+** INSERT INTO f_data(iLang, cFrom, cTo, Cost) VALUES(0, 'b', '', 87);
+** INSERT INTO f_data(iLang, cFrom, cTo, Cost) VALUES(0, 'o', 'oe', 38);
+** INSERT INTO f_data(iLang, cFrom, cTo, Cost) VALUES(0, 'oe', 'o', 40);
+**
+** The first row inserted into the edit-cost-table by the SQL script
+** above indicates that the cost of having an extra 'a' in the vocabulary
+** table that is missing in the user input 100. (All costs are integers.
+** Overall cost must not exceed 16777216.) The second INSERT statement
+** creates a rule saying that the cost of having a single letter 'b' in
+** user input which is missing in the vocabulary table is 87. The third
+** INSERT statement mean that the cost of matching an 'o' in user input
+** against an 'oe' in the vocabulary table is 38. And so forth.
+**
+** The following rules are special:
+**
+** INSERT INTO f_data(iLang, cFrom, cTo, Cost) VALUES(0, '?', '', 97);
+** INSERT INTO f_data(iLang, cFrom, cTo, Cost) VALUES(0, '', '?', 98);
+** INSERT INTO f_data(iLang, cFrom, cTo, Cost) VALUES(0, '?', '?', 99);
+**
+** The '?' to '' rule is the cost of having any single character in the input
+** that is not found in the vocabular. The '' to '?' rule is the cost of
+** having a character in the vocabulary table that is missing from input.
+** And the '?' to '?' rule is the cost of doing an arbitrary character
+** substitution. These three generic rules apply across all languages.
+** In other words, the iLang field is ignored for the generic substitution
+** rules. If more than one cost is given for a generic substitution rule,
+** then the lowest cost is used.
+**
+** Once it has been created, the amatch virtual table can be queried
+** as follows:
+**
+** SELECT word, distance FROM f
+** WHERE word MATCH 'abcdefg'
+** AND distance<200;
+**
+** This query outputs the strings contained in the T(F) field that
+** are close to "abcdefg" and in order of increasing distance. No string
+** is output more than once. If there are multiple ways to transform the
+** target string ("abcdefg") into a string in the vocabulary table then
+** the lowest cost transform is the one that is returned. In this example,
+** the search is limited to strings with a total distance of less than 200.
+**
+** For efficiency, it is important to put tight bounds on the distance.
+** The time and memory space needed to perform this query is exponential
+** in the maximum distance. A good rule of thumb is to limit the distance
+** to no more than 1.5 or 2 times the maximum cost of any rule in the
+** edit-cost-table.
+**
+** The amatch is a read-only table. Any attempt to DELETE, INSERT, or
+** UPDATE on a amatch table will throw an error.
+**
+** It is important to put some kind of a limit on the amatch output. This
+** can be either in the form of a LIMIT clause at the end of the query,
+** or better, a "distance<NNN" constraint where NNN is some number. The
+** running time and memory requirement is exponential in the value of NNN
+** so you want to make sure that NNN is not too big. A value of NNN that
+** is about twice the average transformation cost seems to give good results.
+**
+** The amatch table can be useful for tasks such as spelling correction.
+** Suppose all allowed words are in table vocabulary(w). Then one would create
+** an amatch virtual table like this:
+**
+** CREATE VIRTUAL TABLE ex1 USING amatch(
+** vocabtable=vocabulary,
+** vocabcolumn=w,
+** edit_distances=ec1
+** );
+**
+** Then given an input word $word, look up close spellings this way:
+**
+** SELECT word, distance FROM ex1
+** WHERE word MATCH $word AND distance<200;
+**
+** MULTIPLE LANGUAGES
+**
+** Normally, the "iLang" value associated with all character transformations
+** in the edit-cost-table is zero. However, if required, the amatch
+** virtual table allows multiple languages to be defined. Each query uses
+** only a single iLang value. This allows, for example, a single
+** amatch table to support multiple languages.
+**
+** By default, only the rules with iLang=0 are used. To specify an
+** alternative language, a "language = ?" expression must be added to the
+** WHERE clause of a SELECT, where ? is the integer identifier of the desired
+** language. For example:
+**
+** SELECT word, distance FROM ex1
+** WHERE word MATCH $word
+** AND distance<=200
+** AND language=1 -- Specify use language 1 instead of 0
+**
+** If no "language = ?" constraint is specified in the WHERE clause, language
+** 0 is used.
+**
+** LIMITS
+**
+** The maximum language number is 2147483647. The maximum length of either
+** of the strings in the second or third column of the amatch data table
+** is 50 bytes. The maximum cost on a rule is 1000.
+*/
+#include "sqlite3ext.h"
+SQLITE_EXTENSION_INIT1
+#include <stdlib.h>
+#include <string.h>
+#include <assert.h>
+#include <stdio.h>
+#include <ctype.h>
+
+/*
+** Forward declaration of objects used by this implementation
+*/
+typedef struct amatch_vtab amatch_vtab;
+typedef struct amatch_cursor amatch_cursor;
+typedef struct amatch_rule amatch_rule;
+typedef struct amatch_word amatch_word;
+typedef struct amatch_avl amatch_avl;
+
+
+/*****************************************************************************
+** AVL Tree implementation
+*/
+/*
+** Objects that want to be members of the AVL tree should embedded an
+** instance of this structure.
+*/
+struct amatch_avl {
+ amatch_word *pWord; /* Points to the object being stored in the tree */
+ char *zKey; /* Key. zero-terminated string. Must be unique */
+ amatch_avl *pBefore; /* Other elements less than zKey */
+ amatch_avl *pAfter; /* Other elements greater than zKey */
+ amatch_avl *pUp; /* Parent element */
+ short int height; /* Height of this node. Leaf==1 */
+ short int imbalance; /* Height difference between pBefore and pAfter */
+};
+
+/* Recompute the amatch_avl.height and amatch_avl.imbalance fields for p.
+** Assume that the children of p have correct heights.
+*/
+static void amatchAvlRecomputeHeight(amatch_avl *p){
+ short int hBefore = p->pBefore ? p->pBefore->height : 0;
+ short int hAfter = p->pAfter ? p->pAfter->height : 0;
+ p->imbalance = hBefore - hAfter; /* -: pAfter higher. +: pBefore higher */
+ p->height = (hBefore>hAfter ? hBefore : hAfter)+1;
+}
+
+/*
+** P B
+** / \ / \
+** B Z ==> X P
+** / \ / \
+** X Y Y Z
+**
+*/
+static amatch_avl *amatchAvlRotateBefore(amatch_avl *pP){
+ amatch_avl *pB = pP->pBefore;
+ amatch_avl *pY = pB->pAfter;
+ pB->pUp = pP->pUp;
+ pB->pAfter = pP;
+ pP->pUp = pB;
+ pP->pBefore = pY;
+ if( pY ) pY->pUp = pP;
+ amatchAvlRecomputeHeight(pP);
+ amatchAvlRecomputeHeight(pB);
+ return pB;
+}
+
+/*
+** P A
+** / \ / \
+** X A ==> P Z
+** / \ / \
+** Y Z X Y
+**
+*/
+static amatch_avl *amatchAvlRotateAfter(amatch_avl *pP){
+ amatch_avl *pA = pP->pAfter;
+ amatch_avl *pY = pA->pBefore;
+ pA->pUp = pP->pUp;
+ pA->pBefore = pP;
+ pP->pUp = pA;
+ pP->pAfter = pY;
+ if( pY ) pY->pUp = pP;
+ amatchAvlRecomputeHeight(pP);
+ amatchAvlRecomputeHeight(pA);
+ return pA;
+}
+
+/*
+** Return a pointer to the pBefore or pAfter pointer in the parent
+** of p that points to p. Or if p is the root node, return pp.
+*/
+static amatch_avl **amatchAvlFromPtr(amatch_avl *p, amatch_avl **pp){
+ amatch_avl *pUp = p->pUp;
+ if( pUp==0 ) return pp;
+ if( pUp->pAfter==p ) return &pUp->pAfter;
+ return &pUp->pBefore;
+}
+
+/*
+** Rebalance all nodes starting with p and working up to the root.
+** Return the new root.
+*/
+static amatch_avl *amatchAvlBalance(amatch_avl *p){
+ amatch_avl *pTop = p;
+ amatch_avl **pp;
+ while( p ){
+ amatchAvlRecomputeHeight(p);
+ if( p->imbalance>=2 ){
+ amatch_avl *pB = p->pBefore;
+ if( pB->imbalance<0 ) p->pBefore = amatchAvlRotateAfter(pB);
+ pp = amatchAvlFromPtr(p,&p);
+ p = *pp = amatchAvlRotateBefore(p);
+ }else if( p->imbalance<=(-2) ){
+ amatch_avl *pA = p->pAfter;
+ if( pA->imbalance>0 ) p->pAfter = amatchAvlRotateBefore(pA);
+ pp = amatchAvlFromPtr(p,&p);
+ p = *pp = amatchAvlRotateAfter(p);
+ }
+ pTop = p;
+ p = p->pUp;
+ }
+ return pTop;
+}
+
+/* Search the tree rooted at p for an entry with zKey. Return a pointer
+** to the entry or return NULL.
+*/
+static amatch_avl *amatchAvlSearch(amatch_avl *p, const char *zKey){
+ int c;
+ while( p && (c = strcmp(zKey, p->zKey))!=0 ){
+ p = (c<0) ? p->pBefore : p->pAfter;
+ }
+ return p;
+}
+
+/* Find the first node (the one with the smallest key).
+*/
+static amatch_avl *amatchAvlFirst(amatch_avl *p){
+ if( p ) while( p->pBefore ) p = p->pBefore;
+ return p;
+}
+
+#if 0 /* NOT USED */
+/* Return the node with the next larger key after p.
+*/
+static amatch_avl *amatchAvlNext(amatch_avl *p){
+ amatch_avl *pPrev = 0;
+ while( p && p->pAfter==pPrev ){
+ pPrev = p;
+ p = p->pUp;
+ }
+ if( p && pPrev==0 ){
+ p = amatchAvlFirst(p->pAfter);
+ }
+ return p;
+}
+#endif
+
+#if 0 /* NOT USED */
+/* Verify AVL tree integrity
+*/
+static int amatchAvlIntegrity(amatch_avl *pHead){
+ amatch_avl *p;
+ if( pHead==0 ) return 1;
+ if( (p = pHead->pBefore)!=0 ){
+ assert( p->pUp==pHead );
+ assert( amatchAvlIntegrity(p) );
+ assert( strcmp(p->zKey, pHead->zKey)<0 );
+ while( p->pAfter ) p = p->pAfter;
+ assert( strcmp(p->zKey, pHead->zKey)<0 );
+ }
+ if( (p = pHead->pAfter)!=0 ){
+ assert( p->pUp==pHead );
+ assert( amatchAvlIntegrity(p) );
+ assert( strcmp(p->zKey, pHead->zKey)>0 );
+ p = amatchAvlFirst(p);
+ assert( strcmp(p->zKey, pHead->zKey)>0 );
+ }
+ return 1;
+}
+static int amatchAvlIntegrity2(amatch_avl *pHead){
+ amatch_avl *p, *pNext;
+ for(p=amatchAvlFirst(pHead); p; p=pNext){
+ pNext = amatchAvlNext(p);
+ if( pNext==0 ) break;
+ assert( strcmp(p->zKey, pNext->zKey)<0 );
+ }
+ return 1;
+}
+#endif
+
+/* Insert a new node pNew. Return NULL on success. If the key is not
+** unique, then do not perform the insert but instead leave pNew unchanged
+** and return a pointer to an existing node with the same key.
+*/
+static amatch_avl *amatchAvlInsert(amatch_avl **ppHead, amatch_avl *pNew){
+ int c;
+ amatch_avl *p = *ppHead;
+ if( p==0 ){
+ p = pNew;
+ pNew->pUp = 0;
+ }else{
+ while( p ){
+ c = strcmp(pNew->zKey, p->zKey);
+ if( c<0 ){
+ if( p->pBefore ){
+ p = p->pBefore;
+ }else{
+ p->pBefore = pNew;
+ pNew->pUp = p;
+ break;
+ }
+ }else if( c>0 ){
+ if( p->pAfter ){
+ p = p->pAfter;
+ }else{
+ p->pAfter = pNew;
+ pNew->pUp = p;
+ break;
+ }
+ }else{
+ return p;
+ }
+ }
+ }
+ pNew->pBefore = 0;
+ pNew->pAfter = 0;
+ pNew->height = 1;
+ pNew->imbalance = 0;
+ *ppHead = amatchAvlBalance(p);
+ /* assert( amatchAvlIntegrity(*ppHead) ); */
+ /* assert( amatchAvlIntegrity2(*ppHead) ); */
+ return 0;
+}
+
+/* Remove node pOld from the tree. pOld must be an element of the tree or
+** the AVL tree will become corrupt.
+*/
+static void amatchAvlRemove(amatch_avl **ppHead, amatch_avl *pOld){
+ amatch_avl **ppParent;
+ amatch_avl *pBalance;
+ /* assert( amatchAvlSearch(*ppHead, pOld->zKey)==pOld ); */
+ ppParent = amatchAvlFromPtr(pOld, ppHead);
+ if( pOld->pBefore==0 && pOld->pAfter==0 ){
+ *ppParent = 0;
+ pBalance = pOld->pUp;
+ }else if( pOld->pBefore && pOld->pAfter ){
+ amatch_avl *pX, *pY;
+ pX = amatchAvlFirst(pOld->pAfter);
+ *amatchAvlFromPtr(pX, 0) = pX->pAfter;
+ if( pX->pAfter ) pX->pAfter->pUp = pX->pUp;
+ pBalance = pX->pUp;
+ pX->pAfter = pOld->pAfter;
+ if( pX->pAfter ){
+ pX->pAfter->pUp = pX;
+ }else{
+ assert( pBalance==pOld );
+ pBalance = pX;
+ }
+ pX->pBefore = pY = pOld->pBefore;
+ if( pY ) pY->pUp = pX;
+ pX->pUp = pOld->pUp;
+ *ppParent = pX;
+ }else if( pOld->pBefore==0 ){
+ *ppParent = pBalance = pOld->pAfter;
+ pBalance->pUp = pOld->pUp;
+ }else if( pOld->pAfter==0 ){
+ *ppParent = pBalance = pOld->pBefore;
+ pBalance->pUp = pOld->pUp;
+ }
+ *ppHead = amatchAvlBalance(pBalance);
+ pOld->pUp = 0;
+ pOld->pBefore = 0;
+ pOld->pAfter = 0;
+ /* assert( amatchAvlIntegrity(*ppHead) ); */
+ /* assert( amatchAvlIntegrity2(*ppHead) ); */
+}
+/*
+** End of the AVL Tree implementation
+******************************************************************************/
+
+
+/*
+** Various types.
