Index: Makefile.in ================================================================== --- Makefile.in +++ Makefile.in @@ -359,11 +359,10 @@ $(TOP)/src/test_config.c \ $(TOP)/src/test_demovfs.c \ $(TOP)/src/test_devsym.c \ $(TOP)/src/test_fs.c \ $(TOP)/src/test_func.c \ - $(TOP)/src/test_fuzzer.c \ $(TOP)/src/test_hexio.c \ $(TOP)/src/test_init.c \ $(TOP)/src/test_intarray.c \ $(TOP)/src/test_journal.c \ $(TOP)/src/test_malloc.c \ @@ -371,25 +370,34 @@ $(TOP)/src/test_mutex.c \ $(TOP)/src/test_onefile.c \ $(TOP)/src/test_osinst.c \ $(TOP)/src/test_pcache.c \ $(TOP)/src/test_quota.c \ - $(TOP)/src/test_regexp.c \ $(TOP)/src/test_rtree.c \ $(TOP)/src/test_schema.c \ $(TOP)/src/test_server.c \ $(TOP)/src/test_superlock.c \ $(TOP)/src/test_syscall.c \ $(TOP)/src/test_stat.c \ $(TOP)/src/test_tclvar.c \ $(TOP)/src/test_thread.c \ $(TOP)/src/test_vfs.c \ - $(TOP)/src/test_wholenumber.c \ $(TOP)/src/test_wsd.c \ $(TOP)/ext/fts3/fts3_term.c \ $(TOP)/ext/fts3/fts3_test.c +# 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 + # Source code to the library files needed by the test fixture # TESTSRC2 = \ $(TOP)/src/attach.c \ $(TOP)/src/backup.c \ Index: Makefile.msc ================================================================== --- Makefile.msc +++ Makefile.msc @@ -679,11 +679,10 @@ $(TOP)\src\test_config.c \ $(TOP)\src\test_demovfs.c \ $(TOP)\src\test_devsym.c \ $(TOP)\src\test_fs.c \ $(TOP)\src\test_func.c \ - $(TOP)\src\test_fuzzer.c \ $(TOP)\src\test_hexio.c \ $(TOP)\src\test_init.c \ $(TOP)\src\test_intarray.c \ $(TOP)\src\test_journal.c \ $(TOP)\src\test_malloc.c \ @@ -691,24 +690,34 @@ $(TOP)\src\test_mutex.c \ $(TOP)\src\test_onefile.c \ $(TOP)\src\test_osinst.c \ $(TOP)\src\test_pcache.c \ $(TOP)\src\test_quota.c \ - $(TOP)\src\test_regexp.c \ $(TOP)\src\test_rtree.c \ $(TOP)\src\test_schema.c \ $(TOP)\src\test_server.c \ $(TOP)\src\test_superlock.c \ $(TOP)\src\test_syscall.c \ $(TOP)\src\test_stat.c \ $(TOP)\src\test_tclvar.c \ $(TOP)\src\test_thread.c \ $(TOP)\src\test_vfs.c \ - $(TOP)\src\test_wholenumber.c \ $(TOP)\src\test_wsd.c \ $(TOP)\ext\fts3\fts3_term.c \ $(TOP)\ext\fts3\fts3_test.c + +# 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 + # Source code to the library files needed by the test fixture # TESTSRC2 = \ $(TOP)\src\attach.c \ @@ -1202,12 +1211,12 @@ # hidden when the library is built via the amalgamation). # TESTFIXTURE_FLAGS = -DTCLSH=1 -DSQLITE_TEST=1 -DSQLITE_CRASH_TEST=1 TESTFIXTURE_FLAGS = $(TESTFIXTURE_FLAGS) -DSQLITE_SERVER=1 -DSQLITE_PRIVATE="" -DSQLITE_CORE -TESTFIXTURE_SRC0 = $(TESTSRC2) libsqlite3.lib -TESTFIXTURE_SRC1 = sqlite3.c +TESTFIXTURE_SRC0 = $(TESTEXT) $(TESTSRC2) libsqlite3.lib +TESTFIXTURE_SRC1 = $(TESTEXT) sqlite3.c !IF $(USE_AMALGAMATION)==0 TESTFIXTURE_SRC = $(TESTSRC) $(TOP)\src\tclsqlite.c $(TESTFIXTURE_SRC0) !ELSE TESTFIXTURE_SRC = $(TESTSRC) $(TOP)\src\tclsqlite.c $(TESTFIXTURE_SRC1) !ENDIF ADDED ext/misc/amatch.c Index: ext/misc/amatch.c ================================================================== --- /dev/null +++ ext/misc/amatch.c @@ -0,0 +1,1477 @@ +/* +** 2013-03-14 +** +** The author disclaims copyright to this source code. In place of +** a legal notice, here is a blessing: +** +** May you do good and not evil. +** May you find forgiveness for yourself and forgive others. +** May you share freely, never taking more than you give. +** +************************************************************************* +** +** This file contains code for a demonstration virtual table that finds +** "approximate matches" - strings from a finite set that are nearly the +** same as a single input string. The virtual table is called "amatch". +** +** A amatch virtual table is created like this: +** +** CREATE VIRTUAL TABLE f USING approximate_match( +** vocabulary_table=, -- V +** vocabulary_word=, -- W +** vocabulary_language=, -- L +** edit_distances= +** ); +** +** When it is created, the new amatch table must be supplied with the +** the name of a table V and columns V.W and V.L such that +** +** SELECT W FROM V WHERE L=$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 " 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 +#include +#include +#include +#include + +/* +** 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; ipNext; + pX->pNext = 0; + for(i=0; a[i] && ipRule = 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; iInpVCheck ){ + 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 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; izVocabTab); + 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; inConstraint; 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; +} ADDED ext/misc/closure.c Index: ext/misc/closure.c ================================================================== --- /dev/null +++ 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=, -- T +** idcolumn=, -- X +** parentcolumn= -- 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 +#include +#include +#include +#include + +/* +** 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 = (idid) ? 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->idid ){ + 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; iInzDb); + 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 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; izTableName); + 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; inConstraint; 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; +} ADDED ext/misc/fuzzer.c Index: ext/misc/fuzzer.c ================================================================== --- /dev/null +++ ext/misc/fuzzer.c @@ -0,0 +1,1171 @@ +/* +** 2011 March 24 +** +** 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. +** +************************************************************************* +** +** Code for a demonstration virtual table that generates variations +** on an input word at increasing edit distances from the original. +** +** A fuzzer virtual table is created like this: +** +** CREATE VIRTUAL TABLE f USING fuzzer(); +** +** When it is created, the new fuzzer table must be supplied with the +** name of a "fuzzer data table", which must reside in the same database +** file as the new fuzzer table. The fuzzer data table contains the various +** transformations and their costs that the fuzzer logic uses to generate +** variations. +** +** The fuzzer data table must contain exactly four columns (more precisely, +** the statement "SELECT * FROM " must return records +** that consist of four columns). It does not matter what the columns are +** named. +** +** Each row in the fuzzer data table represents a single character +** transformation. The left most column of the row (column 0) contains an +** integer value - the identifier of the ruleset to which the transformation +** rule belongs (see "MULTIPLE RULE SETS" below). The second column of the +** row (column 0) contains the input character or characters. The third +** column contains the output character or characters. And the fourth column +** contains the integer cost of making the transformation. For example: +** +** CREATE TABLE f_data(ruleset, cFrom, cTo, Cost); +** INSERT INTO f_data(ruleset, cFrom, cTo, Cost) VALUES(0, '', 'a', 100); +** INSERT INTO f_data(ruleset, cFrom, cTo, Cost) VALUES(0, 'b', '', 87); +** INSERT INTO f_data(ruleset, cFrom, cTo, Cost) VALUES(0, 'o', 'oe', 38); +** INSERT INTO f_data(ruleset, cFrom, cTo, Cost) VALUES(0, 'oe', 'o', 40); +** +** The first row inserted into the fuzzer data table by the SQL script +** above indicates that the cost of inserting a letter 'a' is 100. (All +** costs are integers. We recommend that costs be scaled so that the +** average cost is around 100.) The second INSERT statement creates a rule +** saying that the cost of deleting a single letter 'b' is 87. The third +** and fourth INSERT statements mean that the cost of transforming a +** single letter "o" into the two-letter sequence "oe" is 38 and that the +** cost of transforming "oe" back into "o" is 40. +** +** The contents of the fuzzer data table are loaded into main memory when +** a fuzzer table is first created, and may be internally reloaded by the +** system at any subsequent time. Therefore, the fuzzer data table should be +** populated before the fuzzer table is created and not modified thereafter. +** If you do need to modify the contents of the fuzzer data table, it is +** recommended that the associated fuzzer table be dropped, the fuzzer data +** table edited, and the fuzzer table recreated within a single transaction. +** Alternatively, the fuzzer data table can be edited then the database +** connection can be closed and reopened. +** +** Once it has been created, the fuzzer table can be queried as follows: +** +** SELECT word, distance FROM f +** WHERE word MATCH 'abcdefg' +** AND distance<200; +** +** This first query outputs the string "abcdefg" and all strings that +** can be derived from that string by appling the specified transformations. +** The strings are output together with their total transformation cost +** (called "distance") and appear in order of increasing cost. No string +** is output more than once. If there are multiple ways to transform the +** target string into the output string then the lowest cost transform is +** the one that is returned. In the example, the search is limited to +** strings with a total distance of less than 200. +** +** The fuzzer is a read-only table. Any attempt to DELETE, INSERT, or +** UPDATE on a fuzzer table will throw an error. +** +** It is important to put some kind of a limit on the fuzzer output. This +** can be either in the form of a LIMIT clause at the end of the query, +** or better, a "distance +#include +#include +#include + +#ifndef SQLITE_OMIT_VIRTUALTABLE + +/* +** Forward declaration of objects used by this implementation +*/ +typedef struct fuzzer_vtab fuzzer_vtab; +typedef struct fuzzer_cursor fuzzer_cursor; +typedef struct fuzzer_rule fuzzer_rule; +typedef struct fuzzer_seen fuzzer_seen; +typedef struct fuzzer_stem fuzzer_stem; + +/* +** Various types. +** +** fuzzer_cost is the "cost" of an edit operation. +** +** fuzzer_len is the length of a matching string. +** +** fuzzer_ruleid is an ruleset identifier. +*/ +typedef int fuzzer_cost; +typedef signed char fuzzer_len; +typedef int fuzzer_ruleid; + +/* +** Limits +*/ +#define FUZZER_MX_LENGTH 50 /* Maximum length of a rule string */ +#define FUZZER_MX_RULEID 2147483647 /* Maximum rule ID */ +#define FUZZER_MX_COST 1000 /* Maximum single-rule cost */ +#define FUZZER_MX_OUTPUT_LENGTH 100 /* Maximum length of an output string */ + + +/* +** Each transformation rule is stored as an instance of this object. +** All rules are kept on a linked list sorted by rCost. +*/ +struct fuzzer_rule { + fuzzer_rule *pNext; /* Next rule in order of increasing rCost */ + char *zFrom; /* Transform from */ + fuzzer_cost rCost; /* Cost of this transformation */ + fuzzer_len nFrom, nTo; /* Length of the zFrom and zTo strings */ + fuzzer_ruleid iRuleset; /* The rule set to which this rule belongs */ + char zTo[4]; /* Transform to (extra space appended) */ +}; + +/* +** A stem object is used to generate variants. It is also used to record +** previously generated outputs. +** +** Every stem is added to a hash table as it is output. Generation of +** duplicate stems is suppressed. +** +** Active stems (those that might generate new outputs) are kepts on a linked +** list sorted by increasing cost. The cost is the sum of rBaseCost and +** pRule->rCost. +*/ +struct fuzzer_stem { + char *zBasis; /* Word being fuzzed */ + const fuzzer_rule *pRule; /* Current rule to apply */ + fuzzer_stem *pNext; /* Next stem in rCost order */ + fuzzer_stem *pHash; /* Next stem with same hash on zBasis */ + fuzzer_cost rBaseCost; /* Base cost of getting to zBasis */ + fuzzer_cost rCostX; /* Precomputed rBaseCost + pRule->rCost */ + fuzzer_len nBasis; /* Length of the zBasis string */ + fuzzer_len n; /* Apply pRule at this character offset */ +}; + +/* +** A fuzzer virtual-table object +*/ +struct fuzzer_vtab { + sqlite3_vtab base; /* Base class - must be first */ + char *zClassName; /* Name of this class. Default: "fuzzer" */ + fuzzer_rule *pRule; /* All active rules in this fuzzer */ + int nCursor; /* Number of active cursors */ +}; + +#define FUZZER_HASH 4001 /* Hash table size */ +#define FUZZER_NQUEUE 20 /* Number of slots on the stem queue */ + +/* A fuzzer cursor object */ +struct fuzzer_cursor { + sqlite3_vtab_cursor base; /* Base class - must be first */ + sqlite3_int64 iRowid; /* The rowid of the current word */ + fuzzer_vtab *pVtab; /* The virtual table this cursor belongs to */ + fuzzer_cost rLimit; /* Maximum cost of any term */ + fuzzer_stem *pStem; /* Stem with smallest rCostX */ + fuzzer_stem *pDone; /* Stems already processed to completion */ + fuzzer_stem *aQueue[FUZZER_NQUEUE]; /* Queue of stems with higher rCostX */ + int mxQueue; /* Largest used index in aQueue[] */ + char *zBuf; /* Temporary use buffer */ + int nBuf; /* Bytes allocated for zBuf */ + int nStem; /* Number of stems allocated */ + int iRuleset; /* Only process rules from this ruleset */ + fuzzer_rule nullRule; /* Null rule used first */ + fuzzer_stem *apHash[FUZZER_HASH]; /* Hash of previously generated terms */ +}; + +/* +** 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 fuzzer_rule *fuzzerMergeRules(fuzzer_rule *pA, fuzzer_rule *pB){ + fuzzer_rule head; + fuzzer_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 fuzzer data table. This +** function allocates and populates a fuzzer_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 fuzzerLoadOneRule( + fuzzer_vtab *p, /* Fuzzer virtual table handle */ + sqlite3_stmt *pStmt, /* Base rule on statements current row */ + fuzzer_rule **ppRule, /* OUT: New rule object */ + char **pzErr /* OUT: Error message */ +){ + sqlite3_int64 iRuleset = 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); + int nCost = 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 */ + fuzzer_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 ){ + *ppRule = 0; + return SQLITE_OK; + } + + if( nCost<=0 || nCost>FUZZER_MX_COST ){ + *pzErr = sqlite3_mprintf("%s: cost must be between 1 and %d", + p->zClassName, FUZZER_MX_COST + ); + rc = SQLITE_ERROR; + }else + if( nFrom>FUZZER_MX_LENGTH || nTo>FUZZER_MX_LENGTH ){ + *pzErr = sqlite3_mprintf("%s: maximum string length is %d", + p->zClassName, FUZZER_MX_LENGTH + ); + rc = SQLITE_ERROR; + }else + if( iRuleset<0 || iRuleset>FUZZER_MX_RULEID ){ + *pzErr = sqlite3_mprintf("%s: ruleset must be between 0 and %d", + p->zClassName, FUZZER_MX_RULEID + ); + rc = SQLITE_ERROR; + }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 = nCost; + pRule->iRuleset = (int)iRuleset; + } + } + + *ppRule = pRule; + return rc; +} + +/* +** Load the content of the fuzzer data table into memory. +*/ +static int fuzzerLoadRules( + sqlite3 *db, /* Database handle */ + fuzzer_vtab *p, /* Virtual fuzzer table to configure */ + const char *zDb, /* Database containing rules data */ + const char *zData, /* Table containing rules data */ + char **pzErr /* OUT: Error message */ +){ + int rc = SQLITE_OK; /* Return code */ + char *zSql; /* SELECT used to read from rules table */ + fuzzer_rule *pHead = 0; + + zSql = sqlite3_mprintf("SELECT * FROM %Q.%Q", zDb, zData); + 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, zData, sqlite3_column_count(pStmt) + ); + rc = SQLITE_ERROR; + }else{ + while( rc==SQLITE_OK && SQLITE_ROW==sqlite3_step(pStmt) ){ + fuzzer_rule *pRule = 0; + rc = fuzzerLoadOneRule(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 fuzzer_vtab.pRule to point to + ** point to the head of the sorted list. + */ + if( rc==SQLITE_OK ){ + unsigned int i; + fuzzer_rule *pX; + fuzzer_rule *a[15]; + for(i=0; ipNext; + pX->pNext = 0; + for(i=0; a[i] && ipRule = fuzzerMergeRules(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 *fuzzerDequote(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; iInnCursor==0 ); + while( p->pRule ){ + fuzzer_rule *pRule = p->pRule; + p->pRule = pRule->pNext; + sqlite3_free(pRule); + } + sqlite3_free(p); + return SQLITE_OK; +} + +/* +** xConnect/xCreate method for the fuzzer module. Arguments are: +** +** argv[0] -> module name ("fuzzer") +** argv[1] -> database name +** argv[2] -> table name +** argv[3] -> fuzzer rule table name +*/ +static int fuzzerConnect( + sqlite3 *db, + void *pAux, + int argc, const char *const*argv, + sqlite3_vtab **ppVtab, + char **pzErr +){ + int rc = SQLITE_OK; /* Return code */ + fuzzer_vtab *pNew = 0; /* New virtual table */ + const char *zModule = argv[0]; + const char *zDb = argv[1]; + + if( argc!=4 ){ + *pzErr = sqlite3_mprintf( + "%s: wrong number of CREATE VIRTUAL TABLE arguments", zModule + ); + rc = SQLITE_ERROR; + }else{ + int nModule; /* Length of zModule, in bytes */ + + nModule = (int)strlen(zModule); + pNew = sqlite3_malloc( sizeof(*pNew) + nModule + 1); + if( pNew==0 ){ + rc = SQLITE_NOMEM; + }else{ + char *zTab; /* Dequoted name of fuzzer data table */ + + memset(pNew, 0, sizeof(*pNew)); + pNew->zClassName = (char*)&pNew[1]; + memcpy(pNew->zClassName, zModule, nModule+1); + + zTab = fuzzerDequote(argv[3]); + if( zTab==0 ){ + rc = SQLITE_NOMEM; + }else{ + rc = fuzzerLoadRules(db, pNew, zDb, zTab, pzErr); + sqlite3_free(zTab); + } + + if( rc==SQLITE_OK ){ + rc = sqlite3_declare_vtab(db, "CREATE TABLE x(word,distance,ruleset)"); + } + if( rc!=SQLITE_OK ){ + fuzzerDisconnect((sqlite3_vtab *)pNew); + pNew = 0; + } + } + } + + *ppVtab = (sqlite3_vtab *)pNew; + return rc; +} + +/* +** Open a new fuzzer cursor. +*/ +static int fuzzerOpen(sqlite3_vtab *pVTab, sqlite3_vtab_cursor **ppCursor){ + fuzzer_vtab *p = (fuzzer_vtab*)pVTab; + fuzzer_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 all stems in a list. +*/ +static void fuzzerClearStemList(fuzzer_stem *pStem){ + while( pStem ){ + fuzzer_stem *pNext = pStem->pNext; + sqlite3_free(pStem); + pStem = pNext; + } +} + +/* +** Free up all the memory allocated by a cursor. Set it rLimit to 0 +** to indicate that it is at EOF. +*/ +static void fuzzerClearCursor(fuzzer_cursor *pCur, int clearHash){ + int i; + fuzzerClearStemList(pCur->pStem); + fuzzerClearStemList(pCur->pDone); + for(i=0; iaQueue[i]); + pCur->rLimit = (fuzzer_cost)0; + if( clearHash && pCur->nStem ){ + pCur->mxQueue = 0; + pCur->pStem = 0; + pCur->pDone = 0; + memset(pCur->aQueue, 0, sizeof(pCur->aQueue)); + memset(pCur->apHash, 0, sizeof(pCur->apHash)); + } + pCur->nStem = 0; +} + +/* +** Close a fuzzer cursor. +*/ +static int fuzzerClose(sqlite3_vtab_cursor *cur){ + fuzzer_cursor *pCur = (fuzzer_cursor *)cur; + fuzzerClearCursor(pCur, 0); + sqlite3_free(pCur->zBuf); + pCur->pVtab->nCursor--; + sqlite3_free(pCur); + return SQLITE_OK; +} + +/* +** Compute the current output term for a fuzzer_stem. +*/ +static int fuzzerRender( + fuzzer_stem *pStem, /* The stem to be rendered */ + char **pzBuf, /* Write results into this buffer. realloc if needed */ + int *pnBuf /* Size of the buffer */ +){ + const fuzzer_rule *pRule = pStem->pRule; + int n; /* Size of output term without nul-term */ + char *z; /* Buffer to assemble output term in */ + + n = pStem->nBasis + pRule->nTo - pRule->nFrom; + if( (*pnBuf)n; + z = *pzBuf; + if( n<0 ){ + memcpy(z, pStem->zBasis, pStem->nBasis+1); + }else{ + memcpy(z, pStem->zBasis, n); + memcpy(&z[n], pRule->zTo, pRule->nTo); + memcpy(&z[n+pRule->nTo], &pStem->zBasis[n+pRule->nFrom], + pStem->nBasis-n-pRule->nFrom+1); + } + + assert( z[pStem->nBasis + pRule->nTo - pRule->nFrom]==0 ); + return SQLITE_OK; +} + +/* +** Compute a hash on zBasis. +*/ +static unsigned int fuzzerHash(const char *z){ + unsigned int h = 0; + while( *z ){ h = (h<<3) ^ (h>>29) ^ *(z++); } + return h % FUZZER_HASH; +} + +/* +** Current cost of a stem +*/ +static fuzzer_cost fuzzerCost(fuzzer_stem *pStem){ + return pStem->rCostX = pStem->rBaseCost + pStem->pRule->rCost; +} + +#if 0 +/* +** Print a description of a fuzzer_stem on stderr. +*/ +static void fuzzerStemPrint( + const char *zPrefix, + fuzzer_stem *pStem, + const char *zSuffix +){ + if( pStem->n<0 ){ + fprintf(stderr, "%s[%s](%d)-->self%s", + zPrefix, + pStem->zBasis, pStem->rBaseCost, + zSuffix + ); + }else{ + char *zBuf = 0; + int nBuf = 0; + if( fuzzerRender(pStem, &zBuf, &nBuf)!=SQLITE_OK ) return; + fprintf(stderr, "%s[%s](%d)-->{%s}(%d)%s", + zPrefix, + pStem->zBasis, pStem->rBaseCost, zBuf, pStem->, + zSuffix + ); + sqlite3_free(zBuf); + } +} +#endif + +/* +** Return 1 if the string to which the cursor is point has already +** been emitted. Return 0 if not. Return -1 on a memory allocation +** failures. +*/ +static int fuzzerSeen(fuzzer_cursor *pCur, fuzzer_stem *pStem){ + unsigned int h; + fuzzer_stem *pLookup; + + if( fuzzerRender(pStem, &pCur->zBuf, &pCur->nBuf)==SQLITE_NOMEM ){ + return -1; + } + h = fuzzerHash(pCur->zBuf); + pLookup = pCur->apHash[h]; + while( pLookup && strcmp(pLookup->zBasis, pCur->zBuf)!=0 ){ + pLookup = pLookup->pHash; + } + return pLookup!=0; +} + +/* +** If argument pRule is NULL, this function returns false. +** +** Otherwise, it returns true if rule pRule should be skipped. A rule +** should be skipped if it does not belong to rule-set iRuleset, or if +** applying it to stem pStem would create a string longer than +** FUZZER_MX_OUTPUT_LENGTH bytes. +*/ +static int fuzzerSkipRule( + const fuzzer_rule *pRule, /* Determine whether or not to skip this */ + fuzzer_stem *pStem, /* Stem rule may be applied to */ + int iRuleset /* Rule-set used by the current query */ +){ + return pRule && ( + (pRule->iRuleset!=iRuleset) + || (pStem->nBasis + pRule->nTo - pRule->nFrom)>FUZZER_MX_OUTPUT_LENGTH + ); +} + +/* +** Advance a fuzzer_stem to its next value. Return 0 if there are +** no more values that can be generated by this fuzzer_stem. Return +** -1 on a memory allocation failure. +*/ +static int fuzzerAdvance(fuzzer_cursor *pCur, fuzzer_stem *pStem){ + const fuzzer_rule *pRule; + while( (pRule = pStem->pRule)!=0 ){ + assert( pRule==&pCur->nullRule || pRule->iRuleset==pCur->iRuleset ); + while( pStem->n < pStem->nBasis - pRule->nFrom ){ + pStem->n++; + if( pRule->nFrom==0 + || memcmp(&pStem->zBasis[pStem->n], pRule->zFrom, pRule->nFrom)==0 + ){ + /* Found a rewrite case. Make sure it is not a duplicate */ + int rc = fuzzerSeen(pCur, pStem); + if( rc<0 ) return -1; + if( rc==0 ){ + fuzzerCost(pStem); + return 1; + } + } + } + pStem->n = -1; + do{ + pRule = pRule->pNext; + }while( fuzzerSkipRule(pRule, pStem, pCur->iRuleset) ); + pStem->pRule = pRule; + if( pRule && fuzzerCost(pStem)>pCur->rLimit ) pStem->pRule = 0; + } + return 0; +} + +/* +** The two input stem lists are both sorted in order of increasing +** rCostX. Merge them together into a single list, sorted by rCostX, and +** return a pointer to the head of that new list. +*/ +static fuzzer_stem *fuzzerMergeStems(fuzzer_stem *pA, fuzzer_stem *pB){ + fuzzer_stem head; + fuzzer_stem *pTail; + + pTail = &head; + while( pA && pB ){ + if( pA->rCostX<=pB->rCostX ){ + 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; +} + +/* +** Load pCur->pStem with the lowest-cost stem. Return a pointer +** to the lowest-cost stem. +*/ +static fuzzer_stem *fuzzerLowestCostStem(fuzzer_cursor *pCur){ + fuzzer_stem *pBest, *pX; + int iBest; + int i; + + if( pCur->pStem==0 ){ + iBest = -1; + pBest = 0; + for(i=0; i<=pCur->mxQueue; i++){ + pX = pCur->aQueue[i]; + if( pX==0 ) continue; + if( pBest==0 || pBest->rCostX>pX->rCostX ){ + pBest = pX; + iBest = i; + } + } + if( pBest ){ + pCur->aQueue[iBest] = pBest->pNext; + pBest->pNext = 0; + pCur->pStem = pBest; + } + } + return pCur->pStem; +} + +/* +** Insert pNew into queue of pending stems. Then find the stem +** with the lowest rCostX and move it into pCur->pStem. +** list. The insert is done such the pNew is in the correct order +** according to fuzzer_stem.zBaseCost+fuzzer_stem.pRule->rCost. +*/ +static fuzzer_stem *fuzzerInsert(fuzzer_cursor *pCur, fuzzer_stem *pNew){ + fuzzer_stem *pX; + int i; + + /* If pCur->pStem exists and is greater than pNew, then make pNew + ** the new pCur->pStem and insert the old pCur->pStem instead. + */ + if( (pX = pCur->pStem)!=0 && pX->rCostX>pNew->rCostX ){ + pNew->pNext = 0; + pCur->pStem = pNew; + pNew = pX; + } + + /* Insert the new value */ + pNew->pNext = 0; + pX = pNew; + for(i=0; i<=pCur->mxQueue; i++){ + if( pCur->aQueue[i] ){ + pX = fuzzerMergeStems(pX, pCur->aQueue[i]); + pCur->aQueue[i] = 0; + }else{ + pCur->aQueue[i] = pX; + break; + } + } + if( i>pCur->mxQueue ){ + if( imxQueue = i; + pCur->aQueue[i] = pX; + }else{ + assert( pCur->mxQueue==FUZZER_NQUEUE-1 ); + pX = fuzzerMergeStems(pX, pCur->aQueue[FUZZER_NQUEUE-1]); + pCur->aQueue[FUZZER_NQUEUE-1] = pX; + } + } + + return fuzzerLowestCostStem(pCur); +} + +/* +** Allocate a new fuzzer_stem. Add it to the hash table but do not +** link it into either the pCur->pStem or pCur->pDone lists. +*/ +static fuzzer_stem *fuzzerNewStem( + fuzzer_cursor *pCur, + const char *zWord, + fuzzer_cost rBaseCost +){ + fuzzer_stem *pNew; + fuzzer_rule *pRule; + unsigned int h; + + pNew = sqlite3_malloc( sizeof(*pNew) + (int)strlen(zWord) + 1 ); + if( pNew==0 ) return 0; + memset(pNew, 0, sizeof(*pNew)); + pNew->zBasis = (char*)&pNew[1]; + pNew->nBasis = (int)strlen(zWord); + memcpy(pNew->zBasis, zWord, pNew->nBasis+1); + pRule = pCur->pVtab->pRule; + while( fuzzerSkipRule(pRule, pNew, pCur->iRuleset) ){ + pRule = pRule->pNext; + } + pNew->pRule = pRule; + pNew->n = -1; + pNew->rBaseCost = pNew->rCostX = rBaseCost; + h = fuzzerHash(pNew->zBasis); + pNew->pHash = pCur->apHash[h]; + pCur->apHash[h] = pNew; + pCur->nStem++; + return pNew; +} + + +/* +** Advance a cursor to its next row of output +*/ +static int fuzzerNext(sqlite3_vtab_cursor *cur){ + fuzzer_cursor *pCur = (fuzzer_cursor*)cur; + int rc; + fuzzer_stem *pStem, *pNew; + + pCur->iRowid++; + + /* Use the element the cursor is currently point to to create + ** a new stem and insert the new stem into the priority queue. + */ + pStem = pCur->pStem; + if( pStem->rCostX>0 ){ + rc = fuzzerRender(pStem, &pCur->zBuf, &pCur->nBuf); + if( rc==SQLITE_NOMEM ) return SQLITE_NOMEM; + pNew = fuzzerNewStem(pCur, pCur->zBuf, pStem->rCostX); + if( pNew ){ + if( fuzzerAdvance(pCur, pNew)==0 ){ + pNew->pNext = pCur->pDone; + pCur->pDone = pNew; + }else{ + if( fuzzerInsert(pCur, pNew)==pNew ){ + return SQLITE_OK; + } + } + }else{ + return SQLITE_NOMEM; + } + } + + /* Adjust the priority queue so that the first element of the + ** stem list is the next lowest cost word. + */ + while( (pStem = pCur->pStem)!=0 ){ + int res = fuzzerAdvance(pCur, pStem); + if( res<0 ){ + return SQLITE_NOMEM; + }else if( res>0 ){ + pCur->pStem = 0; + pStem = fuzzerInsert(pCur, pStem); + if( (rc = fuzzerSeen(pCur, pStem))!=0 ){ + if( rc<0 ) return SQLITE_NOMEM; + continue; + } + return SQLITE_OK; /* New word found */ + } + pCur->pStem = 0; + pStem->pNext = pCur->pDone; + pCur->pDone = pStem; + if( fuzzerLowestCostStem(pCur) ){ + rc = fuzzerSeen(pCur, pCur->pStem); + if( rc<0 ) return SQLITE_NOMEM; + if( rc==0 ){ + return SQLITE_OK; + } + } + } + + /* Reach this point only if queue has been exhausted and there is + ** nothing left to be output. */ + pCur->rLimit = (fuzzer_cost)0; + 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 fuzzerColumn, fuzzerRowid, or fuzzerEof call. +*/ +static int fuzzerFilter( + sqlite3_vtab_cursor *pVtabCursor, + int idxNum, const char *idxStr, + int argc, sqlite3_value **argv +){ + fuzzer_cursor *pCur = (fuzzer_cursor *)pVtabCursor; + const char *zWord = ""; + fuzzer_stem *pStem; + int idx; + + fuzzerClearCursor(pCur, 1); + pCur->rLimit = 2147483647; + idx = 0; + if( idxNum & 1 ){ + zWord = (const char*)sqlite3_value_text(argv[0]); + idx++; + } + if( idxNum & 2 ){ + pCur->rLimit = (fuzzer_cost)sqlite3_value_int(argv[idx]); + idx++; + } + if( idxNum & 4 ){ + pCur->iRuleset = (fuzzer_cost)sqlite3_value_int(argv[idx]); + idx++; + } + pCur->nullRule.pNext = pCur->pVtab->pRule; + pCur->nullRule.rCost = 0; + pCur->nullRule.nFrom = 0; + pCur->nullRule.nTo = 0; + pCur->nullRule.zFrom = ""; + pCur->iRowid = 1; + assert( pCur->pStem==0 ); + + /* If the query term is longer than FUZZER_MX_OUTPUT_LENGTH bytes, this + ** query will return zero rows. */ + if( (int)strlen(zWord)pStem = pStem = fuzzerNewStem(pCur, zWord, (fuzzer_cost)0); + if( pStem==0 ) return SQLITE_NOMEM; + pStem->pRule = &pCur->nullRule; + pStem->n = pStem->nBasis; + }else{ + pCur->rLimit = 0; + } + + return SQLITE_OK; +} + +/* +** Only the word and distance columns have values. All other columns +** return NULL +*/ +static int fuzzerColumn(sqlite3_vtab_cursor *cur, sqlite3_context *ctx, int i){ + fuzzer_cursor *pCur = (fuzzer_cursor*)cur; + if( i==0 ){ + /* the "word" column */ + if( fuzzerRender(pCur->pStem, &pCur->zBuf, &pCur->nBuf)==SQLITE_NOMEM ){ + return SQLITE_NOMEM; + } + sqlite3_result_text(ctx, pCur->zBuf, -1, SQLITE_TRANSIENT); + }else if( i==1 ){ + /* the "distance" column */ + sqlite3_result_int(ctx, pCur->pStem->rCostX); + }else{ + /* All other columns are NULL */ + sqlite3_result_null(ctx); + } + return SQLITE_OK; +} + +/* +** The rowid. +*/ +static int fuzzerRowid(sqlite3_vtab_cursor *cur, sqlite_int64 *pRowid){ + fuzzer_cursor *pCur = (fuzzer_cursor*)cur; + *pRowid = pCur->iRowid; + return SQLITE_OK; +} + +/* +** When the fuzzer_cursor.rLimit value is 0 or less, that is a signal +** that the cursor has nothing more to output. +*/ +static int fuzzerEof(sqlite3_vtab_cursor *cur){ + fuzzer_cursor *pCur = (fuzzer_cursor*)cur; + return pCur->rLimit<=(fuzzer_cost)0; +} + +/* +** Search for terms of these forms: +** +** (A) word MATCH $str +** (B1) distance < $value +** (B2) distance <= $value +** (C) ruleid == $ruleid +** +** 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 fuzzerBestIndex(sqlite3_vtab *tab, sqlite3_index_info *pIdxInfo){ + int iPlan = 0; + int iDistTerm = -1; + int iRulesetTerm = -1; + int i; + const struct sqlite3_index_constraint *pConstraint; + pConstraint = pIdxInfo->aConstraint; + for(i=0; inConstraint; 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; + iRulesetTerm = 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[iRulesetTerm].