0
0
mirror of https://github.com/tursodatabase/libsql.git synced 2024-11-27 06:58:59 +00:00
libsql/libsql-sqlite3/ext/fts3/fts3_expr.c
2023-10-16 13:58:16 +02:00

1294 lines
40 KiB
C

/*
** 2008 Nov 28
**
** 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 contains code that implements a parser for fts3 query strings
** (the right-hand argument to the MATCH operator). Because the supported
** syntax is relatively simple, the whole tokenizer/parser system is
** hand-coded.
*/
#include "fts3Int.h"
#if !defined(SQLITE_CORE) || defined(SQLITE_ENABLE_FTS3)
/*
** By default, this module parses the legacy syntax that has been
** traditionally used by fts3. Or, if SQLITE_ENABLE_FTS3_PARENTHESIS
** is defined, then it uses the new syntax. The differences between
** the new and the old syntaxes are:
**
** a) The new syntax supports parenthesis. The old does not.
**
** b) The new syntax supports the AND and NOT operators. The old does not.
**
** c) The old syntax supports the "-" token qualifier. This is not
** supported by the new syntax (it is replaced by the NOT operator).
**
** d) When using the old syntax, the OR operator has a greater precedence
** than an implicit AND. When using the new, both implicity and explicit
** AND operators have a higher precedence than OR.
**
** If compiled with SQLITE_TEST defined, then this module exports the
** symbol "int sqlite3_fts3_enable_parentheses". Setting this variable
** to zero causes the module to use the old syntax. If it is set to
** non-zero the new syntax is activated. This is so both syntaxes can
** be tested using a single build of testfixture.
**
** The following describes the syntax supported by the fts3 MATCH
** operator in a similar format to that used by the lemon parser
** generator. This module does not use actually lemon, it uses a
** custom parser.
**
** query ::= andexpr (OR andexpr)*.
**
** andexpr ::= notexpr (AND? notexpr)*.
**
** notexpr ::= nearexpr (NOT nearexpr|-TOKEN)*.
** notexpr ::= LP query RP.
**
** nearexpr ::= phrase (NEAR distance_opt nearexpr)*.
**
** distance_opt ::= .
** distance_opt ::= / INTEGER.
**
** phrase ::= TOKEN.
** phrase ::= COLUMN:TOKEN.
** phrase ::= "TOKEN TOKEN TOKEN...".
*/
#ifdef SQLITE_TEST
int sqlite3_fts3_enable_parentheses = 0;
#else
# ifdef SQLITE_ENABLE_FTS3_PARENTHESIS
# define sqlite3_fts3_enable_parentheses 1
# else
# define sqlite3_fts3_enable_parentheses 0
# endif
#endif
/*
** Default span for NEAR operators.
*/
#define SQLITE_FTS3_DEFAULT_NEAR_PARAM 10
#include <string.h>
#include <assert.h>
/*
** isNot:
** This variable is used by function getNextNode(). When getNextNode() is
** called, it sets ParseContext.isNot to true if the 'next node' is a
** FTSQUERY_PHRASE with a unary "-" attached to it. i.e. "mysql" in the
** FTS3 query "sqlite -mysql". Otherwise, ParseContext.isNot is set to
** zero.
*/
typedef struct ParseContext ParseContext;
struct ParseContext {
sqlite3_tokenizer *pTokenizer; /* Tokenizer module */
int iLangid; /* Language id used with tokenizer */
const char **azCol; /* Array of column names for fts3 table */
int bFts4; /* True to allow FTS4-only syntax */
int nCol; /* Number of entries in azCol[] */
int iDefaultCol; /* Default column to query */
int isNot; /* True if getNextNode() sees a unary - */
sqlite3_context *pCtx; /* Write error message here */
int nNest; /* Number of nested brackets */
};