+**
+** amatch_cost is the "cost" of an edit operation.
+**
+** amatch_len is the length of a matching string.
+**
+** amatch_langid is an ruleset identifier.
+*/
+typedef int amatch_cost;
+typedef signed char amatch_len;
+typedef int amatch_langid;
+
+/*
+** Limits
+*/
+#define AMATCH_MX_LENGTH 50 /* Maximum length of a rule string */
+#define AMATCH_MX_LANGID 2147483647 /* Maximum rule ID */
+#define AMATCH_MX_COST 1000 /* Maximum single-rule cost */
+
+/*
+** A match or partial match
+*/
+struct amatch_word {
+ amatch_word *pNext; /* Next on a list of all amatch_words */
+ amatch_avl sCost; /* Linkage of this node into the cost tree */
+ amatch_avl sWord; /* Linkage of this node into the word tree */
+ amatch_cost rCost; /* Cost of the match so far */
+ int iSeq; /* Sequence number */
+ char zCost[10]; /* Cost key (text rendering of rCost) */
+ short int nMatch; /* Input characters matched */
+ char zWord[4]; /* Text of the word. Extra space appended as needed */
+};
+
+/*
+** Each transformation rule is stored as an instance of this object.
+** All rules are kept on a linked list sorted by rCost.
+*/
+struct amatch_rule {
+ amatch_rule *pNext; /* Next rule in order of increasing rCost */
+ char *zFrom; /* Transform from (a string from user input) */
+ amatch_cost rCost; /* Cost of this transformation */
+ amatch_langid iLang; /* The langauge to which this rule belongs */
+ amatch_len nFrom, nTo; /* Length of the zFrom and zTo strings */
+ char zTo[4]; /* Tranform to V.W value (extra space appended) */
+};
+
+/*
+** A amatch virtual-table object
+*/
+struct amatch_vtab {
+ sqlite3_vtab base; /* Base class - must be first */
+ char *zClassName; /* Name of this class. Default: "amatch" */
+ char *zDb; /* Name of database. (ex: "main") */
+ char *zSelf; /* Name of this virtual table */
+ char *zCostTab; /* Name of edit-cost-table */
+ char *zVocabTab; /* Name of vocabulary table */
+ char *zVocabWord; /* Name of vocabulary table word column */
+ char *zVocabLang; /* Name of vocabulary table language column */
+ amatch_rule *pRule; /* All active rules in this amatch */
+ amatch_cost rIns; /* Generic insertion cost '' -> ? */
+ amatch_cost rDel; /* Generic deletion cost ? -> '' */
+ amatch_cost rSub; /* Generic substitution cost ? -> ? */
+ sqlite3 *db; /* The database connection */
+ sqlite3_stmt *pVCheck; /* Query to check zVocabTab */
+ int nCursor; /* Number of active cursors */
+};
+
+/* A amatch cursor object */
+struct amatch_cursor {
+ sqlite3_vtab_cursor base; /* Base class - must be first */
+ sqlite3_int64 iRowid; /* The rowid of the current word */
+ amatch_langid iLang; /* Use this language ID */
+ amatch_cost rLimit; /* Maximum cost of any term */
+ int nBuf; /* Space allocated for zBuf */
+ int oomErr; /* True following an OOM error */
+ int nWord; /* Number of amatch_word objects */
+ char *zBuf; /* Temp-use buffer space */
+ char *zInput; /* Input word to match against */
+ amatch_vtab *pVtab; /* The virtual table this cursor belongs to */
+ amatch_word *pAllWords; /* List of all amatch_word objects */
+ amatch_word *pCurrent; /* Most recent solution */
+ amatch_avl *pCost; /* amatch_word objects keyed by iCost */
+ amatch_avl *pWord; /* amatch_word objects keyed by zWord */
+};
+
+/*
+** The two input rule lists are both sorted in order of increasing
+** cost. Merge them together into a single list, sorted by cost, and
+** return a pointer to the head of that list.
+*/
+static amatch_rule *amatchMergeRules(amatch_rule *pA, amatch_rule *pB){
+ amatch_rule head;
+ amatch_rule *pTail;
+
+ pTail = &head;
+ while( pA && pB ){
+ if( pA->rCost<=pB->rCost ){
+ pTail->pNext = pA;
+ pTail = pA;
+ pA = pA->pNext;
+ }else{
+ pTail->pNext = pB;
+ pTail = pB;
+ pB = pB->pNext;
+ }
+ }
+ if( pA==0 ){
+ pTail->pNext = pB;
+ }else{
+ pTail->pNext = pA;
+ }
+ return head.pNext;
+}
+
+/*
+** Statement pStmt currently points to a row in the amatch data table. This
+** function allocates and populates a amatch_rule structure according to
+** the content of the row.
+**
+** If successful, *ppRule is set to point to the new object and SQLITE_OK
+** is returned. Otherwise, *ppRule is zeroed, *pzErr may be set to point
+** to an error message and an SQLite error code returned.
+*/
+static int amatchLoadOneRule(
+ amatch_vtab *p, /* Fuzzer virtual table handle */
+ sqlite3_stmt *pStmt, /* Base rule on statements current row */
+ amatch_rule **ppRule, /* OUT: New rule object */
+ char **pzErr /* OUT: Error message */
+){
+ sqlite3_int64 iLang = sqlite3_column_int64(pStmt, 0);
+ const char *zFrom = (const char *)sqlite3_column_text(pStmt, 1);
+ const char *zTo = (const char *)sqlite3_column_text(pStmt, 2);
+ amatch_cost rCost = sqlite3_column_int(pStmt, 3);
+
+ int rc = SQLITE_OK; /* Return code */
+ int nFrom; /* Size of string zFrom, in bytes */
+ int nTo; /* Size of string zTo, in bytes */
+ amatch_rule *pRule = 0; /* New rule object to return */
+
+ if( zFrom==0 ) zFrom = "";
+ if( zTo==0 ) zTo = "";
+ nFrom = (int)strlen(zFrom);
+ nTo = (int)strlen(zTo);
+
+ /* Silently ignore null transformations */
+ if( strcmp(zFrom, zTo)==0 ){
+ if( zFrom[0]=='?' && zFrom[1]==0 ){
+ if( p->rSub==0 || p->rSub>rCost ) p->rSub = rCost;
+ }
+ *ppRule = 0;
+ return SQLITE_OK;
+ }
+
+ if( rCost<=0 || rCost>AMATCH_MX_COST ){
+ *pzErr = sqlite3_mprintf("%s: cost must be between 1 and %d",
+ p->zClassName, AMATCH_MX_COST
+ );
+ rc = SQLITE_ERROR;
+ }else
+ if( nFrom>AMATCH_MX_LENGTH || nTo>AMATCH_MX_LENGTH ){
+ *pzErr = sqlite3_mprintf("%s: maximum string length is %d",
+ p->zClassName, AMATCH_MX_LENGTH
+ );
+ rc = SQLITE_ERROR;
+ }else
+ if( iLang<0 || iLang>AMATCH_MX_LANGID ){
+ *pzErr = sqlite3_mprintf("%s: iLang must be between 0 and %d",
+ p->zClassName, AMATCH_MX_LANGID
+ );
+ rc = SQLITE_ERROR;
+ }else
+ if( strcmp(zFrom,"")==0 && strcmp(zTo,"?")==0 ){
+ if( p->rIns==0 || p->rIns>rCost ) p->rIns = rCost;
+ }else
+ if( strcmp(zFrom,"?")==0 && strcmp(zTo,"")==0 ){
+ if( p->rDel==0 || p->rDel>rCost ) p->rDel = rCost;
+ }else
+ {
+ pRule = sqlite3_malloc( sizeof(*pRule) + nFrom + nTo );
+ if( pRule==0 ){
+ rc = SQLITE_NOMEM;
+ }else{
+ memset(pRule, 0, sizeof(*pRule));
+ pRule->zFrom = &pRule->zTo[nTo+1];
+ pRule->nFrom = nFrom;
+ memcpy(pRule->zFrom, zFrom, nFrom+1);
+ memcpy(pRule->zTo, zTo, nTo+1);
+ pRule->nTo = nTo;
+ pRule->rCost = rCost;
+ pRule->iLang = (int)iLang;
+ }
+ }
+
+ *ppRule = pRule;
+ return rc;
+}
+
+/*
+** Free all the content in the edit-cost-table
+*/
+static void amatchFreeRules(amatch_vtab *p){
+ while( p->pRule ){
+ amatch_rule *pRule = p->pRule;
+ p->pRule = pRule->pNext;
+ sqlite3_free(pRule);
+ }
+ p->pRule = 0;
+}
+
+/*
+** Load the content of the amatch data table into memory.
+*/
+static int amatchLoadRules(
+ sqlite3 *db, /* Database handle */
+ amatch_vtab *p, /* Virtual amatch table to configure */
+ char **pzErr /* OUT: Error message */
+){
+ int rc = SQLITE_OK; /* Return code */
+ char *zSql; /* SELECT used to read from rules table */
+ amatch_rule *pHead = 0;
+
+ zSql = sqlite3_mprintf("SELECT * FROM %Q.%Q", p->zDb, p->zCostTab);
+ if( zSql==0 ){
+ rc = SQLITE_NOMEM;
+ }else{
+ int rc2; /* finalize() return code */
+ sqlite3_stmt *pStmt = 0;
+ rc = sqlite3_prepare_v2(db, zSql, -1, &pStmt, 0);
+ if( rc!=SQLITE_OK ){
+ *pzErr = sqlite3_mprintf("%s: %s", p->zClassName, sqlite3_errmsg(db));
+ }else if( sqlite3_column_count(pStmt)!=4 ){
+ *pzErr = sqlite3_mprintf("%s: %s has %d columns, expected 4",
+ p->zClassName, p->zCostTab, sqlite3_column_count(pStmt)
+ );
+ rc = SQLITE_ERROR;
+ }else{
+ while( rc==SQLITE_OK && SQLITE_ROW==sqlite3_step(pStmt) ){
+ amatch_rule *pRule = 0;
+ rc = amatchLoadOneRule(p, pStmt, &pRule, pzErr);
+ if( pRule ){
+ pRule->pNext = pHead;
+ pHead = pRule;
+ }
+ }
+ }
+ rc2 = sqlite3_finalize(pStmt);
+ if( rc==SQLITE_OK ) rc = rc2;
+ }
+ sqlite3_free(zSql);
+
+ /* All rules are now in a singly linked list starting at pHead. This
+ ** block sorts them by cost and then sets amatch_vtab.pRule to point to
+ ** point to the head of the sorted list.
+ */
+ if( rc==SQLITE_OK ){
+ unsigned int i;
+ amatch_rule *pX;
+ amatch_rule *a[15];
+ for(i=0; i<sizeof(a)/sizeof(a[0]); i++) a[i] = 0;
+ while( (pX = pHead)!=0 ){
+ pHead = pX->pNext;
+ pX->pNext = 0;
+ for(i=0; a[i] && i<sizeof(a)/sizeof(a[0])-1; i++){
+ pX = amatchMergeRules(a[i], pX);
+ a[i] = 0;
+ }
+ a[i] = amatchMergeRules(a[i], pX);
+ }
+ for(pX=a[0], i=1; i<sizeof(a)/sizeof(a[0]); i++){
+ pX = amatchMergeRules(a[i], pX);
+ }
+ p->pRule = amatchMergeRules(p->pRule, pX);
+ }else{
+ /* An error has occurred. Setting p->pRule to point to the head of the
+ ** allocated list ensures that the list will be cleaned up in this case.
+ */
+ assert( p->pRule==0 );
+ p->pRule = pHead;
+ }
+
+ return rc;
+}
+
+/*
+** This function converts an SQL quoted string into an unquoted string
+** and returns a pointer to a buffer allocated using sqlite3_malloc()
+** containing the result. The caller should eventually free this buffer
+** using sqlite3_free.
+**
+** Examples:
+**
+** "abc" becomes abc
+** 'xyz' becomes xyz
+** [pqr] becomes pqr
+** `mno` becomes mno
+*/
+static char *amatchDequote(const char *zIn){
+ int nIn; /* Size of input string, in bytes */
+ char *zOut; /* Output (dequoted) string */
+
+ nIn = (int)strlen(zIn);
+ zOut = sqlite3_malloc(nIn+1);
+ if( zOut ){
+ char q = zIn[0]; /* Quote character (if any ) */
+
+ if( q!='[' && q!= '\'' && q!='"' && q!='`' ){
+ memcpy(zOut, zIn, nIn+1);
+ }else{
+ int iOut = 0; /* Index of next byte to write to output */
+ int iIn; /* Index of next byte to read from input */
+
+ if( q=='[' ) q = ']';
+ for(iIn=1; iIn<nIn; iIn++){
+ if( zIn[iIn]==q ) iIn++;
+ zOut[iOut++] = zIn[iIn];
+ }
+ }
+ assert( (int)strlen(zOut)<=nIn );
+ }
+ return zOut;
+}
+
+/*
+** Deallocate the pVCheck prepared statement.