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; +} + +/* +** A virtual table module that implements the "fuzzer". +*/ +static sqlite3_module fuzzerModule = { + 0, /* iVersion */ + fuzzerConnect, + fuzzerConnect, + fuzzerBestIndex, + fuzzerDisconnect, + fuzzerDisconnect, + fuzzerOpen, /* xOpen - open a cursor */ + fuzzerClose, /* xClose - close a cursor */ + fuzzerFilter, /* xFilter - configure scan constraints */ + fuzzerNext, /* xNext - advance a cursor */ + fuzzerEof, /* xEof - check for end of scan */ + fuzzerColumn, /* xColumn - read data */ + fuzzerRowid, /* xRowid - read data */ + 0, /* xUpdate */ + 0, /* xBegin */ + 0, /* xSync */ + 0, /* xCommit */ + 0, /* xRollback */ + 0, /* xFindMethod */ + 0, /* xRename */ +}; + +#endif /* SQLITE_OMIT_VIRTUALTABLE */ + + +#ifdef _WIN32 +__declspec(dllexport) +#endif +int sqlite3_fuzzer_init( + sqlite3 *db, + char **pzErrMsg, + const sqlite3_api_routines *pApi +){ + int rc = SQLITE_OK; + SQLITE_EXTENSION_INIT2(pApi); + rc = sqlite3_create_module(db, "fuzzer", &fuzzerModule, 0); + return rc; +} ADDED ext/misc/ieee754.c Index: ext/misc/ieee754.c ================================================================== --- /dev/null +++ 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 +#include + +/* +** 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; +} ADDED ext/misc/regexp.c Index: ext/misc/regexp.c ================================================================== --- /dev/null +++ ext/misc/regexp.c @@ -0,0 +1,756 @@ +/* +** 2012-11-13 +** +** 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. +** +****************************************************************************** +** +** The code in this file implements a compact but reasonably +** efficient regular-expression matcher for posix extended regular +** 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 +** X? zero or one occurrences of X +** X{p,q} between p and q occurrences of X +** (X) match X +** X|Y X or Y +** ^X X occurring at the beginning of the string +** X$ X occurring at the end of the string +** . Match any single character +** \c Character c where c is one of \{}()[]|*+?. +** \c C-language escapes for c in afnrtv. ex: \t or \n +** \uXXXX Where XXXX is exactly 4 hex digits, unicode value XXXX +** \xXX Where XX is exactly 2 hex digits, unicode value XX +** [abc] Any single character from the set abc +** [^abc] Any single character not in the set abc +** [a-z] Any single character in the range a-z +** [^a-z] Any single character not in the range a-z +** \b Word boundary +** \w Word character. [A-Za-z0-9_] +** \W Non-word character +** \d Digit +** \D Non-digit +** \s Whitespace character +** \S Non-whitespace character +** +** A nondeterministic finite automaton (NFA) is used for matching, so the +** performance is bounded by O(N*M) where N is the size of the regular +** expression and M is the size of the input string. The matcher never +** exhibits exponential behavior. Note that the X{p,q} operator expands +** to p copies of X following by q-p copies of X? and that the size of the +** regular expression in the O(N*M) performance bound is computed after +** this expansion. +*/ +#include +#include +#include "sqlite3ext.h" +SQLITE_EXTENSION_INIT1 + +/* +** The following #defines change the names of some functions implemented in +** this file to prevent name collisions with C-library functions of the +** same name. +*/ +#define re_match sqlite3re_match +#define re_compile sqlite3re_compile +#define re_free sqlite3re_free + +/* The end-of-input character */ +#define RE_EOF 0 /* End of input */ + +/* The NFA is implemented as sequence of opcodes taken from the following +** set. Each opcode has a single integer argument. +*/ +#define RE_OP_MATCH 1 /* Match the one character in the argument */ +#define RE_OP_ANY 2 /* Match any one character. (Implements ".") */ +#define RE_OP_ANYSTAR 3 /* Special optimized version of .* */ +#define RE_OP_FORK 4 /* Continue to both next and opcode at iArg */ +#define RE_OP_GOTO 5 /* Jump to opcode at iArg */ +#define RE_OP_ACCEPT 6 /* Halt and indicate a successful match */ +#define RE_OP_CC_INC 7 /* Beginning of a [...] character class */ +#define RE_OP_CC_EXC 8 /* Beginning of a [^...] character class */ +#define RE_OP_CC_VALUE 9 /* Single value in a character class */ +#define RE_OP_CC_RANGE 10 /* Range of values in a character class */ +#define RE_OP_WORD 11 /* Perl word character [A-Za-z0-9_] */ +#define RE_OP_NOTWORD 12 /* Not a perl word character */ +#define RE_OP_DIGIT 13 /* digit: [0-9] */ +#define RE_OP_NOTDIGIT 14 /* Not a digit */ +#define RE_OP_SPACE 15 /* space: [ \t\n\r\v\f] */ +#define RE_OP_NOTSPACE 16 /* Not a digit */ +#define RE_OP_BOUNDARY 17 /* Boundary between word and non-word */ + +/* Each opcode is a "state" in the NFA */ +typedef unsigned short ReStateNumber; + +/* Because this is an NFA and not a DFA, multiple states can be active at +** once. An instance of the following object records all active states in +** the NFA. The implementation is optimized for the common case where the +** number of actives states is small. +*/ +typedef struct ReStateSet { + unsigned nState; /* Number of current states */ + ReStateNumber *aState; /* Current states */ +} ReStateSet; + +/* An input string read one character at a time. +*/ +typedef struct ReInput ReInput; +struct ReInput { + const unsigned char *z; /* All text */ + int i; /* Next byte to read */ + int mx; /* EOF when i>=mx */ +}; + +/* A compiled NFA (or an NFA that is in the process of being compiled) is +** an instance of the following object. +*/ +typedef struct ReCompiled ReCompiled; +struct ReCompiled { + ReInput sIn; /* Regular expression text */ + const char *zErr; /* Error message to return */ + char *aOp; /* Operators for the virtual machine */ + int *aArg; /* Arguments to each operator */ + unsigned (*xNextChar)(ReInput*); /* Next character function */ + unsigned char zInit[12]; /* Initial text to match */ + int nInit; /* Number of characters in zInit */ + unsigned nState; /* Number of entries in aOp[] and aArg[] */ + unsigned nAlloc; /* Slots allocated for aOp[] and aArg[] */ +}; + +/* Add a state to the given state set if it is not already there */ +static void re_add_state(ReStateSet *pSet, int newState){ + unsigned i; + for(i=0; inState; i++) if( pSet->aState[i]==newState ) return; + pSet->aState[pSet->nState++] = newState; +} + +/* Extract the next unicode character from *pzIn and return it. Advance +** *pzIn to the first byte past the end of the character returned. To +** be clear: this routine converts utf8 to unicode. This routine is +** optimized for the common case where the next character is a single byte. +*/ +static unsigned re_next_char(ReInput *p){ + unsigned c; + if( p->i>=p->mx ) return 0; + c = p->z[p->i++]; + if( c>=0x80 ){ + if( (c&0xe0)==0xc0 && p->imx && (p->z[p->i]&0xc0)==0x80 ){ + c = (c&0x1f)<<6 | (p->z[p->i++]&0x3f); + if( c<0x80 ) c = 0xfffd; + }else if( (c&0xf0)==0xe0 && p->i+1mx && (p->z[p->i]&0xc0)==0x80 + && (p->z[p->i+1]&0xc0)==0x80 ){ + c = (c&0x0f)<<12 | ((p->z[p->i]&0x3f)<<6) | (p->z[p->i+1]&0x3f); + p->i += 2; + if( c<=0x3ff || (c>=0xd800 && c<=0xdfff) ) c = 0xfffd; + }else if( (c&0xf8)==0xf0 && p->i+3mx && (p->z[p->i]&0xc0)==0x80 + && (p->z[p->i+1]&0xc0)==0x80 && (p->z[p->i+2]&0xc0)==0x80 ){ + c = (c&0x07)<<18 | ((p->z[p->i]&0x3f)<<12) | ((p->z[p->i+1]&0x3f)<<6) + | (p->z[p->i+2]&0x3f); + p->i += 3; + if( c<=0xffff || c>0x10ffff ) c = 0xfffd; + }else{ + c = 0xfffd; + } + } + return c; +} +static unsigned re_next_char_nocase(ReInput *p){ + unsigned c = re_next_char(p); + if( c>='A' && c<='Z' ) c += 'a' - 'A'; + return c; +} + +/* Return true if c is a perl "word" character: [A-Za-z0-9_] */ +static int re_word_char(int c){ + return (c>='0' && c<='9') || (c>='a' && c<='z') + || (c>='A' && c<='Z') || c=='_'; +} + +/* Return true if c is a "digit" character: [0-9] */ +static int re_digit_char(int c){ + return (c>='0' && c<='9'); +} + +/* Return true if c is a perl "space" character: [ \t\r\n\v\f] */ +static int re_space_char(int c){ + return c==' ' || c=='\t' || c=='\n' || c=='\r' || c=='\v' || c=='\f'; +} + +/* Run a compiled regular expression on the zero-terminated input +** string zIn[]. Return true on a match and false if there is no match. +*/ +static int re_match(ReCompiled *pRe, const unsigned char *zIn, int nIn){ + ReStateSet aStateSet[2], *pThis, *pNext; + ReStateNumber aSpace[100]; + ReStateNumber *pToFree; + unsigned int i = 0; + unsigned int iSwap = 0; + int c = RE_EOF+1; + int cPrev = 0; + int rc = 0; + ReInput in; + + in.z = zIn; + in.i = 0; + in.mx = nIn>=0 ? nIn : (int)strlen((char const*)zIn); + + /* Look for the initial prefix match, if there is one. */ + if( pRe->nInit ){ + unsigned char x = pRe->zInit[0]; + while( in.i+pRe->nInit<=in.mx + && (zIn[in.i]!=x || + strncmp((const char*)zIn+in.i, (const char*)pRe->zInit, pRe->nInit)!=0) + ){ + in.i++; + } + if( in.i+pRe->nInit>in.mx ) return 0; + } + + if( pRe->nState<=(sizeof(aSpace)/(sizeof(aSpace[0])*2)) ){ + pToFree = 0; + aStateSet[0].aState = aSpace; + }else{ + pToFree = sqlite3_malloc( sizeof(ReStateNumber)*2*pRe->nState ); + if( pToFree==0 ) return -1; + aStateSet[0].aState = pToFree; + } + aStateSet[1].aState = &aStateSet[0].aState[pRe->nState]; + pNext = &aStateSet[1]; + pNext->nState = 0; + re_add_state(pNext, 0); + while( c!=RE_EOF && pNext->nState>0 ){ + cPrev = c; + c = pRe->xNextChar(&in); + pThis = pNext; + pNext = &aStateSet[iSwap]; + iSwap = 1 - iSwap; + pNext->nState = 0; + for(i=0; inState; i++){ + int x = pThis->aState[i]; + switch( pRe->aOp[x] ){ + case RE_OP_MATCH: { + if( pRe->aArg[x]==c ) re_add_state(pNext, x+1); + break; + } + case RE_OP_ANY: { + re_add_state(pNext, x+1); + break; + } + case RE_OP_WORD: { + if( re_word_char(c) ) re_add_state(pNext, x+1); + break; + } + case RE_OP_NOTWORD: { + if( !re_word_char(c) ) re_add_state(pNext, x+1); + break; + } + case RE_OP_DIGIT: { + if( re_digit_char(c) ) re_add_state(pNext, x+1); + break; + } + case RE_OP_NOTDIGIT: { + if( !re_digit_char(c) ) re_add_state(pNext, x+1); + break; + } + case RE_OP_SPACE: { + if( re_space_char(c) ) re_add_state(pNext, x+1); + break; + } + case RE_OP_NOTSPACE: { + if( !re_space_char(c) ) re_add_state(pNext, x+1); + break; + } + case RE_OP_BOUNDARY: { + if( re_word_char(c)!=re_word_char(cPrev) ) re_add_state(pThis, x+1); + break; + } + case RE_OP_ANYSTAR: { + re_add_state(pNext, x); + re_add_state(pThis, x+1); + break; + } + case RE_OP_FORK: { + re_add_state(pThis, x+pRe->aArg[x]); + re_add_state(pThis, x+1); + break; + } + case RE_OP_GOTO: { + re_add_state(pThis, x+pRe->aArg[x]); + break; + } + case RE_OP_ACCEPT: { + rc = 1; + goto re_match_end; + } + case RE_OP_CC_INC: + case RE_OP_CC_EXC: { + int j = 1; + int n = pRe->aArg[x]; + int hit = 0; + for(j=1; j>0 && jaOp[x+j]==RE_OP_CC_VALUE ){ + if( pRe->aArg[x+j]==c ){ + hit = 1; + j = -1; + } + }else{ + if( pRe->aArg[x+j]<=c && pRe->aArg[x+j+1]>=c ){ + hit = 1; + j = -1; + }else{ + j++; + } + } + } + if( pRe->aOp[x]==RE_OP_CC_EXC ) hit = !hit; + if( hit ) re_add_state(pNext, x+n); + break; + } + } + } + } + for(i=0; inState; i++){ + if( pRe->aOp[pNext->aState[i]]==RE_OP_ACCEPT ){ rc = 1; break; } + } +re_match_end: + sqlite3_free(pToFree); + return rc; +} + +/* Resize the opcode and argument arrays for an RE under construction. +*/ +static int re_resize(ReCompiled *p, int N){ + char *aOp; + int *aArg; + aOp = sqlite3_realloc(p->aOp, N*sizeof(p->aOp[0])); + if( aOp==0 ) return 1; + p->aOp = aOp; + aArg = sqlite3_realloc(p->aArg, N*sizeof(p->aArg[0])); + if( aArg==0 ) return 1; + p->aArg = aArg; + p->nAlloc = N; + return 0; +} + +/* Insert a new opcode and argument into an RE under construction. The +** insertion point is just prior to existing opcode iBefore. +*/ +static int re_insert(ReCompiled *p, int iBefore, int op, int arg){ + int i; + if( p->nAlloc<=p->nState && re_resize(p, p->nAlloc*2) ) return 0; + for(i=p->nState; i>iBefore; i--){ + p->aOp[i] = p->aOp[i-1]; + p->aArg[i] = p->aArg[i-1]; + } + p->nState++; + p->aOp[iBefore] = op; + p->aArg[iBefore] = arg; + return iBefore; +} + +/* Append a new opcode and argument to the end of the RE under construction. +*/ +static int re_append(ReCompiled *p, int op, int arg){ + return re_insert(p, p->nState, op, arg); +} + +/* Make a copy of N opcodes starting at iStart onto the end of the RE +** under construction. +*/ +static void re_copy(ReCompiled *p, int iStart, int N){ + if( p->nState+N>=p->nAlloc && re_resize(p, p->nAlloc*2+N) ) return; + memcpy(&p->aOp[p->nState], &p->aOp[iStart], N*sizeof(p->aOp[0])); + memcpy(&p->aArg[p->nState], &p->aArg[iStart], N*sizeof(p->aArg[0])); + p->nState += N; +} + +/* Return true if c is a hexadecimal digit character: [0-9a-fA-F] +** If c is a hex digit, also set *pV = (*pV)*16 + valueof(c). If +** c is not a hex digit *pV is unchanged. +*/ +static int re_hex(int c, int *pV){ + if( c>='0' && c<='9' ){ + c -= '0'; + }else if( c>='a' && c<='f' ){ + c -= 'a' - 10; + }else if( c>='A' && c<='F' ){ + c -= 'A' - 10; + }else{ + return 0; + } + *pV = (*pV)*16 + (c & 0xff); + return 1; +} + +/* A backslash character has been seen, read the next character and +** return its interpretation. +*/ +static unsigned re_esc_char(ReCompiled *p){ + static const char zEsc[] = "afnrtv\\()*.+?[$^{|}]"; + static const char zTrans[] = "\a\f\n\r\t\v"; + int i, v = 0; + char c; + if( p->sIn.i>=p->sIn.mx ) return 0; + c = p->sIn.z[p->sIn.i]; + if( c=='u' && p->sIn.i+4sIn.mx ){ + const unsigned char *zIn = p->sIn.z + p->sIn.i; + if( re_hex(zIn[1],&v) + && re_hex(zIn[2],&v) + && re_hex(zIn[3],&v) + && re_hex(zIn[4],&v) + ){ + p->sIn.i += 5; + return v; + } + } + if( c=='x' && p->sIn.i+2sIn.mx ){ + const unsigned char *zIn = p->sIn.z + p->sIn.i; + if( re_hex(zIn[1],&v) + && re_hex(zIn[2],&v) + ){ + p->sIn.i += 3; + return v; + } + } + for(i=0; zEsc[i] && zEsc[i]!=c; i++){} + if( zEsc[i] ){ + if( i<6 ) c = zTrans[i]; + p->sIn.i++; + }else{ + p->zErr = "unknown \\ escape"; + } + return c; +} + +/* Forward declaration */ +static const char *re_subcompile_string(ReCompiled*); + +/* Peek at the next byte of input */ +static unsigned char rePeek(ReCompiled *p){ + return p->sIn.isIn.mx ? p->sIn.z[p->sIn.i] : 0; +} + +/* Compile RE text into a sequence of opcodes. Continue up to the +** first unmatched ")" character, then return. If an error is found, +** return a pointer to the error message string. +*/ +static const char *re_subcompile_re(ReCompiled *p){ + const char *zErr; + int iStart, iEnd, iGoto; + iStart = p->nState; + zErr = re_subcompile_string(p); + if( zErr ) return zErr; + while( rePeek(p)=='|' ){ + iEnd = p->nState; + re_insert(p, iStart, RE_OP_FORK, iEnd + 2 - iStart); + iGoto = re_append(p, RE_OP_GOTO, 0); + p->sIn.i++; + zErr = re_subcompile_string(p); + if( zErr ) return zErr; + p->aArg[iGoto] = p->nState - iGoto; + } + return 0; +} + +/* Compile an element of regular expression text (anything that can be +** an operand to the "|" operator). Return NULL on success or a pointer +** to the error message if there is a problem. +*/ +static const char *re_subcompile_string(ReCompiled *p){ + int iPrev = -1; + int iStart; + unsigned c; + const char *zErr; + while( (c = p->xNextChar(&p->sIn))!=0 ){ + iStart = p->nState; + switch( c ){ + case '|': + case '$': + case ')': { + p->sIn.i--; + return 0; + } + case '(': { + zErr = re_subcompile_re(p); + if( zErr ) return zErr; + if( rePeek(p)!=')' ) return "unmatched '('"; + p->sIn.i++; + break; + } + case '.': { + if( rePeek(p)=='*' ){ + re_append(p, RE_OP_ANYSTAR, 0); + p->sIn.i++; + }else{ + re_append(p, RE_OP_ANY, 0); + } + break; + } + case '*': { + if( iPrev<0 ) return "'*' without operand"; + re_insert(p, iPrev, RE_OP_GOTO, p->nState - iPrev + 1); + re_append(p, RE_OP_FORK, iPrev - p->nState + 1); + break; + } + case '+': { + if( iPrev<0 ) return "'+' without operand"; + re_append(p, RE_OP_FORK, iPrev - p->nState); + break; + } + case '?': { + if( iPrev<0 ) return "'?' without operand"; + re_insert(p, iPrev, RE_OP_FORK, p->nState - iPrev+1); + break; + } + case '{': { + int m = 0, n = 0; + int sz, j; + if( iPrev<0 ) return "'{m,n}' without operand"; + while( (c=rePeek(p))>='0' && c<='9' ){ m = m*10 + c - '0'; p->sIn.i++; } + n = m; + if( c==',' ){ + p->sIn.i++; + n = 0; + while( (c=rePeek(p))>='0' && c<='9' ){ n = n*10 + c-'0'; p->sIn.i++; } + } + if( c!='}' ) return "unmatched '{'"; + if( n>0 && nsIn.i++; + sz = p->nState - iPrev; + if( m==0 ){ + if( n==0 ) return "both m and n are zero in '{m,n}'"; + re_insert(p, iPrev, RE_OP_FORK, sz+1); + n--; + }else{ + for(j=1; j0 ){ + re_append(p, RE_OP_FORK, -sz); + } + break; + } + case '[': { + int iFirst = p->nState; + if( rePeek(p)=='^' ){ + re_append(p, RE_OP_CC_EXC, 0); + p->sIn.i++; + }else{ + re_append(p, RE_OP_CC_INC, 0); + } + while( (c = p->xNextChar(&p->sIn))!=0 ){ + if( c=='[' && rePeek(p)==':' ){ + return "POSIX character classes not supported"; + } + if( c=='\\' ) c = re_esc_char(p); + if( rePeek(p)=='-' ){ + re_append(p, RE_OP_CC_RANGE, c); + p->sIn.i++; + c = p->xNextChar(&p->sIn); + if( c=='\\' ) c = re_esc_char(p); + re_append(p, RE_OP_CC_RANGE, c); + }else{ + re_append(p, RE_OP_CC_VALUE, c); + } + if( rePeek(p)==']' ){ p->sIn.i++; break; } + } + if( c==0 ) return "unclosed '['"; + p->aArg[iFirst] = p->nState - iFirst; + break; + } + case '\\': { + int specialOp = 0; + switch( rePeek(p) ){ + case 'b': specialOp = RE_OP_BOUNDARY; break; + case 'd': specialOp = RE_OP_DIGIT; break; + case 'D': specialOp = RE_OP_NOTDIGIT; break; + case 's': specialOp = RE_OP_SPACE; break; + case 'S': specialOp = RE_OP_NOTSPACE; break; + case 'w': specialOp = RE_OP_WORD; break; + case 'W': specialOp = RE_OP_NOTWORD; break; + } + if( specialOp ){ + p->sIn.i++; + re_append(p, specialOp, 0); + }else{ + c = re_esc_char(p); + re_append(p, RE_OP_MATCH, c); + } + break; + } + default: { + re_append(p, RE_OP_MATCH, c); + break; + } + } + iPrev = iStart; + } + return 0; +} + +/* Free and reclaim all the memory used by a previously compiled +** regular expression. Applications should invoke this routine once +** for every call to re_compile() to avoid memory leaks. +*/ +void re_free(ReCompiled *pRe){ + if( pRe ){ + sqlite3_free(pRe->aOp); + sqlite3_free(pRe->aArg); + sqlite3_free(pRe); + } +} + +/* +** Compile a textual regular expression in zIn[] into a compiled regular +** expression suitable for us by re_match() and return a pointer to the +** compiled regular expression in *ppRe. Return NULL on success or an +** error message if something goes wrong. +*/ +const char *re_compile(ReCompiled **ppRe, const char *zIn, int noCase){ + ReCompiled *pRe; + const char *zErr; + int i, j; + + *ppRe = 0; + pRe = sqlite3_malloc( sizeof(*pRe) ); + if( pRe==0 ){ + return "out of memory"; + } + memset(pRe, 0, sizeof(*pRe)); + pRe->xNextChar = noCase ? re_next_char_nocase : re_next_char; + if( re_resize(pRe, 30) ){ + re_free(pRe); + return "out of memory"; + } + if( zIn[0]=='^' ){ + zIn++; + }else{ + re_append(pRe, RE_OP_ANYSTAR, 0); + } + pRe->sIn.z = (unsigned char*)zIn; + pRe->sIn.i = 0; + pRe->sIn.mx = (int)strlen(zIn); + zErr = re_subcompile_re(pRe); + if( zErr ){ + re_free(pRe); + return zErr; + } + if( rePeek(pRe)=='$' && pRe->sIn.i+1>=pRe->sIn.mx ){ + re_append(pRe, RE_OP_MATCH, RE_EOF); + re_append(pRe, RE_OP_ACCEPT, 0); + *ppRe = pRe; + }else if( pRe->sIn.i>=pRe->sIn.mx ){ + re_append(pRe, RE_OP_ACCEPT, 0); + *ppRe = pRe; + }else{ + re_free(pRe); + return "unrecognized character"; + } + + /* The following is a performance optimization. If the regex begins with + ** ".*" (if the input regex lacks an initial "^") and afterwards there are + ** one or more matching characters, enter those matching characters into + ** zInit[]. The re_match() routine can then search ahead in the input + ** string looking for the initial match without having to run the whole + ** regex engine over the string. Do not worry able trying to match + ** unicode characters beyond plane 0 - those are very rare and this is + ** just an optimization. */ + if( pRe->aOp[0]==RE_OP_ANYSTAR ){ + for(j=0, i=1; jzInit)-2 && pRe->aOp[i]==RE_OP_MATCH; i++){ + unsigned x = pRe->aArg[i]; + if( x<=127 ){ + pRe->zInit[j++] = x; + }else if( x<=0xfff ){ + pRe->zInit[j++] = 0xc0 | (x>>6); + pRe->zInit[j++] = 0x80 | (x&0x3f); + }else if( x<=0xffff ){ + pRe->zInit[j++] = 0xd0 | (x>>12); + pRe->zInit[j++] = 0x80 | ((x>>6)&0x3f); + pRe->zInit[j++] = 0x80 | (x&0x3f); + }else{ + break; + } + } + if( j>0 && pRe->zInit[j-1]==0 ) j--; + pRe->nInit = j; + } + return pRe->zErr; +} + +/* +** Implementation of the regexp() SQL function. This function implements +** the build-in REGEXP operator. The first argument to the function is the +** pattern and the second argument is the string. So, the SQL statements: +** +** A REGEXP B +** +** is implemented as regexp(B,A). +*/ +static void re_sql_func( + sqlite3_context *context, + int argc, + sqlite3_value **argv +){ + ReCompiled *pRe; /* Compiled regular expression */ + const char *zPattern; /* The regular expression */ + const unsigned char *zStr;/* String being searched */ + const char *zErr; /* Compile error message */ + + pRe = sqlite3_get_auxdata(context, 0); + if( pRe==0 ){ + zPattern = (const char*)sqlite3_value_text(argv[0]); + if( zPattern==0 ) return; + zErr = re_compile(&pRe, zPattern, 0); + if( zErr ){ + re_free(pRe); + sqlite3_result_error(context, zErr, -1); + return; + } + if( pRe==0 ){ + sqlite3_result_error_nomem(context); + return; + } + sqlite3_set_auxdata(context, 0, pRe, (void(*)(void*))re_free); + } + zStr = (const unsigned char*)sqlite3_value_text(argv[1]); + if( zStr!=0 ){ + sqlite3_result_int(context, re_match(pRe, zStr, -1)); + } +} + +/* +** Invoke this routine to register the regexp() function with the +** SQLite database connection. +*/ +#ifdef _WIN32 +__declspec(dllexport) +#endif +int sqlite3_regexp_init( + sqlite3 *db, + char **pzErrMsg, + const sqlite3_api_routines *pApi +){ + int rc = SQLITE_OK; + SQLITE_EXTENSION_INIT2(pApi); + rc = sqlite3_create_function(db, "regexp", 2, SQLITE_UTF8, 0, + re_sql_func, 0, 0); + return rc; +} ADDED ext/misc/spellfix.c Index: ext/misc/spellfix.c ================================================================== --- /dev/null +++ ext/misc/spellfix.c @@ -0,0 +1,2835 @@ +/* +** 2012 April 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 module implements the spellfix1 VIRTUAL TABLE that can be used +** to search a large vocabulary for close matches. See separate +** documentation files (spellfix1.wiki and editdist3.wiki) for details. +*/ +#include "sqlite3ext.h" +SQLITE_EXTENSION_INIT1 + +#include +#include +#include +#include +#define ALWAYS(X) 1 +#define NEVER(X) 0 +typedef unsigned char u8; +typedef unsigned short u16; +#include + +/* +** Character classes for ASCII characters: +** +** 0 '' Silent letters: H W +** 1 'A' Any vowel: A E I O U (Y) +** 2 'B' A bilabeal stop or fricative: B F P V W +** 3 'C' Other fricatives or back stops: C G J K Q S X Z +** 4 'D' Alveolar stops: D T +** 5 'H' Letter H at the beginning of a word +** 6 'L' Glide: L +** 7 'R' Semivowel: R +** 8 'M' Nasals: M N +** 9 'Y' Letter Y at the beginning of a word. +** 10 '9' Digits: 0 1 2 3 4 5 6 7 8 9 +** 11 ' ' White space +** 12 '?' Other. +*/ +#define CCLASS_SILENT 0 +#define CCLASS_VOWEL 1 +#define CCLASS_B 2 +#define CCLASS_C 3 +#define CCLASS_D 4 +#define CCLASS_H 5 +#define CCLASS_L 6 +#define CCLASS_R 7 +#define CCLASS_M 8 +#define CCLASS_Y 9 +#define CCLASS_DIGIT 10 +#define CCLASS_SPACE 11 +#define CCLASS_OTHER 12 + +/* +** The following table gives the character class for non-initial ASCII +** characters. +*/ +static const unsigned char midClass[] = { + /* */ CCLASS_OTHER, /* */ CCLASS_OTHER, /* */ CCLASS_OTHER, + /* */ CCLASS_OTHER, /* */ CCLASS_OTHER, /* */ CCLASS_OTHER, + /* */ CCLASS_OTHER, /* */ CCLASS_OTHER, /* */ CCLASS_OTHER, + /* */ CCLASS_SPACE, /* */ CCLASS_OTHER, /* */ CCLASS_OTHER, + /* */ CCLASS_SPACE, /* */ CCLASS_SPACE, /* */ CCLASS_OTHER, + /* */ CCLASS_OTHER, /* */ CCLASS_OTHER, /* */ CCLASS_OTHER, + /* */ CCLASS_OTHER, /* */ CCLASS_OTHER, /* */ CCLASS_OTHER, + /* */ CCLASS_OTHER, /* */ CCLASS_OTHER, /* */ CCLASS_OTHER, + /* */ CCLASS_OTHER, /* */ CCLASS_OTHER, /* */ CCLASS_OTHER, + /* */ CCLASS_OTHER, /* */ CCLASS_OTHER, /* */ CCLASS_OTHER, + /* */ CCLASS_OTHER, /* */ CCLASS_OTHER, /* */ CCLASS_SPACE, + /* ! */ CCLASS_OTHER, /* " */ CCLASS_OTHER, /* # */ CCLASS_OTHER, + /* $ */ CCLASS_OTHER, /* % */ CCLASS_OTHER, /* & */ CCLASS_OTHER, + /* ' */ CCLASS_SILENT, /* ( */ CCLASS_OTHER, /* ) */ CCLASS_OTHER, + /* * */ CCLASS_OTHER, /* + */ CCLASS_OTHER, /* , */ CCLASS_OTHER, + /* - */ CCLASS_OTHER, /* . */ CCLASS_OTHER, /* / */ CCLASS_OTHER, + /* 0 */ CCLASS_DIGIT, /* 1 */ CCLASS_DIGIT, /* 2 */ CCLASS_DIGIT, + /* 3 */ CCLASS_DIGIT, /* 4 */ CCLASS_DIGIT, /* 5 */ CCLASS_DIGIT, + /* 6 */ CCLASS_DIGIT, /* 7 */ CCLASS_DIGIT, /* 8 */ CCLASS_DIGIT, + /* 9 */ CCLASS_DIGIT, /* : */ CCLASS_OTHER, /* ; */ CCLASS_OTHER, + /* < */ CCLASS_OTHER, /* = */ CCLASS_OTHER, /* > */ CCLASS_OTHER, + /* ? */ CCLASS_OTHER, /* @ */ CCLASS_OTHER, /* A */ CCLASS_VOWEL, + /* B */ CCLASS_B, /* C */ CCLASS_C, /* D */ CCLASS_D, + /* E */ CCLASS_VOWEL, /* F */ CCLASS_B, /* G */ CCLASS_C, + /* H */ CCLASS_SILENT, /* I */ CCLASS_VOWEL, /* J */ CCLASS_C, + /* K */ CCLASS_C, /* L */ CCLASS_L, /* M */ CCLASS_M, + /* N */ CCLASS_M, /* O */ CCLASS_VOWEL, /* P */ CCLASS_B, + /* Q */ CCLASS_C, /* R */ CCLASS_R, /* S */ CCLASS_C, + /* T */ CCLASS_D, /* U */ CCLASS_VOWEL, /* V */ CCLASS_B, + /* W */ CCLASS_B, /* X */ CCLASS_C, /* Y */ CCLASS_VOWEL, + /* Z */ CCLASS_C, /* [ */ CCLASS_OTHER, /* \ */ CCLASS_OTHER, + /* ] */ CCLASS_OTHER, /* ^ */ CCLASS_OTHER, /* _ */ CCLASS_OTHER, + /* ` */ CCLASS_OTHER, /* a */ CCLASS_VOWEL, /* b */ CCLASS_B, + /* c */ CCLASS_C, /* d */ CCLASS_D, /* e */ CCLASS_VOWEL, + /* f */ CCLASS_B, /* g */ CCLASS_C, /* h */ CCLASS_SILENT, + /* i */ CCLASS_VOWEL, /* j */ CCLASS_C, /* k */ CCLASS_C, + /* l */ CCLASS_L, /* m */ CCLASS_M, /* n */ CCLASS_M, + /* o */ CCLASS_VOWEL, /* p */ CCLASS_B, /* q */ CCLASS_C, + /* r */ CCLASS_R, /* s */ CCLASS_C, /* t */ CCLASS_D, + /* u */ CCLASS_VOWEL, /* v */ CCLASS_B, /* w */ CCLASS_B, + /* x */ CCLASS_C, /* y */ CCLASS_VOWEL, /* z */ CCLASS_C, + /* { */ CCLASS_OTHER, /* | */ CCLASS_OTHER, /* } */ CCLASS_OTHER, + /* ~ */ CCLASS_OTHER, /* */ CCLASS_OTHER, +}; +/* +** This tables gives the character class for ASCII characters that form the +** initial character of a word. The only difference from midClass is with +** the letters H, W, and Y. +*/ +static const unsigned char initClass[] = { + /* */ CCLASS_OTHER, /* */ CCLASS_OTHER, /* */ CCLASS_OTHER, + /* */ CCLASS_OTHER, /* */ CCLASS_OTHER, /* */ CCLASS_OTHER, + /* */ CCLASS_OTHER, /* */ CCLASS_OTHER, /* */ CCLASS_OTHER, + /* */ CCLASS_SPACE, /* */ CCLASS_OTHER, /* */ CCLASS_OTHER, + /* */ CCLASS_SPACE, /* */ CCLASS_SPACE, /* */ CCLASS_OTHER, + /* */ CCLASS_OTHER, /* */ CCLASS_OTHER, /* */ CCLASS_OTHER, + /* */ CCLASS_OTHER, /* */ CCLASS_OTHER, /* */ CCLASS_OTHER, + /* */ CCLASS_OTHER, /* */ CCLASS_OTHER, /* */ CCLASS_OTHER, + /* */ CCLASS_OTHER, /* */ CCLASS_OTHER, /* */ CCLASS_OTHER, + /* */ CCLASS_OTHER, /* */ CCLASS_OTHER, /* */ CCLASS_OTHER, + /* */ CCLASS_OTHER, /* */ CCLASS_OTHER, /* */ CCLASS_SPACE, + /* ! */ CCLASS_OTHER, /* " */ CCLASS_OTHER, /* # */ CCLASS_OTHER, + /* $ */ CCLASS_OTHER, /* % */ CCLASS_OTHER, /* & */ CCLASS_OTHER, + /* ' */ CCLASS_OTHER, /* ( */ CCLASS_OTHER, /* ) */ CCLASS_OTHER, + /* * */ CCLASS_OTHER, /* + */ CCLASS_OTHER, /* , */ CCLASS_OTHER, + /* - */ CCLASS_OTHER, /* . */ CCLASS_OTHER, /* / */ CCLASS_OTHER, + /* 0 */ CCLASS_DIGIT, /* 1 */ CCLASS_DIGIT, /* 2 */ CCLASS_DIGIT, + /* 3 */ CCLASS_DIGIT, /* 4 */ CCLASS_DIGIT, /* 5 */ CCLASS_DIGIT, + /* 6 */ CCLASS_DIGIT, /* 7 */ CCLASS_DIGIT, /* 8 */ CCLASS_DIGIT, + /* 9 */ CCLASS_DIGIT, /* : */ CCLASS_OTHER, /* ; */ CCLASS_OTHER, + /* < */ CCLASS_OTHER, /* = */ CCLASS_OTHER, /* > */ CCLASS_OTHER, + /* ? */ CCLASS_OTHER, /* @ */ CCLASS_OTHER, /* A */ CCLASS_VOWEL, + /* B */ CCLASS_B, /* C */ CCLASS_C, /* D */ CCLASS_D, + /* E */ CCLASS_VOWEL, /* F */ CCLASS_B, /* G */ CCLASS_C, + /* H */ CCLASS_SILENT, /* I */ CCLASS_VOWEL, /* J */ CCLASS_C, + /* K */ CCLASS_C, /* L */ CCLASS_L, /* M */ CCLASS_M, + /* N */ CCLASS_M, /* O */ CCLASS_VOWEL, /* P */ CCLASS_B, + /* Q */ CCLASS_C, /* R */ CCLASS_R, /* S */ CCLASS_C, + /* T */ CCLASS_D, /* U */ CCLASS_VOWEL, /* V */ CCLASS_B, + /* W */ CCLASS_B, /* X */ CCLASS_C, /* Y */ CCLASS_Y, + /* Z */ CCLASS_C, /* [ */ CCLASS_OTHER, /* \ */ CCLASS_OTHER, + /* ] */ CCLASS_OTHER, /* ^ */ CCLASS_OTHER, /* _ */ CCLASS_OTHER, + /* ` */ CCLASS_OTHER, /* a */ CCLASS_VOWEL, /* b */ CCLASS_B, + /* c */ CCLASS_C, /* d */ CCLASS_D, /* e */ CCLASS_VOWEL, + /* f */ CCLASS_B, /* g */ CCLASS_C, /* h */ CCLASS_SILENT, + /* i */ CCLASS_VOWEL, /* j */ CCLASS_C, /* k */ CCLASS_C, + /* l */ CCLASS_L, /* m */ CCLASS_M, /* n */ CCLASS_M, + /* o */ CCLASS_VOWEL, /* p */ CCLASS_B, /* q */ CCLASS_C, + /* r */ CCLASS_R, /* s */ CCLASS_C, /* t */ CCLASS_D, + /* u */ CCLASS_VOWEL, /* v */ CCLASS_B, /* w */ CCLASS_B, + /* x */ CCLASS_C, /* y */ CCLASS_Y, /* z */ CCLASS_C, + /* { */ CCLASS_OTHER, /* | */ CCLASS_OTHER, /* } */ CCLASS_OTHER, + /* ~ */ CCLASS_OTHER, /* */ CCLASS_OTHER, +}; + +/* +** Mapping from the character class number (0-13) to a symbol for each +** character class. Note that initClass[] can be used to map the class +** symbol back into the class number. +*/ +static const unsigned char className[] = ".ABCDHLRMY9 ?"; + +/* +** Generate a "phonetic hash" from a string of ASCII characters +** in zIn[0..nIn-1]. +** +** * Map characters by character class as defined above. +** * Omit double-letters +** * Omit vowels beside R and L +** * Omit T when followed by CH +** * Omit W when followed by R +** * Omit D when followed by J or G +** * Omit K in KN or G in GN at the beginning of a word +** +** Space to hold the result is obtained from sqlite3_malloc() +** +** Return NULL if memory allocation fails. +*/ +static unsigned char *phoneticHash(const unsigned char *zIn, int nIn){ + unsigned char *zOut = sqlite3_malloc( nIn + 1 ); + int i; + int nOut = 0; + char cPrev = 0x77; + char cPrevX = 0x77; + const unsigned char *aClass = initClass; + + if( zOut==0 ) return 0; + if( nIn>2 ){ + switch( zIn[0] ){ + case 'g': + case 'k': { + if( zIn[1]=='n' ){ zIn++; nIn--; } + break; + } + } + } + for(i=0; i=0 ); + if( nOut==0 || c!