/*
** This function is equivalent to the standard isspace() function.
**
** The standard isspace() can be awkward to use safely, because although it
** is defined to accept an argument of type int, its behavior when passed
** an integer that falls outside of the range of the unsigned char type
** is undefined (and sometimes, "undefined" means segfault). This wrapper
** is defined to accept an argument of type char, and always returns 0 for
** any values that fall outside of the range of the unsigned char type (i.e.
** negative values).
*/
static int fts3isspace(char c){
return c==' ' || c=='\t' || c=='\n' || c=='\r' || c=='\v' || c=='\f';
}
/*
** Allocate nByte bytes of memory using sqlite3_malloc(). If successful,
** zero the memory before returning a pointer to it. If unsuccessful,
** return NULL.
*/
void *sqlite3Fts3MallocZero(sqlite3_int64 nByte){
void *pRet = sqlite3_malloc64(nByte);
if( pRet ) memset(pRet, 0, nByte);
return pRet;
}
int sqlite3Fts3OpenTokenizer(
sqlite3_tokenizer *pTokenizer,
int iLangid,
const char *z,
int n,
sqlite3_tokenizer_cursor **ppCsr
){
sqlite3_tokenizer_module const *pModule = pTokenizer->pModule;
sqlite3_tokenizer_cursor *pCsr = 0;
int rc;
rc = pModule->xOpen(pTokenizer, z, n, &pCsr);
assert( rc==SQLITE_OK || pCsr==0 );
if( rc==SQLITE_OK ){
pCsr->pTokenizer = pTokenizer;
if( pModule->iVersion>=1 ){
rc = pModule->xLanguageid(pCsr, iLangid);
if( rc!=SQLITE_OK ){
pModule->xClose(pCsr);
pCsr = 0;
}
}
}
*ppCsr = pCsr;
return rc;
}
/*
** Function getNextNode(), which is called by fts3ExprParse(), may itself
** call fts3ExprParse(). So this forward declaration is required.
*/
static int fts3ExprParse(ParseContext *, const char *, int, Fts3Expr **, int *);
/*
** Extract the next token from buffer z (length n) using the tokenizer
** and other information (column names etc.) in pParse. Create an Fts3Expr
** structure of type FTSQUERY_PHRASE containing a phrase consisting of this
** single token and set *ppExpr to point to it. If the end of the buffer is
** reached before a token is found, set *ppExpr to zero. It is the
** responsibility of the caller to eventually deallocate the allocated
** Fts3Expr structure (if any) by passing it to sqlite3_free().
**
** Return SQLITE_OK if successful, or SQLITE_NOMEM if a memory allocation
** fails.
*/
static int getNextToken(
ParseContext *pParse, /* fts3 query parse context */
int iCol, /* Value for Fts3Phrase.iColumn */
const char *z, int n, /* Input string */
Fts3Expr **ppExpr, /* OUT: expression */
int *pnConsumed /* OUT: Number of bytes consumed */
){
sqlite3_tokenizer *pTokenizer = pParse->pTokenizer;
sqlite3_tokenizer_module const *pModule = pTokenizer->pModule;
int rc;
sqlite3_tokenizer_cursor *pCursor;
Fts3Expr *pRet = 0;
int i = 0;
/* Set variable i to the maximum number of bytes of input to tokenize. */
for(i=0; i<n; i++){
if( sqlite3_fts3_enable_parentheses && (z[i]=='(' || z[i]==')') ) break;
if( z[i]=='"' ) break;
}
*pnConsumed = i;
rc = sqlite3Fts3OpenTokenizer(pTokenizer, pParse->iLangid, z, i, &pCursor);
if( rc==SQLITE_OK ){
const char *zToken;
int nToken = 0, iStart = 0, iEnd = 0, iPosition = 0;
sqlite3_int64 nByte; /* total space to allocate */
rc = pModule->xNext(pCursor, &zToken, &nToken, &iStart, &iEnd, &iPosition);
if( rc==SQLITE_OK ){
nByte = sizeof(Fts3Expr) + sizeof(Fts3Phrase) + nToken;
pRet = (Fts3Expr *)sqlite3Fts3MallocZero(nByte);
if( !pRet ){
rc = SQLITE_NOMEM;
}else{
pRet->eType = FTSQUERY_PHRASE;
pRet->pPhrase = (Fts3Phrase *)&pRet[1];
pRet->pPhrase->nToken = 1;
pRet->pPhrase->iColumn = iCol;
pRet->pPhrase->aToken[0].n = nToken;
pRet->pPhrase->aToken[0].z = (char *)&pRet->pPhrase[1];
memcpy(pRet->pPhrase->aToken[0].z, zToken, nToken);
if( iEnd<n && z[iEnd]=='*' ){
pRet->pPhrase->aToken[0].isPrefix = 1;
iEnd++;
}
while( 1 ){
if( !sqlite3_fts3_enable_parentheses
&& iStart>0 && z[iStart-1]=='-'
){
pParse->isNot = 1;
iStart--;
}else if( pParse->bFts4 && iStart>0 && z[iStart-1]=='^' ){
pRet->pPhrase->aToken[0].bFirst = 1;
iStart--;
}else{
break;
}
}
}
*pnConsumed = iEnd;
}else if( i && rc==SQLITE_DONE ){
rc = SQLITE_OK;
}
pModule->xClose(pCursor);
}
*ppExpr = pRet;
return rc;