+*/
+static void amatchVCheckClear(amatch_vtab *p){
+ if( p->pVCheck ){
+ sqlite3_finalize(p->pVCheck);
+ p->pVCheck = 0;
+ }
+}
+
+/*
+** Deallocate an amatch_vtab object
+*/
+static void amatchFree(amatch_vtab *p){
+ if( p ){
+ amatchFreeRules(p);
+ amatchVCheckClear(p);
+ sqlite3_free(p->zClassName);
+ sqlite3_free(p->zDb);
+ sqlite3_free(p->zCostTab);
+ sqlite3_free(p->zVocabTab);
+ sqlite3_free(p->zVocabWord);
+ sqlite3_free(p->zVocabLang);
+ memset(p, 0, sizeof(*p));
+ sqlite3_free(p);
+ }
+}
+
+/*
+** xDisconnect/xDestroy method for the amatch module.
+*/
+static int amatchDisconnect(sqlite3_vtab *pVtab){
+ amatch_vtab *p = (amatch_vtab*)pVtab;
+ assert( p->nCursor==0 );
+ amatchFree(p);
+ return SQLITE_OK;
+}
+
+/*
+** Check to see if the argument is of the form:
+**
+** KEY = VALUE
+**
+** If it is, return a pointer to the first character of VALUE.
+** If not, return NULL. Spaces around the = are ignored.
+*/
+static const char *amatchValueOfKey(const char *zKey, const char *zStr){
+ int nKey = (int)strlen(zKey);
+ int nStr = (int)strlen(zStr);
+ int i;
+ if( nStr<nKey+1 ) return 0;
+ if( memcmp(zStr, zKey, nKey)!=0 ) return 0;
+ for(i=nKey; isspace(zStr[i]); i++){}
+ if( zStr[i]!='=' ) return 0;
+ i++;
+ while( isspace(zStr[i]) ){ i++; }
+ return zStr+i;
+}
+
+/*
+** xConnect/xCreate method for the amatch module. Arguments are:
+**
+** argv[0] -> module name ("approximate_match")
+** argv[1] -> database name
+** argv[2] -> table name
+** argv[3...] -> arguments
+*/
+static int amatchConnect(
+ sqlite3 *db,
+ void *pAux,
+ int argc, const char *const*argv,
+ sqlite3_vtab **ppVtab,
+ char **pzErr
+){
+ int rc = SQLITE_OK; /* Return code */
+ amatch_vtab *pNew = 0; /* New virtual table */
+ const char *zModule = argv[0];
+ const char *zDb = argv[1];
+ const char *zVal;
+ int i;
+
+ (void)pAux;
+ *ppVtab = 0;
+ pNew = sqlite3_malloc( sizeof(*pNew) );
+ if( pNew==0 ) return SQLITE_NOMEM;
+ rc = SQLITE_NOMEM;
+ memset(pNew, 0, sizeof(*pNew));
+ pNew->db = db;
+ pNew->zClassName = sqlite3_mprintf("%s", zModule);
+ if( pNew->zClassName==0 ) goto amatchConnectError;
+ pNew->zDb = sqlite3_mprintf("%s", zDb);
+ if( pNew->zDb==0 ) goto amatchConnectError;
+ pNew->zSelf = sqlite3_mprintf("%s", argv[2]);
+ if( pNew->zSelf==0 ) goto amatchConnectError;
+ for(i=3; i<argc; i++){
+ zVal = amatchValueOfKey("vocabulary_table", argv[i]);
+ if( zVal ){
+ sqlite3_free(pNew->zVocabTab);
+ pNew->zVocabTab = amatchDequote(zVal);
+ if( pNew->zVocabTab==0 ) goto amatchConnectError;
+ continue;
+ }
+ zVal = amatchValueOfKey("vocabulary_word", argv[i]);
+ if( zVal ){
+ sqlite3_free(pNew->zVocabWord);
+ pNew->zVocabWord = amatchDequote(zVal);
+ if( pNew->zVocabWord==0 ) goto amatchConnectError;
+ continue;
+ }
+ zVal = amatchValueOfKey("vocabulary_language", argv[i]);
+ if( zVal ){
+ sqlite3_free(pNew->zVocabLang);
+ pNew->zVocabLang = amatchDequote(zVal);
+ if( pNew->zVocabLang==0 ) goto amatchConnectError;
+ continue;
+ }
+ zVal = amatchValueOfKey("edit_distances", argv[i]);
+ if( zVal ){
+ sqlite3_free(pNew->zCostTab);
+ pNew->zCostTab = amatchDequote(zVal);
+ if( pNew->zCostTab==0 ) goto amatchConnectError;
+ continue;
+ }
+ *pzErr = sqlite3_mprintf("unrecognized argument: [%s]\n", argv[i]);
+ amatchFree(pNew);
+ *ppVtab = 0;
+ return SQLITE_ERROR;
+ }
+ rc = SQLITE_OK;
+ if( pNew->zCostTab==0 ){
+ *pzErr = sqlite3_mprintf("no edit_distances table specified");
+ rc = SQLITE_ERROR;
+ }else{
+ rc = amatchLoadRules(db, pNew, pzErr);
+ }
+ if( rc==SQLITE_OK ){
+ rc = sqlite3_declare_vtab(db,
+ "CREATE TABLE x(word,distance,language,"
+ "command HIDDEN,nword HIDDEN)"
+ );
+#define AMATCH_COL_WORD 0
+#define AMATCH_COL_DISTANCE 1
+#define AMATCH_COL_LANGUAGE 2
+#define AMATCH_COL_COMMAND 3
+#define AMATCH_COL_NWORD 4
+ }
+ if( rc!=SQLITE_OK ){
+ amatchFree(pNew);
+ }
+ *ppVtab = &pNew->base;
+ return rc;
+
+amatchConnectError:
+ amatchFree(pNew);
+ return rc;
+}
+
+/*
+** Open a new amatch cursor.
+*/
+static int amatchOpen(sqlite3_vtab *pVTab, sqlite3_vtab_cursor **ppCursor){
+ amatch_vtab *p = (amatch_vtab*)pVTab;
+ amatch_cursor *pCur;
+ pCur = sqlite3_malloc( sizeof(*pCur) );
+ if( pCur==0 ) return SQLITE_NOMEM;
+ memset(pCur, 0, sizeof(*pCur));
+ pCur->pVtab = p;
+ *ppCursor = &pCur->base;
+ p->nCursor++;
+ return SQLITE_OK;
+}
+
+/*
+** Free up all the memory allocated by a cursor. Set it rLimit to 0
+** to indicate that it is at EOF.
+*/
+static void amatchClearCursor(amatch_cursor *pCur){
+ amatch_word *pWord, *pNextWord;
+ for(pWord=pCur->pAllWords; pWord; pWord=pNextWord){
+ pNextWord = pWord->pNext;
+ sqlite3_free(pWord);
+ }
+ pCur->pAllWords = 0;
+ sqlite3_free(pCur->zInput);
+ pCur->zInput = 0;
+ pCur->pCost = 0;
+ pCur->pWord = 0;
+ pCur->pCurrent = 0;
+ pCur->rLimit = 1000000;
+ pCur->iLang = 0;
+ pCur->nWord = 0;
+}
+
+/*
+** Close a amatch cursor.
+*/
+static int amatchClose(sqlite3_vtab_cursor *cur){
+ amatch_cursor *pCur = (amatch_cursor *)cur;
+ amatchClearCursor(pCur);
+ pCur->pVtab->nCursor--;
+ sqlite3_free(pCur);
+ return SQLITE_OK;
+}
+
+/*
+** Render a 24-bit unsigned integer as a 4-byte base-64 number.
+*/
+static void amatchEncodeInt(int x, char *z){
+ static const char a[] =
+ "0123456789"
+ "ABCDEFGHIJ"
+ "KLMNOPQRST"
+ "UVWXYZ^abc"
+ "defghijklm"
+ "nopqrstuvw"
+ "xyz~";
+ z[0] = a[(x>>18)&0x3f];
+ z[1] = a[(x>>12)&0x3f];
+ z[2] = a[(x>>6)&0x3f];
+ z[3] = a[x&0x3f];
+}
+
+/*
+** Write the zCost[] field for a amatch_word object
+*/
+static void amatchWriteCost(amatch_word *pWord){
+ amatchEncodeInt(pWord->rCost, pWord->zCost);
+ amatchEncodeInt(pWord->iSeq, pWord->zCost+4);
+ pWord->zCost[8] = 0;
+}
+
+/*
+** Add a new amatch_word object to the queue.
+**
+** If a prior amatch_word object with the same zWord, and nMatch
+** already exists, update its rCost (if the new rCost is less) but
+** otherwise leave it unchanged. Do not add a duplicate.
+**
+** Do nothing if the cost exceeds threshold.
+*/
+static void amatchAddWord(
+ amatch_cursor *pCur,
+ amatch_cost rCost,
+ int nMatch,
+ const char *zWordBase,
+ const char *zWordTail
+){
+ amatch_word *pWord;
+ amatch_avl *pNode;
+ amatch_avl *pOther;
+ int nBase, nTail;
+ char zBuf[4];
+
+ if( rCost>pCur->rLimit ){
+ return;
+ }
+ nBase = (int)strlen(zWordBase);
+ nTail = (int)strlen(zWordTail);
+ if( nBase+nTail+3>pCur->nBuf ){
+ pCur->nBuf = nBase+nTail+100;
+ pCur->zBuf = sqlite3_realloc(pCur->zBuf, pCur->nBuf);
+ if( pCur->zBuf==0 ){
+ pCur->nBuf = 0;
+ return;
+ }
+ }
+ amatchEncodeInt(nMatch, zBuf);
+ memcpy(pCur->zBuf, zBuf+2, 2);
+ memcpy(pCur->zBuf+2, zWordBase, nBase);
+ memcpy(pCur->zBuf+2+nBase, zWordTail, nTail+1);
+ pNode = amatchAvlSearch(pCur->pWord, pCur->zBuf);
+ if( pNode ){
+ pWord = pNode->pWord;
+ if( pWord->rCost>rCost ){
+#ifdef AMATCH_TRACE_1
+ printf("UPDATE [%s][%.*s^%s] %d (\"%s\" \"%s\")\n",
+ pWord->zWord+2, pWord->nMatch, pCur->zInput, pCur->zInput,
+ pWord->rCost, pWord->zWord, pWord->zCost);
+#endif
+ amatchAvlRemove(&pCur->pCost, &pWord->sCost);
+ pWord->rCost = rCost;
+ amatchWriteCost(pWord);
+#ifdef AMATCH_TRACE_1
+ printf(" ---> %d (\"%s\" \"%s\")\n",
+ pWord->rCost, pWord->zWord, pWord->zCost);
+#endif
+ pOther = amatchAvlInsert(&pCur->pCost, &pWord->sCost);
+ assert( pOther==0 ); (void)pOther;
+ }
+ return;
+ }
+ pWord = sqlite3_malloc( sizeof(*pWord) + nBase + nTail - 1 );
+ if( pWord==0 ) return;
+ memset(pWord, 0, sizeof(*pWord));
+ pWord->rCost = rCost;
+ pWord->iSeq = pCur->nWord++;
+ amatchWriteCost(pWord);
+ pWord->nMatch = nMatch;
+ pWord->pNext = pCur->pAllWords;
+ pCur->pAllWords = pWord;
+ pWord->sCost.zKey = pWord->zCost;
+ pWord->sCost.pWord = pWord;
+ pOther = amatchAvlInsert(&pCur->pCost, &pWord->sCost);
+ assert( pOther==0 ); (void)pOther;
+ pWord->sWord.zKey = pWord->zWord;
+ pWord->sWord.pWord = pWord;
+ strcpy(pWord->zWord, pCur->zBuf);
+ pOther = amatchAvlInsert(&pCur->pWord, &pWord->sWord);
+ assert( pOther==0 ); (void)pOther;
+#ifdef AMATCH_TRACE_1
+ printf("INSERT [%s][%.*s^%s] %d (\"%s\" \"%s\")\n", pWord->zWord+2,
+ pWord->nMatch, pCur->zInput, pCur->zInput+pWord->nMatch, rCost,
+ pWord->zWord, pWord->zCost);
+#endif
+}
+
+/*
+** Advance a cursor to its next row of output
+*/
+static int amatchNext(sqlite3_vtab_cursor *cur){
+ amatch_cursor *pCur = (amatch_cursor*)cur;
+ amatch_word *pWord = 0;
+ amatch_avl *pNode;
+ int isMatch = 0;
+ amatch_vtab *p = pCur->pVtab;
+ int nWord;
+ int rc;
+ int i;
+ const char *zW;
+ amatch_rule *pRule;
+ char *zBuf = 0;
+ char nBuf = 0;
+ char zNext[8];
+ char zNextIn[8];
+ int nNextIn;
+
+ if( p->pVCheck==0 ){
+ char *zSql;
+ if( p->zVocabLang && p->zVocabLang[0] ){
+ zSql = sqlite3_mprintf(
+ "SELECT \"%s\" FROM \"%s\"",
+ " WHERE \"%w\">=?1 AND \"%w\"=?2"
+ " ORDER BY 1",
+ p->zVocabWord, p->zVocabTab,
+ p->zVocabWord, p->zVocabLang
+ );
+ }else{
+ zSql = sqlite3_mprintf(
+ "SELECT \"%s\" FROM \"%s\""
+ " WHERE \"%w\">=?1"
+ " ORDER BY 1",
+ p->zVocabWord, p->zVocabTab,
+ p->zVocabWord
+ );
+ }
+ rc = sqlite3_prepare_v2(p->db, zSql, -1, &p->pVCheck, 0);
+ sqlite3_free(zSql);
+ if( rc ) return rc;
+ }
+ sqlite3_bind_int(p->pVCheck, 2, pCur->iLang);
+
+ do{
+ pNode = amatchAvlFirst(pCur->pCost);
+ if( pNode==0 ){
+ pWord = 0;
+ break;
+ }
+ pWord = pNode->pWord;
+ amatchAvlRemove(&pCur->pCost, &pWord->sCost);
+
+#ifdef AMATCH_TRACE_1
+ printf("PROCESS [%s][%.*s^%s] %d (\"%s\" \"%s\")\n",
+ pWord->zWord+2, pWord->nMatch, pCur->zInput, pCur->zInput+pWord->nMatch,
+ pWord->rCost, pWord->zWord, pWord->zCost);
+#endif
+ nWord = (int)strlen(pWord->zWord+2);
+ if( nWord+20>nBuf ){
+ nBuf = nWord+100;
+ zBuf = sqlite3_realloc(zBuf, nBuf);
+ if( zBuf==0 ) return SQLITE_NOMEM;
+ }
+ strcpy(zBuf, pWord->zWord+2);
+ zNext[0] = 0;
+ zNextIn[0] = pCur->zInput[pWord->nMatch];
+ if( zNextIn[0] ){
+ for(i=1; i<=4 && (pCur->zInput[pWord->nMatch+i]&0xc0)==0x80; i++){
+ zNextIn[i] = pCur->zInput[pWord->nMatch+i];
+ }
+ zNextIn[i] = 0;
+ nNextIn = i;
+ }else{
+ nNextIn = 0;
+ }
+
+ if( zNextIn[0] && zNextIn[0]!='*' ){
+ sqlite3_reset(p->pVCheck);
+ strcat(zBuf, zNextIn);
+ sqlite3_bind_text(p->pVCheck, 1, zBuf, nWord+nNextIn, SQLITE_STATIC);
+ rc = sqlite3_step(p->pVCheck);
+ if( rc==SQLITE_ROW ){
+ zW = (const char*)sqlite3_column_text(p->pVCheck, 0);
+ if( strncmp(zBuf, zW, nWord+nNextIn)==0 ){
+ amatchAddWord(pCur, pWord->rCost, pWord->nMatch+nNextIn, zBuf, "");
+ }
+ }
+ zBuf[nWord] = 0;
+ }
+
+ while( 1 ){
+ strcpy(zBuf+nWord, zNext);
+ sqlite3_reset(p->pVCheck);
+ sqlite3_bind_text(p->pVCheck, 1, zBuf, -1, SQLITE_TRANSIENT);
+ rc = sqlite3_step(p->pVCheck);
+ if( rc!=SQLITE_ROW ) break;
+ zW = (const char*)sqlite3_column_text(p->pVCheck, 0);
+ strcpy(zBuf+nWord, zNext);
+ if( strncmp(zW, zBuf, nWord)!=0 ) break;
+ if( (zNextIn[0]=='*' && zNextIn[1]==0)
+ || (zNextIn[0]==0 && zW[nWord]==0)
+ ){
+ isMatch = 1;
+ zNextIn[0] = 0;
+ nNextIn = 0;
+ break;
+ }
+ zNext[0] = zW[nWord];
+ for(i=1; i<=4 && (zW[nWord+i]&0xc0)==0x80; i++){
+ zNext[i] = zW[nWord+i];
+ }
+ zNext[i] = 0;
+ zBuf[nWord] = 0;
+ if( p->rIns>0 ){
+ amatchAddWord(pCur, pWord->rCost+p->rIns, pWord->nMatch,
+ zBuf, zNext);
+ }
+ if( p->rSub>0 ){
+ amatchAddWord(pCur, pWord->rCost+p->rSub, pWord->nMatch+nNextIn,
+ zBuf, zNext);
+ }
+ if( p->rIns<0 && p->rSub<0 ) break;
+ zNext[i-1]++; /* FIX ME */
+ }
+ sqlite3_reset(p->pVCheck);
+
+ if( p->rDel>0 ){
+ zBuf[nWord] = 0;
+ amatchAddWord(pCur, pWord->rCost+p->rDel, pWord->nMatch+nNextIn,
+ zBuf, "");
+ }
+
+ for(pRule=p->pRule; pRule; pRule=pRule->pNext){
+ if( pRule->iLang!=pCur->iLang ) continue;
+ if( strncmp(pRule->zFrom, pCur->zInput+pWord->nMatch, pRule->nFrom)==0 ){
+ amatchAddWord(pCur, pWord->rCost+pRule->rCost,
+ pWord->nMatch+pRule->nFrom, pWord->zWord+2, pRule->zTo);
+ }
+ }
+ }while( !isMatch );
+ pCur->pCurrent = pWord;
+ sqlite3_free(zBuf);
+ return SQLITE_OK;
+}
+
+/*
+** Called to "rewind" a cursor back to the beginning so that
+** it starts its output over again. Always called at least once
+** prior to any amatchColumn, amatchRowid, or amatchEof call.