=zOut[nOut-1] ) zOut[nOut++] = c; + } + zOut[nOut] = 0; + return zOut; +} + +/* +** This is an SQL function wrapper around phoneticHash(). See +** the description of phoneticHash() for additional information. +*/ +static void phoneticHashSqlFunc( + sqlite3_context *context, + int argc, + sqlite3_value **argv +){ + const unsigned char *zIn; + unsigned char *zOut; + + zIn = sqlite3_value_text(argv[0]); + if( zIn==0 ) return; + zOut = phoneticHash(zIn, sqlite3_value_bytes(argv[0])); + if( zOut==0 ){ + sqlite3_result_error_nomem(context); + }else{ + sqlite3_result_text(context, (char*)zOut, -1, sqlite3_free); + } +} + +/* +** Return the character class number for a character given its +** context. +*/ +static char characterClass(char cPrev, char c){ + return cPrev==0 ? initClass[c&0x7f] : midClass[c&0x7f]; +} + +/* +** Return the cost of inserting or deleting character c immediately +** following character cPrev. If cPrev==0, that means c is the first +** character of the word. +*/ +static int insertOrDeleteCost(char cPrev, char c, char cNext){ + char classC = characterClass(cPrev, c); + char classCprev; + + if( classC==CCLASS_SILENT ){ + /* Insert or delete "silent" characters such as H or W */ + return 1; + } + if( cPrev==c ){ + /* Repeated characters, or miss a repeat */ + return 10; + } + if( classC==CCLASS_VOWEL && (cPrev=='r' || cNext=='r') ){ + return 20; /* Insert a vowel before or after 'r' */ + } + classCprev = characterClass(cPrev, cPrev); + if( classC==classCprev ){ + if( classC==CCLASS_VOWEL ){ + /* Remove or add a new vowel to a vowel cluster */ + return 15; + }else{ + /* Remove or add a consonant not in the same class */ + return 50; + } + } + + /* any other character insertion or deletion */ + return 100; +} + +/* +** Divide the insertion cost by this factor when appending to the +** end of the word. +*/ +#define FINAL_INS_COST_DIV 4 + +/* +** Return the cost of substituting cTo in place of cFrom assuming +** the previous character is cPrev. If cPrev==0 then cTo is the first +** character of the word. +*/ +static int substituteCost(char cPrev, char cFrom, char cTo){ + char classFrom, classTo; + if( cFrom==cTo ){ + /* Exact match */ + return 0; + } + if( cFrom==(cTo^0x20) && ((cTo>='A' && cTo<='Z') || (cTo>='a' && cTo<='z')) ){ + /* differ only in case */ + return 0; + } + classFrom = characterClass(cPrev, cFrom); + classTo = characterClass(cPrev, cTo); + if( classFrom==classTo ){ + /* Same character class */ + return 40; + } + if( classFrom>=CCLASS_B && classFrom<=CCLASS_Y + && classTo>=CCLASS_B && classTo<=CCLASS_Y ){ + /* Convert from one consonant to another, but in a different class */ + return 75; + } + /* Any other subsitution */ + return 100; +} + +/* +** Given two strings zA and zB which are pure ASCII, return the cost +** of transforming zA into zB. If zA ends with '*' assume that it is +** a prefix of zB and give only minimal penalty for extra characters +** on the end of zB. +** +** Smaller numbers mean a closer match. +** +** Negative values indicate an error: +** -1 One of the inputs is NULL +** -2 Non-ASCII characters on input +** -3 Unable to allocate memory +** +** If pnMatch is not NULL, then *pnMatch is set to the number of bytes +** of zB that matched the pattern in zA. If zA does not end with a '*', +** then this value is always the number of bytes in zB (i.e. strlen(zB)). +** If zA does end in a '*', then it is the number of bytes in the prefix +** of zB that was deemed to match zA. +*/ +static int editdist1(const char *zA, const char *zB, int *pnMatch){ + int nA, nB; /* Number of characters in zA[] and zB[] */ + int xA, xB; /* Loop counters for zA[] and zB[] */ + char cA, cB; /* Current character of zA and zB */ + char cAprev, cBprev; /* Previous character of zA and zB */ + char cAnext, cBnext; /* Next character in zA and zB */ + int d; /* North-west cost value */ + int dc = 0; /* North-west character value */ + int res; /* Final result */ + int *m; /* The cost matrix */ + char *cx; /* Corresponding character values */ + int *toFree = 0; /* Malloced space */ + int mStack[60+15]; /* Stack space to use if not too much is needed */ + int nMatch = 0; + + /* Early out if either input is NULL */ + if( zA==0 || zB==0 ) return -1; + + /* Skip any common prefix */ + while( zA[0] && zA[0]==zB[0] ){ dc = zA[0]; zA++; zB++; nMatch++; } + if( pnMatch ) *pnMatch = nMatch; + if( zA[0]==0 && zB[0]==0 ) return 0; + +#if 0 + printf("A=\"%s\" B=\"%s\" dc=%c\n", zA, zB, dc?dc:' '); +#endif + + /* Verify input strings and measure their lengths */ + for(nA=0; zA[nA]; nA++){ + if( zA[nA]&0x80 ) return -2; + } + for(nB=0; zB[nB]; nB++){ + if( zB[nB]&0x80 ) return -2; + } + + /* Special processing if either string is empty */ + if( nA==0 ){ + cBprev = dc; + for(xB=res=0; (cB = zB[xB])!=0; xB++){ + res += insertOrDeleteCost(cBprev, cB, zB[xB+1])/FINAL_INS_COST_DIV; + cBprev = cB; + } + return res; + } + if( nB==0 ){ + cAprev = dc; + for(xA=res=0; (cA = zA[xA])!=0; xA++){ + res += insertOrDeleteCost(cAprev, cA, zA[xA+1]); + cAprev = cA; + } + return res; + } + + /* A is a prefix of B */ + if( zA[0]=='*' && zA[1]==0 ) return 0; + + /* Allocate and initialize the Wagner matrix */ + if( nB<(sizeof(mStack)*4)/(sizeof(mStack[0])*5) ){ + m = mStack; + }else{ + m = toFree = sqlite3_malloc( (nB+1)*5*sizeof(m[0])/4 ); + if( m==0 ) return -3; + } + cx = (char*)&m[nB+1]; + + /* Compute the Wagner edit distance */ + m[0] = 0; + cx[0] = dc; + cBprev = dc; + for(xB=1; xB<=nB; xB++){ + cBnext = zB[xB]; + cB = zB[xB-1]; + cx[xB] = cB; + m[xB] = m[xB-1] + insertOrDeleteCost(cBprev, cB, cBnext); + cBprev = cB; + } + cAprev = dc; + for(xA=1; xA<=nA; xA++){ + int lastA = (xA==nA); + cA = zA[xA-1]; + cAnext = zA[xA]; + if( cA=='*' && lastA ) break; + d = m[0]; + dc = cx[0]; + m[0] = d + insertOrDeleteCost(cAprev, cA, cAnext); + cBprev = 0; + for(xB=1; xB<=nB; xB++){ + int totalCost, insCost, delCost, subCost, ncx; + cB = zB[xB-1]; + cBnext = zB[xB]; + + /* Cost to insert cB */ + insCost = insertOrDeleteCost(cx[xB-1], cB, cBnext); + if( lastA ) insCost /= FINAL_INS_COST_DIV; + + /* Cost to delete cA */ + delCost = insertOrDeleteCost(cx[xB], cA, cBnext); + + /* Cost to substitute cA->cB */ + subCost = substituteCost(cx[xB-1], cA, cB); + + /* Best cost */ + totalCost = insCost + m[xB-1]; + ncx = cB; + if( (delCost + m[xB])nLang; i++){ + EditDist3Cost *pCost, *pNext; + pCost = p->a[i].pCost; + while( pCost ){ + pNext = pCost->pNext; + sqlite3_free(pCost); + pCost = pNext; + } + } + sqlite3_free(p->a); + memset(p, 0, sizeof(*p)); +} +static void editDist3ConfigDelete(void *pIn){ + EditDist3Config *p = (EditDist3Config*)pIn; + editDist3ConfigClear(p); + sqlite3_free(p); +} + +/* +** Load all edit-distance weights from a table. +*/ +static int editDist3ConfigLoad( + EditDist3Config *p, /* The edit distance configuration to load */ + sqlite3 *db, /* Load from this database */ + const char *zTable /* Name of the table from which to load */ +){ + sqlite3_stmt *pStmt; + int rc, rc2; + char *zSql; + int iLangPrev = -9999; + EditDist3Lang *pLang = 0; + + zSql = sqlite3_mprintf("SELECT iLang, cFrom, cTo, iCost" + " FROM \"%w\" WHERE iLang>=0 ORDER BY iLang", zTable); + if( zSql==0 ) return SQLITE_NOMEM; + rc = sqlite3_prepare(db, zSql, -1, &pStmt, 0); + sqlite3_free(zSql); + if( rc ) return rc; + editDist3ConfigClear(p); + while( sqlite3_step(pStmt)==SQLITE_ROW ){ + int iLang = sqlite3_column_int(pStmt, 0); + const char *zFrom = (const char*)sqlite3_column_text(pStmt, 1); + int nFrom = zFrom ? sqlite3_column_bytes(pStmt, 1) : 0; + const char *zTo = (const char*)sqlite3_column_text(pStmt, 2); + int nTo = zTo ? sqlite3_column_bytes(pStmt, 2) : 0; + int iCost = sqlite3_column_int(pStmt, 3); + + assert( zFrom!=0 || nFrom==0 ); + assert( zTo!=0 || nTo==0 ); + if( nFrom>100 || nTo>100 ) continue; + if( iCost<0 ) continue; + if( pLang==0 || iLang!=iLangPrev ){ + EditDist3Lang *pNew; + pNew = sqlite3_realloc(p->a, (p->nLang+1)*sizeof(p->a[0])); + if( pNew==0 ){ rc = SQLITE_NOMEM; break; } + p->a = pNew; + pLang = &p->a[p->nLang]; + p->nLang++; + pLang->iLang = iLang; + pLang->iInsCost = 100; + pLang->iDelCost = 100; + pLang->iSubCost = 150; + pLang->pCost = 0; + iLangPrev = iLang; + } + if( nFrom==1 && zFrom[0]=='?' && nTo==0 ){ + pLang->iDelCost = iCost; + }else if( nFrom==0 && nTo==1 && zTo[0]=='?' ){ + pLang->iInsCost = iCost; + }else if( nFrom==1 && nTo==1 && zFrom[0]=='?' && zTo[0]=='?' ){ + pLang->iSubCost = iCost; + }else{ + EditDist3Cost *pCost; + int nExtra = nFrom + nTo - 4; + if( nExtra<0 ) nExtra = 0; + pCost = sqlite3_malloc( sizeof(*pCost) + nExtra ); + if( pCost==0 ){ rc = SQLITE_NOMEM; break; } + pCost->nFrom = nFrom; + pCost->nTo = nTo; + pCost->iCost = iCost; + memcpy(pCost->a, zFrom, nFrom); + memcpy(pCost->a + nFrom, zTo, nTo); + pCost->pNext = pLang->pCost; + pLang->pCost = pCost; + } + } + rc2 = sqlite3_finalize(pStmt); + if( rc==SQLITE_OK ) rc = rc2; + return rc; +} + +/* +** Return the length (in bytes) of a utf-8 character. Or return a maximum +** of N. +*/ +static int utf8Len(unsigned char c, int N){ + int len = 1; + if( c>0x7f ){ + if( (c&0xe0)==0xc0 ){ + len = 2; + }else if( (c&0xf0)==0xe0 ){ + len = 3; + }else{ + len = 4; + } + } + if( len>N ) len = N; + return len; +} + +/* +** Return TRUE (non-zero) if the To side of the given cost matches +** the given string. +*/ +static int matchTo(EditDist3Cost *p, const char *z, int n){ + if( p->nTo>n ) return 0; + if( strncmp(p->a+p->nFrom, z, p->nTo)!=0 ) return 0; + return 1; +} + +/* +** Return TRUE (non-zero) if the From side of the given cost matches +** the given string. +*/ +static int matchFrom(EditDist3Cost *p, const char *z, int n){ + assert( p->nFrom<=n ); + if( strncmp(p->a, z, p->nFrom)!=0 ) return 0; + return 1; +} + +/* +** Return TRUE (non-zero) of the next FROM character and the next TO +** character are the same. +*/ +static int matchFromTo( + EditDist3FromString *pStr, /* Left hand string */ + int n1, /* Index of comparison character on the left */ + const char *z2, /* Right-handl comparison character */ + int n2 /* Bytes remaining in z2[] */ +){ + int b1 = pStr->a[n1].nByte; + if( b1>n2 ) return 0; + if( memcmp(pStr->z+n1, z2, b1)!=0 ) return 0; + return 1; +} + +/* +** Delete an EditDist3FromString objecct +*/ +static void editDist3FromStringDelete(EditDist3FromString *p){ + int i; + if( p ){ + for(i=0; in; i++){ + sqlite3_free(p->a[i].apDel); + sqlite3_free(p->a[i].apSubst); + } + sqlite3_free(p); + } +} + +/* +** Create a EditDist3FromString object. +*/ +static EditDist3FromString *editDist3FromStringNew( + const EditDist3Lang *pLang, + const char *z, + int n +){ + EditDist3FromString *pStr; + EditDist3Cost *p; + int i; + + if( z==0 ) return 0; + if( n<0 ) n = (int)strlen(z); + pStr = sqlite3_malloc( sizeof(*pStr) + sizeof(pStr->a[0])*n + n + 1 ); + if( pStr==0 ) return 0; + pStr->a = (EditDist3From*)&pStr[1]; + memset(pStr->a, 0, sizeof(pStr->a[0])*n); + pStr->n = n; + pStr->z = (char*)&pStr->a[n]; + memcpy(pStr->z, z, n+1); + if( n && z[n-1]=='*' ){ + pStr->isPrefix = 1; + n--; + pStr->n--; + pStr->z[n] = 0; + }else{ + pStr->isPrefix = 0; + } + + for(i=0; ia[i]; + memset(pFrom, 0, sizeof(*pFrom)); + pFrom->nByte = utf8Len((unsigned char)z[i], n-i); + for(p=pLang->pCost; p; p=p->pNext){ + EditDist3Cost **apNew; + if( i+p->nFrom>n ) continue; + if( matchFrom(p, z+i, n-i)==0 ) continue; + if( p->nTo==0 ){ + apNew = sqlite3_realloc(pFrom->apDel, + sizeof(*apNew)*(pFrom->nDel+1)); + if( apNew==0 ) break; + pFrom->apDel = apNew; + apNew[pFrom->nDel++] = p; + }else{ + apNew = sqlite3_realloc(pFrom->apSubst, + sizeof(*apNew)*(pFrom->nSubst+1)); + if( apNew==0 ) break; + pFrom->apSubst = apNew; + apNew[pFrom->nSubst++] = p; + } + } + if( p ){ + editDist3FromStringDelete(pStr); + pStr = 0; + break; + } + } + return pStr; +} + +/* +** Update entry m[i] such that it is the minimum of its current value +** and m[j]+iCost. +** +** If the iCost is 1,000,000 or greater, then consider the cost to be +** infinite and skip the update. +*/ +static void updateCost( + unsigned int *m, + int i, + int j, + int iCost +){ + assert( iCost>=0 ); + if( iCost<10000 ){ + unsigned int b = m[j] + iCost; + if( bpCost; p; p=p->pNext){ + EditDist3Cost **apNew; + if( p->nFrom>0 ) continue; + if( i2+p->nTo>n2 ) continue; + if( matchTo(p, z2+i2, n2-i2)==0 ) continue; + a2[i2].nIns++; + apNew = sqlite3_realloc(a2[i2].apIns, sizeof(*apNew)*a2[i2].nIns); + if( apNew==0 ){ + res = -1; /* Out of memory */ + goto editDist3Abort; + } + a2[i2].apIns = apNew; + a2[i2].apIns[a2[i2].nIns-1] = p; + } + } + + /* Prepare to compute the minimum edit distance */ + szRow = f.n+1; + memset(m, 0x01, (n2+1)*szRow*sizeof(m[0])); + m[0] = 0; + + /* First fill in the top-row of the matrix with FROM deletion costs */ + for(i1=0; i1iDelCost); + for(k=0; knFrom, i1, p->iCost); + } + } + + /* Fill in all subsequent rows, top-to-bottom, left-to-right */ + for(i2=0; i2iInsCost); + for(k=0; knTo), rxp, p->iCost); + } + for(i1=0; i1iDelCost); + for(k=0; knFrom, cxp, p->iCost); + } + updateCost(m, cx, cxu, pLang->iInsCost); + if( matchFromTo(&f, i1, z2+i2, n2-i2) ){ + updateCost(m, cx, cxd, 0); + } + updateCost(m, cx, cxd, pLang->iSubCost); + for(k=0; knFrom+szRow*p->nTo, cxd, p->iCost); + } + } + } + } + +#if 0 /* Enable for debugging */ + printf(" ^"); + for(i1=0; i19999 ) printf(" ****"); + else printf(" %4d", v); + } + printf("\n"); + for(i2=0; i29999 ) printf(" ****"); + else printf(" %4d", v); + } + printf("\n"); + } +#endif + + /* Free memory allocations and return the result */ + res = (int)m[szRow*(n2+1)-1]; + n = n2; + if( f.isPrefix ){ + for(i2=1; i2<=n2; i2++){ + int b = m[szRow*i2-1]; + if( b<=res ){ + res = b; + n = i2 - 1; + } + } + } + if( pnMatch ){ + int nExtra = 0; + for(k=0; knLang; i++){ + if( pConfig->a[i].iLang==iLang ) return &pConfig->a[i]; + } + return &editDist3Lang; +} + +/* +** Function: editdist3(A,B,iLang) +** editdist3(tablename) +** +** Return the cost of transforming string A into string B using edit +** weights for iLang. +** +** The second form loads edit weights into memory from a table. +*/ +static void editDist3SqlFunc( + sqlite3_context *context, + int argc, + sqlite3_value **argv +){ + EditDist3Config *pConfig = (EditDist3Config*)sqlite3_user_data(context); + sqlite3 *db = sqlite3_context_db_handle(context); + int rc; + if( argc==1 ){ + const char *zTable = (const char*)sqlite3_value_text(argv[0]); + rc = editDist3ConfigLoad(pConfig, db, zTable); + if( rc ) sqlite3_result_error_code(context, rc); + }else{ + const char *zA = (const char*)sqlite3_value_text(argv[0]); + const char *zB = (const char*)sqlite3_value_text(argv[1]); + int nA = sqlite3_value_bytes(argv[0]); + int nB = sqlite3_value_bytes(argv[1]); + int iLang = argc==3 ? sqlite3_value_int(argv[2]) : 0; + const EditDist3Lang *pLang = editDist3FindLang(pConfig, iLang); + EditDist3FromString *pFrom; + int dist; + + pFrom = editDist3FromStringNew(pLang, zA, nA); + if( pFrom==0 ){ + sqlite3_result_error_nomem(context); + return; + } + dist = editDist3Core(pFrom, zB, nB, pLang, 0); + editDist3FromStringDelete(pFrom); + if( dist==(-1) ){ + sqlite3_result_error_nomem(context); + }else{ + sqlite3_result_int(context, dist); + } + } +} + +/* +** Register the editDist3 function with SQLite +*/ +static int editDist3Install(sqlite3 *db){ + int rc; + EditDist3Config *pConfig = sqlite3_malloc( sizeof(*pConfig) ); + if( pConfig==0 ) return SQLITE_NOMEM; + memset(pConfig, 0, sizeof(*pConfig)); + rc = sqlite3_create_function_v2(db, "editdist3", + 2, SQLITE_UTF8, pConfig, editDist3SqlFunc, 0, 0, 0); + if( rc==SQLITE_OK ){ + rc = sqlite3_create_function_v2(db, "editdist3", + 3, SQLITE_UTF8, pConfig, editDist3SqlFunc, 0, 0, 0); + } + if( rc==SQLITE_OK ){ + rc = sqlite3_create_function_v2(db, "editdist3", + 1, SQLITE_UTF8, pConfig, editDist3SqlFunc, 0, 0, + editDist3ConfigDelete); + }else{ + sqlite3_free(pConfig); + } + return rc; +} +/* End configurable cost unicode edit distance routines +****************************************************************************** +****************************************************************************** +** Begin transliterate unicode-to-ascii implementation +*/ + +#if !SQLITE_AMALGAMATION +/* +** This lookup table is used to help decode the first byte of +** a multi-byte UTF8 character. +*/ +static const unsigned char sqlite3Utf8Trans1[] = { + 0x00, 0x01, 0x02, 0x03, 0x04, 0x05, 0x06, 0x07, + 0x08, 0x09, 0x0a, 0x0b, 0x0c, 0x0d, 0x0e, 0x0f, + 0x10, 0x11, 0x12, 0x13, 0x14, 0x15, 0x16, 0x17, + 0x18, 0x19, 0x1a, 0x1b, 0x1c, 0x1d, 0x1e, 0x1f, + 0x00, 0x01, 0x02, 0x03, 0x04, 0x05, 0x06, 0x07, + 0x08, 0x09, 0x0a, 0x0b, 0x0c, 0x0d, 0x0e, 0x0f, + 0x00, 0x01, 0x02, 0x03, 0x04, 0x05, 0x06, 0x07, + 0x00, 0x01, 0x02, 0x03, 0x00, 0x01, 0x00, 0x00, +}; +#endif + +/* +** Return the value of the first UTF-8 character in the string. +*/ +static int utf8Read(const unsigned char *z, int n, int *pSize){ + int c, i; + + /* All callers to this routine (in the current implementation) + ** always have n>0. */ + if( NEVER(n==0) ){ + c = i = 0; + }else{ + c = z[0]; + i = 1; + if( c>=0xc0 ){ + c = sqlite3Utf8Trans1[c-0xc0]; + while( i0 ){ + c = utf8Read(zIn, nIn, &sz); + zIn += sz; + nIn -= sz; + if( c<=127 ){ + zOut[nOut++] = c; + }else{ + int xTop, xBtm, x; + xTop = sizeof(translit)/sizeof(translit[0]) - 1; + xBtm = 0; + while( xTop>=xBtm ){ + x = (xTop + xBtm)/2; + if( translit[x].cFrom==c ){ + zOut[nOut++] = translit[x].cTo0; + if( translit[x].cTo1 ){ + zOut[nOut++] = translit[x].cTo1; + /* Add an extra "ch" after the "sh" for Щ and щ */ + if( c==0x0429 || c== 0x0449 ){ + zOut[nOut++] = 'c'; + zOut[nOut++] = 'h'; + } + } + c = 0; + break; + }else if( translit[x].cFrom>c ){ + xTop = x-1; + }else{ + xBtm = x+1; + } + } + if( c ) zOut[nOut++] = '?'; + } + } + zOut[nOut] = 0; + return zOut; +} + +/* +** Return the number of characters in the shortest prefix of the input +** string that transliterates to an ASCII string nTrans bytes or longer. +** Or, if the transliteration of the input string is less than nTrans +** bytes in size, return the number of characters in the input string. +*/ +static int translen_to_charlen(const char *zIn, int nIn, int nTrans){ + int i, c, sz, nOut; + int nChar; + + i = nOut = 0; + for(nChar=0; i=128 ){ + int xTop, xBtm, x; + xTop = sizeof(translit)/sizeof(translit[0]) - 1; + xBtm = 0; + while( xTop>=xBtm ){ + x = (xTop + xBtm)/2; + if( translit[x].cFrom==c ){ + if( translit[x].cTo1 ) nOut++; + if( c==0x0429 || c== 0x0449 ) nOut += 2; + break; + }else if( translit[x].cFrom>c ){ + xTop = x-1; + }else{ + xBtm = x+1; + } + } + } + } + + return nChar; +} + + +/* +** spellfix1_translit(X) +** +** Convert a string that contains non-ASCII Roman characters into +** pure ASCII. +*/ +static void transliterateSqlFunc( + sqlite3_context *context, + int argc, + sqlite3_value **argv +){ + const unsigned char *zIn = sqlite3_value_text(argv[0]); + int nIn = sqlite3_value_bytes(argv[0]); + unsigned char *zOut = transliterate(zIn, nIn); + if( zOut==0 ){ + sqlite3_result_error_nomem(context); + }else{ + sqlite3_result_text(context, (char*)zOut, -1, sqlite3_free); + } +} + +/* +** spellfix1_scriptcode(X) +** +** Try to determine the dominant script used by the word X and return +** its ISO 15924 numeric code. +** +** The current implementation only understands the following scripts: +** +** 215 (Latin) +** 220 (Cyrillic) +** 200 (Greek) +** +** This routine will return 998 if the input X contains characters from +** two or more of the above scripts or 999 if X contains no characters +** from any of the above scripts. +*/ +static void scriptCodeSqlFunc( + sqlite3_context *context, + int argc, + sqlite3_value **argv +){ + const unsigned char *zIn = sqlite3_value_text(argv[0]); + int nIn = sqlite3_value_bytes(argv[0]); + int c, sz; + int scriptMask = 0; + int res; +# define SCRIPT_LATIN 0x0001 +# define SCRIPT_CYRILLIC 0x0002 +# define SCRIPT_GREEK 0x0004 + + while( nIn>0 ){ + c = utf8Read(zIn, nIn, &sz); + zIn += sz; + nIn -= sz; + if( c<0x02af ){ + scriptMask |= SCRIPT_LATIN; + }else if( c>=0x0400 && c<=0x04ff ){ + scriptMask |= SCRIPT_CYRILLIC; + }else if( c>=0x0386 && c<=0x03ce ){ + scriptMask |= SCRIPT_GREEK; + } + } + switch( scriptMask ){ + case 0: res = 999; break; + case SCRIPT_LATIN: res = 215; break; + case SCRIPT_CYRILLIC: res = 220; break; + case SCRIPT_GREEK: res = 200; break; + default: res = 998; break; + } + sqlite3_result_int(context, res); +} + +/* End transliterate +****************************************************************************** +****************************************************************************** +** Begin spellfix1 virtual table. +*/ + +/* Maximum length of a phonehash used for querying the shadow table */ +#define SPELLFIX_MX_HASH 8 + +/* Maximum number of hash strings to examine per query */ +#define SPELLFIX_MX_RUN 1 + +typedef struct spellfix1_vtab spellfix1_vtab; +typedef struct spellfix1_cursor spellfix1_cursor; + +/* Fuzzy-search virtual table object */ +struct spellfix1_vtab { + sqlite3_vtab base; /* Base class - must be first */ + sqlite3 *db; /* Database connection */ + char *zDbName; /* Name of database holding this table */ + char *zTableName; /* Name of the virtual table */ + char *zCostTable; /* Table holding edit-distance cost numbers */ + EditDist3Config *pConfig3; /* Parsed edit distance costs */ +}; + +/* Fuzzy-search cursor object */ +struct spellfix1_cursor { + sqlite3_vtab_cursor base; /* Base class - must be first */ + spellfix1_vtab *pVTab; /* The table to which this cursor belongs */ + char *zPattern; /* rhs of MATCH clause */ + int nRow; /* Number of rows of content */ + int nAlloc; /* Number of allocated rows */ + int iRow; /* Current row of content */ + int iLang; /* Value of the langid= constraint */ + int iTop; /* Value of the top= constraint */ + int iScope; /* Value of the scope= constraint */ + int nSearch; /* Number of vocabulary items checked */ + sqlite3_stmt *pFullScan; /* Shadow query for a full table scan */ + struct spellfix1_row { /* For each row of content */ + sqlite3_int64 iRowid; /* Rowid for this row */ + char *zWord; /* Text for this row */ + int iRank; /* Rank for this row */ + int iDistance; /* Distance from pattern for this row */ + int iScore; /* Score for sorting */ + int iMatchlen; /* Value of matchlen column (or -1) */ + char zHash[SPELLFIX_MX_HASH]; /* the phonehash used for this match */ + } *a; +}; + +/* +** Construct one or more SQL statements from the format string given +** and then evaluate those statements. The success code is written +** into *pRc. +** +** If *pRc is initially non-zero then this routine is a no-op. +*/ +static void spellfix1DbExec( + int *pRc, /* Success code */ + sqlite3 *db, /* Database in which to run SQL */ + const char *zFormat, /* Format string for SQL */ + ... /* Arguments to the format string */ +){ + va_list ap; + char *zSql; + if( *pRc ) return; + va_start(ap, zFormat); + zSql = sqlite3_vmprintf(zFormat, ap); + va_end(ap); + if( zSql==0 ){ + *pRc = SQLITE_NOMEM; + }else{ + *pRc = sqlite3_exec(db, zSql, 0, 0, 0); + sqlite3_free(zSql); + } +} + +/* +** xDisconnect/xDestroy method for the fuzzy-search module. +*/ +static int spellfix1Uninit(int isDestroy, sqlite3_vtab *pVTab){ + spellfix1_vtab *p = (spellfix1_vtab*)pVTab; + int rc = SQLITE_OK; + if( isDestroy ){ + sqlite3 *db = p->db; + spellfix1DbExec(&rc, db, "DROP TABLE IF EXISTS \"%w\".\"%w_vocab\"", + p->zDbName, p->zTableName); + } + if( rc==SQLITE_OK ){ + sqlite3_free(p->zTableName); + editDist3ConfigDelete(p->pConfig3); + sqlite3_free(p->zCostTable); + sqlite3_free(p); + } + return rc; +} +static int spellfix1Disconnect(sqlite3_vtab *pVTab){ + return spellfix1Uninit(0, pVTab); +} +static int spellfix1Destroy(sqlite3_vtab *pVTab){ + return spellfix1Uninit(1, pVTab); +} + +/* +** Make a copy of a string. Remove leading and trailing whitespace +** and dequote it. +*/ +static char *spellfix1Dequote(const char *zIn){ + char *zOut; + int i, j; + char c; + while( isspace(zIn[0]) ) zIn++; + zOut = sqlite3_mprintf("%s", zIn); + if( zOut==0 ) return 0; + i = (int)strlen(zOut); +#if 0 /* The parser will never leave spaces at the end */ + while( i>0 && isspace(zOut[i-1]) ){ i--; } +#endif + zOut[i] = 0; + c = zOut[0]; + if( c=='\'' || c=='"' ){ + for(i=1, j=0; ALWAYS(zOut[i]); i++){ + zOut[j++] = zOut[i]; + if( zOut[i]==c ){ + if( zOut[i+1]==c ){ + i++; + }else{ + zOut[j-1] = 0; + break; + } + } + } + } + return zOut; +} + + +/* +** xConnect/xCreate method for the spellfix1 module. Arguments are: +** +** argv[0] -> module name ("spellfix1") +** argv[1] -> database name +** argv[2] -> table name +** argv[3].. -> optional arguments (i.e. "edit_cost_table" parameter) +*/ +static int spellfix1Init( + int isCreate, + sqlite3 *db, + void *pAux, + int argc, const char *const*argv, + sqlite3_vtab **ppVTab, + char **pzErr +){ + spellfix1_vtab *pNew = 0; + const char *zModule = argv[0]; + const char *zDbName = argv[1]; + const char *zTableName = argv[2]; + int nDbName; + int rc = SQLITE_OK; + int i; + + nDbName = (int)strlen(zDbName); + pNew = sqlite3_malloc( sizeof(*pNew) + nDbName + 1); + if( pNew==0 ){ + rc = SQLITE_NOMEM; + }else{ + memset(pNew, 0, sizeof(*pNew)); + pNew->zDbName = (char*)&pNew[1]; + memcpy(pNew->zDbName, zDbName, nDbName+1); + pNew->zTableName = sqlite3_mprintf("%s", zTableName); + pNew->db = db; + if( pNew->zTableName==0 ){ + rc = SQLITE_NOMEM; + }else{ + rc = sqlite3_declare_vtab(db, + "CREATE TABLE x(word,rank,distance,langid, " + "score, matchlen, phonehash HIDDEN, " + "top HIDDEN, scope HIDDEN, srchcnt HIDDEN, " + "soundslike HIDDEN, command HIDDEN)" + ); +#define SPELLFIX_COL_WORD 0 +#define SPELLFIX_COL_RANK 1 +#define SPELLFIX_COL_DISTANCE 2 +#define SPELLFIX_COL_LANGID 3 +#define SPELLFIX_COL_SCORE 4 +#define SPELLFIX_COL_MATCHLEN 5 +#define SPELLFIX_COL_PHONEHASH 6 +#define SPELLFIX_COL_TOP 7 +#define SPELLFIX_COL_SCOPE 8 +#define SPELLFIX_COL_SRCHCNT 9 +#define SPELLFIX_COL_SOUNDSLIKE 10 +#define SPELLFIX_COL_COMMAND 11 + } + if( rc==SQLITE_OK && isCreate ){ + sqlite3_uint64 r; + spellfix1DbExec(&rc, db, + "CREATE TABLE IF NOT EXISTS \"%w\".\"%w_vocab\"(\n" + " id INTEGER PRIMARY KEY,\n" + " rank INT,\n" + " langid INT,\n" + " word TEXT,\n" + " k1 TEXT,\n" + " k2 TEXT\n" + ");\n", + zDbName, zTableName + ); + sqlite3_randomness(sizeof(r), &r); + spellfix1DbExec(&rc, db, + "CREATE INDEX IF NOT EXISTS \"%w\".