}
/*
** Enlarge a memory allocation. If an out-of-memory allocation occurs,
** then free the old allocation.
*/
static void *fts3ReallocOrFree(void *pOrig, sqlite3_int64 nNew){
void *pRet = sqlite3_realloc64(pOrig, nNew);
if( !pRet ){
sqlite3_free(pOrig);
}
return pRet;
}
/*
** Buffer zInput, length nInput, contains the contents of a quoted string
** that appeared as part of an fts3 query expression. Neither quote character
** is included in the buffer. This function attempts to tokenize the entire
** input buffer and create an Fts3Expr structure of type FTSQUERY_PHRASE
** containing the results.
**
** If successful, SQLITE_OK is returned and *ppExpr set to point at the
** allocated Fts3Expr structure. Otherwise, either SQLITE_NOMEM (out of memory
** error) or SQLITE_ERROR (tokenization error) is returned and *ppExpr set
** to 0.
*/
static int getNextString(
ParseContext *pParse, /* fts3 query parse context */
const char *zInput, int nInput, /* Input string */
Fts3Expr **ppExpr /* OUT: expression */
){
sqlite3_tokenizer *pTokenizer = pParse->pTokenizer;
sqlite3_tokenizer_module const *pModule = pTokenizer->pModule;
int rc;
Fts3Expr *p = 0;
sqlite3_tokenizer_cursor *pCursor = 0;
char *zTemp = 0;
int nTemp = 0;
const int nSpace = sizeof(Fts3Expr) + sizeof(Fts3Phrase);
int nToken = 0;
/* The final Fts3Expr data structure, including the Fts3Phrase,
** Fts3PhraseToken structures token buffers are all stored as a single
** allocation so that the expression can be freed with a single call to
** sqlite3_free(). Setting this up requires a two pass approach.
**
** The first pass, in the block below, uses a tokenizer cursor to iterate
** through the tokens in the expression. This pass uses fts3ReallocOrFree()
** to assemble data in two dynamic buffers:
**
** Buffer p: Points to the Fts3Expr structure, followed by the Fts3Phrase
** structure, followed by the array of Fts3PhraseToken
** structures. This pass only populates the Fts3PhraseToken array.
**
** Buffer zTemp: Contains copies of all tokens.
**
** The second pass, in the block that begins "if( rc==SQLITE_DONE )" below,
** appends buffer zTemp to buffer p, and fills in the Fts3Expr and Fts3Phrase
** structures.
*/
rc = sqlite3Fts3OpenTokenizer(
pTokenizer, pParse->iLangid, zInput, nInput, &pCursor);
if( rc==SQLITE_OK ){
int ii;
for(ii=0; rc==SQLITE_OK; ii++){
const char *zByte;
int nByte = 0, iBegin = 0, iEnd = 0, iPos = 0;
rc = pModule->xNext(pCursor, &zByte, &nByte, &iBegin, &iEnd, &iPos);
if( rc==SQLITE_OK ){
Fts3PhraseToken *pToken;
p = fts3ReallocOrFree(p, nSpace + ii*sizeof(Fts3PhraseToken));
if( !p ) goto no_mem;
zTemp = fts3ReallocOrFree(zTemp, nTemp + nByte);
if( !zTemp ) goto no_mem;
assert( nToken==ii );
pToken = &((Fts3Phrase *)(&p[1]))->aToken[ii];
memset(pToken, 0, sizeof(Fts3PhraseToken));
memcpy(&zTemp[nTemp], zByte, nByte);
nTemp += nByte;
pToken->n = nByte;
pToken->isPrefix = (iEnd<nInput && zInput[iEnd]=='*');
pToken->bFirst = (iBegin>0 && zInput[iBegin-1]=='^');
nToken = ii+1;
}
}
pModule->xClose(pCursor);
pCursor = 0;
}
if( rc==SQLITE_DONE ){
int jj;
char *zBuf = 0;