+*/
+static int amatchFilter(
+ sqlite3_vtab_cursor *pVtabCursor,
+ int idxNum, const char *idxStr,
+ int argc, sqlite3_value **argv
+){
+ amatch_cursor *pCur = (amatch_cursor *)pVtabCursor;
+ const char *zWord = "*";
+ int idx;
+
+ amatchClearCursor(pCur);
+ idx = 0;
+ if( idxNum & 1 ){
+ zWord = (const char*)sqlite3_value_text(argv[0]);
+ idx++;
+ }
+ if( idxNum & 2 ){
+ pCur->rLimit = (amatch_cost)sqlite3_value_int(argv[idx]);
+ idx++;
+ }
+ if( idxNum & 4 ){
+ pCur->iLang = (amatch_cost)sqlite3_value_int(argv[idx]);
+ idx++;
+ }
+ pCur->zInput = sqlite3_mprintf("%s", zWord);
+ if( pCur->zInput==0 ) return SQLITE_NOMEM;
+ amatchAddWord(pCur, 0, 0, "", "");
+ amatchNext(pVtabCursor);
+
+ return SQLITE_OK;
+}
+
+/*
+** Only the word and distance columns have values. All other columns
+** return NULL
+*/
+static int amatchColumn(sqlite3_vtab_cursor *cur, sqlite3_context *ctx, int i){
+ amatch_cursor *pCur = (amatch_cursor*)cur;
+ switch( i ){
+ case AMATCH_COL_WORD: {
+ sqlite3_result_text(ctx, pCur->pCurrent->zWord+2, -1, SQLITE_STATIC);
+ break;
+ }
+ case AMATCH_COL_DISTANCE: {
+ sqlite3_result_int(ctx, pCur->pCurrent->rCost);
+ break;
+ }
+ case AMATCH_COL_LANGUAGE: {
+ sqlite3_result_int(ctx, pCur->iLang);
+ break;
+ }
+ case AMATCH_COL_NWORD: {
+ sqlite3_result_int(ctx, pCur->nWord);
+ break;
+ }
+ default: {
+ sqlite3_result_null(ctx);
+ break;
+ }
+ }
+ return SQLITE_OK;
+}
+
+/*
+** The rowid.
+*/
+static int amatchRowid(sqlite3_vtab_cursor *cur, sqlite_int64 *pRowid){
+ amatch_cursor *pCur = (amatch_cursor*)cur;
+ *pRowid = pCur->iRowid;
+ return SQLITE_OK;
+}
+
+/*
+** EOF indicator
+*/
+static int amatchEof(sqlite3_vtab_cursor *cur){
+ amatch_cursor *pCur = (amatch_cursor*)cur;
+ return pCur->pCurrent==0;
+}
+
+/*
+** Search for terms of these forms:
+**
+** (A) word MATCH $str
+** (B1) distance < $value
+** (B2) distance <= $value
+** (C) language == $language
+**
+** The distance< and distance<= are both treated as distance<=.
+** The query plan number is a bit vector:
+**
+** bit 1: Term of the form (A) found
+** bit 2: Term like (B1) or (B2) found
+** bit 3: Term like (C) found
+**
+** If bit-1 is set, $str is always in filter.argv[0]. If bit-2 is set
+** then $value is in filter.argv[0] if bit-1 is clear and is in
+** filter.argv[1] if bit-1 is set. If bit-3 is set, then $ruleid is
+** in filter.argv[0] if bit-1 and bit-2 are both zero, is in
+** filter.argv[1] if exactly one of bit-1 and bit-2 are set, and is in
+** filter.argv[2] if both bit-1 and bit-2 are set.
+*/
+static int amatchBestIndex(
+ sqlite3_vtab *tab,
+ sqlite3_index_info *pIdxInfo
+){
+ int iPlan = 0;
+ int iDistTerm = -1;
+ int iLangTerm = -1;
+ int i;
+ const struct sqlite3_index_constraint *pConstraint;
+
+ (void)tab;
+ pConstraint = pIdxInfo->aConstraint;
+ for(i=0; i<pIdxInfo->nConstraint; i++, pConstraint++){
+ if( pConstraint->usable==0 ) continue;
+ if( (iPlan & 1)==0
+ && pConstraint->iColumn==0
+ && pConstraint->op==SQLITE_INDEX_CONSTRAINT_MATCH
+ ){
+ iPlan |= 1;
+ pIdxInfo->aConstraintUsage[i].argvIndex = 1;
+ pIdxInfo->aConstraintUsage[i].omit = 1;
+ }
+ if( (iPlan & 2)==0
+ && pConstraint->iColumn==1
+ && (pConstraint->op==SQLITE_INDEX_CONSTRAINT_LT
+ || pConstraint->op==SQLITE_INDEX_CONSTRAINT_LE)
+ ){
+ iPlan |= 2;
+ iDistTerm = i;
+ }
+ if( (iPlan & 4)==0
+ && pConstraint->iColumn==2
+ && pConstraint->op==SQLITE_INDEX_CONSTRAINT_EQ
+ ){
+ iPlan |= 4;
+ pIdxInfo->aConstraintUsage[i].omit = 1;
+ iLangTerm = i;
+ }
+ }
+ if( iPlan & 2 ){
+ pIdxInfo->aConstraintUsage[iDistTerm].argvIndex = 1+((iPlan&1)!=0);
+ }
+ if( iPlan & 4 ){
+ int idx = 1;
+ if( iPlan & 1 ) idx++;
+ if( iPlan & 2 ) idx++;
+ pIdxInfo->aConstraintUsage[iLangTerm].argvIndex = idx;
+ }
+ pIdxInfo->idxNum = iPlan;
+ if( pIdxInfo->nOrderBy==1
+ && pIdxInfo->aOrderBy[0].iColumn==1
+ && pIdxInfo->aOrderBy[0].desc==0
+ ){
+ pIdxInfo->orderByConsumed = 1;
+ }
+ pIdxInfo->estimatedCost = (double)10000;
+
+ return SQLITE_OK;
+}
+
+/*
+** The xUpdate() method.
+**
+** This implementation disallows DELETE and UPDATE. The only thing
+** allowed is INSERT into the "command" column.
+*/
+static int amatchUpdate(
+ sqlite3_vtab *pVTab,
+ int argc,
+ sqlite3_value **argv,
+ sqlite_int64 *pRowid
+){
+ amatch_vtab *p = (amatch_vtab*)pVTab;
+ const unsigned char *zCmd;
+ (void)pRowid;
+ if( argc==1 ){
+ pVTab->zErrMsg = sqlite3_mprintf("DELETE from %s is not allowed",
+ p->zSelf);
+ return SQLITE_ERROR;
+ }
+ if( sqlite3_value_type(argv[0])!=SQLITE_NULL ){
+ pVTab->zErrMsg = sqlite3_mprintf("UPDATE of %s is not allowed",
+ p->zSelf);
+ return SQLITE_ERROR;
+ }
+ if( sqlite3_value_type(argv[2+AMATCH_COL_WORD])!=SQLITE_NULL
+ || sqlite3_value_type(argv[2+AMATCH_COL_DISTANCE])!=SQLITE_NULL
+ || sqlite3_value_type(argv[2+AMATCH_COL_LANGUAGE])!=SQLITE_NULL
+ ){
+ pVTab->zErrMsg = sqlite3_mprintf(
+ "INSERT INTO %s allowed for column [command] only", p->zSelf);
+ return SQLITE_ERROR;
+ }
+ zCmd = sqlite3_value_text(argv[2+AMATCH_COL_COMMAND]);
+ if( zCmd==0 ) return SQLITE_OK;
+
+ return SQLITE_OK;
+}
+
+/*
+** A virtual table module that implements the "approximate_match".
+*/
+static sqlite3_module amatchModule = {
+ 0, /* iVersion */
+ amatchConnect, /* xCreate */
+ amatchConnect, /* xConnect */
+ amatchBestIndex, /* xBestIndex */
+ amatchDisconnect, /* xDisconnect */
+ amatchDisconnect, /* xDestroy */
+ amatchOpen, /* xOpen - open a cursor */
+ amatchClose, /* xClose - close a cursor */
+ amatchFilter, /* xFilter - configure scan constraints */
+ amatchNext, /* xNext - advance a cursor */
+ amatchEof, /* xEof - check for end of scan */
+ amatchColumn, /* xColumn - read data */
+ amatchRowid, /* xRowid - read data */
+ amatchUpdate, /* xUpdate */
+ 0, /* xBegin */
+ 0, /* xSync */
+ 0, /* xCommit */
+ 0, /* xRollback */
+ 0, /* xFindMethod */
+ 0, /* xRename */
+ 0, /* xSavepoint */
+ 0, /* xRelease */
+ 0 /* xRollbackTo */
+};
+
+/*
+** Register the amatch virtual table
+*/
+#ifdef _WIN32
+__declspec(dllexport)
+#endif
+int sqlite3_amatch_init(
+ sqlite3 *db,
+ char **pzErrMsg,
+ const sqlite3_api_routines *pApi
+){
+ int rc = SQLITE_OK;
+ SQLITE_EXTENSION_INIT2(pApi);
+ (void)pzErrMsg; /* Not used */
+ rc = sqlite3_create_module(db, "approximate_match", &amatchModule, 0);
+ return rc;
+}
diff --git a/ext/misc/closure.c b/ext/misc/closure.c
new file mode 100644
index 0000000..7b3d064
--- /dev/null
+++ b/ext/misc/closure.c
@@ -0,0 +1,942 @@
+/*
+** 2013-04-16
+**
+** 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 file contains code for a virtual table that finds the transitive
+** closure of a parent/child relationship in a real table. The virtual
+** table is called "transitive_closure".
+**
+** A transitive_closure virtual table is created like this:
+**
+** CREATE VIRTUAL TABLE x USING transitive_closure(
+** tablename=<tablename>, -- T
+** idcolumn=<columnname>, -- X
+** parentcolumn=<columnname> -- P
+** );
+**
+** When it is created, the new transitive_closure table may be supplied
+** with default values for the name of a table T and columns T.X and T.P.