\"%w_index_%llx\" " + "ON \"%w_vocab\"(langid,k2);", + zDbName, zModule, r, zTableName + ); + } + for(i=3; rc==SQLITE_OK && ibase); + }else{ + *ppVTab = (sqlite3_vtab *)pNew; + } + return rc; +} + +/* +** The xConnect and xCreate methods +*/ +static int spellfix1Connect( + sqlite3 *db, + void *pAux, + int argc, const char *const*argv, + sqlite3_vtab **ppVTab, + char **pzErr +){ + return spellfix1Init(0, db, pAux, argc, argv, ppVTab, pzErr); +} +static int spellfix1Create( + sqlite3 *db, + void *pAux, + int argc, const char *const*argv, + sqlite3_vtab **ppVTab, + char **pzErr +){ + return spellfix1Init(1, db, pAux, argc, argv, ppVTab, pzErr); +} + +/* +** Clear all of the content from a cursor. +*/ +static void spellfix1ResetCursor(spellfix1_cursor *pCur){ + int i; + for(i=0; inRow; i++){ + sqlite3_free(pCur->a[i].zWord); + } + pCur->nRow = 0; + pCur->iRow = 0; + pCur->nSearch = 0; + if( pCur->pFullScan ){ + sqlite3_finalize(pCur->pFullScan); + pCur->pFullScan = 0; + } +} + +/* +** Resize the cursor to hold up to N rows of content +*/ +static void spellfix1ResizeCursor(spellfix1_cursor *pCur, int N){ + struct spellfix1_row *aNew; + assert( N>=pCur->nRow ); + aNew = sqlite3_realloc(pCur->a, sizeof(pCur->a[0])*N); + if( aNew==0 && N>0 ){ + spellfix1ResetCursor(pCur); + sqlite3_free(pCur->a); + pCur->nAlloc = 0; + pCur->a = 0; + }else{ + pCur->nAlloc = N; + pCur->a = aNew; + } +} + + +/* +** Close a fuzzy-search cursor. +*/ +static int spellfix1Close(sqlite3_vtab_cursor *cur){ + spellfix1_cursor *pCur = (spellfix1_cursor *)cur; + spellfix1ResetCursor(pCur); + spellfix1ResizeCursor(pCur, 0); + sqlite3_free(pCur->zPattern); + sqlite3_free(pCur); + return SQLITE_OK; +} + +/* +** Search for terms of these forms: +** +** (A) word MATCH $str +** (B) langid == $langid +** (C) top = $top +** (D) scope = $scope +** (E) distance < $distance +** (F) distance <= $distance +** +** The plan number is a bit mask formed with these bits: +** +** 0x01 (A) is found +** 0x02 (B) is found +** 0x04 (C) is found +** 0x08 (D) is found +** 0x10 (E) is found +** 0x20 (F) is found +** +** filter.argv[*] values contains $str, $langid, $top, and $scope, +** if specified and in that order. +*/ +static int spellfix1BestIndex(sqlite3_vtab *tab, sqlite3_index_info *pIdxInfo){ + int iPlan = 0; + int iLangTerm = -1; + int iTopTerm = -1; + int iScopeTerm = -1; + int iDistTerm = -1; + int i; + const struct sqlite3_index_constraint *pConstraint; + pConstraint = pIdxInfo->aConstraint; + for(i=0; inConstraint; i++, pConstraint++){ + if( pConstraint->usable==0 ) continue; + + /* Terms of the form: word MATCH $str */ + if( (iPlan & 1)==0 + && pConstraint->iColumn==SPELLFIX_COL_WORD + && pConstraint->op==SQLITE_INDEX_CONSTRAINT_MATCH + ){ + iPlan |= 1; + pIdxInfo->aConstraintUsage[i].argvIndex = 1; + pIdxInfo->aConstraintUsage[i].omit = 1; + } + + /* Terms of the form: langid = $langid */ + if( (iPlan & 2)==0 + && pConstraint->iColumn==SPELLFIX_COL_LANGID + && pConstraint->op==SQLITE_INDEX_CONSTRAINT_EQ + ){ + iPlan |= 2; + iLangTerm = i; + } + + /* Terms of the form: top = $top */ + if( (iPlan & 4)==0 + && pConstraint->iColumn==SPELLFIX_COL_TOP + && pConstraint->op==SQLITE_INDEX_CONSTRAINT_EQ + ){ + iPlan |= 4; + iTopTerm = i; + } + + /* Terms of the form: scope = $scope */ + if( (iPlan & 8)==0 + && pConstraint->iColumn==SPELLFIX_COL_SCOPE + && pConstraint->op==SQLITE_INDEX_CONSTRAINT_EQ + ){ + iPlan |= 8; + iScopeTerm = i; + } + + /* Terms of the form: distance < $dist or distance <= $dist */ + if( (iPlan & (16|32))==0 + && pConstraint->iColumn==SPELLFIX_COL_DISTANCE + && (pConstraint->op==SQLITE_INDEX_CONSTRAINT_LT + || pConstraint->op==SQLITE_INDEX_CONSTRAINT_LE) + ){ + iPlan |= pConstraint->op==SQLITE_INDEX_CONSTRAINT_LT ? 16 : 32; + iDistTerm = i; + } + } + if( iPlan&1 ){ + int idx = 2; + pIdxInfo->idxNum = iPlan; + if( pIdxInfo->nOrderBy==1 + && pIdxInfo->aOrderBy[0].iColumn==SPELLFIX_COL_SCORE + && pIdxInfo->aOrderBy[0].desc==0 + ){ + pIdxInfo->orderByConsumed = 1; /* Default order by iScore */ + } + if( iPlan&2 ){ + pIdxInfo->aConstraintUsage[iLangTerm].argvIndex = idx++; + pIdxInfo->aConstraintUsage[iLangTerm].omit = 1; + } + if( iPlan&4 ){ + pIdxInfo->aConstraintUsage[iTopTerm].argvIndex = idx++; + pIdxInfo->aConstraintUsage[iTopTerm].omit = 1; + } + if( iPlan&8 ){ + pIdxInfo->aConstraintUsage[iScopeTerm].argvIndex = idx++; + pIdxInfo->aConstraintUsage[iScopeTerm].omit = 1; + } + if( iPlan&(16|32) ){ + pIdxInfo->aConstraintUsage[iDistTerm].argvIndex = idx++; + pIdxInfo->aConstraintUsage[iDistTerm].omit = 1; + } + pIdxInfo->estimatedCost = (double)10000; + }else{ + pIdxInfo->idxNum = 0; + pIdxInfo->estimatedCost = (double)10000000; + } + return SQLITE_OK; +} + +/* +** Open a new fuzzy-search cursor. +*/ +static int spellfix1Open(sqlite3_vtab *pVTab, sqlite3_vtab_cursor **ppCursor){ + spellfix1_vtab *p = (spellfix1_vtab*)pVTab; + spellfix1_cursor *pCur; + pCur = sqlite3_malloc( sizeof(*pCur) ); + if( pCur==0 ) return SQLITE_NOMEM; + memset(pCur, 0, sizeof(*pCur)); + pCur->pVTab = p; + *ppCursor = &pCur->base; + return SQLITE_OK; +} + +/* +** Adjust a distance measurement by the words rank in order to show +** preference to common words. +*/ +static int spellfix1Score(int iDistance, int iRank){ + int iLog2; + for(iLog2=0; iRank>0; iLog2++, iRank>>=1){} + return iDistance + 32 - iLog2; +} + +/* +** Compare two spellfix1_row objects for sorting purposes in qsort() such +** that they sort in order of increasing distance. +*/ +static int spellfix1RowCompare(const void *A, const void *B){ + const struct spellfix1_row *a = (const struct spellfix1_row*)A; + const struct spellfix1_row *b = (const struct spellfix1_row*)B; + return a->iScore - b->iScore; +} + +/* +** A structure used to pass information from spellfix1FilterForMatch() +** into spellfix1RunQuery(). +*/ +typedef struct MatchQuery { + spellfix1_cursor *pCur; /* The cursor being queried */ + sqlite3_stmt *pStmt; /* shadow table query statment */ + char zHash[SPELLFIX_MX_HASH]; /* The current phonehash for zPattern */ + const char *zPattern; /* Transliterated input string */ + int nPattern; /* Length of zPattern */ + EditDist3FromString *pMatchStr3; /* Original unicode string */ + EditDist3Config *pConfig3; /* Edit-distance cost coefficients */ + const EditDist3Lang *pLang; /* The selected language coefficients */ + int iLang; /* The language id */ + int iScope; /* Default scope */ + int iMaxDist; /* Maximum allowed edit distance, or -1 */ + int rc; /* Error code */ + int nRun; /* Number of prior runs for the same zPattern */ + char azPrior[SPELLFIX_MX_RUN][SPELLFIX_MX_HASH]; /* Prior hashes */ +} MatchQuery; + +/* +** Run a query looking for the best matches against zPattern using +** zHash as the character class seed hash. +*/ +static void spellfix1RunQuery(MatchQuery *p, const char *zQuery, int nQuery){ + const char *zK1; + const char *zWord; + int iDist; + int iRank; + int iScore; + int iWorst = 0; + int idx; + int idxWorst = -1; + int i; + int iScope = p->iScope; + spellfix1_cursor *pCur = p->pCur; + sqlite3_stmt *pStmt = p->pStmt; + char zHash1[SPELLFIX_MX_HASH]; + char zHash2[SPELLFIX_MX_HASH]; + char *zClass; + int nClass; + int rc; + + if( pCur->a==0 || p->rc ) return; /* Prior memory allocation failure */ + zClass = (char*)phoneticHash((unsigned char*)zQuery, nQuery); + if( zClass==0 ){ + p->rc = SQLITE_NOMEM; + return; + } + nClass = (int)strlen(zClass); + if( nClass>SPELLFIX_MX_HASH-2 ){ + nClass = SPELLFIX_MX_HASH-2; + zClass[nClass] = 0; + } + if( nClass<=iScope ){ + if( nClass>2 ){ + iScope = nClass-1; + }else{ + iScope = nClass; + } + } + memcpy(zHash1, zClass, iScope); + sqlite3_free(zClass); + zHash1[iScope] = 0; + memcpy(zHash2, zHash1, iScope); + zHash2[iScope] = 'Z'; + zHash2[iScope+1] = 0; +#if SPELLFIX_MX_RUN>1 + for(i=0; inRun; i++){ + if( strcmp(p->azPrior[i], zHash1)==0 ) return; + } +#endif + assert( p->nRunazPrior[p->nRun++], zHash1, iScope+1); + if( sqlite3_bind_text(pStmt, 1, zHash1, -1, SQLITE_STATIC)==SQLITE_NOMEM + || sqlite3_bind_text(pStmt, 2, zHash2, -1, SQLITE_STATIC)==SQLITE_NOMEM + ){ + p->rc = SQLITE_NOMEM; + return; + } +#if SPELLFIX_MX_RUN>1 + for(i=0; inRow; i++){ + if( pCur->a[i].iScore>iWorst ){ + iWorst = pCur->a[i].iScore; + idxWorst = i; + } + } +#endif + while( sqlite3_step(pStmt)==SQLITE_ROW ){ + int iMatchlen = -1; + iRank = sqlite3_column_int(pStmt, 2); + if( p->pMatchStr3 ){ + int nWord = sqlite3_column_bytes(pStmt, 1); + zWord = (const char*)sqlite3_column_text(pStmt, 1); + iDist = editDist3Core(p->pMatchStr3, zWord, nWord, p->pLang, &iMatchlen); + }else{ + zK1 = (const char*)sqlite3_column_text(pStmt, 3); + if( zK1==0 ) continue; + iDist = editdist1(p->zPattern, zK1, 0); + } + if( iDist<0 ){ + p->rc = SQLITE_NOMEM; + break; + } + pCur->nSearch++; + iScore = spellfix1Score(iDist,iRank); + if( p->iMaxDist>=0 ){ + if( iDist>p->iMaxDist ) continue; + if( pCur->nRow>=pCur->nAlloc-1 ){ + spellfix1ResizeCursor(pCur, pCur->nAlloc*2 + 10); + if( pCur->a==0 ) break; + } + idx = pCur->nRow; + }else if( pCur->nRownAlloc ){ + idx = pCur->nRow; + }else if( iScorea[idx].zWord); + }else{ + continue; + } + pCur->a[idx].zWord = sqlite3_mprintf("%s", sqlite3_column_text(pStmt, 1)); + if( pCur->a[idx].zWord==0 ){ + p->rc = SQLITE_NOMEM; + break; + } + pCur->a[idx].iRowid = sqlite3_column_int64(pStmt, 0); + pCur->a[idx].iRank = iRank; + pCur->a[idx].iDistance = iDist; + pCur->a[idx].iScore = iScore; + pCur->a[idx].iMatchlen = iMatchlen; + memcpy(pCur->a[idx].zHash, zHash1, iScope+1); + if( pCur->nRownAlloc ) pCur->nRow++; + if( pCur->nRow==pCur->nAlloc ){ + iWorst = pCur->a[0].iScore; + idxWorst = 0; + for(i=1; inRow; i++){ + iScore = pCur->a[i].iScore; + if( iWorstrc = rc; +} + +/* +** This version of the xFilter method work if the MATCH term is present +** and we are doing a scan. +*/ +static int spellfix1FilterForMatch( + spellfix1_cursor *pCur, + int idxNum, + int argc, + sqlite3_value **argv +){ + const unsigned char *zMatchThis; /* RHS of the MATCH operator */ + EditDist3FromString *pMatchStr3 = 0; /* zMatchThis as an editdist string */ + char *zPattern; /* Transliteration of zMatchThis */ + int nPattern; /* Length of zPattern */ + int iLimit = 20; /* Max number of rows of output */ + int iScope = 3; /* Use this many characters of zClass */ + int iLang = 0; /* Language code */ + char *zSql; /* SQL of shadow table query */ + sqlite3_stmt *pStmt = 0; /* Shadow table query */ + int rc; /* Result code */ + int idx = 1; /* Next available filter parameter */ + spellfix1_vtab *p = pCur->pVTab; /* The virtual table that owns pCur */ + MatchQuery x; /* For passing info to RunQuery() */ + + /* Load the cost table if we have not already done so */ + if( p->zCostTable!=0 && p->pConfig3==0 ){ + p->pConfig3 = sqlite3_malloc( sizeof(p->pConfig3[0]) ); + if( p->pConfig3==0 ) return SQLITE_NOMEM; + memset(p->pConfig3, 0, sizeof(p->pConfig3[0])); + rc = editDist3ConfigLoad(p->pConfig3, p->db, p->zCostTable); + if( rc ) return rc; + } + memset(&x, 0, sizeof(x)); + x.iScope = 3; /* Default scope if none specified by "WHERE scope=N" */ + x.iMaxDist = -1; /* Maximum allowed edit distance */ + + if( idxNum&2 ){ + iLang = sqlite3_value_int(argv[idx++]); + } + if( idxNum&4 ){ + iLimit = sqlite3_value_int(argv[idx++]); + if( iLimit<1 ) iLimit = 1; + } + if( idxNum&8 ){ + x.iScope = sqlite3_value_int(argv[idx++]); + if( x.iScope<1 ) x.iScope = 1; + if( x.iScope>SPELLFIX_MX_HASH-2 ) x.iScope = SPELLFIX_MX_HASH-2; + } + if( idxNum&(16|32) ){ + x.iMaxDist = sqlite3_value_int(argv[idx++]); + if( idxNum&16 ) x.iMaxDist--; + if( x.iMaxDist<0 ) x.iMaxDist = 0; + } + spellfix1ResetCursor(pCur); + spellfix1ResizeCursor(pCur, iLimit); + zMatchThis = sqlite3_value_text(argv[0]); + if( zMatchThis==0 ) return SQLITE_OK; + if( p->pConfig3 ){ + x.pLang = editDist3FindLang(p->pConfig3, iLang); + pMatchStr3 = editDist3FromStringNew(x.pLang, (const char*)zMatchThis, -1); + if( pMatchStr3==0 ){ + x.rc = SQLITE_NOMEM; + goto filter_exit; + } + }else{ + x.pLang = 0; + } + zPattern = (char*)transliterate(zMatchThis, sqlite3_value_bytes(argv[0])); + sqlite3_free(pCur->zPattern); + pCur->zPattern = zPattern; + if( zPattern==0 ){ + x.rc = SQLITE_NOMEM; + goto filter_exit; + } + nPattern = (int)strlen(zPattern); + if( zPattern[nPattern-1]=='*' ) nPattern--; + zSql = sqlite3_mprintf( + "SELECT id, word, rank, k1" + " FROM \"%w\".\"%w_vocab\"" + " WHERE langid=%d AND k2>=?1 AND k2zDbName, p->zTableName, iLang + ); + if( zSql==0 ){ + x.rc = SQLITE_NOMEM; + pStmt = 0; + goto filter_exit; + } + rc = sqlite3_prepare_v2(p->db, zSql, -1, &pStmt, 0); + sqlite3_free(zSql); + pCur->iLang = iLang; + x.pCur = pCur; + x.pStmt = pStmt; + x.zPattern = zPattern; + x.nPattern = nPattern; + x.pMatchStr3 = pMatchStr3; + x.iLang = iLang; + x.rc = rc; + x.pConfig3 = p->pConfig3; + if( x.rc==SQLITE_OK ){ + spellfix1RunQuery(&x, zPattern, nPattern); + } + + if( pCur->a ){ + qsort(pCur->a, pCur->nRow, sizeof(pCur->a[0]), spellfix1RowCompare); + pCur->iTop = iLimit; + pCur->iScope = iScope; + }else{ + x.rc = SQLITE_NOMEM; + } + +filter_exit: + sqlite3_finalize(pStmt); + editDist3FromStringDelete(pMatchStr3); + return x.rc; +} + +/* +** This version of xFilter handles a full-table scan case +*/ +static int spellfix1FilterForFullScan( + spellfix1_cursor *pCur, + int idxNum, + int argc, + sqlite3_value **argv +){ + int rc; + char *zSql; + spellfix1_vtab *pVTab = pCur->pVTab; + spellfix1ResetCursor(pCur); + zSql = sqlite3_mprintf( + "SELECT word, rank, NULL, langid, id FROM \"%w\".\"%w_vocab\"", + pVTab->zDbName, pVTab->zTableName); + if( zSql==0 ) return SQLITE_NOMEM; + rc = sqlite3_prepare_v2(pVTab->db, zSql, -1, &pCur->pFullScan, 0); + sqlite3_free(zSql); + pCur->nRow = pCur->iRow = 0; + if( rc==SQLITE_OK ){ + rc = sqlite3_step(pCur->pFullScan); + if( rc==SQLITE_ROW ){ pCur->iRow = -1; rc = SQLITE_OK; } + if( rc==SQLITE_DONE ){ rc = SQLITE_OK; } + }else{ + pCur->iRow = 0; + } + return rc; +} + + +/* +** 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 spellfix1Column, spellfix1Rowid, or spellfix1Eof call. +*/ +static int spellfix1Filter( + sqlite3_vtab_cursor *cur, + int idxNum, const char *idxStr, + int argc, sqlite3_value **argv +){ + spellfix1_cursor *pCur = (spellfix1_cursor *)cur; + int rc; + if( idxNum & 1 ){ + rc = spellfix1FilterForMatch(pCur, idxNum, argc, argv); + }else{ + rc = spellfix1FilterForFullScan(pCur, idxNum, argc, argv); + } + return rc; +} + + +/* +** Advance a cursor to its next row of output +*/ +static int spellfix1Next(sqlite3_vtab_cursor *cur){ + spellfix1_cursor *pCur = (spellfix1_cursor *)cur; + int rc = SQLITE_OK; + if( pCur->iRow < pCur->nRow ){ + if( pCur->pFullScan ){ + rc = sqlite3_step(pCur->pFullScan); + if( rc!=SQLITE_ROW ) pCur->iRow = pCur->nRow; + if( rc==SQLITE_ROW || rc==SQLITE_DONE ) rc = SQLITE_OK; + }else{ + pCur->iRow++; + } + } + return rc; +} + +/* +** Return TRUE if we are at the end-of-file +*/ +static int spellfix1Eof(sqlite3_vtab_cursor *cur){ + spellfix1_cursor *pCur = (spellfix1_cursor *)cur; + return pCur->iRow>=pCur->nRow; +} + +/* +** Return columns from the current row. +*/ +static int spellfix1Column( + sqlite3_vtab_cursor *cur, + sqlite3_context *ctx, + int i +){ + spellfix1_cursor *pCur = (spellfix1_cursor*)cur; + if( pCur->pFullScan ){ + if( i<=SPELLFIX_COL_LANGID ){ + sqlite3_result_value(ctx, sqlite3_column_value(pCur->pFullScan, i)); + }else{ + sqlite3_result_null(ctx); + } + return SQLITE_OK; + } + switch( i ){ + case SPELLFIX_COL_WORD: { + sqlite3_result_text(ctx, pCur->a[pCur->iRow].zWord, -1, SQLITE_STATIC); + break; + } + case SPELLFIX_COL_RANK: { + sqlite3_result_int(ctx, pCur->a[pCur->iRow].iRank); + break; + } + case SPELLFIX_COL_DISTANCE: { + sqlite3_result_int(ctx, pCur->a[pCur->iRow].iDistance); + break; + } + case SPELLFIX_COL_LANGID: { + sqlite3_result_int(ctx, pCur->iLang); + break; + } + case SPELLFIX_COL_SCORE: { + sqlite3_result_int(ctx, pCur->a[pCur->iRow].iScore); + break; + } + case SPELLFIX_COL_MATCHLEN: { + int iMatchlen = pCur->a[pCur->iRow].iMatchlen; + if( iMatchlen<0 ){ + int nPattern = (int)strlen(pCur->zPattern); + char *zWord = pCur->a[pCur->iRow].zWord; + int nWord = (int)strlen(zWord); + + if( nPattern>0 && pCur->zPattern[nPattern-1]=='*' ){ + char *zTranslit; + int res; + zTranslit = (char *)transliterate((unsigned char *)zWord, nWord); + if( !zTranslit ) return SQLITE_NOMEM; + res = editdist1(pCur->zPattern, zTranslit, &iMatchlen); + sqlite3_free(zTranslit); + if( res<0 ) return SQLITE_NOMEM; + iMatchlen = translen_to_charlen(zWord, nWord, iMatchlen); + }else{ + iMatchlen = utf8Charlen(zWord, nWord); + } + } + + sqlite3_result_int(ctx, iMatchlen); + break; + } + case SPELLFIX_COL_PHONEHASH: { + sqlite3_result_text(ctx, pCur->a[pCur->iRow].zHash, -1, SQLITE_STATIC); + break; + } + case SPELLFIX_COL_TOP: { + sqlite3_result_int(ctx, pCur->iTop); + break; + } + case SPELLFIX_COL_SCOPE: { + sqlite3_result_int(ctx, pCur->iScope); + break; + } + case SPELLFIX_COL_SRCHCNT: { + sqlite3_result_int(ctx, pCur->nSearch); + break; + } + default: { + sqlite3_result_null(ctx); + break; + } + } + return SQLITE_OK; +} + +/* +** The rowid. +*/ +static int spellfix1Rowid(sqlite3_vtab_cursor *cur, sqlite_int64 *pRowid){ + spellfix1_cursor *pCur = (spellfix1_cursor*)cur; + if( pCur->pFullScan ){ + *pRowid = sqlite3_column_int64(pCur->pFullScan, 4); + }else{ + *pRowid = pCur->a[pCur->iRow].iRowid; + } + return SQLITE_OK; +} + +/* +** The xUpdate() method. +*/ +static int spellfix1Update( + sqlite3_vtab *pVTab, + int argc, + sqlite3_value **argv, + sqlite_int64 *pRowid +){ + int rc = SQLITE_OK; + sqlite3_int64 rowid, newRowid; + spellfix1_vtab *p = (spellfix1_vtab*)pVTab; + sqlite3 *db = p->db; + + if( argc==1 ){ + /* A delete operation on the rowid given by argv[0] */ + rowid = *pRowid = sqlite3_value_int64(argv[0]); + spellfix1DbExec(&rc, db, "DELETE FROM \"%w\".\"%w_vocab\" " + " WHERE id=%lld", + p->zDbName, p->zTableName, rowid); + }else{ + const unsigned char *zWord = sqlite3_value_text(argv[SPELLFIX_COL_WORD+2]); + int nWord = sqlite3_value_bytes(argv[SPELLFIX_COL_WORD+2]); + int iLang = sqlite3_value_int(argv[SPELLFIX_COL_LANGID+2]); + int iRank = sqlite3_value_int(argv[SPELLFIX_COL_RANK+2]); + const unsigned char *zSoundslike = + sqlite3_value_text(argv[SPELLFIX_COL_SOUNDSLIKE+2]); + int nSoundslike = sqlite3_value_bytes(argv[SPELLFIX_COL_SOUNDSLIKE+2]); + char *zK1, *zK2; + int i; + char c; + + if( zWord==0 ){ + /* Inserts of the form: INSERT INTO table(command) VALUES('xyzzy'); + ** cause zWord to be NULL, so we look at the "command" column to see + ** what special actions to take */ + const char *zCmd = + (const char*)sqlite3_value_text(argv[SPELLFIX_COL_COMMAND+2]); + if( zCmd==0 ){ + pVTab->zErrMsg = sqlite3_mprintf("%s.word may not be NULL", + p->zTableName); + return SQLITE_CONSTRAINT_NOTNULL; + } + if( strcmp(zCmd,"reset")==0 ){ + /* Reset the edit cost table (if there is one). */ + editDist3ConfigDelete(p->pConfig3); + p->pConfig3 = 0; + return SQLITE_OK; + } + if( strncmp(zCmd,"edit_cost_table=",16)==0 ){ + editDist3ConfigDelete(p->pConfig3); + p->pConfig3 = 0; + sqlite3_free(p->zCostTable); + p->zCostTable = spellfix1Dequote(zCmd+16); + if( p->zCostTable==0 ) return SQLITE_NOMEM; + if( p->zCostTable[0]==0 || sqlite3_stricmp(p->zCostTable,"null")==0 ){ + sqlite3_free(p->zCostTable); + p->zCostTable = 0; + } + return SQLITE_OK; + } + pVTab->zErrMsg = sqlite3_mprintf("unknown value for %s.command: \"%w\"", + p->zTableName, zCmd); + return SQLITE_ERROR; + } + if( iRank<1 ) iRank = 1; + if( zSoundslike ){ + zK1 = (char*)transliterate(zSoundslike, nSoundslike); + }else{ + zK1 = (char*)transliterate(zWord, nWord); + } + if( zK1==0 ) return SQLITE_NOMEM; + for(i=0; (c = zK1[i])!=0; i++){ + if( c>='A' && c<='Z' ) zK1[i] += 'a' - 'A'; + } + zK2 = (char*)phoneticHash((const unsigned char*)zK1, i); + if( zK2==0 ){ + sqlite3_free(zK1); + return SQLITE_NOMEM; + } + if( sqlite3_value_type(argv[0])==SQLITE_NULL ){ + spellfix1DbExec(&rc, db, + "INSERT INTO \"%w\".\"%w_vocab\"(rank,langid,word,k1,k2) " + "VALUES(%d,%d,%Q,%Q,%Q)", + p->zDbName, p->zTableName, + iRank, iLang, zWord, zK1, zK2 + ); + *pRowid = sqlite3_last_insert_rowid(db); + }else{ + rowid = sqlite3_value_int64(argv[0]); + newRowid = *pRowid = sqlite3_value_int64(argv[1]); + spellfix1DbExec(&rc, db, + "UPDATE \"%w\".\"%w_vocab\" SET id=%lld, rank=%d, langid=%d," + " word=%Q, k1=%Q, k2=%Q WHERE id=%lld", + p->zDbName, p->zTableName, newRowid, iRank, iLang, + zWord, zK1, zK2, rowid + ); + } + sqlite3_free(zK1); + sqlite3_free(zK2); + } + return rc; +} + +/* +** Rename the spellfix1 table. +*/ +static int spellfix1Rename(sqlite3_vtab *pVTab, const char *zNew){ + spellfix1_vtab *p = (spellfix1_vtab*)pVTab; + sqlite3 *db = p->db; + int rc = SQLITE_OK; + char *zNewName = sqlite3_mprintf("%s", zNew); + if( zNewName==0 ){ + return SQLITE_NOMEM; + } + spellfix1DbExec(&rc, db, + "ALTER TABLE \"%w\".\"%w_vocab\" RENAME TO \"%w_vocab\"", + p->zDbName, p->zTableName, zNewName + ); + if( rc==SQLITE_OK ){ + sqlite3_free(p->zTableName); + p->zTableName = zNewName; + }else{ + sqlite3_free(zNewName); + } + return rc; +} + + +/* +** A virtual table module that provides fuzzy search. +*/ +static sqlite3_module spellfix1Module = { + 0, /* iVersion */ + spellfix1Create, /* xCreate - handle CREATE VIRTUAL TABLE */ + spellfix1Connect, /* xConnect - reconnected to an existing table */ + spellfix1BestIndex, /* xBestIndex - figure out how to do a query */ + spellfix1Disconnect, /* xDisconnect - close a connection */ + spellfix1Destroy, /* xDestroy - handle DROP TABLE */ + spellfix1Open, /* xOpen - open a cursor */ + spellfix1Close, /* xClose - close a cursor */ + spellfix1Filter, /* xFilter - configure scan constraints */ + spellfix1Next, /* xNext - advance a cursor */ + spellfix1Eof, /* xEof - check for end of scan */ + spellfix1Column, /* xColumn - read data */ + spellfix1Rowid, /* xRowid - read data */ + spellfix1Update, /* xUpdate */ + 0, /* xBegin */ + 0, /* xSync */ + 0, /* xCommit */ + 0, /* xRollback */ + 0, /* xFindMethod */ + spellfix1Rename, /* xRename */ +}; + +/* +** Register the various functions and the virtual table. +*/ +static int spellfix1Register(sqlite3 *db){ + int rc = SQLITE_OK; + int i; + rc = sqlite3_create_function(db, "spellfix1_translit", 1, SQLITE_UTF8, 0, + transliterateSqlFunc, 0, 0); + if( rc==SQLITE_OK ){ + rc = sqlite3_create_function(db, "spellfix1_editdist", 2, SQLITE_UTF8, 0, + editdistSqlFunc, 0, 0); + } + if( rc==SQLITE_OK ){ + rc = sqlite3_create_function(db, "spellfix1_phonehash", 1, SQLITE_UTF8, 0, + phoneticHashSqlFunc, 0, 0); + } + if( rc==SQLITE_OK ){ + rc = sqlite3_create_function(db, "spellfix1_scriptcode", 1, SQLITE_UTF8, 0, + scriptCodeSqlFunc, 0, 0); + } + if( rc==SQLITE_OK ){ + rc = sqlite3_create_module(db, "spellfix1", &spellfix1Module, 0); + } + if( rc==SQLITE_OK ){ + rc = editDist3Install(db); + } + + /* Verify sanity of the translit[] table */ + for(i=0; i +#include + +#ifndef SQLITE_OMIT_VIRTUALTABLE + + +/* A wholenumber cursor object */ +typedef struct wholenumber_cursor wholenumber_cursor; +struct wholenumber_cursor { + sqlite3_vtab_cursor base; /* Base class - must be first */ + sqlite3_int64 iValue; /* Current value */ + sqlite3_int64 mxValue; /* Maximum value */ +}; + +/* Methods for the wholenumber module */ +static int wholenumberConnect( + sqlite3 *db, + void *pAux, + int argc, const char *const*argv, + sqlite3_vtab **ppVtab, + char **pzErr +){ + sqlite3_vtab *pNew; + pNew = *ppVtab = sqlite3_malloc( sizeof(*pNew) ); + if( pNew==0 ) return SQLITE_NOMEM; + sqlite3_declare_vtab(db, "CREATE TABLE x(value)"); + memset(pNew, 0, sizeof(*pNew)); + return SQLITE_OK; +} +/* Note that for this virtual table, the xCreate and xConnect +** methods are identical. */ + +static int wholenumberDisconnect(sqlite3_vtab *pVtab){ + sqlite3_free(pVtab); + return SQLITE_OK; +} +/* The xDisconnect and xDestroy methods are also the same */ + + +/* +** Open a new wholenumber cursor. +*/ +static int wholenumberOpen(sqlite3_vtab *p, sqlite3_vtab_cursor **ppCursor){ + wholenumber_cursor *pCur; + pCur = sqlite3_malloc( sizeof(*pCur) ); + if( pCur==0 ) return SQLITE_NOMEM; + memset(pCur, 0, sizeof(*pCur)); + *ppCursor = &pCur->base; + return SQLITE_OK; +} + +/* +** Close a wholenumber cursor. +*/ +static int wholenumberClose(sqlite3_vtab_cursor *cur){ + sqlite3_free(cur); + return SQLITE_OK; +} + + +/* +** Advance a cursor to its next row of output +*/ +static int wholenumberNext(sqlite3_vtab_cursor *cur){ + wholenumber_cursor *pCur = (wholenumber_cursor*)cur; + pCur->iValue++; + return SQLITE_OK; +} + +/* +** Return the value associated with a wholenumber. +*/ +static int wholenumberColumn( + sqlite3_vtab_cursor *cur, + sqlite3_context *ctx, + int i +){ + wholenumber_cursor *pCur = (wholenumber_cursor*)cur; + sqlite3_result_int64(ctx, pCur->iValue); + return SQLITE_OK; +} + +/* +** The rowid. +*/ +static int wholenumberRowid(sqlite3_vtab_cursor *cur, sqlite_int64 *pRowid){ + wholenumber_cursor *pCur = (wholenumber_cursor*)cur; + *pRowid = pCur->iValue; + return SQLITE_OK; +} + +/* +** When the wholenumber_cursor.rLimit value is 0 or less, that is a signal +** that the cursor has nothing more to output. +*/ +static int wholenumberEof(sqlite3_vtab_cursor *cur){ + wholenumber_cursor *pCur = (wholenumber_cursor*)cur; + return pCur->iValue>pCur->mxValue || pCur->iValue==0; +} + +/* +** 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 wholenumberColumn, wholenumberRowid, or wholenumberEof call. +** +** idxNum Constraints +** ------ --------------------- +** 0 (none) +** 1 value > $argv0 +** 2 value >= $argv0 +** 4 value < $argv0 +** 8 value <= $argv0 +** +** 5 value > $argv0 AND value < $argv1 +** 6 value >= $argv0 AND value < $argv1 +** 9 value > $argv0 AND value <= $argv1 +** 10 value >= $argv0 AND value <= $argv1 +*/ +static int wholenumberFilter( + sqlite3_vtab_cursor *pVtabCursor, + int idxNum, const char *idxStr, + int argc, sqlite3_value **argv +){ + wholenumber_cursor *pCur = (wholenumber_cursor *)pVtabCursor; + sqlite3_int64 v; + int i = 0; + pCur->iValue = 1; + pCur->mxValue = 0xffffffff; /* 4294967295 */ + if( idxNum & 3 ){ + v = sqlite3_value_int64(argv[0]) + (idxNum&1); + if( v>pCur->iValue && v<=pCur->mxValue ) pCur->iValue = v; + i++; + } + if( idxNum & 12 ){ + v = sqlite3_value_int64(argv[i]) - ((idxNum>>2)&1); + if( v>=pCur->iValue && vmxValue ) pCur->mxValue = v; + } + return SQLITE_OK; +} + +/* +** Search for terms of these forms: +** +** (1) value > $value +** (2) value >= $value +** (4) value < $value +** (8) value <= $value +** +** idxNum is an ORed combination of 1 or 2 with 4 or 8. +*/ +static int wholenumberBestIndex( + sqlite3_vtab *tab, + sqlite3_index_info *pIdxInfo +){ + int i; + int idxNum = 0; + int argvIdx = 1; + int ltIdx = -1; + int gtIdx = -1; + const struct sqlite3_index_constraint *pConstraint; + pConstraint = pIdxInfo->aConstraint; + for(i=0; inConstraint; i++, pConstraint++){ + if( pConstraint->usable==0 ) continue; + if( (idxNum & 3)==0 && pConstraint->op==SQLITE_INDEX_CONSTRAINT_GT ){ + idxNum |= 1; + ltIdx = i; + } + if( (idxNum & 3)==0 && pConstraint->op==SQLITE_INDEX_CONSTRAINT_GE ){ + idxNum |= 2; + ltIdx = i; + } + if( (idxNum & 12)==0 && pConstraint->op==SQLITE_INDEX_CONSTRAINT_LT ){ + idxNum |= 4; + gtIdx = i; + } + if( (idxNum & 12)==0 && pConstraint->op==SQLITE_INDEX_CONSTRAINT_LE ){ + idxNum |= 8; + gtIdx = i; + } + } + pIdxInfo->idxNum = idxNum; + if( ltIdx>=0 ){ + pIdxInfo->aConstraintUsage[ltIdx].argvIndex = argvIdx++; + pIdxInfo->aConstraintUsage[ltIdx].omit = 1; + } + if( gtIdx>=0 ){ + pIdxInfo->aConstraintUsage[gtIdx].argvIndex = argvIdx; + pIdxInfo->aConstraintUsage[gtIdx].omit = 1; + } + if( pIdxInfo->nOrderBy==1 + && pIdxInfo->aOrderBy[0].desc==0 + ){ + pIdxInfo->orderByConsumed = 1; + } + pIdxInfo->estimatedCost = (double)1; + return SQLITE_OK; +} + +/* +** A virtual table module that provides read-only access to a +** Tcl global variable namespace. +*/ +static sqlite3_module wholenumberModule = { + 0, /* iVersion */ + wholenumberConnect, + wholenumberConnect, + wholenumberBestIndex, + wholenumberDisconnect, + wholenumberDisconnect, + wholenumberOpen, /* xOpen - open a cursor */ + wholenumberClose, /* xClose - close a cursor */ + wholenumberFilter, /* xFilter - configure scan constraints */ + wholenumberNext, /* xNext - advance a cursor */ + wholenumberEof, /* xEof - check for end of scan */ + wholenumberColumn, /* xColumn - read data */ + wholenumberRowid, /* xRowid - read data */ + 0, /* xUpdate */ + 0, /* xBegin */ + 0, /* xSync */ + 0, /* xCommit */ + 0, /* xRollback */ + 0, /* xFindMethod */ + 0, /* xRename */ +}; + +#endif /* SQLITE_OMIT_VIRTUALTABLE */ + +#ifdef _WIN32 +__declspec(dllexport) +#endif +int sqlite3_wholenumber_init( + sqlite3 *db, + char **pzErrMsg, + const sqlite3_api_routines *pApi +){ + int rc = SQLITE_OK; + SQLITE_EXTENSION_INIT2(pApi); +#ifndef SQLITE_OMIT_VIRTUALTABLE + rc = sqlite3_create_module(db, "wholenumber", &wholenumberModule, 0); +#endif + return rc; +} Index: main.mk ================================================================== --- main.mk +++ main.mk @@ -242,11 +242,10 @@ $(TOP)/src/test_config.c \ $(TOP)/src/test_demovfs.c \ $(TOP)/src/test_devsym.c \ $(TOP)/src/test_fs.c \ $(TOP)/src/test_func.c \ - $(TOP)/src/test_fuzzer.c \ $(TOP)/src/test_hexio.c \ $(TOP)/src/test_init.c \ $(TOP)/src/test_intarray.c \ $(TOP)/src/test_journal.c \ $(TOP)/src/test_malloc.c \ @@ -254,11 +253,10 @@ $(TOP)/src/test_mutex.c \ $(TOP)/src/test_onefile.c \ $(TOP)/src/test_osinst.c \ $(TOP)/src/test_pcache.c \ $(TOP)/src/test_quota.c \ - $(TOP)/src/test_regexp.c \ $(TOP)/src/test_rtree.c \ $(TOP)/src/test_schema.c \ $(TOP)/src/test_server.c \ $(TOP)/src/test_stat.c \ $(TOP)/src/test_sqllog.c \ @@ -265,12 +263,23 @@ $(TOP)/src/test_superlock.c \ $(TOP)/src/test_syscall.c \ $(TOP)/src/test_tclvar.c \ $(TOP)/src/test_thread.c \ $(TOP)/src/test_vfs.c \ - $(TOP)/src/test_wholenumber.c \ $(TOP)/src/test_wsd.c + +# 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 + #TESTSRC += $(TOP)/ext/fts2/fts2_tokenizer.c #TESTSRC += $(TOP)/ext/fts3/fts3_tokenizer.c TESTSRC2 = \ Index: src/shell.c ================================================================== --- src/shell.c +++ src/shell.c @@ -1478,22 +1478,10 @@ p->zDbFilename, sqlite3_errmsg(db)); exit(1); } #ifndef SQLITE_OMIT_LOAD_EXTENSION sqlite3_enable_load_extension(p->db, 1); -#endif -#ifdef SQLITE_ENABLE_REGEXP - { - extern int sqlite3_add_regexp_func(sqlite3*); - sqlite3_add_regexp_func(db); - } -#endif -#ifdef SQLITE_ENABLE_SPELLFIX - { - extern int sqlite3_spellfix1_register(sqlite3*); - sqlite3_spellfix1_register(db); - } #endif } } /* Index: src/tclsqlite.c ================================================================== --- src/tclsqlite.c +++ src/tclsqlite.c @@ -3681,13 +3681,10 @@ extern int Sqlitetestrtree_Init(Tcl_Interp*); extern int Sqlitequota_Init(Tcl_Interp*); extern int Sqlitemultiplex_Init(Tcl_Interp*); extern int SqliteSuperlock_Init(Tcl_Interp*); extern int SqlitetestSyscall_Init(Tcl_Interp*); - extern int Sqlitetestfuzzer_Init(Tcl_Interp*); - extern int Sqlitetestwholenumber_Init(Tcl_Interp*); - extern int Sqlitetestregexp_Init(Tcl_Interp*); #if defined(SQLITE_ENABLE_FTS3) || defined(SQLITE_ENABLE_FTS4) extern int Sqlitetestfts3_Init(Tcl_Interp *interp); #endif @@ -3726,13 +3723,10 @@ Sqlitetestrtree_Init(interp); Sqlitequota_Init(interp); Sqlitemultiplex_Init(interp); SqliteSuperlock_Init(interp); SqlitetestSyscall_Init(interp); - Sqlitetestfuzzer_Init(interp); - Sqlitetestwholenumber_Init(interp); - Sqlitetestregexp_Init(interp); #if defined(SQLITE_ENABLE_FTS3) || defined(SQLITE_ENABLE_FTS4) Sqlitetestfts3_Init(interp); #endif Index: src/test1.c ================================================================== --- src/test1.c +++ src/test1.c @@ -6042,10 +6042,69 @@ return TCL_ERROR; } sqlite3_test_control(SQLITE_TESTCTRL_OPTIMIZATIONS, db, mask); return TCL_OK; } + +typedef struct sqlite3_api_routines sqlite3_api_routines; +/* +** load_static_extension DB NAME +** +** Load an extension that is statically linked. +*/ +static int tclLoadStaticExtensionCmd( + void * clientData, + Tcl_Interp *interp, + 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*); + static const struct { + 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 }, + }; + sqlite3 *db; + const char *zName; + int i, rc; + char *zErrMsg = 0; + if( objc!