p = fts3ReallocOrFree(p, nSpace + nToken*sizeof(Fts3PhraseToken) + nTemp);
if( !p ) goto no_mem;
memset(p, 0, (char *)&(((Fts3Phrase *)&p[1])->aToken[0])-(char *)p);
p->eType = FTSQUERY_PHRASE;
p->pPhrase = (Fts3Phrase *)&p[1];
p->pPhrase->iColumn = pParse->iDefaultCol;
p->pPhrase->nToken = nToken;
zBuf = (char *)&p->pPhrase->aToken[nToken];
if( zTemp ){
memcpy(zBuf, zTemp, nTemp);
sqlite3_free(zTemp);
}else{
assert( nTemp==0 );
}
for(jj=0; jj<p->pPhrase->nToken; jj++){
p->pPhrase->aToken[jj].z = zBuf;
zBuf += p->pPhrase->aToken[jj].n;
}
rc = SQLITE_OK;
}
*ppExpr = p;
return rc;
no_mem:
if( pCursor ){
pModule->xClose(pCursor);
}
sqlite3_free(zTemp);
sqlite3_free(p);
*ppExpr = 0;
return SQLITE_NOMEM;
}
/*
** The output variable *ppExpr is populated with an allocated Fts3Expr
** structure, or set to 0 if the end of the input buffer is reached.
**
** Returns an SQLite error code. SQLITE_OK if everything works, SQLITE_NOMEM
** if a malloc failure occurs, or SQLITE_ERROR if a parse error is encountered.
** If SQLITE_ERROR is returned, pContext is populated with an error message.
*/
static int getNextNode(
ParseContext *pParse, /* fts3 query parse context */
const char *z, int n, /* Input string */
Fts3Expr **ppExpr, /* OUT: expression */
int *pnConsumed /* OUT: Number of bytes consumed */
){
static const struct Fts3Keyword {
char *z; /* Keyword text */
unsigned char n; /* Length of the keyword */
unsigned char parenOnly; /* Only valid in paren mode */
unsigned char eType; /* Keyword code */
} aKeyword[] = {
{ "OR" , 2, 0, FTSQUERY_OR },
{ "AND", 3, 1, FTSQUERY_AND },
{ "NOT", 3, 1, FTSQUERY_NOT },
{ "NEAR", 4, 0, FTSQUERY_NEAR }
};
int ii;
int iCol;
int iColLen;
int rc;
Fts3Expr *pRet = 0;
const char *zInput = z;
int nInput = n;
pParse->isNot = 0;
/* Skip over any whitespace before checking for a keyword, an open or
** close bracket, or a quoted string.
*/
while( nInput>0 && fts3isspace(*zInput) ){
nInput--;
zInput++;
}
if( nInput==0 ){
return SQLITE_DONE;
}
/* See if we are dealing with a keyword. */
for(ii=0; ii<(int)(sizeof(aKeyword)/sizeof(struct Fts3Keyword)); ii++){
const struct Fts3Keyword *pKey = &aKeyword[ii];
if( (pKey->parenOnly & ~sqlite3_fts3_enable_parentheses)!=0 ){
continue;
}
if( nInput>=pKey->n && 0==memcmp(zInput, pKey->z, pKey->n) ){
int nNear = SQLITE_FTS3_DEFAULT_NEAR_PARAM;
int nKey = pKey->n;
char cNext;
/* If this is a "NEAR" keyword, check for an explicit nearness. */
if( pKey->eType==FTSQUERY_NEAR ){
assert( nKey==4 );
if( zInput[4]=='/' && zInput[5]>='0' && zInput[5]<='9' ){
nKey += 1+sqlite3Fts3ReadInt(&zInput[nKey+1], &nNear);
}
}
/* At this point this is probably a keyword. But for that to be true,
** the next byte must contain either whitespace, an open or close
** parenthesis, a quote character, or EOF.
*/
cNext = zInput[nKey];
if( fts3isspace(cNext)
|| cNext=='"' || cNext=='(' || cNext==')' || cNext==0
){
pRet = (Fts3Expr *)sqlite3Fts3MallocZero(sizeof(Fts3Expr));
if( !pRet ){
return SQLITE_NOMEM;
}
pRet->eType = pKey->eType;
pRet->nNear = nNear;
*ppExpr = pRet;
*pnConsumed = (int)((zInput - z) + nKey);
return SQLITE_OK;