+** The T.X and T.P columns must contain integers. The ideal case is for
+** T.X to be the INTEGER PRIMARY KEY. The T.P column should reference
+** the T.X column. The row referenced by T.P is the parent of the current row.
+**
+** The tablename, idcolumn, and parentcolumn supplied by the CREATE VIRTUAL
+** TABLE statement may be overridden in individual queries by including
+** terms like tablename='newtable', idcolumn='id2', or
+** parentcolumn='parent3' in the WHERE clause of the query.
+**
+** For efficiency, it is essential that there be an index on the P column:
+**
+** CREATE Tidx1 ON T(P)
+**
+** Suppose a specific instance of the closure table is as follows:
+**
+** CREATE VIRTUAL TABLE ct1 USING transitive_closure(
+** tablename='group',
+** idcolumn='groupId',
+** parentcolumn='parentId'
+** );
+**
+** Such an instance of the transitive_closure virtual table would be
+** appropriate for walking a tree defined using a table like this, for example:
+**
+** CREATE TABLE group(
+** groupId INTEGER PRIMARY KEY,
+** parentId INTEGER REFERENCES group
+** );
+** CREATE INDEX group_idx1 ON group(parentId);
+**
+** The group table above would presumably have other application-specific
+** fields. The key point here is that rows of the group table form a
+** tree. The purpose of the ct1 virtual table is to easily extract
+** branches of that tree.
+**
+** Once it has been created, the ct1 virtual table can be queried
+** as follows:
+**
+** SELECT * FROM element
+** WHERE element.groupId IN (SELECT id FROM ct1 WHERE root=?1);
+**
+** The above query will return all elements that are part of group ?1
+** or children of group ?1 or grand-children of ?1 and so forth for all
+** descendents of group ?1. The same query can be formulated as a join:
+**
+** SELECT element.* FROM element, ct1
+** WHERE element.groupid=ct1.id
+** AND ct1.root=?1;
+**
+** The depth of the transitive_closure (the number of generations of
+** parent/child relations to follow) can be limited by setting "depth"
+** column in the WHERE clause. So, for example, the following query
+** finds only children and grandchildren but no further descendents:
+**
+** SELECT element.* FROM element, ct1
+** WHERE element.groupid=ct1.id
+** AND ct1.root=?1
+** AND ct1.depth<=2;
+**
+** The "ct1.depth<=2" term could be a strict equality "ct1.depth=2" in
+** order to find only the grandchildren of ?1, not ?1 itself or the
+** children of ?1.
+**
+** The root=?1 term must be supplied in WHERE clause or else the query
+** of the ct1 virtual table will return an empty set. The tablename,
+** idcolumn, and parentcolumn attributes can be overridden in the WHERE
+** clause if desired. So, for example, the ct1 table could be repurposed
+** to find ancestors rather than descendents by inverting the roles of
+** the idcolumn and parentcolumn:
+**
+** SELECT element.* FROM element, ct1
+** WHERE element.groupid=ct1.id
+** AND ct1.root=?1
+** AND ct1.idcolumn='parentId'
+** AND ct1.parentcolumn='groupId';
+**
+** Multiple calls to ct1 could be combined. For example, the following
+** query finds all elements that "cousins" of groupId ?1. That is to say
+** elements where the groupId is a grandchild of the grandparent of ?1.
+** (This definition of "cousins" also includes siblings and self.)
+**
+** SELECT element.* FROM element, ct1
+** WHERE element.groupId=ct1.id
+** AND ct1.depth=2
+** AND ct1.root IN (SELECT id FROM ct1
+** WHERE root=?1
+** AND depth=2
+** AND idcolumn='parentId'
+** AND parentcolumn='groupId');
+**
+** In our example, the group.groupId column is unique and thus the
+** subquery will return exactly one row. For that reason, the IN
+** operator could be replaced by "=" to get the same result. But
+** in the general case where the idcolumn is not unique, an IN operator
+** would be required for this kind of query.
+**
+** Note that because the tablename, idcolumn, and parentcolumn can
+** all be specified in the query, it is possible for an application
+** to define a single transitive_closure virtual table for use on lots
+** of different hierarchy tables. One might say:
+**
+** CREATE VIRTUAL TABLE temp.closure USING transitive_closure;
+**
+** As each database connection is being opened. Then the application
+** would always have a "closure" virtual table handy to use for querying.
+**
+** SELECT element.* FROM element, closure
+** WHERE element.groupid=ct1.id
+** AND closure.root=?1
+** AND closure.tablename='group'
+** AND closure.idname='groupId'
+** AND closure.parentname='parentId';
+**
+** See the documentation at http://www.sqlite.org/loadext.html for information
+** on how to compile and use loadable extensions such as this one.
+*/
+#include "sqlite3ext.h"
+SQLITE_EXTENSION_INIT1
+#include <stdlib.h>
+#include <string.h>
+#include <assert.h>
+#include <stdio.h>
+#include <ctype.h>
+
+/*
+** Forward declaration of objects used by this implementation
+*/
+typedef struct closure_vtab closure_vtab;
+typedef struct closure_cursor closure_cursor;
+typedef struct closure_queue closure_queue;
+typedef struct closure_avl closure_avl;
+
+/*****************************************************************************
+** AVL Tree implementation
+*/
+/*
+** Objects that want to be members of the AVL tree should embedded an
+** instance of this structure.
+*/
+struct closure_avl {
+ sqlite3_int64 id; /* Id of this entry in the table */
+ int iGeneration; /* Which generation is this entry part of */
+ closure_avl *pList; /* A linked list of nodes */
+ closure_avl *pBefore; /* Other elements less than id */
+ closure_avl *pAfter; /* Other elements greater than id */
+ closure_avl *pUp; /* Parent element */
+ short int height; /* Height of this node. Leaf==1 */
+ short int imbalance; /* Height difference between pBefore and pAfter */
+};
+
+/* Recompute the closure_avl.height and closure_avl.imbalance fields for p.
+** Assume that the children of p have correct heights.
+*/
+static void closureAvlRecomputeHeight(closure_avl *p){
+ short int hBefore = p->pBefore ? p->pBefore->height : 0;
+ short int hAfter = p->pAfter ? p->pAfter->height : 0;
+ p->imbalance = hBefore - hAfter; /* -: pAfter higher. +: pBefore higher */
+ p->height = (hBefore>hAfter ? hBefore : hAfter)+1;
+}
+
+/*
+** P B
+** / \ / \
+** B Z ==> X P
+** / \ / \
+** X Y Y Z
+**
+*/
+static closure_avl *closureAvlRotateBefore(closure_avl *pP){
+ closure_avl *pB = pP->pBefore;
+ closure_avl *pY = pB->pAfter;
+ pB->pUp = pP->pUp;
+ pB->pAfter = pP;
+ pP->pUp = pB;
+ pP->pBefore = pY;
+ if( pY ) pY->pUp = pP;
+ closureAvlRecomputeHeight(pP);
+ closureAvlRecomputeHeight(pB);
+ return pB;
+}
+
+/*
+** P A
+** / \ / \
+** X A ==> P Z
+** / \ / \
+** Y Z X Y
+**
+*/
+static closure_avl *closureAvlRotateAfter(closure_avl *pP){
+ closure_avl *pA = pP->pAfter;
+ closure_avl *pY = pA->pBefore;
+ pA->pUp = pP->pUp;
+ pA->pBefore = pP;
+ pP->pUp = pA;
+ pP->pAfter = pY;
+ if( pY ) pY->pUp = pP;
+ closureAvlRecomputeHeight(pP);
+ closureAvlRecomputeHeight(pA);
+ return pA;
+}
+
+/*
+** Return a pointer to the pBefore or pAfter pointer in the parent
+** of p that points to p. Or if p is the root node, return pp.
+*/
+static closure_avl **closureAvlFromPtr(closure_avl *p, closure_avl **pp){
+ closure_avl *pUp = p->pUp;
+ if( pUp==0 ) return pp;
+ if( pUp->pAfter==p ) return &pUp->pAfter;
+ return &pUp->pBefore;
+}
+
+/*
+** Rebalance all nodes starting with p and working up to the root.
+** Return the new root.
+*/
+static closure_avl *closureAvlBalance(closure_avl *p){
+ closure_avl *pTop = p;
+ closure_avl **pp;
+ while( p ){
+ closureAvlRecomputeHeight(p);
+ if( p->imbalance>=2 ){
+ closure_avl *pB = p->pBefore;
+ if( pB->imbalance<0 ) p->pBefore = closureAvlRotateAfter(pB);
+ pp = closureAvlFromPtr(p,&p);
+ p = *pp = closureAvlRotateBefore(p);
+ }else if( p->imbalance<=(-2) ){
+ closure_avl *pA = p->pAfter;
+ if( pA->imbalance>0 ) p->pAfter = closureAvlRotateBefore(pA);
+ pp = closureAvlFromPtr(p,&p);
+ p = *pp = closureAvlRotateAfter(p);
+ }
+ pTop = p;
+ p = p->pUp;
+ }
+ return pTop;
+}
+
+/* Search the tree rooted at p for an entry with id. Return a pointer
+** to the entry or return NULL.
+*/
+static closure_avl *closureAvlSearch(closure_avl *p, sqlite3_int64 id){
+ while( p && id!=p->id ){
+ p = (id<p->id) ? p->pBefore : p->pAfter;
+ }
+ return p;
+}
+
+/* Find the first node (the one with the smallest key).
+*/
+static closure_avl *closureAvlFirst(closure_avl *p){
+ if( p ) while( p->pBefore ) p = p->pBefore;
+ return p;
+}
+
+/* Return the node with the next larger key after p.
+*/
+closure_avl *closureAvlNext(closure_avl *p){
+ closure_avl *pPrev = 0;
+ while( p && p->pAfter==pPrev ){
+ pPrev = p;
+ p = p->pUp;
+ }
+ if( p && pPrev==0 ){
+ p = closureAvlFirst(p->pAfter);
+ }
+ return p;
+}
+
+/* Insert a new node pNew. Return NULL on success. If the key is not
+** unique, then do not perform the insert but instead leave pNew unchanged
+** and return a pointer to an existing node with the same key.
+*/
+static closure_avl *closureAvlInsert(
+ closure_avl **ppHead, /* Head of the tree */
+ closure_avl *pNew /* New node to be inserted */
+){
+ closure_avl *p = *ppHead;
+ if( p==0 ){
+ p = pNew;
+ pNew->pUp = 0;
+ }else{
+ while( p ){
+ if( pNew->id<p->id ){
+ if( p->pBefore ){
+ p = p->pBefore;
+ }else{
+ p->pBefore = pNew;
+ pNew->pUp = p;
+ break;
+ }
+ }else if( pNew->id>p->id ){
+ if( p->pAfter ){
+ p = p->pAfter;
+ }else{
+ p->pAfter = pNew;
+ pNew->pUp = p;
+ break;
+ }
+ }else{
+ return p;
+ }
+ }
+ }
+ pNew->pBefore = 0;
+ pNew->pAfter = 0;
+ pNew->height = 1;
+ pNew->imbalance = 0;
+ *ppHead = closureAvlBalance(p);
+ return 0;
+}
+
+/* Walk the tree can call xDestroy on each node
+*/
+static void closureAvlDestroy(closure_avl *p, void (*xDestroy)(closure_avl*)){
+ if( p ){
+ closureAvlDestroy(p->pBefore, xDestroy);
+ closureAvlDestroy(p->pAfter, xDestroy);
+ xDestroy(p);
+ }
+}
+/*
+** End of the AVL Tree implementation
+******************************************************************************/
+
+/*
+** A closure virtual-table object
+*/
+struct closure_vtab {
+ sqlite3_vtab base; /* Base class - must be first */
+ char *zDb; /* Name of database. (ex: "main") */
+ char *zSelf; /* Name of this virtual table */
+ char *zTableName; /* Name of table holding parent/child relation */
+ char *zIdColumn; /* Name of ID column of zTableName */
+ char *zParentColumn; /* Name of PARENT column in zTableName */
+ sqlite3 *db; /* The database connection */
+ int nCursor; /* Number of pending cursors */
+};
+
+/* A closure cursor object */
+struct closure_cursor {
+ sqlite3_vtab_cursor base; /* Base class - must be first */
+ closure_vtab *pVtab; /* The virtual table this cursor belongs to */
+ char *zTableName; /* Name of table holding parent/child relation */
+ char *zIdColumn; /* Name of ID column of zTableName */
+ char *zParentColumn; /* Name of PARENT column in zTableName */
+ closure_avl *pCurrent; /* Current element of output */
+ closure_avl *pClosure; /* The complete closure tree */
+};
+
+/* A queue of AVL nodes */
+struct closure_queue {
+ closure_avl *pFirst; /* Oldest node on the queue */
+ closure_avl *pLast; /* Youngest node on the queue */
+};
+
+/*
+** Add a node to the end of the queue
+*/
+static void queuePush(closure_queue *pQueue, closure_avl *pNode){
+ pNode->pList = 0;
+ if( pQueue->pLast ){
+ pQueue->pLast->pList = pNode;
+ }else{
+ pQueue->pFirst = pNode;
+ }
+ pQueue->pLast = pNode;
+}
+
+/*
+** Extract the oldest element (the front element) from the queue.
+*/
+static closure_avl *queuePull(closure_queue *pQueue){
+ closure_avl *p = pQueue->pFirst;
+ if( p ){
+ pQueue->pFirst = p->pList;
+ if( pQueue->pFirst==0 ) pQueue->pLast = 0;
+ }
+ return p;
+}
+
+/*
+** This function converts an SQL quoted string into an unquoted string
+** and returns a pointer to a buffer allocated using sqlite3_malloc()
+** containing the result. The caller should eventually free this buffer
+** using sqlite3_free.