=3 ){ + Tcl_WrongNumArgs(interp, 1, objv, "DB NAME"); + return TCL_ERROR; + } + if( getDbPointer(interp, Tcl_GetString(objv[1]), &db) ) return TCL_ERROR; + zName = Tcl_GetString(objv[2]); + for(i=0; i=ArraySize(aExtension) ){ + Tcl_AppendResult(interp, "no such extension: ", zName, (char*)0); + return TCL_ERROR; + } + rc = aExtension[i].pInit(db, &zErrMsg, 0); + if( rc!=SQLITE_OK || zErrMsg ){ + Tcl_AppendResult(interp, "initialization of ", zName, " failed: ", zErrMsg, + (char*)0); + sqlite3_free(zErrMsg); + return TCL_ERROR; + } + return TCL_OK; +} + /* ** Register commands with the TCL interpreter. */ int Sqlitetest1_Init(Tcl_Interp *interp){ @@ -6264,10 +6323,11 @@ #endif { "sqlite3_test_control", test_test_control }, #if SQLITE_OS_UNIX { "getrusage", test_getrusage }, #endif + { "load_static_extension", tclLoadStaticExtensionCmd }, }; static int bitmask_size = sizeof(Bitmask)*8; int i; extern int sqlite3_sync_count, sqlite3_fullsync_count; extern int sqlite3_opentemp_count; Index: src/test8.c ================================================================== --- src/test8.c +++ src/test8.c @@ -1368,33 +1368,10 @@ return TCL_ERROR; } return TCL_OK; } -#include "test_spellfix.c" - -/* -** Register the spellfix virtual table module. -*/ -static int register_spellfix_module( - ClientData clientData, - Tcl_Interp *interp, - int objc, - Tcl_Obj *CONST objv[] -){ - sqlite3 *db; - - if( objc!=2 ){ - Tcl_WrongNumArgs(interp, 1, objv, "DB"); - return TCL_ERROR; - } - if( getDbPointer(interp, Tcl_GetString(objv[1]), &db) ) return TCL_ERROR; - - sqlite3_spellfix1_register(db); - return TCL_OK; -} - #endif /* ifndef SQLITE_OMIT_VIRTUALTABLE */ /* ** Register commands with the TCL interpreter. */ @@ -1404,11 +1381,10 @@ char *zName; Tcl_ObjCmdProc *xProc; void *clientData; } aObjCmd[] = { { "register_echo_module", register_echo_module, 0 }, - { "register_spellfix_module", register_spellfix_module, 0 }, { "sqlite3_declare_vtab", declare_vtab, 0 }, }; int i; for(i=0; i); -** -** When it is created, the new fuzzer table must be supplied with the -** name of a "fuzzer data table", which must reside in the same database -** file as the new fuzzer table. The fuzzer data table contains the various -** transformations and their costs that the fuzzer logic uses to generate -** variations. -** -** The fuzzer data table must contain exactly four columns (more precisely, -** the statement "SELECT * FROM " must return records -** that consist of four columns). It does not matter what the columns are -** named. -** -** Each row in the fuzzer data table represents a single character -** transformation. The left most column of the row (column 0) contains an -** integer value - the identifier of the ruleset to which the transformation -** rule belongs (see "MULTIPLE RULE SETS" below). The second column of the -** row (column 0) contains the input character or characters. The third -** column contains the output character or characters. And the fourth column -** contains the integer cost of making the transformation. For example: -** -** CREATE TABLE f_data(ruleset, cFrom, cTo, Cost); -** INSERT INTO f_data(ruleset, cFrom, cTo, Cost) VALUES(0, '', 'a', 100); -** INSERT INTO f_data(ruleset, cFrom, cTo, Cost) VALUES(0, 'b', '', 87); -** INSERT INTO f_data(ruleset, cFrom, cTo, Cost) VALUES(0, 'o', 'oe', 38); -** INSERT INTO f_data(ruleset, cFrom, cTo, Cost) VALUES(0, 'oe', 'o', 40); -** -** The first row inserted into the fuzzer data table by the SQL script -** above indicates that the cost of inserting a letter 'a' is 100. (All -** costs are integers. We recommend that costs be scaled so that the -** average cost is around 100.) The second INSERT statement creates a rule -** saying that the cost of deleting a single letter 'b' is 87. The third -** and fourth INSERT statements mean that the cost of transforming a -** single letter "o" into the two-letter sequence "oe" is 38 and that the -** cost of transforming "oe" back into "o" is 40. -** -** The contents of the fuzzer data table are loaded into main memory when -** a fuzzer table is first created, and may be internally reloaded by the -** system at any subsequent time. Therefore, the fuzzer data table should be -** populated before the fuzzer table is created and not modified thereafter. -** If you do need to modify the contents of the fuzzer data table, it is -** recommended that the associated fuzzer table be dropped, the fuzzer data -** table edited, and the fuzzer table recreated within a single transaction. -** Alternatively, the fuzzer data table can be edited then the database -** connection can be closed and reopened. -** -** Once it has been created, the fuzzer table can be queried as follows: -** -** SELECT word, distance FROM f -** WHERE word MATCH 'abcdefg' -** AND distance<200; -** -** This first query outputs the string "abcdefg" and all strings that -** can be derived from that string by appling the specified transformations. -** The strings are output together with their total transformation cost -** (called "distance") and appear in order of increasing cost. No string -** is output more than once. If there are multiple ways to transform the -** target string into the output string then the lowest cost transform is -** the one that is returned. In the example, the search is limited to -** strings with a total distance of less than 200. -** -** The fuzzer is a read-only table. Any attempt to DELETE, INSERT, or -** UPDATE on a fuzzer table will throw an error. -** -** It is important to put some kind of a limit on the fuzzer output. This -** can be either in the form of a LIMIT clause at the end of the query, -** or better, a "distance -#include -#include -#include - -#ifndef SQLITE_OMIT_VIRTUALTABLE - -/* -** Forward declaration of objects used by this implementation -*/ -typedef struct fuzzer_vtab fuzzer_vtab; -typedef struct fuzzer_cursor fuzzer_cursor; -typedef struct fuzzer_rule fuzzer_rule; -typedef struct fuzzer_seen fuzzer_seen; -typedef struct fuzzer_stem fuzzer_stem; - -/* -** Various types. -** -** fuzzer_cost is the "cost" of an edit operation. -** -** fuzzer_len is the length of a matching string. -** -** fuzzer_ruleid is an ruleset identifier. -*/ -typedef int fuzzer_cost; -typedef signed char fuzzer_len; -typedef int fuzzer_ruleid; - -/* -** Limits -*/ -#define FUZZER_MX_LENGTH 50 /* Maximum length of a rule string */ -#define FUZZER_MX_RULEID 2147483647 /* Maximum rule ID */ -#define FUZZER_MX_COST 1000 /* Maximum single-rule cost */ -#define FUZZER_MX_OUTPUT_LENGTH 100 /* Maximum length of an output string */ - - -/* -** Each transformation rule is stored as an instance of this object. -** All rules are kept on a linked list sorted by rCost. -*/ -struct fuzzer_rule { - fuzzer_rule *pNext; /* Next rule in order of increasing rCost */ - char *zFrom; /* Transform from */ - fuzzer_cost rCost; /* Cost of this transformation */ - fuzzer_len nFrom, nTo; /* Length of the zFrom and zTo strings */ - fuzzer_ruleid iRuleset; /* The rule set to which this rule belongs */ - char zTo[4]; /* Transform to (extra space appended) */ -}; - -/* -** A stem object is used to generate variants. It is also used to record -** previously generated outputs. -** -** Every stem is added to a hash table as it is output. Generation of -** duplicate stems is suppressed. -** -** Active stems (those that might generate new outputs) are kepts on a linked -** list sorted by increasing cost. The cost is the sum of rBaseCost and -** pRule->rCost. -*/ -struct fuzzer_stem { - char *zBasis; /* Word being fuzzed */ - const fuzzer_rule *pRule; /* Current rule to apply */ - fuzzer_stem *pNext; /* Next stem in rCost order */ - fuzzer_stem *pHash; /* Next stem with same hash on zBasis */ - fuzzer_cost rBaseCost; /* Base cost of getting to zBasis */ - fuzzer_cost rCostX; /* Precomputed rBaseCost + pRule->rCost */ - fuzzer_len nBasis; /* Length of the zBasis string */ - fuzzer_len n; /* Apply pRule at this character offset */ -}; - -/* -** A fuzzer virtual-table object -*/ -struct fuzzer_vtab { - sqlite3_vtab base; /* Base class - must be first */ - char *zClassName; /* Name of this class. Default: "fuzzer" */ - fuzzer_rule *pRule; /* All active rules in this fuzzer */ - int nCursor; /* Number of active cursors */ -}; - -#define FUZZER_HASH 4001 /* Hash table size */ -#define FUZZER_NQUEUE 20 /* Number of slots on the stem queue */ - -/* A fuzzer cursor object */ -struct fuzzer_cursor { - sqlite3_vtab_cursor base; /* Base class - must be first */ - sqlite3_int64 iRowid; /* The rowid of the current word */ - fuzzer_vtab *pVtab; /* The virtual table this cursor belongs to */ - fuzzer_cost rLimit; /* Maximum cost of any term */ - fuzzer_stem *pStem; /* Stem with smallest rCostX */ - fuzzer_stem *pDone; /* Stems already processed to completion */ - fuzzer_stem *aQueue[FUZZER_NQUEUE]; /* Queue of stems with higher rCostX */ - int mxQueue; /* Largest used index in aQueue[] */ - char *zBuf; /* Temporary use buffer */ - int nBuf; /* Bytes allocated for zBuf */ - int nStem; /* Number of stems allocated */ - int iRuleset; /* Only process rules from this ruleset */ - fuzzer_rule nullRule; /* Null rule used first */ - fuzzer_stem *apHash[FUZZER_HASH]; /* Hash of previously generated terms */ -}; - -/* -** 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 fuzzer_rule *fuzzerMergeRules(fuzzer_rule *pA, fuzzer_rule *pB){ - fuzzer_rule head; - fuzzer_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 fuzzer data table. This -** function allocates and populates a fuzzer_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 fuzzerLoadOneRule( - fuzzer_vtab *p, /* Fuzzer virtual table handle */ - sqlite3_stmt *pStmt, /* Base rule on statements current row */ - fuzzer_rule **ppRule, /* OUT: New rule object */ - char **pzErr /* OUT: Error message */ -){ - sqlite3_int64 iRuleset = 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); - int nCost = 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 */ - fuzzer_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 ){ - *ppRule = 0; - return SQLITE_OK; - } - - if( nCost<=0 || nCost>FUZZER_MX_COST ){ - *pzErr = sqlite3_mprintf("%s: cost must be between 1 and %d", - p->zClassName, FUZZER_MX_COST - ); - rc = SQLITE_ERROR; - }else - if( nFrom>FUZZER_MX_LENGTH || nTo>FUZZER_MX_LENGTH ){ - *pzErr = sqlite3_mprintf("%s: maximum string length is %d", - p->zClassName, FUZZER_MX_LENGTH - ); - rc = SQLITE_ERROR; - }else - if( iRuleset<0 || iRuleset>FUZZER_MX_RULEID ){ - *pzErr = sqlite3_mprintf("%s: ruleset must be between 0 and %d", - p->zClassName, FUZZER_MX_RULEID - ); - rc = SQLITE_ERROR; - }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 = nCost; - pRule->iRuleset = (int)iRuleset; - } - } - - *ppRule = pRule; - return rc; -} - -/* -** Load the content of the fuzzer data table into memory. -*/ -static int fuzzerLoadRules( - sqlite3 *db, /* Database handle */ - fuzzer_vtab *p, /* Virtual fuzzer table to configure */ - const char *zDb, /* Database containing rules data */ - const char *zData, /* Table containing rules data */ - char **pzErr /* OUT: Error message */ -){ - int rc = SQLITE_OK; /* Return code */ - char *zSql; /* SELECT used to read from rules table */ - fuzzer_rule *pHead = 0; - - zSql = sqlite3_mprintf("SELECT * FROM %Q.%Q", zDb, zData); - 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, zData, sqlite3_column_count(pStmt) - ); - rc = SQLITE_ERROR; - }else{ - while( rc==SQLITE_OK && SQLITE_ROW==sqlite3_step(pStmt) ){ - fuzzer_rule *pRule = 0; - rc = fuzzerLoadOneRule(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 fuzzer_vtab.pRule to point to - ** point to the head of the sorted list. - */ - if( rc==SQLITE_OK ){ - unsigned int i; - fuzzer_rule *pX; - fuzzer_rule *a[15]; - for(i=0; ipNext; - pX->pNext = 0; - for(i=0; a[i] && ipRule = fuzzerMergeRules(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 *fuzzerDequote(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; iInnCursor==0 ); - while( p->pRule ){ - fuzzer_rule *pRule = p->pRule; - p->pRule = pRule->pNext; - sqlite3_free(pRule); - } - sqlite3_free(p); - return SQLITE_OK; -} - -/* -** xConnect/xCreate method for the fuzzer module. Arguments are: -** -** argv[0] -> module name ("fuzzer") -** argv[1] -> database name -** argv[2] -> table name -** argv[3] -> fuzzer rule table name -*/ -static int fuzzerConnect( - sqlite3 *db, - void *pAux, - int argc, const char *const*argv, - sqlite3_vtab **ppVtab, - char **pzErr -){ - int rc = SQLITE_OK; /* Return code */ - fuzzer_vtab *pNew = 0; /* New virtual table */ - const char *zModule = argv[0]; - const char *zDb = argv[1]; - - if( argc!=4 ){ - *pzErr = sqlite3_mprintf( - "%s: wrong number of CREATE VIRTUAL TABLE arguments", zModule - ); - rc = SQLITE_ERROR; - }else{ - int nModule; /* Length of zModule, in bytes */ - - nModule = (int)strlen(zModule); - pNew = sqlite3_malloc( sizeof(*pNew) + nModule + 1); - if( pNew==0 ){ - rc = SQLITE_NOMEM; - }else{ - char *zTab; /* Dequoted name of fuzzer data table */ - - memset(pNew, 0, sizeof(*pNew)); - pNew->zClassName = (char*)&pNew[1]; - memcpy(pNew->zClassName, zModule, nModule+1); - - zTab = fuzzerDequote(argv[3]); - if( zTab==0 ){ - rc = SQLITE_NOMEM; - }else{ - rc = fuzzerLoadRules(db, pNew, zDb, zTab, pzErr); - sqlite3_free(zTab); - } - - if( rc==SQLITE_OK ){ - rc = sqlite3_declare_vtab(db, "CREATE TABLE x(word,distance,ruleset)"); - } - if( rc!=SQLITE_OK ){ - fuzzerDisconnect((sqlite3_vtab *)pNew); - pNew = 0; - } - } - } - - *ppVtab = (sqlite3_vtab *)pNew; - return rc; -} - -/* -** Open a new fuzzer cursor. -*/ -static int fuzzerOpen(sqlite3_vtab *pVTab, sqlite3_vtab_cursor **ppCursor){ - fuzzer_vtab *p = (fuzzer_vtab*)pVTab; - fuzzer_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 all stems in a list. -*/ -static void fuzzerClearStemList(fuzzer_stem *pStem){ - while( pStem ){ - fuzzer_stem *pNext = pStem->pNext; - sqlite3_free(pStem); - pStem = pNext; - } -} - -/* -** Free up all the memory allocated by a cursor. Set it rLimit to 0 -** to indicate that it is at EOF. -*/ -static void fuzzerClearCursor(fuzzer_cursor *pCur, int clearHash){ - int i; - fuzzerClearStemList(pCur->pStem); - fuzzerClearStemList(pCur->pDone); - for(i=0; iaQueue[i]); - pCur->rLimit = (fuzzer_cost)0; - if( clearHash && pCur->nStem ){ - pCur->mxQueue = 0; - pCur->pStem = 0; - pCur->pDone = 0; - memset(pCur->aQueue, 0, sizeof(pCur->aQueue)); - memset(pCur->apHash, 0, sizeof(pCur->apHash)); - } - pCur->nStem = 0; -} - -/* -** Close a fuzzer cursor. -*/ -static int fuzzerClose(sqlite3_vtab_cursor *cur){ - fuzzer_cursor *pCur = (fuzzer_cursor *)cur; - fuzzerClearCursor(pCur, 0); - sqlite3_free(pCur->zBuf); - pCur->pVtab->nCursor--; - sqlite3_free(pCur); - return SQLITE_OK; -} - -/* -** Compute the current output term for a fuzzer_stem. -*/ -static int fuzzerRender( - fuzzer_stem *pStem, /* The stem to be rendered */ - char **pzBuf, /* Write results into this buffer. realloc if needed */ - int *pnBuf /* Size of the buffer */ -){ - const fuzzer_rule *pRule = pStem->pRule; - int n; /* Size of output term without nul-term */ - char *z; /* Buffer to assemble output term in */ - - n = pStem->nBasis + pRule->nTo - pRule->nFrom; - if( (*pnBuf)n; - z = *pzBuf; - if( n<0 ){ - memcpy(z, pStem->zBasis, pStem->nBasis+1); - }else{ - memcpy(z, pStem->zBasis, n); - memcpy(&z[n], pRule->zTo, pRule->nTo); - memcpy(&z[n+pRule->nTo], &pStem->zBasis[n+pRule->nFrom], - pStem->nBasis-n-pRule->nFrom+1); - } - - assert( z[pStem->nBasis + pRule->nTo - pRule->nFrom]==0 ); - return SQLITE_OK; -} - -/* -** Compute a hash on zBasis. -*/ -static unsigned int fuzzerHash(const char *z){ - unsigned int h = 0; - while( *z ){ h = (h<<3) ^ (h>>29) ^ *(z++); } - return h % FUZZER_HASH; -} - -/* -** Current cost of a stem -*/ -static fuzzer_cost fuzzerCost(fuzzer_stem *pStem){ - return pStem->rCostX = pStem->rBaseCost + pStem->pRule->rCost; -} - -#if 0 -/* -** Print a description of a fuzzer_stem on stderr. -*/ -static void fuzzerStemPrint( - const char *zPrefix, - fuzzer_stem *pStem, - const char *zSuffix -){ - if( pStem->n<0 ){ - fprintf(stderr, "%s[%s](%d)-->self%s", - zPrefix, - pStem->zBasis, pStem->rBaseCost, - zSuffix - ); - }else{ - char *zBuf = 0; - int nBuf = 0; - if( fuzzerRender(pStem, &zBuf, &nBuf)!=SQLITE_OK ) return; - fprintf(stderr, "%s[%s](%d)-->{%s}(%d)%s", - zPrefix, - pStem->zBasis, pStem->rBaseCost, zBuf, pStem->, - zSuffix - ); - sqlite3_free(zBuf); - } -} -#endif - -/* -** Return 1 if the string to which the cursor is point has already -** been emitted. Return 0 if not. Return -1 on a memory allocation -** failures. -*/ -static int fuzzerSeen(fuzzer_cursor *pCur, fuzzer_stem *pStem){ - unsigned int h; - fuzzer_stem *pLookup; - - if( fuzzerRender(pStem, &pCur->zBuf, &pCur->nBuf)==SQLITE_NOMEM ){ - return -1; - } - h = fuzzerHash(pCur->zBuf); - pLookup = pCur->apHash[h]; - while( pLookup && strcmp(pLookup->zBasis, pCur->zBuf)!=0 ){ - pLookup = pLookup->pHash; - } - return pLookup!=0; -} - -/* -** If argument pRule is NULL, this function returns false. -** -** Otherwise, it returns true if rule pRule should be skipped. A rule -** should be skipped if it does not belong to rule-set iRuleset, or if -** applying it to stem pStem would create a string longer than -** FUZZER_MX_OUTPUT_LENGTH bytes. -*/ -static int fuzzerSkipRule( - const fuzzer_rule *pRule, /* Determine whether or not to skip this */ - fuzzer_stem *pStem, /* Stem rule may be applied to */ - int iRuleset /* Rule-set used by the current query */ -){ - return pRule && ( - (pRule->iRuleset!=iRuleset) - || (pStem->nBasis + pRule->nTo - pRule->nFrom)>FUZZER_MX_OUTPUT_LENGTH - ); -} - -/* -** Advance a fuzzer_stem to its next value. Return 0 if there are -** no more values that can be generated by this fuzzer_stem. Return -** -1 on a memory allocation failure. -*/ -static int fuzzerAdvance(fuzzer_cursor *pCur, fuzzer_stem *pStem){ - const fuzzer_rule *pRule; - while( (pRule = pStem->pRule)!=0 ){ - assert( pRule==&pCur->nullRule || pRule->iRuleset==pCur->iRuleset ); - while( pStem->n < pStem->nBasis - pRule->nFrom ){ - pStem->n++; - if( pRule->nFrom==0 - || memcmp(&pStem->zBasis[pStem->n], pRule->zFrom, pRule->nFrom)==0 - ){ - /* Found a rewrite case. Make sure it is not a duplicate */ - int rc = fuzzerSeen(pCur, pStem); - if( rc<0 ) return -1; - if( rc==0 ){ - fuzzerCost(pStem); - return 1; - } - } - } - pStem->n = -1; - do{ - pRule = pRule->pNext; - }while( fuzzerSkipRule(pRule, pStem, pCur->iRuleset) ); - pStem->pRule = pRule; - if( pRule && fuzzerCost(pStem)>pCur->rLimit ) pStem->pRule = 0; - } - return 0; -} - -/* -** The two input stem lists are both sorted in order of increasing -** rCostX. Merge them together into a single list, sorted by rCostX, and -** return a pointer to the head of that new list. -*/ -static fuzzer_stem *fuzzerMergeStems(fuzzer_stem *pA, fuzzer_stem *pB){ - fuzzer_stem head; - fuzzer_stem *pTail; - - pTail = &head; - while( pA && pB ){ - if( pA->rCostX<=pB->rCostX ){ - 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; -} - -/* -** Load pCur->pStem with the lowest-cost stem. Return a pointer -** to the lowest-cost stem. -*/ -static fuzzer_stem *fuzzerLowestCostStem(fuzzer_cursor *pCur){ - fuzzer_stem *pBest, *pX; - int iBest; - int i; - - if( pCur->pStem==0 ){ - iBest = -1; - pBest = 0; - for(i=0; i<=pCur->mxQueue; i++){ - pX = pCur->aQueue[i]; - if( pX==0 ) continue; - if( pBest==0 || pBest->rCostX>pX->rCostX ){ - pBest = pX; - iBest = i; - } - } - if( pBest ){ - pCur->aQueue[iBest] = pBest->pNext; - pBest->pNext = 0; - pCur->pStem = pBest; - } - } - return pCur->pStem; -} - -/* -** Insert pNew into queue of pending stems. Then find the stem -** with the lowest rCostX and move it into pCur->pStem. -** list. The insert is done such the pNew is in the correct order -** according to fuzzer_stem.zBaseCost+fuzzer_stem.pRule->rCost. -*/ -static fuzzer_stem *fuzzerInsert(fuzzer_cursor *pCur, fuzzer_stem *pNew){ - fuzzer_stem *pX; - int i; - - /* If pCur->pStem exists and is greater than pNew, then make pNew - ** the new pCur->pStem and insert the old pCur->pStem instead. - */ - if( (pX = pCur->pStem)!=0 && pX->rCostX>pNew->rCostX ){ - pNew->pNext = 0; - pCur->pStem = pNew; - pNew = pX; - } - - /* Insert the new value */ - pNew->pNext = 0; - pX = pNew; - for(i=0; i<=pCur->mxQueue; i++){ - if( pCur->aQueue[i] ){ - pX = fuzzerMergeStems(pX, pCur->aQueue[i]); - pCur->aQueue[i] = 0; - }else{ - pCur->aQueue[i] = pX; - break; - } - } - if( i>pCur->mxQueue ){ - if( imxQueue = i; - pCur->aQueue[i] = pX; - }else{ - assert( pCur->mxQueue==FUZZER_NQUEUE-1 ); - pX = fuzzerMergeStems(pX, pCur->aQueue[FUZZER_NQUEUE-1]); - pCur->aQueue[FUZZER_NQUEUE-1] = pX; - } - } - - return fuzzerLowestCostStem(pCur); -} - -/* -** Allocate a new fuzzer_stem. Add it to the hash table but do not -** link it into either the pCur->pStem or pCur->pDone lists. -*/ -static fuzzer_stem *fuzzerNewStem( - fuzzer_cursor *pCur, - const char *zWord, - fuzzer_cost rBaseCost -){ - fuzzer_stem *pNew; - fuzzer_rule *pRule; - unsigned int h; - - pNew = sqlite3_malloc( sizeof(*pNew) + (int)strlen(zWord) + 1 ); - if( pNew==0 ) return 0; - memset(pNew, 0, sizeof(*pNew)); - pNew->zBasis = (char*)&pNew[1]; - pNew->nBasis = (int)strlen(zWord); - memcpy(pNew->zBasis, zWord, pNew->nBasis+1); - pRule = pCur->pVtab->pRule; - while( fuzzerSkipRule(pRule, pNew, pCur->iRuleset) ){ - pRule = pRule->pNext; - } - pNew->pRule = pRule; - pNew->n = -1; - pNew->rBaseCost = pNew->rCostX = rBaseCost; - h = fuzzerHash(pNew->zBasis); - pNew->pHash = pCur->apHash[h]; - pCur->apHash[h] = pNew; - pCur->nStem++; - return pNew; -} - - -/* -** Advance a cursor to its next row of output -*/ -static int fuzzerNext(sqlite3_vtab_cursor *cur){ - fuzzer_cursor *pCur = (fuzzer_cursor*)cur; - int rc; - fuzzer_stem *pStem, *pNew; - - pCur->iRowid++; - - /* Use the element the cursor is currently point to to create - ** a new stem and insert the new stem into the priority queue. - */ - pStem = pCur->pStem; - if( pStem->rCostX>0 ){ - rc = fuzzerRender(pStem, &pCur->zBuf, &pCur->nBuf); - if( rc==SQLITE_NOMEM ) return SQLITE_NOMEM; - pNew = fuzzerNewStem(pCur, pCur->zBuf, pStem->rCostX); - if( pNew ){ - if( fuzzerAdvance(pCur, pNew)==0 ){ - pNew->pNext = pCur->pDone; - pCur->pDone = pNew; - }else{ - if( fuzzerInsert(pCur, pNew)==pNew ){ - return SQLITE_OK; - } - } - }else{ - return SQLITE_NOMEM; - } - } - - /* Adjust the priority queue so that the first element of the - ** stem list is the next lowest cost word. - */ - while( (pStem = pCur->pStem)!=0 ){ - int res = fuzzerAdvance(pCur, pStem); - if( res<0 ){ - return SQLITE_NOMEM; - }else if( res>0 ){ - pCur->pStem = 0; - pStem = fuzzerInsert(pCur, pStem); - if( (rc = fuzzerSeen(pCur, pStem))!=0 ){ - if( rc<0 ) return SQLITE_NOMEM; - continue; - } - return SQLITE_OK; /* New word found */ - } - pCur->pStem = 0; - pStem->pNext = pCur->pDone; - pCur->pDone = pStem; - if( fuzzerLowestCostStem(pCur) ){ - rc = fuzzerSeen(pCur, pCur->pStem); - if( rc<0 ) return SQLITE_NOMEM; - if( rc==0 ){ - return SQLITE_OK; - } - } - } - - /* Reach this point only if queue has been exhausted and there is - ** nothing left to be output. */ - pCur->rLimit = (fuzzer_cost)0; - 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 fuzzerColumn, fuzzerRowid, or fuzzerEof call. -*/ -static int fuzzerFilter( - sqlite3_vtab_cursor *pVtabCursor, - int idxNum, const char *idxStr, - int argc, sqlite3_value **argv -){ - fuzzer_cursor *pCur = (fuzzer_cursor *)pVtabCursor; - const char *zWord = ""; - fuzzer_stem *pStem; - int idx; - - fuzzerClearCursor(pCur, 1); - pCur->rLimit = 2147483647; - idx = 0; - if( idxNum & 1 ){ - zWord = (const char*)sqlite3_value_text(argv[0]); - idx++; - } - if( idxNum & 2 ){ - pCur->rLimit = (fuzzer_cost)sqlite3_value_int(argv[idx]); - idx++; - } - if( idxNum & 4 ){ - pCur->iRuleset = (fuzzer_cost)sqlite3_value_int(argv[idx]); - idx++; - } - pCur->nullRule.pNext = pCur->pVtab->pRule; - pCur->nullRule.rCost = 0; - pCur->nullRule.nFrom = 0; - pCur->nullRule.nTo = 0; - pCur->nullRule.zFrom = ""; - pCur->iRowid = 1; - assert( pCur->pStem==0 ); - - /* If the query term is longer than FUZZER_MX_OUTPUT_LENGTH bytes, this - ** query will return zero rows. */ - if( (int)strlen(zWord)pStem = pStem = fuzzerNewStem(pCur, zWord, (fuzzer_cost)0); - if( pStem==0 ) return SQLITE_NOMEM; - pStem->pRule = &pCur->nullRule; - pStem->n = pStem->nBasis; - }else{ - pCur->rLimit = 0; - } - - return SQLITE_OK; -} - -/* -** Only the word and distance columns have values. All other columns -** return NULL -*/ -static int fuzzerColumn(sqlite3_vtab_cursor *cur, sqlite3_context *ctx, int i){ - fuzzer_cursor *pCur = (fuzzer_cursor*)cur; - if( i==0 ){ - /* the "word" column */ - if( fuzzerRender(pCur->pStem, &pCur->zBuf, &pCur->nBuf)==SQLITE_NOMEM ){ - return SQLITE_NOMEM; - } - sqlite3_result_text(ctx, pCur->zBuf, -1, SQLITE_TRANSIENT); - }else if( i==1 ){ - /* the "distance" column */ - sqlite3_result_int(ctx, pCur->pStem->rCostX); - }else{ - /* All other columns are NULL */ - sqlite3_result_null(ctx); - } - return SQLITE_OK; -} - -/* -** The rowid. -*/ -static int fuzzerRowid(sqlite3_vtab_cursor *cur, sqlite_int64 *pRowid){ - fuzzer_cursor *pCur = (fuzzer_cursor*)cur; - *pRowid = pCur->iRowid; - return SQLITE_OK; -} - -/* -** When the fuzzer_cursor.rLimit value is 0 or less, that is a signal -** that the cursor has nothing more to output. -*/ -static int fuzzerEof(sqlite3_vtab_cursor *cur){ - fuzzer_cursor *pCur = (fuzzer_cursor*)cur; - return pCur->rLimit<=(fuzzer_cost)0; -} - -/* -** Search for terms of these forms: -** -** (A) word MATCH $str -** (B1) distance < $value -** (B2) distance <= $value -** (C) ruleid == $ruleid -** -** 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 fuzzerBestIndex(sqlite3_vtab *tab, sqlite3_index_info *pIdxInfo){ - int iPlan = 0; - int iDistTerm = -1; - int iRulesetTerm = -1; - int i; - const struct sqlite3_index_constraint *pConstraint; - pConstraint = pIdxInfo->aConstraint; - for(i=0; inConstraint; 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; - iRulesetTerm = 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[iRulesetTerm].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; -} - -/* -** A virtual table module that implements the "fuzzer". -*/ -static sqlite3_module fuzzerModule = { - 0, /* iVersion */ - fuzzerConnect, - fuzzerConnect, - fuzzerBestIndex, - fuzzerDisconnect, - fuzzerDisconnect, - fuzzerOpen, /* xOpen - open a cursor */ - fuzzerClose, /* xClose - close a cursor */ - fuzzerFilter, /* xFilter - configure scan constraints */ - fuzzerNext, /* xNext - advance a cursor */ - fuzzerEof, /* xEof - check for end of scan */ - fuzzerColumn, /* xColumn - read data */ - fuzzerRowid, /* xRowid - read data */ - 0, /* xUpdate */ - 0, /* xBegin */ - 0, /* xSync */ - 0, /* xCommit */ - 0, /* xRollback */ - 0, /* xFindMethod */ - 0, /* xRename */ -}; - -#endif /* SQLITE_OMIT_VIRTUALTABLE */ - - -/* -** Register the fuzzer virtual table -*/ -int fuzzer_register(sqlite3 *db){ - int rc = SQLITE_OK; -#ifndef SQLITE_OMIT_VIRTUALTABLE - rc = sqlite3_create_module(db, "fuzzer", &fuzzerModule, 0); -#endif - return rc; -} - -#ifdef SQLITE_TEST -#include -/* -** Decode a pointer to an sqlite3 object. -*/ -extern int getDbPointer(Tcl_Interp *interp, const char *zA, sqlite3 **ppDb); - -/* -** Register the echo virtual table module. -*/ -static int register_fuzzer_module( - ClientData clientData, /* Pointer to sqlite3_enable_XXX function */ - Tcl_Interp *interp, /* The TCL interpreter that invoked this command */ - int objc, /* Number of arguments */ - Tcl_Obj *CONST objv[] /* Command arguments */ -){ - sqlite3 *db; - if( objc!=2 ){ - Tcl_WrongNumArgs(interp, 1, objv, "DB"); - return TCL_ERROR; - } - getDbPointer(interp, Tcl_GetString(objv[1]), &db); - fuzzer_register(db); - return TCL_OK; -} - - -/* -** Register commands with the TCL interpreter. -*/ -int Sqlitetestfuzzer_Init(Tcl_Interp *interp){ - static struct { - char *zName; - Tcl_ObjCmdProc *xProc; - void *clientData; - } aObjCmd[] = { - { "register_fuzzer_module", register_fuzzer_module, 0 }, - }; - int i; - for(i=0; i -#include -#include "sqlite3.h" - -/* The end-of-input character */ -#define RE_EOF 0 /* End of input */ - -/* The NFA is implemented as sequence of opcodes taken from the following -** set. Each opcode has a single integer argument. -*/ -#define RE_OP_MATCH 1 /* Match the one character in the argument */ -#define RE_OP_ANY 2 /* Match any one character. (Implements ".") */ -#define RE_OP_ANYSTAR 3 /* Special optimized version of .* */ -#define RE_OP_FORK 4 /* Continue to both next and opcode at iArg */ -#define RE_OP_GOTO 5 /* Jump to opcode at iArg */ -#define RE_OP_ACCEPT 6 /* Halt and indicate a successful match */ -#define RE_OP_CC_INC 7 /* Beginning of a [...] character class */ -#define RE_OP_CC_EXC 8 /* Beginning of a [^...] character class */ -#define RE_OP_CC_VALUE 9 /* Single value in a character class */ -#define RE_OP_CC_RANGE 10 /* Range of values in a character class */ -#define RE_OP_WORD 11 /* Perl word character [A-Za-z0-9_] */ -#define RE_OP_NOTWORD 12 /* Not a perl word character */ -#define RE_OP_DIGIT 13 /* digit: [0-9] */ -#define RE_OP_NOTDIGIT 14 /* Not a digit */ -#define RE_OP_SPACE 15 /* space: [ \t\n\r\v\f] */ -#define RE_OP_NOTSPACE 16 /* Not a digit */ -#define RE_OP_BOUNDARY 17 /* Boundary between word and non-word */ - -/* Each opcode is a "state" in the NFA */ -typedef unsigned short ReStateNumber; - -/* Because this is an NFA and not a DFA, multiple states can be active at -** once. An instance of the following object records all active states in -** the NFA. The implementation is optimized for the common case where the -** number of actives states is small. -*/ -typedef struct ReStateSet { - unsigned nState; /* Number of current states */ - ReStateNumber *aState; /* Current states */ -} ReStateSet; - -/* An input string read one character at a time. -*/ -typedef struct ReInput ReInput; -struct ReInput { - const unsigned char *z; /* All text */ - int i; /* Next byte to read */ - int mx; /* EOF when i>=mx */ -}; - -/* A compiled NFA (or an NFA that is in the process of being compiled) is -** an instance of the following object. -*/ -typedef struct ReCompiled ReCompiled; -struct ReCompiled { - ReInput sIn; /* Regular expression text */ - const char *zErr; /* Error message to return */ - char *aOp; /* Operators for the virtual machine */ - int *aArg; /* Arguments to each operator */ - unsigned (*xNextChar)(ReInput*); /* Next character function */ - unsigned char zInit[12]; /* Initial text to match */ - int nInit; /* Number of characters in zInit */ - unsigned nState; /* Number of entries in aOp[] and aArg[] */ - unsigned nAlloc; /* Slots allocated for aOp[] and aArg[] */ -}; - -/* Add a state to the given state set if it is not already there */ -static void re_add_state(ReStateSet *pSet, int newState){ - unsigned i; - for(i=0; inState; i++) if( pSet->aState[i]==newState ) return; - pSet->aState[pSet->nState++] = newState; -} - -/* Extract the next unicode character from *pzIn and return it. Advance -** *pzIn to the first byte past the end of the character returned. To -** be clear: this routine converts utf8 to unicode. This routine is -** optimized for the common case where the next character is a single byte. -*/ -static unsigned re_next_char(ReInput *p){ - unsigned c; - if( p->i>=p->mx ) return 0; - c = p->z[p->i++]; - if( c>=0x80 ){ - if( (c&0xe0)==0xc0 && p->imx && (p->z[p->i]&0xc0)==0x80 ){ - c = (c&0x1f)<<6 | (p->z[p->i++]&0x3f); - if( c<0x80 ) c = 0xfffd; - }else if( (c&0xf0)==0xe0 && p->i+1mx && (p->z[p->i]&0xc0)==0x80 - && (p->z[p->i+1]&0xc0)==0x80 ){ - c = (c&0x0f)<<12 | ((p->z[p->i]&0x3f)<<6) | (p->z[p->i+1]&0x3f); - p->i += 2; - if( c<=0x3ff || (c>=0xd800 && c<=0xdfff) ) c = 0xfffd; - }else if( (c&0xf8)==0xf0 && p->i+3mx && (p->z[p->i]&0xc0)==0x80 - && (p->z[p->i+1]&0xc0)==0x80 && (p->z[p->i+2]&0xc0)==0x80 ){ - c = (c&0x07)<<18 | ((p->z[p->i]&0x3f)<<12) | ((p->z[p->i+1]&0x3f)<<6) - | (p->z[p->i+2]&0x3f); - p->i += 3; - if( c<=0xffff || c>0x10ffff ) c = 0xfffd; - }else{ - c = 0xfffd; - } - } - return c; -} -static unsigned re_next_char_nocase(ReInput *p){ - unsigned c = re_next_char(p); - if( c>='A' && c<='Z' ) c += 'a' - 'A'; - return c; -} - -/* Return true if c is a perl "word" character: [A-Za-z0-9_] */ -static int re_word_char(int c){ - return (c>='0' && c<='9') || (c>='a' && c<='z') - || (c>='A' && c<='Z') || c=='_'; -} - -/* Return true if c is a "digit" character: [0-9] */ -static int re_digit_char(int c){ - return (c>='0' && c<='9'); -} - -/* Return true if c is a perl "space" character: [ \t\r\n\v\f] */ -static int re_space_char(int c){ - return c==' ' || c=='\t' || c=='\n' || c=='\r' || c=='\v' || c=='\f'; -} - -/* Run a compiled regular expression on the zero-terminated input -** string zIn[]. Return true on a match and false if there is no match. -*/ -int re_match(ReCompiled *pRe, const unsigned char *zIn, int nIn){ - ReStateSet aStateSet[2], *pThis, *pNext; - ReStateNumber aSpace[100]; - ReStateNumber *pToFree; - unsigned int i = 0; - unsigned int iSwap = 0; - int c = RE_EOF+1; - int cPrev = 0; - int rc = 0; - ReInput in; - - in.z = zIn; - in.i = 0; - in.mx = nIn>=0 ? nIn : (int)strlen((char const*)zIn); - - /* Look for the initial prefix match, if there is one. */ - if( pRe->nInit ){ - unsigned char x = pRe->zInit[0]; - while( in.i+pRe->nInit<=in.mx - && (zIn[in.i]!=x || - strncmp((const char*)zIn+in.i, (const char*)pRe->zInit, pRe->nInit)!=0) - ){ - in.i++; - } - if( in.i+pRe->nInit>in.mx ) return 0; - } - - if( pRe->nState<=(sizeof(aSpace)/(sizeof(aSpace[0])*2)) ){ - pToFree = 0; - aStateSet[0].aState = aSpace; - }else{ - pToFree = sqlite3_malloc( sizeof(ReStateNumber)*2*pRe->nState ); - if( pToFree==0 ) return -1; - aStateSet[0].aState = pToFree; - } - aStateSet[1].aState = &aStateSet[0].aState[pRe->nState]; - pNext = &aStateSet[1]; - pNext->nState = 0; - re_add_state(pNext, 0); - while( c!=RE_EOF && pNext->nState>0 ){ - cPrev = c; - c = pRe->xNextChar(&in); - pThis = pNext; - pNext = &aStateSet[iSwap]; - iSwap = 1 - iSwap; - pNext->nState = 0; - for(i=0; inState; i++){ - int x = pThis->aState[i]; - switch( pRe->aOp[x] ){ - case RE_OP_MATCH: { - if( pRe->aArg[x]==c ) re_add_state(pNext, x+1); - break; - } - case RE_OP_ANY: { - re_add_state(pNext, x+1); - break; - } - case RE_OP_WORD: { - if( re_word_char(c) ) re_add_state(pNext, x+1); - break; - } - case RE_OP_NOTWORD: { - if( !re_word_char(c) ) re_add_state(pNext, x+1); - break; - } - case RE_OP_DIGIT: { - if( re_digit_char(c) ) re_add_state(pNext, x+1); - break; - } - case RE_OP_NOTDIGIT: { - if( !re_digit_char(c) ) re_add_state(pNext, x+1); - break; - } - case RE_OP_SPACE: { - if( re_space_char(c) ) re_add_state(pNext, x+1); - break; - } - case RE_OP_NOTSPACE: { - if( !re_space_char(c) ) re_add_state(pNext, x+1); - break; - } - case RE_OP_BOUNDARY: { - if( re_word_char(c)!=re_word_char(cPrev) ) re_add_state(pThis, x+1); - break; - } - case RE_OP_ANYSTAR: { - re_add_state(pNext, x); - re_add_state(pThis, x+1); - break; - } - case RE_OP_FORK: { - re_add_state(pThis, x+pRe->aArg[x]); - re_add_state(pThis, x+1); - break; - } - case RE_OP_GOTO: { - re_add_state(pThis, x+pRe->aArg[x]); - break; - } - case RE_OP_ACCEPT: { - rc = 1; - goto re_match_end; - } - case RE_OP_CC_INC: - case RE_OP_CC_EXC: { - int j = 1; - int n = pRe->aArg[x]; - int hit = 0; - for(j=1; j>0 && jaOp[x+j]==RE_OP_CC_VALUE ){ - if( pRe->aArg[x+j]==c ){ - hit = 1; - j = -1; - } - }else{ - if( pRe->aArg[x+j]<=c && pRe->aArg[x+j+1]>=c ){ - hit = 1; - j = -1; - }else{ - j++; - } - } - } - if( pRe->aOp[x]==RE_OP_CC_EXC ) hit = !hit; - if( hit ) re_add_state(pNext, x+n); - break; - } - } - } - } - for(i=0; inState; i++){ - if( pRe->aOp[pNext->aState[i]]==RE_OP_ACCEPT ){ rc = 1; break; } - } -re_match_end: - sqlite3_free(pToFree); - return rc; -} - -/* Resize the opcode and argument arrays for an RE under construction. -*/ -static int re_resize(ReCompiled *p, int N){ - char *aOp; - int *aArg; - aOp = sqlite3_realloc(p->aOp, N*sizeof(p->aOp[0])); - if( aOp==0 ) return 1; - p->aOp = aOp; - aArg = sqlite3_realloc(p->aArg, N*sizeof(p->aArg[0])); - if( aArg==0 ) return 1; - p->aArg = aArg; - p->nAlloc = N; - return 0; -} - -/* Insert a new opcode and argument into an RE under construction. The -** insertion point is just prior to existing opcode iBefore. -*/ -static int re_insert(ReCompiled *p, int iBefore, int op, int arg){ - int i; - if( p->nAlloc<=p->nState && re_resize(p, p->nAlloc*2) ) return 0; - for(i=p->nState; i>iBefore; i--){ - p->aOp[i] = p->aOp[i-1]; - p->aArg[i] = p->aArg[i-1]; - } - p->nState++; - p->aOp[iBefore] = op; - p->aArg[iBefore] = arg; - return iBefore; -} - -/* Append a new opcode and argument to the end of the RE under construction. -*/ -static int re_append(ReCompiled *p, int op, int arg){ - return re_insert(p, p->nState, op, arg); -} - -/* Make a copy of N opcodes starting at iStart onto the end of the RE -** under construction. -*/ -static void re_copy(ReCompiled *p, int iStart, int N){ - if( p->nState+N>=p->nAlloc && re_resize(p, p->nAlloc*2+N) ) return; - memcpy(&p->aOp[p->nState], &p->aOp[iStart], N*sizeof(p->aOp[0])); - memcpy(&p->aArg[p->nState], &p->aArg[iStart], N*sizeof(p->aArg[0])); - p->nState += N; -} - -/* Return true if c is a hexadecimal digit character: [0-9a-fA-F] -** If c is a hex digit, also set *pV = (*pV)*16 + valueof(c). If -** c is not a hex digit *pV is unchanged. -*/ -static int re_hex(int c, int *pV){ - if( c>='0' && c<='9' ){ - c -= '0'; - }else if( c>='a' && c<='f' ){ - c -= 'a' - 10; - }else if( c>='A' && c<='F' ){ - c -= 'A' - 10; - }else{ - return 0; - } - *pV = (*pV)*16 + (c & 0xff); - return 1; -} - -/* A backslash character has been seen, read the next character and -** return its interpretation. -*/ -static unsigned re_esc_char(ReCompiled *p){ - static const char zEsc[] = "afnrtv\\()*.+?[$^{|}]"; - static const char zTrans[] = "\a\f\n\r\t\v"; - int i, v = 0; - char c; - if( p->sIn.i>=p->sIn.mx ) return 0; - c = p->sIn.z[p->sIn.i]; - if( c=='u' && p->sIn.i+4sIn.mx ){ - const unsigned char *zIn = p->sIn.z + p->sIn.i; - if( re_hex(zIn[1],&v) - && re_hex(zIn[2],&v) - && re_hex(zIn[3],&v) - && re_hex(zIn[4],&v) - ){ - p->sIn.i += 5; - return v; - } - } - if( c=='x' && p->sIn.i+2sIn.mx ){ - const unsigned char *zIn = p->sIn.z + p->sIn.i; - if( re_hex(zIn[1],&v) - && re_hex(zIn[2],&v) - ){ - p->sIn.i += 3; - return v; - } - } - for(i=0; zEsc[i] && zEsc[i]!=c; i++){} - if( zEsc[i] ){ - if( i<6 ) c = zTrans[i]; - p->sIn.i++; - }else{ - p->zErr = "unknown \\ escape"; - } - return c; -} - -/* Forward declaration */ -static const char *re_subcompile_string(ReCompiled*); - -/* Peek at the next byte of input */ -static unsigned char rePeek(ReCompiled *p){ - return p->sIn.isIn.mx ? p->sIn.z[p->sIn.i] : 0; -} - -/* Compile RE text into a sequence of opcodes. Continue up to the -** first unmatched ")" character, then return. If an error is found, -** return a pointer to the error message string. -*/ -static const char *re_subcompile_re(ReCompiled *p){ - const char *zErr; - int iStart, iEnd, iGoto; - iStart = p->nState; - zErr = re_subcompile_string(p); - if( zErr ) return zErr; - while( rePeek(p)=='|' ){ - iEnd = p->nState; - re_insert(p, iStart, RE_OP_FORK, iEnd + 2 - iStart); - iGoto = re_append(p, RE_OP_GOTO, 0); - p->sIn.i++; - zErr = re_subcompile_string(p); - if( zErr ) return zErr; - p->aArg[iGoto] = p->nState - iGoto; - } - return 0; -} - -/* Compile an element of regular expression text (anything that can be -** an operand to the "|" operator). Return NULL on success or a pointer -** to the error message if there is a problem. -*/ -static const char *re_subcompile_string(ReCompiled *p){ - int iPrev = -1; - int iStart; - unsigned c; - const char *zErr; - while( (c = p->xNextChar(&p->sIn))!=0 ){ - iStart = p->nState; - switch( c ){ - case '|': - case '$': - case ')': { - p->sIn.i--; - return 0; - } - case '(': { - zErr = re_subcompile_re(p); - if( zErr ) return zErr; - if( rePeek(p)!=')' ) return "unmatched '('"; - p->sIn.i++; - break; - } - case '.': { - if( rePeek(p)=='*' ){ - re_append(p, RE_OP_ANYSTAR, 0); - p->sIn.i++; - }else{ - re_append(p, RE_OP_ANY, 0); - } - break; - } - case '*': { - if( iPrev<0 ) return "'*' without operand"; - re_insert(p, iPrev, RE_OP_GOTO, p->nState - iPrev + 1); - re_append(p, RE_OP_FORK, iPrev - p->nState + 1); - break; - } - case '+': { - if( iPrev<0 ) return "'+' without operand"; - re_append(p, RE_OP_FORK, iPrev - p->nState); - break; - } - case '?': { - if( iPrev<0 ) return "'?' without operand"; - re_insert(p, iPrev, RE_OP_FORK, p->nState - iPrev+1); - break; - } - case '{': { - int m = 0, n = 0; - int sz, j; - if( iPrev<0 ) return "'{m,n}' without operand"; - while( (c=rePeek(p))>='0' && c<='9' ){ m = m*10 + c - '0'; p->sIn.i++; } - n = m; - if( c==',' ){ - p->sIn.i++; - n = 0; - while( (c=rePeek(p))>='0' && c<='9' ){ n = n*10 + c-'0'; p->sIn.i++; } - } - if( c!='}' ) return "unmatched '{'"; - if( n>0 && nsIn.i++; - sz = p->nState - iPrev; - if( m==0 ){ - if( n==0 ) return "both m and n are zero in '{m,n}'"; - re_insert(p, iPrev, RE_OP_FORK, sz+1); - n--; - }else{ - for(j=1; j0 ){ - re_append(p, RE_OP_FORK, -sz); - } - break; - } - case '[': { - int iFirst = p->nState; - if( rePeek(p)=='^' ){ - re_append(p, RE_OP_CC_EXC, 0); - p->sIn.i++; - }else{ - re_append(p, RE_OP_CC_INC, 0); - } - while( (c = p->xNextChar(&p->sIn))!=0 ){ - if( c=='[' && rePeek(p)==':' ){ - return "POSIX character classes not supported"; - } - if( c=='\\' ) c = re_esc_char(p); - if( rePeek(p)=='-' ){ - re_append(p, RE_OP_CC_RANGE, c); - p->sIn.i++; - c = p->xNextChar(&p->sIn); - if( c=='\\' ) c = re_esc_char(p); - re_append(p, RE_OP_CC_RANGE, c); - }else{ - re_append(p, RE_OP_CC_VALUE, c); - } - if( rePeek(p)==']' ){ p->sIn.i++; break; } - } - if( c==0 ) return "unclosed '['"; - p->aArg[iFirst] = p->nState - iFirst; - break; - } - case '\\': { - int specialOp = 0; - switch( rePeek(p) ){ - case 'b': specialOp = RE_OP_BOUNDARY; break; - case 'd': specialOp = RE_OP_DIGIT; break; - case 'D': specialOp = RE_OP_NOTDIGIT; break; - case 's': specialOp = RE_OP_SPACE; break; - case 'S': specialOp = RE_OP_NOTSPACE; break; - case 'w': specialOp = RE_OP_WORD; break; - case 'W': specialOp = RE_OP_NOTWORD; break; - } - if( specialOp ){ - p->sIn.i++; - re_append(p, specialOp, 0); - }else{ - c = re_esc_char(p); - re_append(p, RE_OP_MATCH, c); - } - break; - } - default: { - re_append(p, RE_OP_MATCH, c); - break; - } - } - iPrev = iStart; - } - return 0; -} - -/* Free and reclaim all the memory used by a previously compiled -** regular expression. Applications should invoke this routine once -** for every call to re_compile() to avoid memory leaks. -*/ -void re_free(ReCompiled *pRe){ - if( pRe ){ - sqlite3_free(pRe->aOp); - sqlite3_free(pRe->aArg); - sqlite3_free(pRe); - } -} - -/* -** Compile a textual regular expression in zIn[] into a compiled regular -** expression suitable for us by re_match() and return a pointer to the -** compiled regular expression in *ppRe. Return NULL on success or an -** error message if something goes wrong. -*/ -const char *re_compile(ReCompiled **ppRe, const char *zIn, int noCase){ - ReCompiled *pRe; - const char *zErr; - int i, j; - - *ppRe = 0; - pRe = sqlite3_malloc( sizeof(*pRe) ); - if( pRe==0 ){ - return "out of memory"; - } - memset(pRe, 0, sizeof(*pRe)); - pRe->xNextChar = noCase ? re_next_char_nocase : re_next_char; - if( re_resize(pRe, 30) ){ - re_free(pRe); - return "out of memory"; - } - if( zIn[0]=='^' ){ - zIn++; - }else{ - re_append(pRe, RE_OP_ANYSTAR, 0); - } - pRe->sIn.z = (unsigned char*)zIn; - pRe->sIn.i = 0; - pRe->sIn.mx = (int)strlen(zIn); - zErr = re_subcompile_re(pRe); - if( zErr ){ - re_free(pRe); - return zErr; - } - if( rePeek(pRe)=='$' && pRe->sIn.i+1>=pRe->sIn.mx ){ - re_append(pRe, RE_OP_MATCH, RE_EOF); - re_append(pRe, RE_OP_ACCEPT, 0); - *ppRe = pRe; - }else if( pRe->sIn.i>=pRe->sIn.mx ){ - re_append(pRe, RE_OP_ACCEPT, 0); - *ppRe = pRe; - }else{ - re_free(pRe); - return "unrecognized character"; - } - - /* The following is a performance optimization. If the regex begins with - ** ".*" (if the input regex lacks an initial "^") and afterwards there are - ** one or more matching characters, enter those matching characters into - ** zInit[]. The re_match() routine can then search ahead in the input - ** string looking for the initial match without having to run the whole - ** regex engine over the string. Do not worry able trying to match - ** unicode characters beyond plane 0 - those are very rare and this is - ** just an optimization. */ - if( pRe->aOp[0]==RE_OP_ANYSTAR ){ - for(j=0, i=1; jzInit)-2 && pRe->aOp[i]==RE_OP_MATCH; i++){ - unsigned x = pRe->aArg[i]; - if( x<=127 ){ - pRe->zInit[j++] = x; - }else if( x<=0xfff ){ - pRe->zInit[j++] = 0xc0 | (x>>6); - pRe->zInit[j++] = 0x80 | (x&0x3f); - }else if( x<=0xffff ){ - pRe->zInit[j++] = 0xd0 | (x>>12); - pRe->zInit[j++] = 0x80 | ((x>>6)&0x3f); - pRe->zInit[j++] = 0x80 | (x&0x3f); - }else{ - break; - } - } - if( j>0 && pRe->zInit[j-1]==0 ) j--; - pRe->nInit = j; - } - return pRe->zErr; -} - -/* -** Implementation of the regexp() SQL function. This function implements -** the build-in REGEXP operator. The first argument to the function is the -** pattern and the second argument is the string. So, the SQL statements: -** -** A REGEXP B -** -** is implemented as regexp(B,A). -*/ -static void re_sql_func( - sqlite3_context *context, - int argc, - sqlite3_value **argv -){ - ReCompiled *pRe; /* Compiled regular expression */ - const char *zPattern; /* The regular expression */ - const unsigned char *zStr;/* String being searched */ - const char *zErr; /* Compile error message */ - - pRe = sqlite3_get_auxdata(context, 0); - if( pRe==0 ){ - zPattern = (const char*)sqlite3_value_text(argv[0]); - if( zPattern==0 ) return; - zErr = re_compile(&pRe, zPattern, 0); - if( zErr ){ - re_free(pRe); - sqlite3_result_error(context, zErr, -1); - return; - } - if( pRe==0 ){ - sqlite3_result_error_nomem(context); - return; - } - sqlite3_set_auxdata(context, 0, pRe, (void(*)(void*))re_free); - } - zStr = (const unsigned char*)sqlite3_value_text(argv[1]); - if( zStr!=0 ){ - sqlite3_result_int(context, re_match(pRe, zStr, -1)); - } -} - -/* -** Invoke this routine in order to install the REGEXP function in an -** SQLite database connection. -** -** Use: -** -** sqlite3_auto_extension(sqlite3_add_regexp_func); -** -** to cause this extension to be automatically loaded into each new -** database connection. -*/ -int sqlite3_add_regexp_func(sqlite3 *db){ - return sqlite3_create_function(db, "regexp", 2, SQLITE_UTF8, 0, - re_sql_func, 0, 0); -} - - -/***************************** Test Code ***********************************/ -#ifdef SQLITE_TEST -#include -extern int getDbPointer(Tcl_Interp *interp, const char *zA, sqlite3 **ppDb); - -/* Implementation of the TCL command: -** -** sqlite3_add_regexp_func $DB -*/ -static int tclSqlite3AddRegexpFunc( - void * clientData, - Tcl_Interp *interp, - int objc, - Tcl_Obj *CONST objv[] -){ - sqlite3 *db; - if( objc!=2 ){ - Tcl_WrongNumArgs(interp, 1, objv, "DB"); - return TCL_ERROR; - } - if( getDbPointer(interp, Tcl_GetString(objv[1]), &db) ) return TCL_ERROR; - sqlite3_add_regexp_func(db); - return TCL_OK; -} - -/* Register the sqlite3_add_regexp_func TCL command with the TCL interpreter. -*/ -int Sqlitetestregexp_Init(Tcl_Interp *interp){ - Tcl_CreateObjCommand(interp, "sqlite3_add_regexp_func", - tclSqlite3AddRegexpFunc, 0, 0); - return TCL_OK; -} -#endif /* SQLITE_TEST */ -/**************************** End Of Test Code *******************************/ DELETED src/test_spellfix.c Index: src/test_spellfix.c ================================================================== --- src/test_spellfix.c +++ /dev/null @@ -1,2847 +0,0 @@ -/* -** 2012 April 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 module implements the spellfix1 VIRTUAL TABLE that can be used -** to search a large vocabulary for close matches. See separate -** documentation files (spellfix1.wiki and editdist3.wiki) for details. -*/ -#if SQLITE_CORE -# include "sqliteInt.h" -#else -# include -# include -# include -# include "sqlite3ext.h" -# include -# define ALWAYS(X) 1 -# define NEVER(X) 0 - typedef unsigned char u8; - typedef unsigned short u16; - SQLITE_EXTENSION_INIT1 -#endif /* !SQLITE_CORE */ -#include - -/* -** Character classes for ASCII characters: -** -** 0 '' Silent letters: H W -** 1 'A' Any vowel: A E I O U (Y) -** 2 'B' A bilabeal stop or fricative: B F P V W -** 3 'C' Other fricatives or back stops: C G J K Q S X Z -** 4 'D' Alveolar stops: D T -** 5 'H' Letter H at the beginning of a word -** 6 'L' Glide: L -** 7 'R' Semivowel: R -** 8 'M' Nasals: M N -** 9 'Y' Letter Y at the beginning of a word. -** 10 '9' Digits: 0 1 2 3 4 5 6 7 8 9 -** 11 ' ' White space -** 12 '?' Other. -*/ -#define CCLASS_SILENT 0 -#define CCLASS_VOWEL 1 -#define CCLASS_B 2 -#define CCLASS_C 3 -#define CCLASS_D 4 -#define CCLASS_H 5 -#define CCLASS_L 6 -#define CCLASS_R 7 -#define CCLASS_M 8 -#define CCLASS_Y 9 -#define CCLASS_DIGIT 10 -#define CCLASS_SPACE 11 -#define CCLASS_OTHER 12 - -/* -** The following table gives the character class for non-initial ASCII -** characters. -*/ -static const unsigned char midClass[] = { - /* */ CCLASS_OTHER, /* */ CCLASS_OTHER, /* */ CCLASS_OTHER, - /* */ CCLASS_OTHER, /* */ CCLASS_OTHER, /* */ CCLASS_OTHER, - /* */ CCLASS_OTHER, /* */ CCLASS_OTHER, /* */ CCLASS_OTHER, - /* */ CCLASS_SPACE, /* */ CCLASS_OTHER, /* */ CCLASS_OTHER, - /* */ CCLASS_SPACE, /* */ CCLASS_SPACE, /* */ CCLASS_OTHER, - /* */ CCLASS_OTHER, /* */ CCLASS_OTHER, /* */ CCLASS_OTHER, - /* */ CCLASS_OTHER, /* */ CCLASS_OTHER, /* */ CCLASS_OTHER, - /* */ CCLASS_OTHER, /* */ CCLASS_OTHER, /* */ CCLASS_OTHER, - /* */ CCLASS_OTHER, /* */ CCLASS_OTHER, /* */ CCLASS_OTHER, - /* */ CCLASS_OTHER, /* */ CCLASS_OTHER, /* */ CCLASS_OTHER, - /* */ CCLASS_OTHER, /* */ CCLASS_OTHER, /* */ CCLASS_SPACE, - /* ! */ CCLASS_OTHER, /* " */ CCLASS_OTHER, /* # */ CCLASS_OTHER, - /* $ */ CCLASS_OTHER, /* % */ CCLASS_OTHER, /* & */ CCLASS_OTHER, - /* ' */ CCLASS_SILENT, /* ( */ CCLASS_OTHER, /* ) */ CCLASS_OTHER, - /* * */ CCLASS_OTHER, /* + */ CCLASS_OTHER, /* , */ CCLASS_OTHER, - /* - */ CCLASS_OTHER, /* . */ CCLASS_OTHER, /* / */ CCLASS_OTHER, - /* 0 */ CCLASS_DIGIT, /* 1 */ CCLASS_DIGIT, /* 2 */ CCLASS_DIGIT, - /* 3 */ CCLASS_DIGIT, /* 4 */ CCLASS_DIGIT, /* 5 */ CCLASS_DIGIT, - /* 6 */ CCLASS_DIGIT, /* 7 */ CCLASS_DIGIT, /* 8 */ CCLASS_DIGIT, - /* 9 */ CCLASS_DIGIT, /* : */ CCLASS_OTHER, /* ; */ CCLASS_OTHER, - /* < */ CCLASS_OTHER, /* = */ CCLASS_OTHER, /* > */ CCLASS_OTHER, - /* ? */ CCLASS_OTHER, /* @ */ CCLASS_OTHER, /* A */ CCLASS_VOWEL, - /* B */ CCLASS_B, /* C */ CCLASS_C, /* D */ CCLASS_D, - /* E */ CCLASS_VOWEL, /* F */ CCLASS_B, /* G */ CCLASS_C, - /* H */ CCLASS_SILENT, /* I */ CCLASS_VOWEL, /* J */ CCLASS_C, - /* K */ CCLASS_C, /* L */ CCLASS_L, /* M */ CCLASS_M, - /* N */ CCLASS_M, /* O */ CCLASS_VOWEL, /* P */ CCLASS_B, - /* Q */ CCLASS_C, /* R */ CCLASS_R, /* S */ CCLASS_C, - /* T */ CCLASS_D, /* U */ CCLASS_VOWEL, /* V */ CCLASS_B, - /* W */ CCLASS_B, /* X */ CCLASS_C, /* Y */ CCLASS_VOWEL, - /* Z */ CCLASS_C, /* [ */ CCLASS_OTHER, /* \ */ CCLASS_OTHER, - /* ] */ CCLASS_OTHER, /* ^ */ CCLASS_OTHER, /* _ */ CCLASS_OTHER, - /* ` */ CCLASS_OTHER, /* a */ CCLASS_VOWEL, /* b */ CCLASS_B, - /* c */ CCLASS_C, /* d */ CCLASS_D, /* e */ CCLASS_VOWEL, - /* f */ CCLASS_B, /* g */ CCLASS_C, /* h */ CCLASS_SILENT, - /* i */ CCLASS_VOWEL, /* j */ CCLASS_C, /* k */ CCLASS_C, - /* l */ CCLASS_L, /* m */ CCLASS_M, /* n */ CCLASS_M, - /* o */ CCLASS_VOWEL, /* p */ CCLASS_B, /* q */ CCLASS_C, - /* r */ CCLASS_R, /* s */ CCLASS_C, /* t */ CCLASS_D, - /* u */ CCLASS_VOWEL, /* v */ CCLASS_B, /* w */ CCLASS_B, - /* x */ CCLASS_C, /* y */ CCLASS_VOWEL, /* z */ CCLASS_C, - /* { */ CCLASS_OTHER, /* | */ CCLASS_OTHER, /* } */ CCLASS_OTHER, - /* ~ */ CCLASS_OTHER, /* */ CCLASS_OTHER, -}; -/* -** This tables gives the character class for ASCII characters that form the -** initial character of a word. The only difference from midClass is with -** the letters H, W, and Y. -*/ -static const unsigned char initClass[] = { - /* */ CCLASS_OTHER, /* */ CCLASS_OTHER, /* */ CCLASS_OTHER, - /* */ CCLASS_OTHER, /* */ CCLASS_OTHER, /* */ CCLASS_OTHER, - /* */ CCLASS_OTHER, /* */ CCLASS_OTHER, /* */ CCLASS_OTHER, - /* */ CCLASS_SPACE, /* */ CCLASS_OTHER, /* */ CCLASS_OTHER, - /* */ CCLASS_SPACE, /* */ CCLASS_SPACE, /* */ CCLASS_OTHER, - /* */ CCLASS_OTHER, /* */ CCLASS_OTHER, /* */ CCLASS_OTHER, - /* */ CCLASS_OTHER, /* */ CCLASS_OTHER, /* */ CCLASS_OTHER, - /* */ CCLASS_OTHER, /* */ CCLASS_OTHER, /* */ CCLASS_OTHER, - /* */ CCLASS_OTHER, /* */ CCLASS_OTHER, /* */ CCLASS_OTHER, - /* */ CCLASS_OTHER, /* */ CCLASS_OTHER, /* */ CCLASS_OTHER, - /* */ CCLASS_OTHER, /* */ CCLASS_OTHER, /* */ CCLASS_SPACE, - /* ! */ CCLASS_OTHER, /* " */ CCLASS_OTHER, /* # */ CCLASS_OTHER, - /* $ */ CCLASS_OTHER, /* % */ CCLASS_OTHER, /* & */ CCLASS_OTHER, - /* ' */ CCLASS_OTHER, /* ( */ CCLASS_OTHER, /* ) */ CCLASS_OTHER, - /* * */ CCLASS_OTHER, /* + */ CCLASS_OTHER, /* , */ CCLASS_OTHER, - /* - */ CCLASS_OTHER, /* . */ CCLASS_OTHER, /* / */ CCLASS_OTHER, - /* 0 */ CCLASS_DIGIT, /* 1 */ CCLASS_DIGIT, /* 2 */ CCLASS_DIGIT, - /* 3 */ CCLASS_DIGIT, /* 4 */ CCLASS_DIGIT, /* 5 */ CCLASS_DIGIT, - /* 6 */ CCLASS_DIGIT, /* 7 */ CCLASS_DIGIT, /* 8 */ CCLASS_DIGIT, - /* 9 */ CCLASS_DIGIT, /* : */ CCLASS_OTHER, /* ; */ CCLASS_OTHER, - /* < */ CCLASS_OTHER, /* = */ CCLASS_OTHER, /* > */ CCLASS_OTHER, - /* ? */ CCLASS_OTHER, /* @ */ CCLASS_OTHER, /* A */ CCLASS_VOWEL, - /* B */ CCLASS_B, /* C */ CCLASS_C, /* D */ CCLASS_D, - /* E */ CCLASS_VOWEL, /* F */ CCLASS_B, /* G */ CCLASS_C, - /* H */ CCLASS_SILENT, /* I */ CCLASS_VOWEL, /* J */ CCLASS_C, - /* K */ CCLASS_C, /* L */ CCLASS_L, /* M */ CCLASS_M, - /* N */ CCLASS_M, /* O */ CCLASS_VOWEL, /* P */ CCLASS_B, - /* Q */ CCLASS_C, /* R */ CCLASS_R, /* S */ CCLASS_C, - /* T */ CCLASS_D, /* U */ CCLASS_VOWEL, /* V */ CCLASS_B, - /* W */ CCLASS_B, /* X */ CCLASS_C, /* Y */ CCLASS_Y, - /* Z */ CCLASS_C, /* [ */ CCLASS_OTHER, /* \ */ CCLASS_OTHER, - /* ] */ CCLASS_OTHER, /* ^ */ CCLASS_OTHER, /* _ */ CCLASS_OTHER, - /* ` */ CCLASS_OTHER, /* a */ CCLASS_VOWEL, /* b */ CCLASS_B, - /* c */ CCLASS_C, /* d */ CCLASS_D, /* e */ CCLASS_VOWEL, - /* f */ CCLASS_B, /* g */ CCLASS_C, /* h */ CCLASS_SILENT, - /* i */ CCLASS_VOWEL, /* j */ CCLASS_C, /* k */ CCLASS_C, - /* l */ CCLASS_L, /* m */ CCLASS_M, /* n */ CCLASS_M, - /* o */ CCLASS_VOWEL, /* p */ CCLASS_B, /* q */ CCLASS_C, - /* r */ CCLASS_R, /* s */ CCLASS_C, /* t */ CCLASS_D, - /* u */ CCLASS_VOWEL, /* v */ CCLASS_B, /* w */ CCLASS_B, - /* x */ CCLASS_C, /* y */ CCLASS_Y, /* z */ CCLASS_C, - /* { */ CCLASS_OTHER, /* | */ CCLASS_OTHER, /* } */ CCLASS_OTHER, - /* ~ */ CCLASS_OTHER, /* */ CCLASS_OTHER, -}; - -/* -** Mapping from the character class number (0-13) to a symbol for each -** character class. Note that initClass[] can be used to map the class -** symbol back into the class number. -*/ -static const unsigned char className[] = ".ABCDHLRMY9 ?"; - -/* -** Generate a "phonetic hash" from a string of ASCII characters -** in zIn[0..nIn-1]. -** -** * Map characters by character class as defined above. -** * Omit double-letters -** * Omit vowels beside R and L -** * Omit T when followed by CH -** * Omit W when followed by R -** * Omit D when followed by J or G -** * Omit K in KN or G in GN at the beginning of a word -** -** Space to hold the result is obtained from sqlite3_malloc() -** -** Return NULL if memory allocation fails. -*/ -static unsigned char *phoneticHash(const unsigned char *zIn, int nIn){ - unsigned char *zOut = sqlite3_malloc( nIn + 1 ); - int i; - int nOut = 0; - char cPrev = 0x77; - char cPrevX = 0x77; - const unsigned char *aClass = initClass; - - if( zOut==0 ) return 0; - if( nIn>2 ){ - switch( zIn[0] ){ - case 'g': - case 'k': { - if( zIn[1]=='n' ){ zIn++; nIn--; } - break; - } - } - } - for(i=0; i=0 ); - if( nOut==0 || c!