+**
+** Examples:
+**
+** "abc" becomes abc
+** 'xyz' becomes xyz
+** [pqr] becomes pqr
+** `mno` becomes mno
+*/
+static char *closureDequote(const char *zIn){
+ int nIn; /* Size of input string, in bytes */
+ char *zOut; /* Output (dequoted) string */
+
+ nIn = (int)strlen(zIn);
+ zOut = sqlite3_malloc(nIn+1);
+ if( zOut ){
+ char q = zIn[0]; /* Quote character (if any ) */
+
+ if( q!='[' && q!= '\'' && q!='"' && q!='`' ){
+ memcpy(zOut, zIn, nIn+1);
+ }else{
+ int iOut = 0; /* Index of next byte to write to output */
+ int iIn; /* Index of next byte to read from input */
+
+ if( q=='[' ) q = ']';
+ for(iIn=1; iIn<nIn; iIn++){
+ if( zIn[iIn]==q ) iIn++;
+ zOut[iOut++] = zIn[iIn];
+ }
+ }
+ assert( (int)strlen(zOut)<=nIn );
+ }
+ return zOut;
+}
+
+/*
+** Deallocate an closure_vtab object
+*/
+static void closureFree(closure_vtab *p){
+ if( p ){
+ sqlite3_free(p->zDb);
+ sqlite3_free(p->zSelf);
+ sqlite3_free(p->zTableName);
+ sqlite3_free(p->zIdColumn);
+ sqlite3_free(p->zParentColumn);
+ memset(p, 0, sizeof(*p));
+ sqlite3_free(p);
+ }
+}
+
+/*
+** xDisconnect/xDestroy method for the closure module.
+*/
+static int closureDisconnect(sqlite3_vtab *pVtab){
+ closure_vtab *p = (closure_vtab*)pVtab;
+ assert( p->nCursor==0 );
+ closureFree(p);
+ return SQLITE_OK;
+}
+
+/*
+** Check to see if the argument is of the form:
+**
+** KEY = VALUE
+**
+** If it is, return a pointer to the first character of VALUE.
+** If not, return NULL. Spaces around the = are ignored.
+*/
+static const char *closureValueOfKey(const char *zKey, const char *zStr){
+ int nKey = (int)strlen(zKey);
+ int nStr = (int)strlen(zStr);
+ int i;
+ if( nStr<nKey+1 ) return 0;
+ if( memcmp(zStr, zKey, nKey)!=0 ) return 0;
+ for(i=nKey; isspace(zStr[i]); i++){}
+ if( zStr[i]!='=' ) return 0;
+ i++;
+ while( isspace(zStr[i]) ){ i++; }
+ return zStr+i;
+}
+
+/*
+** xConnect/xCreate method for the closure module. Arguments are:
+**
+** argv[0] -> module name ("approximate_match")
+** argv[1] -> database name
+** argv[2] -> table name
+** argv[3...] -> arguments
+*/
+static int closureConnect(
+ sqlite3 *db,
+ void *pAux,
+ int argc, const char *const*argv,
+ sqlite3_vtab **ppVtab,
+ char **pzErr
+){
+ int rc = SQLITE_OK; /* Return code */
+ closure_vtab *pNew = 0; /* New virtual table */
+ const char *zDb = argv[1];
+ const char *zVal;
+ int i;
+
+ (void)pAux;
+ *ppVtab = 0;
+ pNew = sqlite3_malloc( sizeof(*pNew) );
+ if( pNew==0 ) return SQLITE_NOMEM;
+ rc = SQLITE_NOMEM;
+ memset(pNew, 0, sizeof(*pNew));
+ pNew->db = db;
+ pNew->zDb = sqlite3_mprintf("%s", zDb);
+ if( pNew->zDb==0 ) goto closureConnectError;
+ pNew->zSelf = sqlite3_mprintf("%s", argv[2]);
+ if( pNew->zSelf==0 ) goto closureConnectError;
+ for(i=3; i<argc; i++){
+ zVal = closureValueOfKey("tablename", argv[i]);
+ if( zVal ){
+ sqlite3_free(pNew->zTableName);
+ pNew->zTableName = closureDequote(zVal);
+ if( pNew->zTableName==0 ) goto closureConnectError;
+ continue;
+ }
+ zVal = closureValueOfKey("idcolumn", argv[i]);
+ if( zVal ){
+ sqlite3_free(pNew->zIdColumn);
+ pNew->zIdColumn = closureDequote(zVal);
+ if( pNew->zIdColumn==0 ) goto closureConnectError;
+ continue;
+ }
+ zVal = closureValueOfKey("parentcolumn", argv[i]);
+ if( zVal ){
+ sqlite3_free(pNew->zParentColumn);
+ pNew->zParentColumn = closureDequote(zVal);
+ if( pNew->zParentColumn==0 ) goto closureConnectError;
+ continue;
+ }
+ *pzErr = sqlite3_mprintf("unrecognized argument: [%s]\n", argv[i]);
+ closureFree(pNew);
+ *ppVtab = 0;
+ return SQLITE_ERROR;
+ }
+ rc = sqlite3_declare_vtab(db,
+ "CREATE TABLE x(id,depth,root HIDDEN,tablename HIDDEN,"
+ "idcolumn HIDDEN,parentcolumn HIDDEN)"
+ );
+#define CLOSURE_COL_ID 0
+#define CLOSURE_COL_DEPTH 1
+#define CLOSURE_COL_ROOT 2
+#define CLOSURE_COL_TABLENAME 3
+#define CLOSURE_COL_IDCOLUMN 4
+#define CLOSURE_COL_PARENTCOLUMN 5
+ if( rc!=SQLITE_OK ){
+ closureFree(pNew);
+ }
+ *ppVtab = &pNew->base;
+ return rc;
+
+closureConnectError:
+ closureFree(pNew);
+ return rc;
+}
+
+/*
+** Open a new closure cursor.
+*/
+static int closureOpen(sqlite3_vtab *pVTab, sqlite3_vtab_cursor **ppCursor){
+ closure_vtab *p = (closure_vtab*)pVTab;
+ closure_cursor *pCur;
+ pCur = sqlite3_malloc( sizeof(*pCur) );
+ if( pCur==0 ) return SQLITE_NOMEM;
+ memset(pCur, 0, sizeof(*pCur));
+ pCur->pVtab = p;
+ *ppCursor = &pCur->base;
+ p->nCursor++;
+ return SQLITE_OK;
+}
+
+/*
+** Free up all the memory allocated by a cursor. Set it rLimit to 0
+** to indicate that it is at EOF.
+*/
+static void closureClearCursor(closure_cursor *pCur){
+ closureAvlDestroy(pCur->pClosure, (void(*)(closure_avl*))sqlite3_free);
+ sqlite3_free(pCur->zTableName);
+ sqlite3_free(pCur->zIdColumn);
+ sqlite3_free(pCur->zParentColumn);
+ pCur->zTableName = 0;
+ pCur->zIdColumn = 0;
+ pCur->zParentColumn = 0;
+ pCur->pCurrent = 0;
+ pCur->pClosure = 0;
+}
+
+/*
+** Close a closure cursor.
+*/
+static int closureClose(sqlite3_vtab_cursor *cur){
+ closure_cursor *pCur = (closure_cursor *)cur;
+ closureClearCursor(pCur);
+ pCur->pVtab->nCursor--;
+ sqlite3_free(pCur);
+ return SQLITE_OK;
+}
+
+/*
+** Advance a cursor to its next row of output
+*/
+static int closureNext(sqlite3_vtab_cursor *cur){
+ closure_cursor *pCur = (closure_cursor*)cur;
+ pCur->pCurrent = closureAvlNext(pCur->pCurrent);
+ return SQLITE_OK;
+}
+
+/*
+** Allocate and insert a node
+*/
+static int closureInsertNode(
+ closure_queue *pQueue, /* Add new node to this queue */
+ closure_cursor *pCur, /* The cursor into which to add the node */
+ sqlite3_int64 id, /* The node ID */
+ int iGeneration /* The generation number for this node */
+){
+ closure_avl *pNew = sqlite3_malloc( sizeof(*pNew) );
+ if( pNew==0 ) return SQLITE_NOMEM;
+ memset(pNew, 0, sizeof(*pNew));
+ pNew->id = id;
+ pNew->iGeneration = iGeneration;
+ closureAvlInsert(&pCur->pClosure, pNew);
+ queuePush(pQueue, pNew);
+ return SQLITE_OK;
+}
+
+/*
+** Called to "rewind" a cursor back to the beginning so that
+** it starts its output over again. Always called at least once
+** prior to any closureColumn, closureRowid, or closureEof call.
+**
+** This routine actually computes the closure.
+**
+** See the comment at the beginning of closureBestIndex() for a
+** description of the meaning of idxNum. The idxStr parameter is
+** not used.
+*/
+static int closureFilter(
+ sqlite3_vtab_cursor *pVtabCursor,
+ int idxNum, const char *idxStr,
+ int argc, sqlite3_value **argv
+){
+ closure_cursor *pCur = (closure_cursor *)pVtabCursor;
+ closure_vtab *pVtab = pCur->pVtab;
+ sqlite3_int64 iRoot;
+ int mxGen = 999999999;
+ char *zSql;
+ sqlite3_stmt *pStmt;
+ closure_avl *pAvl;
+ int rc = SQLITE_OK;
+ const char *zTableName = pVtab->zTableName;
+ const char *zIdColumn = pVtab->zIdColumn;
+ const char *zParentColumn = pVtab->zParentColumn;
+ closure_queue sQueue;
+
+ (void)idxStr; /* Unused parameter */
+ (void)argc; /* Unused parameter */
+ closureClearCursor(pCur);
+ memset(&sQueue, 0, sizeof(sQueue));
+ if( (idxNum & 1)==0 ){
+ /* No root=$root in the WHERE clause. Return an empty set */
+ return SQLITE_OK;
+ }
+ iRoot = sqlite3_value_int64(argv[0]);
+ if( (idxNum & 0x000f0)!=0 ){
+ mxGen = sqlite3_value_int(argv[(idxNum>>4)&0x0f]);
+ if( (idxNum & 0x00002)!=0 ) mxGen--;
+ }
+ if( (idxNum & 0x00f00)!=0 ){
+ zTableName = (const char*)sqlite3_value_text(argv[(idxNum>>8)&0x0f]);
+ pCur->zTableName = sqlite3_mprintf("%s", zTableName);
+ }
+ if( (idxNum & 0x0f000)!=0 ){
+ zIdColumn = (const char*)sqlite3_value_text(argv[(idxNum>>12)&0x0f]);
+ pCur->zIdColumn = sqlite3_mprintf("%s", zIdColumn);
+ }
+ if( (idxNum & 0x0f0000)!=0 ){
+ zParentColumn = (const char*)sqlite3_value_text(argv[(idxNum>>16)&0x0f]);
+ pCur->zParentColumn = sqlite3_mprintf("%s", zParentColumn);
+ }
+
+ zSql = sqlite3_mprintf(
+ "SELECT \"%w\".\"%w\" FROM \"%w\" WHERE \"%w\".\"%w\"=?1",
+ zTableName, zIdColumn, zTableName, zTableName, zParentColumn);
+ if( zSql==0 ){
+ return SQLITE_NOMEM;
+ }else{
+ rc = sqlite3_prepare_v2(pVtab->db, zSql, -1, &pStmt, 0);
+ sqlite3_free(zSql);
+ if( rc ){
+ sqlite3_free(pVtab->base.zErrMsg);
+ pVtab->base.zErrMsg = sqlite3_mprintf("%s", sqlite3_errmsg(pVtab->db));
+ return rc;
+ }
+ }
+ if( rc==SQLITE_OK ){
+ rc = closureInsertNode(&sQueue, pCur, iRoot, 0);
+ }
+ while( (pAvl = queuePull(&sQueue))!=0 ){
+ if( pAvl->iGeneration>=mxGen ) continue;
+ sqlite3_bind_int64(pStmt, 1, pAvl->id);
+ while( rc==SQLITE_OK && sqlite3_step(pStmt)==SQLITE_ROW ){
+ if( sqlite3_column_type(pStmt,0)==SQLITE_INTEGER ){
+ sqlite3_int64 iNew = sqlite3_column_int64(pStmt, 0);
+ if( closureAvlSearch(pCur->pClosure, iNew)==0 ){
+ rc = closureInsertNode(&sQueue, pCur, iNew, pAvl->iGeneration+1);
+ }
+ }
+ }
+ sqlite3_reset(pStmt);
+ }
+ sqlite3_finalize(pStmt);
+ if( rc==SQLITE_OK ){
+ pCur->pCurrent = closureAvlFirst(pCur->pClosure);
+ }
+
+ return rc;
+}
+
+/*
+** Only the word and distance columns have values. All other columns
+** return NULL
+*/
+static int closureColumn(sqlite3_vtab_cursor *cur, sqlite3_context *ctx, int i){
+ closure_cursor *pCur = (closure_cursor*)cur;
+ switch( i ){
+ case CLOSURE_COL_ID: {
+ sqlite3_result_int64(ctx, pCur->pCurrent->id);
+ break;
+ }
+ case CLOSURE_COL_DEPTH: {
+ sqlite3_result_int(ctx, pCur->pCurrent->iGeneration);
+ break;
+ }
+ case CLOSURE_COL_ROOT: {
+ sqlite3_result_null(ctx);
+ break;
+ }
+ case CLOSURE_COL_TABLENAME: {
+ sqlite3_result_text(ctx,
+ pCur->zTableName ? pCur->zTableName : pCur->pVtab->zTableName,
+ -1, SQLITE_TRANSIENT);
+ break;
+ }
+ case CLOSURE_COL_IDCOLUMN: {
+ sqlite3_result_text(ctx,
+ pCur->zIdColumn ? pCur->zIdColumn : pCur->pVtab->zIdColumn,
+ -1, SQLITE_TRANSIENT);
+ break;
+ }
+ case CLOSURE_COL_PARENTCOLUMN: {
+ sqlite3_result_text(ctx,
+ pCur->zParentColumn ? pCur->zParentColumn : pCur->pVtab->zParentColumn,
+ -1, SQLITE_TRANSIENT);
+ break;
+ }
+ }
+ return SQLITE_OK;
+}
+
+/*
+** The rowid. For the closure table, this is the same as the "id" column.