=zOut[nOut-1] ) zOut[nOut++] = c; - } - zOut[nOut] = 0; - return zOut; -} - -/* -** This is an SQL function wrapper around phoneticHash(). See -** the description of phoneticHash() for additional information. -*/ -static void phoneticHashSqlFunc( - sqlite3_context *context, - int argc, - sqlite3_value **argv -){ - const unsigned char *zIn; - unsigned char *zOut; - - zIn = sqlite3_value_text(argv[0]); - if( zIn==0 ) return; - zOut = phoneticHash(zIn, sqlite3_value_bytes(argv[0])); - if( zOut==0 ){ - sqlite3_result_error_nomem(context); - }else{ - sqlite3_result_text(context, (char*)zOut, -1, sqlite3_free); - } -} - -/* -** Return the character class number for a character given its -** context. -*/ -static char characterClass(char cPrev, char c){ - return cPrev==0 ? initClass[c&0x7f] : midClass[c&0x7f]; -} - -/* -** Return the cost of inserting or deleting character c immediately -** following character cPrev. If cPrev==0, that means c is the first -** character of the word. -*/ -static int insertOrDeleteCost(char cPrev, char c, char cNext){ - char classC = characterClass(cPrev, c); - char classCprev; - - if( classC==CCLASS_SILENT ){ - /* Insert or delete "silent" characters such as H or W */ - return 1; - } - if( cPrev==c ){ - /* Repeated characters, or miss a repeat */ - return 10; - } - if( classC==CCLASS_VOWEL && (cPrev=='r' || cNext=='r') ){ - return 20; /* Insert a vowel before or after 'r' */ - } - classCprev = characterClass(cPrev, cPrev); - if( classC==classCprev ){ - if( classC==CCLASS_VOWEL ){ - /* Remove or add a new vowel to a vowel cluster */ - return 15; - }else{ - /* Remove or add a consonant not in the same class */ - return 50; - } - } - - /* any other character insertion or deletion */ - return 100; -} - -/* -** Divide the insertion cost by this factor when appending to the -** end of the word. -*/ -#define FINAL_INS_COST_DIV 4 - -/* -** Return the cost of substituting cTo in place of cFrom assuming -** the previous character is cPrev. If cPrev==0 then cTo is the first -** character of the word. -*/ -static int substituteCost(char cPrev, char cFrom, char cTo){ - char classFrom, classTo; - if( cFrom==cTo ){ - /* Exact match */ - return 0; - } - if( cFrom==(cTo^0x20) && ((cTo>='A' && cTo<='Z') || (cTo>='a' && cTo<='z')) ){ - /* differ only in case */ - return 0; - } - classFrom = characterClass(cPrev, cFrom); - classTo = characterClass(cPrev, cTo); - if( classFrom==classTo ){ - /* Same character class */ - return 40; - } - if( classFrom>=CCLASS_B && classFrom<=CCLASS_Y - && classTo>=CCLASS_B && classTo<=CCLASS_Y ){ - /* Convert from one consonant to another, but in a different class */ - return 75; - } - /* Any other subsitution */ - return 100; -} - -/* -** Given two strings zA and zB which are pure ASCII, return the cost -** of transforming zA into zB. If zA ends with '*' assume that it is -** a prefix of zB and give only minimal penalty for extra characters -** on the end of zB. -** -** Smaller numbers mean a closer match. -** -** Negative values indicate an error: -** -1 One of the inputs is NULL -** -2 Non-ASCII characters on input -** -3 Unable to allocate memory -** -** If pnMatch is not NULL, then *pnMatch is set to the number of bytes -** of zB that matched the pattern in zA. If zA does not end with a '*', -** then this value is always the number of bytes in zB (i.e. strlen(zB)). -** If zA does end in a '*', then it is the number of bytes in the prefix -** of zB that was deemed to match zA. -*/ -static int editdist1(const char *zA, const char *zB, int *pnMatch){ - int nA, nB; /* Number of characters in zA[] and zB[] */ - int xA, xB; /* Loop counters for zA[] and zB[] */ - char cA, cB; /* Current character of zA and zB */ - char cAprev, cBprev; /* Previous character of zA and zB */ - char cAnext, cBnext; /* Next character in zA and zB */ - int d; /* North-west cost value */ - int dc = 0; /* North-west character value */ - int res; /* Final result */ - int *m; /* The cost matrix */ - char *cx; /* Corresponding character values */ - int *toFree = 0; /* Malloced space */ - int mStack[60+15]; /* Stack space to use if not too much is needed */ - int nMatch = 0; - - /* Early out if either input is NULL */ - if( zA==0 || zB==0 ) return -1; - - /* Skip any common prefix */ - while( zA[0] && zA[0]==zB[0] ){ dc = zA[0]; zA++; zB++; nMatch++; } - if( pnMatch ) *pnMatch = nMatch; - if( zA[0]==0 && zB[0]==0 ) return 0; - -#if 0 - printf("A=\"%s\" B=\"%s\" dc=%c\n", zA, zB, dc?dc:' '); -#endif - - /* Verify input strings and measure their lengths */ - for(nA=0; zA[nA]; nA++){ - if( zA[nA]&0x80 ) return -2; - } - for(nB=0; zB[nB]; nB++){ - if( zB[nB]&0x80 ) return -2; - } - - /* Special processing if either string is empty */ - if( nA==0 ){ - cBprev = dc; - for(xB=res=0; (cB = zB[xB])!=0; xB++){ - res += insertOrDeleteCost(cBprev, cB, zB[xB+1])/FINAL_INS_COST_DIV; - cBprev = cB; - } - return res; - } - if( nB==0 ){ - cAprev = dc; - for(xA=res=0; (cA = zA[xA])!=0; xA++){ - res += insertOrDeleteCost(cAprev, cA, zA[xA+1]); - cAprev = cA; - } - return res; - } - - /* A is a prefix of B */ - if( zA[0]=='*' && zA[1]==0 ) return 0; - - /* Allocate and initialize the Wagner matrix */ - if( nB<(sizeof(mStack)*4)/(sizeof(mStack[0])*5) ){ - m = mStack; - }else{ - m = toFree = sqlite3_malloc( (nB+1)*5*sizeof(m[0])/4 ); - if( m==0 ) return -3; - } - cx = (char*)&m[nB+1]; - - /* Compute the Wagner edit distance */ - m[0] = 0; - cx[0] = dc; - cBprev = dc; - for(xB=1; xB<=nB; xB++){ - cBnext = zB[xB]; - cB = zB[xB-1]; - cx[xB] = cB; - m[xB] = m[xB-1] + insertOrDeleteCost(cBprev, cB, cBnext); - cBprev = cB; - } - cAprev = dc; - for(xA=1; xA<=nA; xA++){ - int lastA = (xA==nA); - cA = zA[xA-1]; - cAnext = zA[xA]; - if( cA=='*' && lastA ) break; - d = m[0]; - dc = cx[0]; - m[0] = d + insertOrDeleteCost(cAprev, cA, cAnext); - cBprev = 0; - for(xB=1; xB<=nB; xB++){ - int totalCost, insCost, delCost, subCost, ncx; - cB = zB[xB-1]; - cBnext = zB[xB]; - - /* Cost to insert cB */ - insCost = insertOrDeleteCost(cx[xB-1], cB, cBnext); - if( lastA ) insCost /= FINAL_INS_COST_DIV; - - /* Cost to delete cA */ - delCost = insertOrDeleteCost(cx[xB], cA, cBnext); - - /* Cost to substitute cA->cB */ - subCost = substituteCost(cx[xB-1], cA, cB); - - /* Best cost */ - totalCost = insCost + m[xB-1]; - ncx = cB; - if( (delCost + m[xB])nLang; i++){ - EditDist3Cost *pCost, *pNext; - pCost = p->a[i].pCost; - while( pCost ){ - pNext = pCost->pNext; - sqlite3_free(pCost); - pCost = pNext; - } - } - sqlite3_free(p->a); - memset(p, 0, sizeof(*p)); -} -static void editDist3ConfigDelete(void *pIn){ - EditDist3Config *p = (EditDist3Config*)pIn; - editDist3ConfigClear(p); - sqlite3_free(p); -} - -/* -** Load all edit-distance weights from a table. -*/ -static int editDist3ConfigLoad( - EditDist3Config *p, /* The edit distance configuration to load */ - sqlite3 *db, /* Load from this database */ - const char *zTable /* Name of the table from which to load */ -){ - sqlite3_stmt *pStmt; - int rc, rc2; - char *zSql; - int iLangPrev = -9999; - EditDist3Lang *pLang = 0; - - zSql = sqlite3_mprintf("SELECT iLang, cFrom, cTo, iCost" - " FROM \"%w\" WHERE iLang>=0 ORDER BY iLang", zTable); - if( zSql==0 ) return SQLITE_NOMEM; - rc = sqlite3_prepare(db, zSql, -1, &pStmt, 0); - sqlite3_free(zSql); - if( rc ) return rc; - editDist3ConfigClear(p); - while( sqlite3_step(pStmt)==SQLITE_ROW ){ - int iLang = sqlite3_column_int(pStmt, 0); - const char *zFrom = (const char*)sqlite3_column_text(pStmt, 1); - int nFrom = zFrom ? sqlite3_column_bytes(pStmt, 1) : 0; - const char *zTo = (const char*)sqlite3_column_text(pStmt, 2); - int nTo = zTo ? sqlite3_column_bytes(pStmt, 2) : 0; - int iCost = sqlite3_column_int(pStmt, 3); - - assert( zFrom!=0 || nFrom==0 ); - assert( zTo!=0 || nTo==0 ); - if( nFrom>100 || nTo>100 ) continue; - if( iCost<0 ) continue; - if( pLang==0 || iLang!=iLangPrev ){ - EditDist3Lang *pNew; - pNew = sqlite3_realloc(p->a, (p->nLang+1)*sizeof(p->a[0])); - if( pNew==0 ){ rc = SQLITE_NOMEM; break; } - p->a = pNew; - pLang = &p->a[p->nLang]; - p->nLang++; - pLang->iLang = iLang; - pLang->iInsCost = 100; - pLang->iDelCost = 100; - pLang->iSubCost = 150; - pLang->pCost = 0; - iLangPrev = iLang; - } - if( nFrom==1 && zFrom[0]=='?' && nTo==0 ){ - pLang->iDelCost = iCost; - }else if( nFrom==0 && nTo==1 && zTo[0]=='?' ){ - pLang->iInsCost = iCost; - }else if( nFrom==1 && nTo==1 && zFrom[0]=='?' && zTo[0]=='?' ){ - pLang->iSubCost = iCost; - }else{ - EditDist3Cost *pCost; - int nExtra = nFrom + nTo - 4; - if( nExtra<0 ) nExtra = 0; - pCost = sqlite3_malloc( sizeof(*pCost) + nExtra ); - if( pCost==0 ){ rc = SQLITE_NOMEM; break; } - pCost->nFrom = nFrom; - pCost->nTo = nTo; - pCost->iCost = iCost; - memcpy(pCost->a, zFrom, nFrom); - memcpy(pCost->a + nFrom, zTo, nTo); - pCost->pNext = pLang->pCost; - pLang->pCost = pCost; - } - } - rc2 = sqlite3_finalize(pStmt); - if( rc==SQLITE_OK ) rc = rc2; - return rc; -} - -/* -** Return the length (in bytes) of a utf-8 character. Or return a maximum -** of N. -*/ -static int utf8Len(unsigned char c, int N){ - int len = 1; - if( c>0x7f ){ - if( (c&0xe0)==0xc0 ){ - len = 2; - }else if( (c&0xf0)==0xe0 ){ - len = 3; - }else{ - len = 4; - } - } - if( len>N ) len = N; - return len; -} - -/* -** Return TRUE (non-zero) if the To side of the given cost matches -** the given string. -*/ -static int matchTo(EditDist3Cost *p, const char *z, int n){ - if( p->nTo>n ) return 0; - if( strncmp(p->a+p->nFrom, z, p->nTo)!=0 ) return 0; - return 1; -} - -/* -** Return TRUE (non-zero) if the From side of the given cost matches -** the given string. -*/ -static int matchFrom(EditDist3Cost *p, const char *z, int n){ - assert( p->nFrom<=n ); - if( strncmp(p->a, z, p->nFrom)!=0 ) return 0; - return 1; -} - -/* -** Return TRUE (non-zero) of the next FROM character and the next TO -** character are the same. -*/ -static int matchFromTo( - EditDist3FromString *pStr, /* Left hand string */ - int n1, /* Index of comparison character on the left */ - const char *z2, /* Right-handl comparison character */ - int n2 /* Bytes remaining in z2[] */ -){ - int b1 = pStr->a[n1].nByte; - if( b1>n2 ) return 0; - if( memcmp(pStr->z+n1, z2, b1)!=0 ) return 0; - return 1; -} - -/* -** Delete an EditDist3FromString objecct -*/ -static void editDist3FromStringDelete(EditDist3FromString *p){ - int i; - if( p ){ - for(i=0; in; i++){ - sqlite3_free(p->a[i].apDel); - sqlite3_free(p->a[i].apSubst); - } - sqlite3_free(p); - } -} - -/* -** Create a EditDist3FromString object. -*/ -static EditDist3FromString *editDist3FromStringNew( - const EditDist3Lang *pLang, - const char *z, - int n -){ - EditDist3FromString *pStr; - EditDist3Cost *p; - int i; - - if( z==0 ) return 0; - if( n<0 ) n = (int)strlen(z); - pStr = sqlite3_malloc( sizeof(*pStr) + sizeof(pStr->a[0])*n + n + 1 ); - if( pStr==0 ) return 0; - pStr->a = (EditDist3From*)&pStr[1]; - memset(pStr->a, 0, sizeof(pStr->a[0])*n); - pStr->n = n; - pStr->z = (char*)&pStr->a[n]; - memcpy(pStr->z, z, n+1); - if( n && z[n-1]=='*' ){ - pStr->isPrefix = 1; - n--; - pStr->n--; - pStr->z[n] = 0; - }else{ - pStr->isPrefix = 0; - } - - for(i=0; ia[i]; - memset(pFrom, 0, sizeof(*pFrom)); - pFrom->nByte = utf8Len((unsigned char)z[i], n-i); - for(p=pLang->pCost; p; p=p->pNext){ - EditDist3Cost **apNew; - if( i+p->nFrom>n ) continue; - if( matchFrom(p, z+i, n-i)==0 ) continue; - if( p->nTo==0 ){ - apNew = sqlite3_realloc(pFrom->apDel, - sizeof(*apNew)*(pFrom->nDel+1)); - if( apNew==0 ) break; - pFrom->apDel = apNew; - apNew[pFrom->nDel++] = p; - }else{ - apNew = sqlite3_realloc(pFrom->apSubst, - sizeof(*apNew)*(pFrom->nSubst+1)); - if( apNew==0 ) break; - pFrom->apSubst = apNew; - apNew[pFrom->nSubst++] = p; - } - } - if( p ){ - editDist3FromStringDelete(pStr); - pStr = 0; - break; - } - } - return pStr; -} - -/* -** Update entry m[i] such that it is the minimum of its current value -** and m[j]+iCost. -** -** If the iCost is 1,000,000 or greater, then consider the cost to be -** infinite and skip the update. -*/ -static void updateCost( - unsigned int *m, - int i, - int j, - int iCost -){ - assert( iCost>=0 ); - if( iCost<10000 ){ - unsigned int b = m[j] + iCost; - if( bpCost; p; p=p->pNext){ - EditDist3Cost **apNew; - if( p->nFrom>0 ) continue; - if( i2+p->nTo>n2 ) continue; - if( matchTo(p, z2+i2, n2-i2)==0 ) continue; - a2[i2].nIns++; - apNew = sqlite3_realloc(a2[i2].apIns, sizeof(*apNew)*a2[i2].nIns); - if( apNew==0 ){ - res = -1; /* Out of memory */ - goto editDist3Abort; - } - a2[i2].apIns = apNew; - a2[i2].apIns[a2[i2].nIns-1] = p; - } - } - - /* Prepare to compute the minimum edit distance */ - szRow = f.n+1; - memset(m, 0x01, (n2+1)*szRow*sizeof(m[0])); - m[0] = 0; - - /* First fill in the top-row of the matrix with FROM deletion costs */ - for(i1=0; i1iDelCost); - for(k=0; knFrom, i1, p->iCost); - } - } - - /* Fill in all subsequent rows, top-to-bottom, left-to-right */ - for(i2=0; i2iInsCost); - for(k=0; knTo), rxp, p->iCost); - } - for(i1=0; i1iDelCost); - for(k=0; knFrom, cxp, p->iCost); - } - updateCost(m, cx, cxu, pLang->iInsCost); - if( matchFromTo(&f, i1, z2+i2, n2-i2) ){ - updateCost(m, cx, cxd, 0); - } - updateCost(m, cx, cxd, pLang->iSubCost); - for(k=0; knFrom+szRow*p->nTo, cxd, p->iCost); - } - } - } - } - -#if 0 /* Enable for debugging */ - printf(" ^"); - for(i1=0; i19999 ) printf(" ****"); - else printf(" %4d", v); - } - printf("\n"); - for(i2=0; i29999 ) printf(" ****"); - else printf(" %4d", v); - } - printf("\n"); - } -#endif - - /* Free memory allocations and return the result */ - res = (int)m[szRow*(n2+1)-1]; - n = n2; - if( f.isPrefix ){ - for(i2=1; i2<=n2; i2++){ - int b = m[szRow*i2-1]; - if( b<=res ){ - res = b; - n = i2 - 1; - } - } - } - if( pnMatch ){ - int nExtra = 0; - for(k=0; knLang; i++){ - if( pConfig->a[i].iLang==iLang ) return &pConfig->a[i]; - } - return &editDist3Lang; -} - -/* -** Function: editdist3(A,B,iLang) -** editdist3(tablename) -** -** Return the cost of transforming string A into string B using edit -** weights for iLang. -** -** The second form loads edit weights into memory from a table. -*/ -static void editDist3SqlFunc( - sqlite3_context *context, - int argc, - sqlite3_value **argv -){ - EditDist3Config *pConfig = (EditDist3Config*)sqlite3_user_data(context); - sqlite3 *db = sqlite3_context_db_handle(context); - int rc; - if( argc==1 ){ - const char *zTable = (const char*)sqlite3_value_text(argv[0]); - rc = editDist3ConfigLoad(pConfig, db, zTable); - if( rc ) sqlite3_result_error_code(context, rc); - }else{ - const char *zA = (const char*)sqlite3_value_text(argv[0]); - const char *zB = (const char*)sqlite3_value_text(argv[1]); - int nA = sqlite3_value_bytes(argv[0]); - int nB = sqlite3_value_bytes(argv[1]); - int iLang = argc==3 ? sqlite3_value_int(argv[2]) : 0; - const EditDist3Lang *pLang = editDist3FindLang(pConfig, iLang); - EditDist3FromString *pFrom; - int dist; - - pFrom = editDist3FromStringNew(pLang, zA, nA); - if( pFrom==0 ){ - sqlite3_result_error_nomem(context); - return; - } - dist = editDist3Core(pFrom, zB, nB, pLang, 0); - editDist3FromStringDelete(pFrom); - if( dist==(-1) ){ - sqlite3_result_error_nomem(context); - }else{ - sqlite3_result_int(context, dist); - } - } -} - -/* -** Register the editDist3 function with SQLite -*/ -static int editDist3Install(sqlite3 *db){ - int rc; - EditDist3Config *pConfig = sqlite3_malloc( sizeof(*pConfig) ); - if( pConfig==0 ) return SQLITE_NOMEM; - memset(pConfig, 0, sizeof(*pConfig)); - rc = sqlite3_create_function_v2(db, "editdist3", - 2, SQLITE_UTF8, pConfig, editDist3SqlFunc, 0, 0, 0); - if( rc==SQLITE_OK ){ - rc = sqlite3_create_function_v2(db, "editdist3", - 3, SQLITE_UTF8, pConfig, editDist3SqlFunc, 0, 0, 0); - } - if( rc==SQLITE_OK ){ - rc = sqlite3_create_function_v2(db, "editdist3", - 1, SQLITE_UTF8, pConfig, editDist3SqlFunc, 0, 0, - editDist3ConfigDelete); - }else{ - sqlite3_free(pConfig); - } - return rc; -} -/* End configurable cost unicode edit distance routines -****************************************************************************** -****************************************************************************** -** Begin transliterate unicode-to-ascii implementation -*/ - -#if !SQLITE_AMALGAMATION -/* -** This lookup table is used to help decode the first byte of -** a multi-byte UTF8 character. -*/ -static const unsigned char sqlite3Utf8Trans1[] = { - 0x00, 0x01, 0x02, 0x03, 0x04, 0x05, 0x06, 0x07, - 0x08, 0x09, 0x0a, 0x0b, 0x0c, 0x0d, 0x0e, 0x0f, - 0x10, 0x11, 0x12, 0x13, 0x14, 0x15, 0x16, 0x17, - 0x18, 0x19, 0x1a, 0x1b, 0x1c, 0x1d, 0x1e, 0x1f, - 0x00, 0x01, 0x02, 0x03, 0x04, 0x05, 0x06, 0x07, - 0x08, 0x09, 0x0a, 0x0b, 0x0c, 0x0d, 0x0e, 0x0f, - 0x00, 0x01, 0x02, 0x03, 0x04, 0x05, 0x06, 0x07, - 0x00, 0x01, 0x02, 0x03, 0x00, 0x01, 0x00, 0x00, -}; -#endif - -/* -** Return the value of the first UTF-8 character in the string. -*/ -static int utf8Read(const unsigned char *z, int n, int *pSize){ - int c, i; - - /* All callers to this routine (in the current implementation) - ** always have n>0. */ - if( NEVER(n==0) ){ - c = i = 0; - }else{ - c = z[0]; - i = 1; - if( c>=0xc0 ){ - c = sqlite3Utf8Trans1[c-0xc0]; - while( i0 ){ - c = utf8Read(zIn, nIn, &sz); - zIn += sz; - nIn -= sz; - if( c<=127 ){ - zOut[nOut++] = c; - }else{ - int xTop, xBtm, x; - xTop = sizeof(translit)/sizeof(translit[0]) - 1; - xBtm = 0; - while( xTop>=xBtm ){ - x = (xTop + xBtm)/2; - if( translit[x].cFrom==c ){ - zOut[nOut++] = translit[x].cTo0; - if( translit[x].cTo1 ){ - zOut[nOut++] = translit[x].cTo1; - /* Add an extra "ch" after the "sh" for Щ and щ */ - if( c==0x0429 || c== 0x0449 ){ - zOut[nOut++] = 'c'; - zOut[nOut++] = 'h'; - } - } - c = 0; - break; - }else if( translit[x].cFrom>c ){ - xTop = x-1; - }else{ - xBtm = x+1; - } - } - if( c ) zOut[nOut++] = '?'; - } - } - zOut[nOut] = 0; - return zOut; -} - -/* -** Return the number of characters in the shortest prefix of the input -** string that transliterates to an ASCII string nTrans bytes or longer. -** Or, if the transliteration of the input string is less than nTrans -** bytes in size, return the number of characters in the input string. -*/ -static int translen_to_charlen(const char *zIn, int nIn, int nTrans){ - int i, c, sz, nOut; - int nChar; - - i = nOut = 0; - for(nChar=0; i=128 ){ - int xTop, xBtm, x; - xTop = sizeof(translit)/sizeof(translit[0]) - 1; - xBtm = 0; - while( xTop>=xBtm ){ - x = (xTop + xBtm)/2; - if( translit[x].cFrom==c ){ - if( translit[x].cTo1 ) nOut++; - if( c==0x0429 || c== 0x0449 ) nOut += 2; - break; - }else if( translit[x].cFrom>c ){ - xTop = x-1; - }else{ - xBtm = x+1; - } - } - } - } - - return nChar; -} - - -/* -** spellfix1_translit(X) -** -** Convert a string that contains non-ASCII Roman characters into -** pure ASCII. -*/ -static void transliterateSqlFunc( - sqlite3_context *context, - int argc, - sqlite3_value **argv -){ - const unsigned char *zIn = sqlite3_value_text(argv[0]); - int nIn = sqlite3_value_bytes(argv[0]); - unsigned char *zOut = transliterate(zIn, nIn); - if( zOut==0 ){ - sqlite3_result_error_nomem(context); - }else{ - sqlite3_result_text(context, (char*)zOut, -1, sqlite3_free); - } -} - -/* -** spellfix1_scriptcode(X) -** -** Try to determine the dominant script used by the word X and return -** its ISO 15924 numeric code. -** -** The current implementation only understands the following scripts: -** -** 215 (Latin) -** 220 (Cyrillic) -** 200 (Greek) -** -** This routine will return 998 if the input X contains characters from -** two or more of the above scripts or 999 if X contains no characters -** from any of the above scripts. -*/ -static void scriptCodeSqlFunc( - sqlite3_context *context, - int argc, - sqlite3_value **argv -){ - const unsigned char *zIn = sqlite3_value_text(argv[0]); - int nIn = sqlite3_value_bytes(argv[0]); - int c, sz; - int scriptMask = 0; - int res; -# define SCRIPT_LATIN 0x0001 -# define SCRIPT_CYRILLIC 0x0002 -# define SCRIPT_GREEK 0x0004 - - while( nIn>0 ){ - c = utf8Read(zIn, nIn, &sz); - zIn += sz; - nIn -= sz; - if( c<0x02af ){ - scriptMask |= SCRIPT_LATIN; - }else if( c>=0x0400 && c<=0x04ff ){ - scriptMask |= SCRIPT_CYRILLIC; - }else if( c>=0x0386 && c<=0x03ce ){ - scriptMask |= SCRIPT_GREEK; - } - } - switch( scriptMask ){ - case 0: res = 999; break; - case SCRIPT_LATIN: res = 215; break; - case SCRIPT_CYRILLIC: res = 220; break; - case SCRIPT_GREEK: res = 200; break; - default: res = 998; break; - } - sqlite3_result_int(context, res); -} - -/* End transliterate -****************************************************************************** -****************************************************************************** -** Begin spellfix1 virtual table. -*/ - -/* Maximum length of a phonehash used for querying the shadow table */ -#define SPELLFIX_MX_HASH 8 - -/* Maximum number of hash strings to examine per query */ -#define SPELLFIX_MX_RUN 1 - -typedef struct spellfix1_vtab spellfix1_vtab; -typedef struct spellfix1_cursor spellfix1_cursor; - -/* Fuzzy-search virtual table object */ -struct spellfix1_vtab { - sqlite3_vtab base; /* Base class - must be first */ - sqlite3 *db; /* Database connection */ - char *zDbName; /* Name of database holding this table */ - char *zTableName; /* Name of the virtual table */ - char *zCostTable; /* Table holding edit-distance cost numbers */ - EditDist3Config *pConfig3; /* Parsed edit distance costs */ -}; - -/* Fuzzy-search cursor object */ -struct spellfix1_cursor { - sqlite3_vtab_cursor base; /* Base class - must be first */ - spellfix1_vtab *pVTab; /* The table to which this cursor belongs */ - char *zPattern; /* rhs of MATCH clause */ - int nRow; /* Number of rows of content */ - int nAlloc; /* Number of allocated rows */ - int iRow; /* Current row of content */ - int iLang; /* Value of the langid= constraint */ - int iTop; /* Value of the top= constraint */ - int iScope; /* Value of the scope= constraint */ - int nSearch; /* Number of vocabulary items checked */ - sqlite3_stmt *pFullScan; /* Shadow query for a full table scan */ - struct spellfix1_row { /* For each row of content */ - sqlite3_int64 iRowid; /* Rowid for this row */ - char *zWord; /* Text for this row */ - int iRank; /* Rank for this row */ - int iDistance; /* Distance from pattern for this row */ - int iScore; /* Score for sorting */ - int iMatchlen; /* Value of matchlen column (or -1) */ - char zHash[SPELLFIX_MX_HASH]; /* the phonehash used for this match */ - } *a; -}; - -/* -** Construct one or more SQL statements from the format string given -** and then evaluate those statements. The success code is written -** into *pRc. -** -** If *pRc is initially non-zero then this routine is a no-op. -*/ -static void spellfix1DbExec( - int *pRc, /* Success code */ - sqlite3 *db, /* Database in which to run SQL */ - const char *zFormat, /* Format string for SQL */ - ... /* Arguments to the format string */ -){ - va_list ap; - char *zSql; - if( *pRc ) return; - va_start(ap, zFormat); - zSql = sqlite3_vmprintf(zFormat, ap); - va_end(ap); - if( zSql==0 ){ - *pRc = SQLITE_NOMEM; - }else{ - *pRc = sqlite3_exec(db, zSql, 0, 0, 0); - sqlite3_free(zSql); - } -} - -/* -** xDisconnect/xDestroy method for the fuzzy-search module. -*/ -static int spellfix1Uninit(int isDestroy, sqlite3_vtab *pVTab){ - spellfix1_vtab *p = (spellfix1_vtab*)pVTab; - int rc = SQLITE_OK; - if( isDestroy ){ - sqlite3 *db = p->db; - spellfix1DbExec(&rc, db, "DROP TABLE IF EXISTS \"%w\".\"%w_vocab\"", - p->zDbName, p->zTableName); - } - if( rc==SQLITE_OK ){ - sqlite3_free(p->zTableName); - editDist3ConfigDelete(p->pConfig3); - sqlite3_free(p->zCostTable); - sqlite3_free(p); - } - return rc; -} -static int spellfix1Disconnect(sqlite3_vtab *pVTab){ - return spellfix1Uninit(0, pVTab); -} -static int spellfix1Destroy(sqlite3_vtab *pVTab){ - return spellfix1Uninit(1, pVTab); -} - -/* -** Make a copy of a string. Remove leading and trailing whitespace -** and dequote it. -*/ -static char *spellfix1Dequote(const char *zIn){ - char *zOut; - int i, j; - char c; - while( isspace(zIn[0]) ) zIn++; - zOut = sqlite3_mprintf("%s", zIn); - if( zOut==0 ) return 0; - i = (int)strlen(zOut); -#if 0 /* The parser will never leave spaces at the end */ - while( i>0 && isspace(zOut[i-1]) ){ i--; } -#endif - zOut[i] = 0; - c = zOut[0]; - if( c=='\'' || c=='"' ){ - for(i=1, j=0; ALWAYS(zOut[i]); i++){ - zOut[j++] = zOut[i]; - if( zOut[i]==c ){ - if( zOut[i+1]==c ){ - i++; - }else{ - zOut[j-1] = 0; - break; - } - } - } - } - return zOut; -} - - -/* -** xConnect/xCreate method for the spellfix1 module. Arguments are: -** -** argv[0] -> module name ("spellfix1") -** argv[1] -> database name -** argv[2] -> table name -** argv[3].. -> optional arguments (i.e. "edit_cost_table" parameter) -*/ -static int spellfix1Init( - int isCreate, - sqlite3 *db, - void *pAux, - int argc, const char *const*argv, - sqlite3_vtab **ppVTab, - char **pzErr -){ - spellfix1_vtab *pNew = 0; - const char *zModule = argv[0]; - const char *zDbName = argv[1]; - const char *zTableName = argv[2]; - int nDbName; - int rc = SQLITE_OK; - int i; - - nDbName = (int)strlen(zDbName); - pNew = sqlite3_malloc( sizeof(*pNew) + nDbName + 1); - if( pNew==0 ){ - rc = SQLITE_NOMEM; - }else{ - memset(pNew, 0, sizeof(*pNew)); - pNew->zDbName = (char*)&pNew[1]; - memcpy(pNew->zDbName, zDbName, nDbName+1); - pNew->zTableName = sqlite3_mprintf("%s", zTableName); - pNew->db = db; - if( pNew->zTableName==0 ){ - rc = SQLITE_NOMEM; - }else{ - rc = sqlite3_declare_vtab(db, - "CREATE TABLE x(word,rank,distance,langid, " - "score, matchlen, phonehash HIDDEN, " - "top HIDDEN, scope HIDDEN, srchcnt HIDDEN, " - "soundslike HIDDEN, command HIDDEN)" - ); -#define SPELLFIX_COL_WORD 0 -#define SPELLFIX_COL_RANK 1 -#define SPELLFIX_COL_DISTANCE 2 -#define SPELLFIX_COL_LANGID 3 -#define SPELLFIX_COL_SCORE 4 -#define SPELLFIX_COL_MATCHLEN 5 -#define SPELLFIX_COL_PHONEHASH 6 -#define SPELLFIX_COL_TOP 7 -#define SPELLFIX_COL_SCOPE 8 -#define SPELLFIX_COL_SRCHCNT 9 -#define SPELLFIX_COL_SOUNDSLIKE 10 -#define SPELLFIX_COL_COMMAND 11 - } - if( rc==SQLITE_OK && isCreate ){ - sqlite3_uint64 r; - spellfix1DbExec(&rc, db, - "CREATE TABLE IF NOT EXISTS \"%w\".\"%w_vocab\"(\n" - " id INTEGER PRIMARY KEY,\n" - " rank INT,\n" - " langid INT,\n" - " word TEXT,\n" - " k1 TEXT,\n" - " k2 TEXT\n" - ");\n", - zDbName, zTableName - ); - sqlite3_randomness(sizeof(r), &r); - spellfix1DbExec(&rc, db, - "CREATE INDEX IF NOT EXISTS \"%w\".\"%w_index_%llx\" " - "ON \"%w_vocab\"(langid,k2);", - zDbName, zModule, r, zTableName - ); - } - for(i=3; rc==SQLITE_OK && ibase); - }else{ - *ppVTab = (sqlite3_vtab *)pNew; - } - return rc; -} - -/* -** The xConnect and xCreate methods -*/ -static int spellfix1Connect( - sqlite3 *db, - void *pAux, - int argc, const char *const*argv, - sqlite3_vtab **ppVTab, - char **pzErr -){ - return spellfix1Init(0, db, pAux, argc, argv, ppVTab, pzErr); -} -static int spellfix1Create( - sqlite3 *db, - void *pAux, - int argc, const char *const*argv, - sqlite3_vtab **ppVTab, - char **pzErr -){ - return spellfix1Init(1, db, pAux, argc, argv, ppVTab, pzErr); -} - -/* -** Clear all of the content from a cursor. -*/ -static void spellfix1ResetCursor(spellfix1_cursor *pCur){ - int i; - for(i=0; inRow; i++){ - sqlite3_free(pCur->a[i].zWord); - } - pCur->nRow = 0; - pCur->iRow = 0; - pCur->nSearch = 0; - if( pCur->pFullScan ){ - sqlite3_finalize(pCur->pFullScan); - pCur->pFullScan = 0; - } -} - -/* -** Resize the cursor to hold up to N rows of content -*/ -static void spellfix1ResizeCursor(spellfix1_cursor *pCur, int N){ - struct spellfix1_row *aNew; - assert( N>=pCur->nRow ); - aNew = sqlite3_realloc(pCur->a, sizeof(pCur->a[0])*N); - if( aNew==0 && N>0 ){ - spellfix1ResetCursor(pCur); - sqlite3_free(pCur->a); - pCur->nAlloc = 0; - pCur->a = 0; - }else{ - pCur->nAlloc = N; - pCur->a = aNew; - } -} - - -/* -** Close a fuzzy-search cursor. -*/ -static int spellfix1Close(sqlite3_vtab_cursor *cur){ - spellfix1_cursor *pCur = (spellfix1_cursor *)cur; - spellfix1ResetCursor(pCur); - spellfix1ResizeCursor(pCur, 0); - sqlite3_free(pCur->zPattern); - sqlite3_free(pCur); - return SQLITE_OK; -} - -/* -** Search for terms of these forms: -** -** (A) word MATCH $str -** (B) langid == $langid -** (C) top = $top -** (D) scope = $scope -** (E) distance < $distance -** (F) distance <= $distance -** -** The plan number is a bit mask formed with these bits: -** -** 0x01 (A) is found -** 0x02 (B) is found -** 0x04 (C) is found -** 0x08 (D) is found -** 0x10 (E) is found -** 0x20 (F) is found -** -** filter.argv[*] values contains $str, $langid, $top, and $scope, -** if specified and in that order. -*/ -static int spellfix1BestIndex(sqlite3_vtab *tab, sqlite3_index_info *pIdxInfo){ - int iPlan = 0; - int iLangTerm = -1; - int iTopTerm = -1; - int iScopeTerm = -1; - int iDistTerm = -1; - int i; - const struct sqlite3_index_constraint *pConstraint; - pConstraint = pIdxInfo->aConstraint; - for(i=0; inConstraint; i++, pConstraint++){ - if( pConstraint->usable==0 ) continue; - - /* Terms of the form: word MATCH $str */ - if( (iPlan & 1)==0 - && pConstraint->iColumn==SPELLFIX_COL_WORD - && pConstraint->op==SQLITE_INDEX_CONSTRAINT_MATCH - ){ - iPlan |= 1; - pIdxInfo->aConstraintUsage[i].argvIndex = 1; - pIdxInfo->aConstraintUsage[i].omit = 1; - } - - /* Terms of the form: langid = $langid */ - if( (iPlan & 2)==0 - && pConstraint->iColumn==SPELLFIX_COL_LANGID - && pConstraint->op==SQLITE_INDEX_CONSTRAINT_EQ - ){ - iPlan |= 2; - iLangTerm = i; - } - - /* Terms of the form: top = $top */ - if( (iPlan & 4)==0 - && pConstraint->iColumn==SPELLFIX_COL_TOP - && pConstraint->op==SQLITE_INDEX_CONSTRAINT_EQ - ){ - iPlan |= 4; - iTopTerm = i; - } - - /* Terms of the form: scope = $scope */ - if( (iPlan & 8)==0 - && pConstraint->iColumn==SPELLFIX_COL_SCOPE - && pConstraint->op==SQLITE_INDEX_CONSTRAINT_EQ - ){ - iPlan |= 8; - iScopeTerm = i; - } - - /* Terms of the form: distance < $dist or distance <= $dist */ - if( (iPlan & (16|32))==0 - && pConstraint->iColumn==SPELLFIX_COL_DISTANCE - && (pConstraint->op==SQLITE_INDEX_CONSTRAINT_LT - || pConstraint->op==SQLITE_INDEX_CONSTRAINT_LE) - ){ - iPlan |= pConstraint->op==SQLITE_INDEX_CONSTRAINT_LT ? 