+*/
+static int closureRowid(sqlite3_vtab_cursor *cur, sqlite_int64 *pRowid){
+ closure_cursor *pCur = (closure_cursor*)cur;
+ *pRowid = pCur->pCurrent->id;
+ return SQLITE_OK;
+}
+
+/*
+** EOF indicator
+*/
+static int closureEof(sqlite3_vtab_cursor *cur){
+ closure_cursor *pCur = (closure_cursor*)cur;
+ return pCur->pCurrent==0;
+}
+
+/*
+** Search for terms of these forms:
+**
+** (A) root = $root
+** (B1) depth < $depth
+** (B2) depth <= $depth
+** (B3) depth = $depth
+** (C) tablename = $tablename
+** (D) idcolumn = $idcolumn
+** (E) parentcolumn = $parentcolumn
+**
+**
+**
+** idxNum meaning
+** ---------- ------------------------------------------------------
+** 0x00000001 Term of the form (A) found
+** 0x00000002 The term of bit-2 is like (B1)
+** 0x000000f0 Index in filter.argv[] of $depth. 0 if not used.
+** 0x00000f00 Index in filter.argv[] of $tablename. 0 if not used.
+** 0x0000f000 Index in filter.argv[] of $idcolumn. 0 if not used
+** 0x000f0000 Index in filter.argv[] of $parentcolumn. 0 if not used.
+**
+** There must be a term of type (A). If there is not, then the index type
+** is 0 and the query will return an empty set.
+*/
+static int closureBestIndex(
+ sqlite3_vtab *pTab, /* The virtual table */
+ sqlite3_index_info *pIdxInfo /* Information about the query */
+){
+ int iPlan = 0;
+ int i;
+ int idx = 1;
+ const struct sqlite3_index_constraint *pConstraint;
+ closure_vtab *pVtab = (closure_vtab*)pTab;
+
+ pConstraint = pIdxInfo->aConstraint;
+ for(i=0; i<pIdxInfo->nConstraint; i++, pConstraint++){
+ if( pConstraint->usable==0 ) continue;
+ if( (iPlan & 1)==0
+ && pConstraint->iColumn==CLOSURE_COL_ROOT
+ && pConstraint->op==SQLITE_INDEX_CONSTRAINT_EQ
+ ){
+ iPlan |= 1;
+ pIdxInfo->aConstraintUsage[i].argvIndex = 1;
+ pIdxInfo->aConstraintUsage[i].omit = 1;
+ }
+ if( (iPlan & 0x0000f0)==0
+ && pConstraint->iColumn==CLOSURE_COL_DEPTH
+ && (pConstraint->op==SQLITE_INDEX_CONSTRAINT_LT
+ || pConstraint->op==SQLITE_INDEX_CONSTRAINT_LE
+ || pConstraint->op==SQLITE_INDEX_CONSTRAINT_EQ)
+ ){
+ iPlan |= idx<<4;
+ pIdxInfo->aConstraintUsage[i].argvIndex = ++idx;
+ if( pConstraint->op==SQLITE_INDEX_CONSTRAINT_LT ) iPlan |= 0x000002;
+ }
+ if( (iPlan & 0x000f00)==0
+ && pConstraint->iColumn==CLOSURE_COL_TABLENAME
+ && pConstraint->op==SQLITE_INDEX_CONSTRAINT_EQ
+ ){
+ iPlan |= idx<<8;
+ pIdxInfo->aConstraintUsage[i].argvIndex = ++idx;
+ pIdxInfo->aConstraintUsage[i].omit = 1;
+ }
+ if( (iPlan & 0x00f000)==0
+ && pConstraint->iColumn==CLOSURE_COL_IDCOLUMN
+ && pConstraint->op==SQLITE_INDEX_CONSTRAINT_EQ
+ ){
+ iPlan |= idx<<12;
+ pIdxInfo->aConstraintUsage[i].argvIndex = ++idx;
+ pIdxInfo->aConstraintUsage[i].omit = 1;
+ }
+ if( (iPlan & 0x0f0000)==0
+ && pConstraint->iColumn==CLOSURE_COL_PARENTCOLUMN
+ && pConstraint->op==SQLITE_INDEX_CONSTRAINT_EQ
+ ){
+ iPlan |= idx<<16;
+ pIdxInfo->aConstraintUsage[i].argvIndex = ++idx;
+ pIdxInfo->aConstraintUsage[i].omit = 1;
+ }
+ }
+ if( (pVtab->zTableName==0 && (iPlan & 0x000f00)==0)
+ || (pVtab->zIdColumn==0 && (iPlan & 0x00f000)==0)
+ || (pVtab->zParentColumn==0 && (iPlan & 0x0f0000)==0)
+ ){
+ /* All of tablename, idcolumn, and parentcolumn must be specified
+ ** in either the CREATE VIRTUAL TABLE or in the WHERE clause constraints
+ ** or else the result is an empty set. */
+ iPlan = 0;
+ }
+ pIdxInfo->idxNum = iPlan;
+ if( pIdxInfo->nOrderBy==1
+ && pIdxInfo->aOrderBy[0].iColumn==CLOSURE_COL_ID
+ && pIdxInfo->aOrderBy[0].desc==0
+ ){
+ pIdxInfo->orderByConsumed = 1;
+ }
+ pIdxInfo->estimatedCost = (double)10000;
+
+ return SQLITE_OK;
+}
+
+/*
+** A virtual table module that implements the "approximate_match".
+*/
+static sqlite3_module closureModule = {
+ 0, /* iVersion */
+ closureConnect, /* xCreate */
+ closureConnect, /* xConnect */
+ closureBestIndex, /* xBestIndex */
+ closureDisconnect, /* xDisconnect */
+ closureDisconnect, /* xDestroy */
+ closureOpen, /* xOpen - open a cursor */
+ closureClose, /* xClose - close a cursor */
+ closureFilter, /* xFilter - configure scan constraints */
+ closureNext, /* xNext - advance a cursor */
+ closureEof, /* xEof - check for end of scan */
+ closureColumn, /* xColumn - read data */
+ closureRowid, /* xRowid - read data */
+ 0, /* xUpdate */
+ 0, /* xBegin */
+ 0, /* xSync */
+ 0, /* xCommit */
+ 0, /* xRollback */
+ 0, /* xFindMethod */
+ 0, /* xRename */
+ 0, /* xSavepoint */
+ 0, /* xRelease */
+ 0 /* xRollbackTo */
+};
+
+/*
+** Register the closure virtual table
+*/
+#ifdef _WIN32
+__declspec(dllexport)
+#endif
+int sqlite3_closure_init(
+ sqlite3 *db,
+ char **pzErrMsg,
+ const sqlite3_api_routines *pApi
+){
+ int rc = SQLITE_OK;
+ SQLITE_EXTENSION_INIT2(pApi);
+ (void)pzErrMsg;
+ rc = sqlite3_create_module(db, "transitive_closure", &closureModule, 0);
+ return rc;
+}
diff --git a/ext/misc/ieee754.c b/ext/misc/ieee754.c
new file mode 100644
index 0000000..436b11e
--- /dev/null
+++ b/ext/misc/ieee754.c
@@ -0,0 +1,131 @@
+/*
+** 2013-04-17
+**
+** 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 SQLite extension implements functions for the exact display
+** and input of IEEE754 Binary64 floating-point numbers.
+**
+** ieee754(X)
+** ieee754(Y,Z)
+**
+** In the first form, the value X should be a floating-point number.
+** The function will return a string of the form 'ieee754(Y,Z)' where
+** Y and Z are integers such that X==Y*pow(w.0,Z).
+**
+** In the second form, Y and Z are integers which are the mantissa and
+** base-2 exponent of a new floating point number. The function returns
+** a floating-point value equal to Y*pow(2.0,Z).
+**
+** Examples:
+**
+** ieee754(2.0) -> 'ieee754(2,0)'
+** ieee754(45.25) -> 'ieee754(181,-2)'
+** ieee754(2, 0) -> 2.0
+** ieee754(181, -2) -> 45.25
+*/
+#include "sqlite3ext.h"
+SQLITE_EXTENSION_INIT1
+#include <assert.h>
+#include <string.h>
+
+/*
+** Implementation of the ieee754() function
+*/
+static void ieee754func(
+ sqlite3_context *context,
+ int argc,
+ sqlite3_value **argv
+){
+ if( argc==1 ){
+ sqlite3_int64 m, a;
+ double r;
+ int e;
+ int isNeg;
+ char zResult[100];
+ assert( sizeof(m)==sizeof(r) );
+ if( sqlite3_value_type(argv[0])!=SQLITE_FLOAT ) return;
+ r = sqlite3_value_double(argv[0]);
+ if( r<0.0 ){
+ isNeg = 1;
+ r = -r;
+ }else{
+ isNeg = 0;
+ }
+ memcpy(&a,&r,sizeof(a));
+ if( a==0 ){
+ e = 0;
+ m = 0;
+ }else{
+ e = a>>52;
+ m = a & ((((sqlite3_int64)1)<<52)-1);
+ m |= ((sqlite3_int64)1)<<52;
+ while( e<1075 && m>0 && (m&1)==0 ){
+ m >>= 1;
+ e++;
+ }
+ if( isNeg ) m = -m;
+ }
+ sqlite3_snprintf(sizeof(zResult), zResult, "ieee754(%lld,%d)",
+ m, e-1075);
+ sqlite3_result_text(context, zResult, -1, SQLITE_TRANSIENT);
+ }else if( argc==2 ){
+ sqlite3_int64 m, e, a;
+ double r;
+ int isNeg = 0;
+ m = sqlite3_value_int64(argv[0]);
+ e = sqlite3_value_int64(argv[1]);
+ if( m<0 ){
+ isNeg = 1;
+ m = -m;
+ if( m<0 ) return;
+ }else if( m==0 && e>1000 && e<1000 ){
+ sqlite3_result_double(context, 0.0);
+ return;
+ }
+ while( (m>>32)&0xffe00000 ){
+ m >>= 1;
+ e++;
+ }
+ while( ((m>>32)&0xfff00000)==0 ){
+ m <<= 1;
+ e--;
+ }
+ e += 1075;
+ if( e<0 ) e = m = 0;
+ if( e>0x7ff ) m = 0;
+ a = m & ((((sqlite3_int64)1)<<52)-1);
+ a |= e<<52;
+ if( isNeg ) a |= ((sqlite3_int64)1)<<63;
+ memcpy(&r, &a, sizeof(r));
+ sqlite3_result_double(context, r);
+ }
+}
+
+
+#ifdef _WIN32
+__declspec(dllexport)
+#endif
+int sqlite3_ieee_init(
+ sqlite3 *db,
+ char **pzErrMsg,
+ const sqlite3_api_routines *pApi
+){
+ int rc = SQLITE_OK;
+ SQLITE_EXTENSION_INIT2(pApi);
+ (void)pzErrMsg; /* Unused parameter */
+ rc = sqlite3_create_function(db, "ieee754", 1, SQLITE_UTF8, 0,
+ ieee754func, 0, 0);
+ if( rc==SQLITE_OK ){
+ rc = sqlite3_create_function(db, "ieee754", 2, SQLITE_UTF8, 0,
+ ieee754func, 0, 0);
+ }
+ return rc;
+}
diff --git a/ext/misc/regexp.c b/ext/misc/regexp.c
index cb39956..16fa7d0 100644
--- a/ext/misc/regexp.c
+++ b/ext/misc/regexp.c
@@ -12,7 +12,16 @@
**
** The code in this file implements a compact but reasonably
** efficient regular-expression matcher for posix extended regular
-** expressions against UTF8 text. The following syntax is supported:
+** expressions against UTF8 text.
+**
+** This file is an SQLite extension. It registers a single function
+** named "regexp(A,B)" where A is the regular expression and B is the
+** string to be matched. By registering this function, SQLite will also
+** then implement the "B regexp A" operator. Note that with the function
+** the regular expression comes first, but with the operator it comes
+** second.
+**
+** The following regular expression syntax is supported:
**
** X* zero or more occurrences of X
** X+ one or more occurrences of X
diff --git a/main.mk b/main.mk
index eb3da5e..90571b0 100644
--- a/main.mk
+++ b/main.mk
@@ -270,7 +270,10 @@
# Extensions to be statically loaded.
#
TESTSRC += \
+ $(TOP)/ext/misc/amatch.c \
+ $(TOP)/ext/misc/closure.c \
$(TOP)/ext/misc/fuzzer.c \
+ $(TOP)/ext/misc/ieee754.c \
$(TOP)/ext/misc/regexp.c \
$(TOP)/ext/misc/spellfix.c \
$(TOP)/ext/misc/wholenumber.c
diff --git a/manifest b/manifest
index 4c3502f..1e2aa91 100644
--- a/manifest
+++ b/manifest
@@ -1,9 +1,9 @@
-C Move\sthe\stest_spellfix.c\smodule\sto\sext/misc/spellfix.c.
-D 2013-04-25T14:59:01.366
+C Added\sthe\stransitive_closure,\sieee754,\sand\samatch\sextensions.
+D 2013-04-25T16:42:55.478
F Makefile.arm-wince-mingw32ce-gcc d6df77f1f48d690bd73162294bbba7f59507c72f
-F Makefile.in 10c635460b6c3a20741d71c3a1b65b0ebec7558b
+F Makefile.in 38a45083d4df568eefd31d27d67381850f35a016
F Makefile.linux-gcc 91d710bdc4998cb015f39edf3cb314ec4f4d7e23
-F Makefile.msc 97a23e910afb8c2d23db0217c343481d59045acb
+F Makefile.msc d03cde9073d5752d6ddcaa6807a8bdda277e48f2
F Makefile.vxworks db21ed42a01d5740e656b16f92cb5d8d5e5dd315
F README cd04a36fbc7ea56932a4052d7d0b7f09f27c33d6
F VERSION 05c7bd63b96f31cfdef5c766ed91307ac121f5aa
@@ -83,9 +83,12 @@
F ext/icu/README.txt d9fbbad0c2f647c3fdf715fc9fd64af53aedfc43
F ext/icu/icu.c eb9ae1d79046bd7871aa97ee6da51eb770134b5a
F ext/icu/sqliteicu.h 728867a802baa5a96de7495e9689a8e01715ef37
+F ext/misc/amatch.c 3369b2b544066e620d986f0085d039c77d1ef17f
+F ext/misc/closure.c fec0c8537c69843e0b7631d500a14c0527962cd6
F ext/misc/fuzzer.c fb64a15af978ae73fa9075b9b1dfbe82b8defc6f
-F ext/misc/regexp.c c0fdb8af86981ff9890d776cfb97fe66297cc3b2
-F ext/misc/spellfix.c 8bb699116e36cc5e68d7ddf1810b638a3090c744 w src/test_spellfix.c
+F ext/misc/ieee754.c 2565ce373d842977efe0922dc50b8a41b3289556
+F ext/misc/regexp.c c25c65fe775f5d9801fb8573e36ebe73f2c0c2e0
+F ext/misc/spellfix.c 8bb699116e36cc5e68d7ddf1810b638a3090c744
F ext/misc/wholenumber.c ce362368b9381ea48cbd951ade8df867eeeab014
F ext/rtree/README 6315c0d73ebf0ec40dedb5aa0e942bc8b54e3761
F ext/rtree/rtree.c 757abea591d4ff67c0ff4e8f9776aeda86b18c14
@@ -108,7 +111,7 @@
F ext/rtree/viewrtree.tcl eea6224b3553599ae665b239bd827e182b466024
F install-sh 9d4de14ab9fb0facae2f48780b874848cbf2f895 x
F ltmain.sh 3ff0879076df340d2e23ae905484d8c15d5fdea8
-F main.mk a6183110637d782988fdf3735f7d125b19639967
+F main.mk 9535835509ac58a1902ba85ed77bcce9840d62d4
F mkdll.sh 7d09b23c05d56532e9d44a50868eb4b12ff4f74a
F mkextu.sh 416f9b7089d80e5590a29692c9d9280a10dbad9f
F mkextw.sh 4123480947681d9b434a5e7b1ee08135abe409ac
@@ -195,7 +198,7 @@
F src/status.c bedc37ec1a6bb9399944024d63f4c769971955a9
F src/table.c 2cd62736f845d82200acfa1287e33feb3c15d62e
F src/tclsqlite.c 2ecec9937e69bc17560ad886da35195daa7261b8
-F src/test1.c 1c7fa0a36703508130d7ed65325883a9e453bf72
+F src/test1.c e9562428421bf0cdbf11bf5f60ab0cdabee45885
F src/test2.c 29e7154112f7448d64204e8d31179cf497ecf425
F src/test3.c 96aed72a8e1d542fed127e3e8350ae357712fa82
F src/test4.c cea2c55110241e4674e66d476d29c914627999f5
@@ -332,6 +335,7 @@
F test/capi3e.test f7408dda65c92b9056199fdc180f893015f83dde
F test/cast.test 4c275cbdc8202d6f9c54a3596701719868ac7dc3
F test/check.test 2eb93611139a7dfaed3be80067c7dc5ceb5fb287
+F test/closure01.test 6194a899cdbba561d0439c0d6cc7bcdf4fc413e7
F test/coalesce.test cee0dccb9fbd2d494b77234bccf9dc6c6786eb91
F test/collate1.test fd02c4d8afc71879c4bb952586389961a21fb0ce
F test/collate2.test 04cebe4a033be319d6ddbb3bbc69464e01700b49
@@ -1054,7 +1058,7 @@
F tool/warnings-clang.sh f6aa929dc20ef1f856af04a730772f59283631d4
F tool/warnings.sh fbc018d67fd7395f440c28f33ef0f94420226381
F tool/win/sqlite.vsix 97894c2790eda7b5bce3cc79cb2a8ec2fde9b3ac
-P 680822e892f3efdb702eea3b321bc5785239dd56
-R f6325e95c7057e7590cfe9f4379a5b36
+P de556add10150140981a2e34b3712e96a7c262e3
+R 9481b1895542c76cd21821bef5dac0b1
U drh
-Z 2d983c030ad222dceb5985301e41e8c4
+Z 540c4ff1b6a2b03d38328f53c71b1af7
diff --git a/manifest.uuid b/manifest.uuid
index 2730c8b..31b1041 100644
--- a/manifest.uuid
+++ b/manifest.uuid
@@ -1 +1 @@
-de556add10150140981a2e34b3712e96a7c262e3
\ No newline at end of file
+84018099c8715b982cd24ce9221f93c7379e8c08
\ No newline at end of file
diff --git a/src/test1.c b/src/test1.c
index 7c30962..3cbe2bd 100644
--- a/src/test1.c
+++ b/src/test1.c
@@ -6057,7 +6057,10 @@
int objc,
Tcl_Obj *CONST objv[]
){
+ extern int sqlite3_amatch_init(sqlite3*,char**,const sqlite3_api_routines*);
+ extern int sqlite3_closure_init(sqlite3*,char**,const sqlite3_api_routines*);
extern int sqlite3_fuzzer_init(sqlite3*,char**,const sqlite3_api_routines*);
+ extern int sqlite3_ieee_init(sqlite3*,char**,const sqlite3_api_routines*);
extern int sqlite3_regexp_init(sqlite3*,char**,const sqlite3_api_routines*);
extern int sqlite3_spellfix_init(sqlite3*,char**,const sqlite3_api_routines*);
extern int sqlite3_wholenumber_init(sqlite3*,char**,const sqlite3_api_routines*);
@@ -6065,7 +6068,10 @@
const char *zExtName;
int (*pInit)(sqlite3*,char**,const sqlite3_api_routines*);
} aExtension[] = {
+ { "amatch", sqlite3_amatch_init },
+ { "closure", sqlite3_closure_init },
{ "fuzzer", sqlite3_fuzzer_init },
+ { "ieee754", sqlite3_ieee_init },
{ "regexp", sqlite3_regexp_init },
{ "spellfix", sqlite3_spellfix_init },
{ "wholenumber", sqlite3_wholenumber_init },
diff --git a/test/closure01.test b/test/closure01.test
new file mode 100644
index 0000000..abae85c
--- /dev/null
+++ b/test/closure01.test
@@ -0,0 +1,224 @@
+# 2013-04-25
+#
+# 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.
+#
+#***********************************************************************
+#
+# Test cases for transitive_closure virtual table.
+
+set testdir [file dirname $argv0]
+source $testdir/tester.tcl
+set testprefix closure01
+
+load_static_extension db closure
+
+do_execsql_test 1.0 {
+ BEGIN;
+ CREATE TABLE t1(x INTEGER PRIMARY KEY, y INTEGER);
+ CREATE INDEX t1y ON t1(y);
+ INSERT INTO t1(x) VALUES(1),(2);
+ INSERT INTO t1(x) SELECT x+2 FROM t1;
+ INSERT INTO t1(x) SELECT x+4 FROM t1;
+ INSERT INTO t1(x) SELECT x+8 FROM t1;
+ INSERT INTO t1(x) SELECT x+16 FROM t1;
+ INSERT INTO t1(x) SELECT x+32 FROM t1;
+ INSERT INTO t1(x) SELECT x+64 FROM t1;
+ INSERT INTO t1(x) SELECT x+128 FROM t1;
+ INSERT INTO t1(x) SELECT x+256 FROM t1;
+ INSERT INTO t1(x) SELECT x+512 FROM t1;
+ INSERT INTO t1(x) SELECT x+1024 FROM t1;
+ INSERT INTO t1(x) SELECT x+2048 FROM t1;
+ INSERT INTO t1(x) SELECT x+4096 FROM t1;
+ INSERT INTO t1(x) SELECT x+8192 FROM t1;
+ INSERT INTO t1(x) SELECT x+16384 FROM t1;
+ INSERT INTO t1(x) SELECT x+32768 FROM t1;
+ INSERT INTO t1(x) SELECT x+65536 FROM t1;
+ UPDATE t1 SET y=x/2 WHERE x>1;
+ COMMIT;
+ CREATE VIRTUAL TABLE cx
+ USING transitive_closure(tablename=t1, idcolumn=x, parentcolumn=y);
+} {}
+
+# The entire table
+do_execsql_test 1.1 {
+ SELECT count(*), depth FROM cx WHERE root=1 GROUP BY depth ORDER BY 1;
+} {/1 0 1 17 2 1 4 2 8 3 16 4 .* 65536 16/}
+
+# descendents of 32768
+do_execsql_test 1.2 {
+ SELECT * FROM cx WHERE root=32768 ORDER BY id;
+} {32768 0 65536 1 65537 1 131072 2}
+
+# descendents of 16384
+do_execsql_test 1.3 {
+ SELECT * FROM cx WHERE root=16384 AND depth<=2 ORDER BY id;
+} {16384 0 32768 1 32769 1 65536 2 65537 2 65538 2 65539 2}
+
+# children of 16384
+do_execsql_test 1.4 {
+ SELECT id, depth, root, tablename, idcolumn, parentcolumn FROM cx
+ WHERE root=16384
+ AND depth=1
+ ORDER BY id;
+} {32768 1 {} t1 x y 32769 1 {} t1 x y}
+
+# great-grandparent of 16384
+do_execsql_test 1.5 {
+ SELECT id, depth, root, tablename, idcolumn, parentcolumn FROM cx
+ WHERE root=16384
+ AND depth=3
+ AND idcolumn='Y'
+ AND parentcolumn='X';
+} {2048 3 {} t1 Y X}
+
+# depth<5
+do_execsql_test 1.6 {
+ SELECT count(*), depth FROM cx WHERE root=1 AND depth<5
+ GROUP BY depth ORDER BY 1;
+} {1 0 2 1 4 2 8 3 16 4}
+
+# depth<=5
+do_execsql_test 1.7 {
+ SELECT count(*), depth FROM cx WHERE root=1 AND depth<=5
+ GROUP BY depth ORDER BY 1;
+} {1 0 2 1 4 2 8 3 16 4 32 5}
+
+# depth==5
+do_execsql_test 1.8 {
+ SELECT count(*), depth FROM cx WHERE root=1 AND depth=5
+ GROUP BY depth ORDER BY 1;
+} {32 5}
+
+# depth BETWEEN 3 AND 5
+do_execsql_test 1.9 {
+ SELECT count(*), depth FROM cx WHERE root=1 AND depth BETWEEN 3 AND 5
+ GROUP BY depth ORDER BY 1;
+} {8 3 16 4 32 5}
+
+# depth==5 with min() and max()
+do_execsql_test 1.10 {
+ SELECT count(*), min(id), max(id) FROM cx WHERE root=1 AND depth=5;
+} {32 32 63}
+
+# Create a much smaller table t2 with only 32 elements
+db eval {
+ CREATE TABLE t2(x INTEGER PRIMARY KEY, y INTEGER);
+ INSERT INTO t2 SELECT x, y FROM t1 WHERE x<32;
+ CREATE INDEX t2y ON t2(y);
+ CREATE VIRTUAL TABLE c2
+ USING transitive_closure(tablename=t2, idcolumn=x, parentcolumn=y);
+}
+
+# t2 full-table
+do_execsql_test 2.1 {
+ SELECT count(*), min(id), max(id) FROM c2 WHERE root=1;
+} {31 1 31}
+# t2 root=10
+do_execsql_test 2.2 {
+ SELECT id FROM c2 WHERE root=10;
+} {10 20 21}
+# t2 root=11
+do_execsql_test 2.3 {
+ SELECT id FROM c2 WHERE root=12;
+} {12 24 25}
+# t2 root IN [10,12]
+do_execsql_test 2.4 {
+ SELECT id FROM c2 WHERE root IN (10,12) ORDER BY id;
+} {10 12 20 21 24 25}
+# t2 root IN [10,12] (sorted)
+do_execsql_test 2.5 {
+ SELECT id FROM c2 WHERE root IN (10,12) ORDER BY +id;
+} {10 12 20 21 24 25}
+
+# t2 c2up from 20
+do_execsql_test 3.0 {
+ CREATE VIRTUAL TABLE c2up USING transitive_closure(
+ tablename = t2,
+ idcolumn = y,
+ parentcolumn = x
+ );
+ SELECT id FROM c2up WHERE root=20;
+} {1 2 5 10 20}
+
+# cx as c2up
+do_execsql_test 3.1 {
+ SELECT id FROM cx
+ WHERE root=20
+ AND tablename='t2'
+ AND idcolumn='y'
+ AND parentcolumn='x';
+} {1 2 5 10 20}
+
+# t2 first cousins of 20
+do_execsql_test 3.2 {
+ SELECT DISTINCT id FROM c2
+ WHERE root IN (SELECT id FROM c2up
+ WHERE root=20 AND depth<=2)
+ ORDER BY id;
+} {5 10 11 20 21 22 23}
+
+# t2 first cousins of 20
+do_execsql_test 3.3 {
+ SELECT id FROM c2
+ WHERE root=(SELECT id FROM c2up
+ WHERE root=20 AND depth=2)
+ AND depth=2
+ EXCEPT
+ SELECT id FROM c2
+ WHERE root=(SELECT id FROM c2up
+ WHERE root=20 AND depth=1)
+ AND depth<=1
+ ORDER BY id;
+} {22 23}
+
+# missing tablename.
+do_test 4.1 {
+ catchsql {
+ SELECT id FROM cx
+ WHERE root=20
+ AND tablename='t3'
+ AND idcolumn='y'
+ AND parentcolumn='x';
+ }
+} {1 {no such table: t3}}
+
+# missing idcolumn
+do_test 4.2 {
+ catchsql {
+ SELECT id FROM cx
+ WHERE root=20
+ AND tablename='t2'
+ AND idcolumn='xyz'
+ AND parentcolumn='x';
+ }
+} {1 {no such column: t2.xyz}}
+
+# missing parentcolumn
+do_test 4.3 {
+ catchsql {
+ SELECT id FROM cx
+ WHERE root=20
+ AND tablename='t2'
+ AND idcolumn='x'
+ AND parentcolumn='pqr';
+ }
+} {1 {no such column: t2.pqr}}
+
+# generic closure
+do_execsql_test 5.1 {
+ CREATE VIRTUAL TABLE temp.closure USING transitive_closure;
+ SELECT id FROM closure
+ WHERE root=1
+ AND depth=3
+ AND tablename='t1'
+ AND idcolumn='x'
+ AND parentcolumn='y'
+ ORDER BY id;
+} {8 9 10 11 12 13 14 15}
+
+finish_test