16 : 32; - iDistTerm = i; - } - } - if( iPlan&1 ){ - int idx = 2; - pIdxInfo->idxNum = iPlan; - if( pIdxInfo->nOrderBy==1 - && pIdxInfo->aOrderBy[0].iColumn==SPELLFIX_COL_SCORE - && pIdxInfo->aOrderBy[0].desc==0 - ){ - pIdxInfo->orderByConsumed = 1; /* Default order by iScore */ - } - if( iPlan&2 ){ - pIdxInfo->aConstraintUsage[iLangTerm].argvIndex = idx++; - pIdxInfo->aConstraintUsage[iLangTerm].omit = 1; - } - if( iPlan&4 ){ - pIdxInfo->aConstraintUsage[iTopTerm].argvIndex = idx++; - pIdxInfo->aConstraintUsage[iTopTerm].omit = 1; - } - if( iPlan&8 ){ - pIdxInfo->aConstraintUsage[iScopeTerm].argvIndex = idx++; - pIdxInfo->aConstraintUsage[iScopeTerm].omit = 1; - } - if( iPlan&(16|32) ){ - pIdxInfo->aConstraintUsage[iDistTerm].argvIndex = idx++; - pIdxInfo->aConstraintUsage[iDistTerm].omit = 1; - } - pIdxInfo->estimatedCost = (double)10000; - }else{ - pIdxInfo->idxNum = 0; - pIdxInfo->estimatedCost = (double)10000000; - } - return SQLITE_OK; -} - -/* -** Open a new fuzzy-search cursor. -*/ -static int spellfix1Open(sqlite3_vtab *pVTab, sqlite3_vtab_cursor **ppCursor){ - spellfix1_vtab *p = (spellfix1_vtab*)pVTab; - spellfix1_cursor *pCur; - pCur = sqlite3_malloc( sizeof(*pCur) ); - if( pCur==0 ) return SQLITE_NOMEM; - memset(pCur, 0, sizeof(*pCur)); - pCur->pVTab = p; - *ppCursor = &pCur->base; - return SQLITE_OK; -} - -/* -** Adjust a distance measurement by the words rank in order to show -** preference to common words. -*/ -static int spellfix1Score(int iDistance, int iRank){ - int iLog2; - for(iLog2=0; iRank>0; iLog2++, iRank>>=1){} - return iDistance + 32 - iLog2; -} - -/* -** Compare two spellfix1_row objects for sorting purposes in qsort() such -** that they sort in order of increasing distance. -*/ -static int spellfix1RowCompare(const void *A, const void *B){ - const struct spellfix1_row *a = (const struct spellfix1_row*)A; - const struct spellfix1_row *b = (const struct spellfix1_row*)B; - return a->iScore - b->iScore; -} - -/* -** A structure used to pass information from spellfix1FilterForMatch() -** into spellfix1RunQuery(). -*/ -typedef struct MatchQuery { - spellfix1_cursor *pCur; /* The cursor being queried */ - sqlite3_stmt *pStmt; /* shadow table query statment */ - char zHash[SPELLFIX_MX_HASH]; /* The current phonehash for zPattern */ - const char *zPattern; /* Transliterated input string */ - int nPattern; /* Length of zPattern */ - EditDist3FromString *pMatchStr3; /* Original unicode string */ - EditDist3Config *pConfig3; /* Edit-distance cost coefficients */ - const EditDist3Lang *pLang; /* The selected language coefficients */ - int iLang; /* The language id */ - int iScope; /* Default scope */ - int iMaxDist; /* Maximum allowed edit distance, or -1 */ - int rc; /* Error code */ - int nRun; /* Number of prior runs for the same zPattern */ - char azPrior[SPELLFIX_MX_RUN][SPELLFIX_MX_HASH]; /* Prior hashes */ -} MatchQuery; - -/* -** Run a query looking for the best matches against zPattern using -** zHash as the character class seed hash. -*/ -static void spellfix1RunQuery(MatchQuery *p, const char *zQuery, int nQuery){ - const char *zK1; - const char *zWord; - int iDist; - int iRank; - int iScore; - int iWorst = 0; - int idx; - int idxWorst = -1; - int i; - int iScope = p->iScope; - spellfix1_cursor *pCur = p->pCur; - sqlite3_stmt *pStmt = p->pStmt; - char zHash1[SPELLFIX_MX_HASH]; - char zHash2[SPELLFIX_MX_HASH]; - char *zClass; - int nClass; - int rc; - - if( pCur->a==0 || p->rc ) return; /* Prior memory allocation failure */ - zClass = (char*)phoneticHash((unsigned char*)zQuery, nQuery); - if( zClass==0 ){ - p->rc = SQLITE_NOMEM; - return; - } - nClass = (int)strlen(zClass); - if( nClass>SPELLFIX_MX_HASH-2 ){ - nClass = SPELLFIX_MX_HASH-2; - zClass[nClass] = 0; - } - if( nClass<=iScope ){ - if( nClass>2 ){ - iScope = nClass-1; - }else{ - iScope = nClass; - } - } - memcpy(zHash1, zClass, iScope); - sqlite3_free(zClass); - zHash1[iScope] = 0; - memcpy(zHash2, zHash1, iScope); - zHash2[iScope] = 'Z'; - zHash2[iScope+1] = 0; -#if SPELLFIX_MX_RUN>1 - for(i=0; inRun; i++){ - if( strcmp(p->azPrior[i], zHash1)==0 ) return; - } -#endif - assert( p->nRunazPrior[p->nRun++], zHash1, iScope+1); - if( sqlite3_bind_text(pStmt, 1, zHash1, -1, SQLITE_STATIC)==SQLITE_NOMEM - || sqlite3_bind_text(pStmt, 2, zHash2, -1, SQLITE_STATIC)==SQLITE_NOMEM - ){ - p->rc = SQLITE_NOMEM; - return; - } -#if SPELLFIX_MX_RUN>1 - for(i=0; inRow; i++){ - if( pCur->a[i].iScore>iWorst ){ - iWorst = pCur->a[i].iScore; - idxWorst = i; - } - } -#endif - while( sqlite3_step(pStmt)==SQLITE_ROW ){ - int iMatchlen = -1; - iRank = sqlite3_column_int(pStmt, 2); - if( p->pMatchStr3 ){ - int nWord = sqlite3_column_bytes(pStmt, 1); - zWord = (const char*)sqlite3_column_text(pStmt, 1); - iDist = editDist3Core(p->pMatchStr3, zWord, nWord, p->pLang, &iMatchlen); - }else{ - zK1 = (const char*)sqlite3_column_text(pStmt, 3); - if( zK1==0 ) continue; - iDist = editdist1(p->zPattern, zK1, 0); - } - if( iDist<0 ){ - p->rc = SQLITE_NOMEM; - break; - } - pCur->nSearch++; - iScore = spellfix1Score(iDist,iRank); - if( p->iMaxDist>=0 ){ - if( iDist>p->iMaxDist ) continue; - if( pCur->nRow>=pCur->nAlloc-1 ){ - spellfix1ResizeCursor(pCur, pCur->nAlloc*2 + 10); - if( pCur->a==0 ) break; - } - idx = pCur->nRow; - }else if( pCur->nRownAlloc ){ - idx = pCur->nRow; - }else if( iScorea[idx].zWord); - }else{ - continue; - } - pCur->a[idx].zWord = sqlite3_mprintf("%s", sqlite3_column_text(pStmt, 1)); - if( pCur->a[idx].zWord==0 ){ - p->rc = SQLITE_NOMEM; - break; - } - pCur->a[idx].iRowid = sqlite3_column_int64(pStmt, 0); - pCur->a[idx].iRank = iRank; - pCur->a[idx].iDistance = iDist; - pCur->a[idx].iScore = iScore; - pCur->a[idx].iMatchlen = iMatchlen; - memcpy(pCur->a[idx].zHash, zHash1, iScope+1); - if( pCur->nRownAlloc ) pCur->nRow++; - if( pCur->nRow==pCur->nAlloc ){ - iWorst = pCur->a[0].iScore; - idxWorst = 0; - for(i=1; inRow; i++){ - iScore = pCur->a[i].iScore; - if( iWorstrc = rc; -} - -/* -** This version of the xFilter method work if the MATCH term is present -** and we are doing a scan. -*/ -static int spellfix1FilterForMatch( - spellfix1_cursor *pCur, - int idxNum, - int argc, - sqlite3_value **argv -){ - const unsigned char *zMatchThis; /* RHS of the MATCH operator */ - EditDist3FromString *pMatchStr3 = 0; /* zMatchThis as an editdist string */ - char *zPattern; /* Transliteration of zMatchThis */ - int nPattern; /* Length of zPattern */ - int iLimit = 20; /* Max number of rows of output */ - int iScope = 3; /* Use this many characters of zClass */ - int iLang = 0; /* Language code */ - char *zSql; /* SQL of shadow table query */ - sqlite3_stmt *pStmt = 0; /* Shadow table query */ - int rc; /* Result code */ - int idx = 1; /* Next available filter parameter */ - spellfix1_vtab *p = pCur->pVTab; /* The virtual table that owns pCur */ - MatchQuery x; /* For passing info to RunQuery() */ - - /* Load the cost table if we have not already done so */ - if( p->zCostTable!=0 && p->pConfig3==0 ){ - p->pConfig3 = sqlite3_malloc( sizeof(p->pConfig3[0]) ); - if( p->pConfig3==0 ) return SQLITE_NOMEM; - memset(p->pConfig3, 0, sizeof(p->pConfig3[0])); - rc = editDist3ConfigLoad(p->pConfig3, p->db, p->zCostTable); - if( rc ) return rc; - } - memset(&x, 0, sizeof(x)); - x.iScope = 3; /* Default scope if none specified by "WHERE scope=N" */ - x.iMaxDist = -1; /* Maximum allowed edit distance */ - - if( idxNum&2 ){ - iLang = sqlite3_value_int(argv[idx++]); - } - if( idxNum&4 ){ - iLimit = sqlite3_value_int(argv[idx++]); - if( iLimit<1 ) iLimit = 1; - } - if( idxNum&8 ){ - x.iScope = sqlite3_value_int(argv[idx++]); - if( x.iScope<1 ) x.iScope = 1; - if( x.iScope>SPELLFIX_MX_HASH-2 ) x.iScope = SPELLFIX_MX_HASH-2; - } - if( idxNum&(16|32) ){ - x.iMaxDist = sqlite3_value_int(argv[idx++]); - if( idxNum&16 ) x.iMaxDist--; - if( x.iMaxDist<0 ) x.iMaxDist = 0; - } - spellfix1ResetCursor(pCur); - spellfix1ResizeCursor(pCur, iLimit); - zMatchThis = sqlite3_value_text(argv[0]); - if( zMatchThis==0 ) return SQLITE_OK; - if( p->pConfig3 ){ - x.pLang = editDist3FindLang(p->pConfig3, iLang); - pMatchStr3 = editDist3FromStringNew(x.pLang, (const char*)zMatchThis, -1); - if( pMatchStr3==0 ){ - x.rc = SQLITE_NOMEM; - goto filter_exit; - } - }else{ - x.pLang = 0; - } - zPattern = (char*)transliterate(zMatchThis, sqlite3_value_bytes(argv[0])); - sqlite3_free(pCur->zPattern); - pCur->zPattern = zPattern; - if( zPattern==0 ){ - x.rc = SQLITE_NOMEM; - goto filter_exit; - } - nPattern = (int)strlen(zPattern); - if( zPattern[nPattern-1]=='*' ) nPattern--; - zSql = sqlite3_mprintf( - "SELECT id, word, rank, k1" - " FROM \"%w\".\"%w_vocab\"" - " WHERE langid=%d AND k2>=?1 AND k2zDbName, p->zTableName, iLang - ); - if( zSql==0 ){ - x.rc = SQLITE_NOMEM; - pStmt = 0; - goto filter_exit; - } - rc = sqlite3_prepare_v2(p->db, zSql, -1, &pStmt, 0); - sqlite3_free(zSql); - pCur->iLang = iLang; - x.pCur = pCur; - x.pStmt = pStmt; - x.zPattern = zPattern; - x.nPattern = nPattern; - x.pMatchStr3 = pMatchStr3; - x.iLang = iLang; - x.rc = rc; - x.pConfig3 = p->pConfig3; - if( x.rc==SQLITE_OK ){ - spellfix1RunQuery(&x, zPattern, nPattern); - } - - if( pCur->a ){ - qsort(pCur->a, pCur->nRow, sizeof(pCur->a[0]), spellfix1RowCompare); - pCur->iTop = iLimit; - pCur->iScope = iScope; - }else{ - x.rc = SQLITE_NOMEM; - } - -filter_exit: - sqlite3_finalize(pStmt); - editDist3FromStringDelete(pMatchStr3); - return x.rc; -} - -/* -** This version of xFilter handles a full-table scan case -*/ -static int spellfix1FilterForFullScan( - spellfix1_cursor *pCur, - int idxNum, - int argc, - sqlite3_value **argv -){ - int rc; - char *zSql; - spellfix1_vtab *pVTab = pCur->pVTab; - spellfix1ResetCursor(pCur); - zSql = sqlite3_mprintf( - "SELECT word, rank, NULL, langid, id FROM \"%w\".\"%w_vocab\"", - pVTab->zDbName, pVTab->zTableName); - if( zSql==0 ) return SQLITE_NOMEM; - rc = sqlite3_prepare_v2(pVTab->db, zSql, -1, &pCur->pFullScan, 0); - sqlite3_free(zSql); - pCur->nRow = pCur->iRow = 0; - if( rc==SQLITE_OK ){ - rc = sqlite3_step(pCur->pFullScan); - if( rc==SQLITE_ROW ){ pCur->iRow = -1; rc = SQLITE_OK; } - if( rc==SQLITE_DONE ){ rc = SQLITE_OK; } - }else{ - pCur->iRow = 0; - } - return rc; -} - - -/* -** 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 spellfix1Column, spellfix1Rowid, or spellfix1Eof call. -*/ -static int spellfix1Filter( - sqlite3_vtab_cursor *cur, - int idxNum, const char *idxStr, - int argc, sqlite3_value **argv -){ - spellfix1_cursor *pCur = (spellfix1_cursor *)cur; - int rc; - if( idxNum & 1 ){ - rc = spellfix1FilterForMatch(pCur, idxNum, argc, argv); - }else{ - rc = spellfix1FilterForFullScan(pCur, idxNum, argc, argv); - } - return rc; -} - - -/* -** Advance a cursor to its next row of output -*/ -static int spellfix1Next(sqlite3_vtab_cursor *cur){ - spellfix1_cursor *pCur = (spellfix1_cursor *)cur; - int rc = SQLITE_OK; - if( pCur->iRow < pCur->nRow ){ - if( pCur->pFullScan ){ - rc = sqlite3_step(pCur->pFullScan); - if( rc!=SQLITE_ROW ) pCur->iRow = pCur->nRow; - if( rc==SQLITE_ROW || rc==SQLITE_DONE ) rc = SQLITE_OK; - }else{ - pCur->iRow++; - } - } - return rc; -} - -/* -** Return TRUE if we are at the end-of-file -*/ -static int spellfix1Eof(sqlite3_vtab_cursor *cur){ - spellfix1_cursor *pCur = (spellfix1_cursor *)cur; - return pCur->iRow>=pCur->nRow; -} - -/* -** Return columns from the current row. -*/ -static int spellfix1Column( - sqlite3_vtab_cursor *cur, - sqlite3_context *ctx, - int i -){ - spellfix1_cursor *pCur = (spellfix1_cursor*)cur; - if( pCur->pFullScan ){ - if( i<=SPELLFIX_COL_LANGID ){ - sqlite3_result_value(ctx, sqlite3_column_value(pCur->pFullScan, i)); - }else{ - sqlite3_result_null(ctx); - } - return SQLITE_OK; - } - switch( i ){ - case SPELLFIX_COL_WORD: { - sqlite3_result_text(ctx, pCur->a[pCur->iRow].zWord, -1, SQLITE_STATIC); - break; - } - case SPELLFIX_COL_RANK: { - sqlite3_result_int(ctx, pCur->a[pCur->iRow].iRank); - break; - } - case SPELLFIX_COL_DISTANCE: { - sqlite3_result_int(ctx, pCur->a[pCur->iRow].iDistance); - break; - } - case SPELLFIX_COL_LANGID: { - sqlite3_result_int(ctx, pCur->iLang); - break; - } - case SPELLFIX_COL_SCORE: { - sqlite3_result_int(ctx, pCur->a[pCur->iRow].iScore); - break; - } - case SPELLFIX_COL_MATCHLEN: { - int iMatchlen = pCur->a[pCur->iRow].iMatchlen; - if( iMatchlen<0 ){ - int nPattern = (int)strlen(pCur->zPattern); - char *zWord = pCur->a[pCur->iRow].zWord; - int nWord = (int)strlen(zWord); - - if( nPattern>0 && pCur->zPattern[nPattern-1]=='*' ){ - char *zTranslit; - int res; - zTranslit = (char *)transliterate((unsigned char *)zWord, nWord); - if( !zTranslit ) return SQLITE_NOMEM; - res = editdist1(pCur->zPattern, zTranslit, &iMatchlen); - sqlite3_free(zTranslit); - if( res<0 ) return SQLITE_NOMEM; - iMatchlen = translen_to_charlen(zWord, nWord, iMatchlen); - }else{ - iMatchlen = utf8Charlen(zWord, nWord); - } - } - - sqlite3_result_int(ctx, iMatchlen); - break; - } - case SPELLFIX_COL_PHONEHASH: { - sqlite3_result_text(ctx, pCur->a[pCur->iRow].zHash, -1, SQLITE_STATIC); - break; - } - case SPELLFIX_COL_TOP: { - sqlite3_result_int(ctx, pCur->iTop); - break; - } - case SPELLFIX_COL_SCOPE: { - sqlite3_result_int(ctx, pCur->iScope); - break; - } - case SPELLFIX_COL_SRCHCNT: { - sqlite3_result_int(ctx, pCur->nSearch); - break; - } - default: { - sqlite3_result_null(ctx); - break; - } - } - return SQLITE_OK; -} - -/* -** The rowid. -*/ -static int spellfix1Rowid(sqlite3_vtab_cursor *cur, sqlite_int64 *pRowid){ - spellfix1_cursor *pCur = (spellfix1_cursor*)cur; - if( pCur->pFullScan ){ - *pRowid = sqlite3_column_int64(pCur->pFullScan, 4); - }else{ - *pRowid = pCur->a[pCur->iRow].iRowid; - } - return SQLITE_OK; -} - -/* -** The xUpdate() method. -*/ -static int spellfix1Update( - sqlite3_vtab *pVTab, - int argc, - sqlite3_value **argv, - sqlite_int64 *pRowid -){ - int rc = SQLITE_OK; - sqlite3_int64 rowid, newRowid; - spellfix1_vtab *p = (spellfix1_vtab*)pVTab; - sqlite3 *db = p->db; - - if( argc==1 ){ - /* A delete operation on the rowid given by argv[0] */ - rowid = *pRowid = sqlite3_value_int64(argv[0]); - spellfix1DbExec(&rc, db, "DELETE FROM \"%w\".\"%w_vocab\" " - " WHERE id=%lld", - p->zDbName, p->zTableName, rowid); - }else{ - const unsigned char *zWord = sqlite3_value_text(argv[SPELLFIX_COL_WORD+2]); - int nWord = sqlite3_value_bytes(argv[SPELLFIX_COL_WORD+2]); - int iLang = sqlite3_value_int(argv[SPELLFIX_COL_LANGID+2]); - int iRank = sqlite3_value_int(argv[SPELLFIX_COL_RANK+2]); - const unsigned char *zSoundslike = - sqlite3_value_text(argv[SPELLFIX_COL_SOUNDSLIKE+2]); - int nSoundslike = sqlite3_value_bytes(argv[SPELLFIX_COL_SOUNDSLIKE+2]); - char *zK1, *zK2; - int i; - char c; - - if( zWord==0 ){ - /* Inserts of the form: INSERT INTO table(command) VALUES('xyzzy'); - ** cause zWord to be NULL, so we look at the "command" column to see - ** what special actions to take */ - const char *zCmd = - (const char*)sqlite3_value_text(argv[SPELLFIX_COL_COMMAND+2]); - if( zCmd==0 ){ - pVTab->zErrMsg = sqlite3_mprintf("%s.word may not be NULL", - p->zTableName); - return SQLITE_CONSTRAINT_NOTNULL; - } - if( strcmp(zCmd,"reset")==0 ){ - /* Reset the edit cost table (if there is one). */ - editDist3ConfigDelete(p->pConfig3); - p->pConfig3 = 0; - return SQLITE_OK; - } - if( strncmp(zCmd,"edit_cost_table=",16)==0 ){ - editDist3ConfigDelete(p->pConfig3); - p->pConfig3 = 0; - sqlite3_free(p->zCostTable); - p->zCostTable = spellfix1Dequote(zCmd+16); - if( p->zCostTable==0 ) return SQLITE_NOMEM; - if( p->zCostTable[0]==0 || sqlite3_stricmp(p->zCostTable,"null")==0 ){ - sqlite3_free(p->zCostTable); - p->zCostTable = 0; - } - return SQLITE_OK; - } - pVTab->zErrMsg = sqlite3_mprintf("unknown value for %s.command: \"%w\"", - p->zTableName, zCmd); - return SQLITE_ERROR; - } - if( iRank<1 ) iRank = 1; - if( zSoundslike ){ - zK1 = (char*)transliterate(zSoundslike, nSoundslike); - }else{ - zK1 = (char*)transliterate(zWord, nWord); - } - if( zK1==0 ) return SQLITE_NOMEM; - for(i=0; (c = zK1[i])!=0; i++){ - if( c>='A' && c<='Z' ) zK1[i] += 'a' - 'A'; - } - zK2 = (char*)phoneticHash((const unsigned char*)zK1, i); - if( zK2==0 ){ - sqlite3_free(zK1); - return SQLITE_NOMEM; - } - if( sqlite3_value_type(argv[0])==SQLITE_NULL ){ - spellfix1DbExec(&rc, db, - "INSERT INTO \"%w\".\"%w_vocab\"(rank,langid,word,k1,k2) " - "VALUES(%d,%d,%Q,%Q,%Q)", - p->zDbName, p->zTableName, - iRank, iLang, zWord, zK1, zK2 - ); - *pRowid = sqlite3_last_insert_rowid(db); - }else{ - rowid = sqlite3_value_int64(argv[0]); - newRowid = *pRowid = sqlite3_value_int64(argv[1]); - spellfix1DbExec(&rc, db, - "UPDATE \"%w\".\"%w_vocab\" SET id=%lld, rank=%d, langid=%d," - " word=%Q, k1=%Q, k2=%Q WHERE id=%lld", - p->zDbName, p->zTableName, newRowid, iRank, iLang, - zWord, zK1, zK2, rowid - ); - } - sqlite3_free(zK1); - sqlite3_free(zK2); - } - return rc; -} - -/* -** Rename the spellfix1 table. -*/ -static int spellfix1Rename(sqlite3_vtab *pVTab, const char *zNew){ - spellfix1_vtab *p = (spellfix1_vtab*)pVTab; - sqlite3 *db = p->db; - int rc = SQLITE_OK; - char *zNewName = sqlite3_mprintf("%s", zNew); - if( zNewName==0 ){ - return SQLITE_NOMEM; - } - spellfix1DbExec(&rc, db, - "ALTER TABLE \"%w\".\"%w_vocab\" RENAME TO \"%w_vocab\"", - p->zDbName, p->zTableName, zNewName - ); - if( rc==SQLITE_OK ){ - sqlite3_free(p->zTableName); - p->zTableName = zNewName; - }else{ - sqlite3_free(zNewName); - } - return rc; -} - - -/* -** A virtual table module that provides fuzzy search. -*/ -static sqlite3_module spellfix1Module = { - 0, /* iVersion */ - spellfix1Create, /* xCreate - handle CREATE VIRTUAL TABLE */ - spellfix1Connect, /* xConnect - reconnected to an existing table */ - spellfix1BestIndex, /* xBestIndex - figure out how to do a query */ - spellfix1Disconnect, /* xDisconnect - close a connection */ - spellfix1Destroy, /* xDestroy - handle DROP TABLE */ - spellfix1Open, /* xOpen - open a cursor */ - spellfix1Close, /* xClose - close a cursor */ - spellfix1Filter, /* xFilter - configure scan constraints */ - spellfix1Next, /* xNext - advance a cursor */ - spellfix1Eof, /* xEof - check for end of scan */ - spellfix1Column, /* xColumn - read data */ - spellfix1Rowid, /* xRowid - read data */ - spellfix1Update, /* xUpdate */ - 0, /* xBegin */ - 0, /* xSync */ - 0, /* xCommit */ - 0, /* xRollback */ - 0, /* xFindMethod */ - spellfix1Rename, /* xRename */ -}; - -/* -** Register the various functions and the virtual table. -*/ -static int spellfix1Register(sqlite3 *db){ - int rc = SQLITE_OK; - int i; - rc = sqlite3_create_function(db, "spellfix1_translit", 1, SQLITE_UTF8, 0, - transliterateSqlFunc, 0, 0); - if( rc==SQLITE_OK ){ - rc = sqlite3_create_function(db, "spellfix1_editdist", 2, SQLITE_UTF8, 0, - editdistSqlFunc, 0, 0); - } - if( rc==SQLITE_OK ){ - rc = sqlite3_create_function(db, "spellfix1_phonehash", 1, SQLITE_UTF8, 0, - phoneticHashSqlFunc, 0, 0); - } - if( rc==SQLITE_OK ){ - rc = sqlite3_create_function(db, "spellfix1_scriptcode", 1, SQLITE_UTF8, 0, - scriptCodeSqlFunc, 0, 0); - } - if( rc==SQLITE_OK ){ - rc = sqlite3_create_module(db, "spellfix1", &spellfix1Module, 0); - } - if( rc==SQLITE_OK ){ - rc = editDist3Install(db); - } - - /* Verify sanity of the translit[] table */ - for(i=0; i -#include - -#ifndef SQLITE_OMIT_VIRTUALTABLE - - -/* A wholenumber cursor object */ -typedef struct wholenumber_cursor wholenumber_cursor; -struct wholenumber_cursor { - sqlite3_vtab_cursor base; /* Base class - must be first */ - sqlite3_int64 iValue; /* Current value */ - sqlite3_int64 mxValue; /* Maximum value */ -}; - -/* Methods for the wholenumber module */ -static int wholenumberConnect( - sqlite3 *db, - void *pAux, - int argc, const char *const*argv, - sqlite3_vtab **ppVtab, - char **pzErr -){ - sqlite3_vtab *pNew; - pNew = *ppVtab = sqlite3_malloc( sizeof(*pNew) ); - if( pNew==0 ) return SQLITE_NOMEM; - sqlite3_declare_vtab(db, "CREATE TABLE x(value)"); - memset(pNew, 0, sizeof(*pNew)); - return SQLITE_OK; -} -/* Note that for this virtual table, the xCreate and xConnect -** methods are identical. */ - -static int wholenumberDisconnect(sqlite3_vtab *pVtab){ - sqlite3_free(pVtab); - return SQLITE_OK; -} -/* The xDisconnect and xDestroy methods are also the same */ - - -/* -** Open a new wholenumber cursor. -*/ -static int wholenumberOpen(sqlite3_vtab *p, sqlite3_vtab_cursor **ppCursor){ - wholenumber_cursor *pCur; - pCur = sqlite3_malloc( sizeof(*pCur) ); - if( pCur==0 ) return SQLITE_NOMEM; - memset(pCur, 0, sizeof(*pCur)); - *ppCursor = &pCur->base; - return SQLITE_OK; -} - -/* -** Close a wholenumber cursor. -*/ -static int wholenumberClose(sqlite3_vtab_cursor *cur){ - sqlite3_free(cur); - return SQLITE_OK; -} - - -/* -** Advance a cursor to its next row of output -*/ -static int wholenumberNext(sqlite3_vtab_cursor *cur){ - wholenumber_cursor *pCur = (wholenumber_cursor*)cur; - pCur->iValue++; - return SQLITE_OK; -} - -/* -** Return the value associated with a wholenumber. -*/ -static int wholenumberColumn( - sqlite3_vtab_cursor *cur, - sqlite3_context *ctx, - int i -){ - wholenumber_cursor *pCur = (wholenumber_cursor*)cur; - sqlite3_result_int64(ctx, pCur->iValue); - return SQLITE_OK; -} - -/* -** The rowid. -*/ -static int wholenumberRowid(sqlite3_vtab_cursor *cur, sqlite_int64 *pRowid){ - wholenumber_cursor *pCur = (wholenumber_cursor*)cur; - *pRowid = pCur->iValue; - return SQLITE_OK; -} - -/* -** When the wholenumber_cursor.rLimit value is 0 or less, that is a signal -** that the cursor has nothing more to output. -*/ -static int wholenumberEof(sqlite3_vtab_cursor *cur){ - wholenumber_cursor *pCur = (wholenumber_cursor*)cur; - return pCur->iValue>pCur->mxValue || pCur->iValue==0; -} - -/* -** 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 wholenumberColumn, wholenumberRowid, or wholenumberEof call. -** -** idxNum Constraints -** ------ --------------------- -** 0 (none) -** 1 value > $argv0 -** 2 value >= $argv0 -** 4 value < $argv0 -** 8 value <= $argv0 -** -** 5 value > $argv0 AND value < $argv1 -** 6 value >= $argv0 AND value < $argv1 -** 9 value > $argv0 AND value <= $argv1 -** 10 value >= $argv0 AND value <= $argv1 -*/ -static int wholenumberFilter( - sqlite3_vtab_cursor *pVtabCursor, - int idxNum, const char *idxStr, - int argc, sqlite3_value **argv -){ - wholenumber_cursor *pCur = (wholenumber_cursor *)pVtabCursor; - sqlite3_int64 v; - int i = 0; - pCur->iValue = 1; - pCur->mxValue = 0xffffffff; /* 4294967295 */ - if( idxNum & 3 ){ - v = sqlite3_value_int64(argv[0]) + (idxNum&1); - if( v>pCur->iValue && v<=pCur->mxValue ) pCur->iValue = v; - i++; - } - if( idxNum & 12 ){ - v = sqlite3_value_int64(argv[i]) - ((idxNum>>2)&1); - if( v>=pCur->iValue && vmxValue ) pCur->mxValue = v; - } - return SQLITE_OK; -} - -/* -** Search for terms of these forms: -** -** (1) value > $value -** (2) value >= $value -** (4) value < $value -** (8) value <= $value -** -** idxNum is an ORed combination of 1 or 2 with 4 or 8. -*/ -static int wholenumberBestIndex( - sqlite3_vtab *tab, - sqlite3_index_info *pIdxInfo -){ - int i; - int idxNum = 0; - int argvIdx = 1; - int ltIdx = -1; - int gtIdx = -1; - const struct sqlite3_index_constraint *pConstraint; - pConstraint = pIdxInfo->aConstraint; - for(i=0; inConstraint; i++, pConstraint++){ - if( pConstraint->usable==0 ) continue; - if( (idxNum & 3)==0 && pConstraint->op==SQLITE_INDEX_CONSTRAINT_GT ){ - idxNum |= 1; - ltIdx = i; - } - if( (idxNum & 3)==0 && pConstraint->op==SQLITE_INDEX_CONSTRAINT_GE ){ - idxNum |= 2; - ltIdx = i; - } - if( (idxNum & 12)==0 && pConstraint->op==SQLITE_INDEX_CONSTRAINT_LT ){ - idxNum |= 4; - gtIdx = i; - } - if( (idxNum & 12)==0 && pConstraint->op==SQLITE_INDEX_CONSTRAINT_LE ){ - idxNum |= 8; - gtIdx = i; - } - } - pIdxInfo->idxNum = idxNum; - if( ltIdx>=0 ){ - pIdxInfo->aConstraintUsage[ltIdx].argvIndex = argvIdx++; - pIdxInfo->aConstraintUsage[ltIdx].omit = 1; - } - if( gtIdx>=0 ){ - pIdxInfo->aConstraintUsage[gtIdx].argvIndex = argvIdx; - pIdxInfo->aConstraintUsage[gtIdx].omit = 1; - } - if( pIdxInfo->nOrderBy==1 - && pIdxInfo->aOrderBy[0].desc==0 - ){ - pIdxInfo->orderByConsumed = 1; - } - pIdxInfo->estimatedCost = (double)1; - return SQLITE_OK; -} - -/* -** A virtual table module that provides read-only access to a -** Tcl global variable namespace. -*/ -static sqlite3_module wholenumberModule = { - 0, /* iVersion */ - wholenumberConnect, - wholenumberConnect, - wholenumberBestIndex, - wholenumberDisconnect, - wholenumberDisconnect, - wholenumberOpen, /* xOpen - open a cursor */ - wholenumberClose, /* xClose - close a cursor */ - wholenumberFilter, /* xFilter - configure scan constraints */ - wholenumberNext, /* xNext - advance a cursor */ - wholenumberEof, /* xEof - check for end of scan */ - wholenumberColumn, /* xColumn - read data */ - wholenumberRowid, /* xRowid - read data */ - 0, /* xUpdate */ - 0, /* xBegin */ - 0, /* xSync */ - 0, /* xCommit */ - 0, /* xRollback */ - 0, /* xFindMethod */ - 0, /* xRename */ -}; - -#endif /* SQLITE_OMIT_VIRTUALTABLE */ - - -/* -** Register the wholenumber virtual table -*/ -int wholenumber_register(sqlite3 *db){ - int rc = SQLITE_OK; -#ifndef SQLITE_OMIT_VIRTUALTABLE - rc = sqlite3_create_module(db, "wholenumber", &wholenumberModule, 0); -#endif - return rc; -} - -#ifdef SQLITE_TEST -#include -/* -** Decode a pointer to an sqlite3 object. -*/ -extern int getDbPointer(Tcl_Interp *interp, const char *zA, sqlite3 **ppDb); - -/* -** Register the echo virtual table module. -*/ -static int register_wholenumber_module( - ClientData clientData, /* Pointer to sqlite3_enable_XXX function */ - Tcl_Interp *interp, /* The TCL interpreter that invoked this command */ - int objc, /* Number of arguments */ - Tcl_Obj *CONST objv[] /* Command arguments */ -){ - sqlite3 *db; - if( objc!=2 ){ - Tcl_WrongNumArgs(interp, 1, objv, "DB"); - return TCL_ERROR; - } - if( getDbPointer(interp, Tcl_GetString(objv[1]), &db) ) return TCL_ERROR; - wholenumber_register(db); - return TCL_OK; -} - - -/* -** Register commands with the TCL interpreter. -*/ -int Sqlitetestwholenumber_Init(Tcl_Interp *interp){ - static struct { - char *zName; - Tcl_ObjCmdProc *xProc; - void *clientData; - } aObjCmd[] = { - { "register_wholenumber_module", register_wholenumber_module, 0 }, - }; - int i; - for(i=0; i1; + 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 Index: test/fuzzer1.test ================================================================== --- test/fuzzer1.test +++ test/fuzzer1.test @@ -22,16 +22,11 @@ return } set ::testprefix fuzzer1 -# Test of test code. Only here to make the coverage metric better. -do_test 0.1 { - list [catch { register_fuzzer_module a b c } msg] $msg -} {1 {wrong # args: should be "register_fuzzer_module DB"}} - -register_fuzzer_module db +load_static_extension db fuzzer # Check configuration errors. # do_catchsql_test fuzzer1-1.1 { CREATE VIRTUAL TABLE f USING fuzzer; Index: test/fuzzerfault.test ================================================================== --- test/fuzzerfault.test +++ test/fuzzerfault.test @@ -15,11 +15,11 @@ set testdir [file dirname $argv0] source $testdir/tester.tcl ifcapable !vtab { finish_test ; return } set ::testprefix fuzzerfault -register_fuzzer_module db +load_static_extension db fuzzer do_test 1-pre1 { execsql { CREATE TABLE x1_rules(ruleset, cFrom, cTo, cost); INSERT INTO x1_rules VALUES(0, 'a', 'b', 1); Index: test/memdb.test ================================================================== --- test/memdb.test +++ test/memdb.test @@ -363,11 +363,11 @@ } } {} ifcapable subquery&&vtab { do_test memdb-7.1 { - register_wholenumber_module db + load_static_extension db wholenumber execsql { CREATE TABLE t6(x); CREATE VIRTUAL TABLE nums USING wholenumber; INSERT INTO t6 SELECT value FROM nums WHERE value BETWEEN 1 AND 256; SELECT count(*) FROM (SELECT DISTINCT x FROM t6); Index: test/regexp1.test ================================================================== --- test/regexp1.test +++ test/regexp1.test @@ -14,11 +14,11 @@ set testdir [file dirname $argv0] source $testdir/tester.tcl do_test regexp1-1.1 { - sqlite3_add_regexp_func db + load_static_extension db regexp db eval { CREATE TABLE t1(x INTEGER PRIMARY KEY, y TEXT); INSERT INTO t1 VALUES(1, 'For since by man came death,'); INSERT INTO t1 VALUES(2, 'by man came also the resurrection of the dead.'); INSERT INTO t1 VALUES(3, 'For as in Adam all die,'); Index: test/spellfix.test ================================================================== --- test/spellfix.test +++ test/spellfix.test @@ -14,11 +14,11 @@ source $testdir/tester.tcl set testprefix spellfix ifcapable !vtab { finish_test ; return } -register_spellfix_module db +load_static_extension db spellfix set vocab { rabbi rabbit rabbits rabble rabid rabies raccoon raccoons race raced racer racers races racetrack racial racially racing rack racked racket racketeer racketeering racketeers rackets racking racks radar radars radial radially Index: test/tkt-2d1a5c67d.test ================================================================== --- test/tkt-2d1a5c67d.test +++ test/tkt-2d1a5c67d.test @@ -44,11 +44,11 @@ } db close forcedelete test.db test.db-wal sqlite3 db test.db -register_wholenumber_module db +load_static_extension db wholenumber db eval { PRAGMA journal_mode=WAL; CREATE TABLE t1(a,b); CREATE INDEX t1b ON t1(b); CREATE TABLE t2(x,y); Index: test/zerodamage.test ================================================================== --- test/zerodamage.test +++ test/zerodamage.test @@ -57,11 +57,11 @@ set sz [file size $file] if {$sz>$::max_journal_size} {set ::max_journal_size $sz} } tv filter xDelete tv script xDeleteCallback - register_wholenumber_module db + load_static_extension db wholenumber db eval { PRAGMA page_size=1024; PRAGMA journal_mode=DELETE; PRAGMA cache_size=5; CREATE VIRTUAL TABLE nums USING wholenumber;