}
/* Turns out that wasn't a keyword after all. This happens if the
** user has supplied a token such as "ORacle". Continue.
*/
}
}
/* See if we are dealing with a quoted phrase. If this is the case, then
** search for the closing quote and pass the whole string to getNextString()
** for processing. This is easy to do, as fts3 has no syntax for escaping
** a quote character embedded in a string.
*/
if( *zInput=='"' ){
for(ii=1; ii<nInput && zInput[ii]!='"'; ii++);
*pnConsumed = (int)((zInput - z) + ii + 1);
if( ii==nInput ){
return SQLITE_ERROR;
}
return getNextString(pParse, &zInput[1], ii-1, ppExpr);
}
if( sqlite3_fts3_enable_parentheses ){
if( *zInput=='(' ){
int nConsumed = 0;
pParse->nNest++;
#if !defined(SQLITE_MAX_EXPR_DEPTH)
if( pParse->nNest>1000 ) return SQLITE_ERROR;
#elif SQLITE_MAX_EXPR_DEPTH>0
if( pParse->nNest>SQLITE_MAX_EXPR_DEPTH ) return SQLITE_ERROR;
#endif
rc = fts3ExprParse(pParse, zInput+1, nInput-1, ppExpr, &nConsumed);
*pnConsumed = (int)(zInput - z) + 1 + nConsumed;
return rc;
}else if( *zInput==')' ){
pParse->nNest--;
*pnConsumed = (int)((zInput - z) + 1);
*ppExpr = 0;
return SQLITE_DONE;
}
}
/* If control flows to this point, this must be a regular token, or
** the end of the input. Read a regular token using the sqlite3_tokenizer
** interface. Before doing so, figure out if there is an explicit
** column specifier for the token.
**
** TODO: Strangely, it is not possible to associate a column specifier
** with a quoted phrase, only with a single token. Not sure if this was
** an implementation artifact or an intentional decision when fts3 was
** first implemented. Whichever it was, this module duplicates the
** limitation.
*/
iCol = pParse->iDefaultCol;
iColLen = 0;
for(ii=0; ii<pParse->nCol; ii++){
const char *zStr = pParse->azCol[ii];
int nStr = (int)strlen(zStr);
if( nInput>nStr && zInput[nStr]==':'
&& sqlite3_strnicmp(zStr, zInput, nStr)==0
){
iCol = ii;
iColLen = (int)((zInput - z) + nStr + 1);
break;
}
}
rc = getNextToken(pParse, iCol, &z[iColLen], n-iColLen, ppExpr, pnConsumed);
*pnConsumed += iColLen;
return rc;
}
/*
** The argument is an Fts3Expr structure for a binary operator (any type
** except an FTSQUERY_PHRASE). Return an integer value representing the
** precedence of the operator. Lower values have a higher precedence (i.e.
** group more tightly). For example, in the C language, the == operator
** groups more tightly than ||, and would therefore have a higher precedence.
**
** When using the new fts3 query syntax (when SQLITE_ENABLE_FTS3_PARENTHESIS
** is defined), the order of the operators in precedence from highest to
** lowest is:
**
** NEAR
** NOT
** AND (including implicit ANDs)
** OR
**
** Note that when using the old query syntax, the OR operator has a higher
** precedence than the AND operator.
*/
static int opPrecedence(Fts3Expr *p){
assert( p->eType!=FTSQUERY_PHRASE );
if( sqlite3_fts3_enable_parentheses ){
return p->eType;
}else if( p->eType==FTSQUERY_NEAR ){
return 1;
}else if( p->eType==FTSQUERY_OR ){
return 2;
}
assert( p->eType==FTSQUERY_AND );
return 3;
}
/*
** Argument ppHead contains a pointer to the current head of a query
** expression tree being parsed. pPrev is the expression node most recently
** inserted into the tree. This function adds pNew, which is always a binary
** operator node, into the expression tree based on the relative precedence
** of pNew and the existing nodes of the tree. This may result in the head
** of the tree changing, in which case *ppHead is set to the new root node.
*/
static void insertBinaryOperator(
Fts3Expr **ppHead, /* Pointer to the root node of a tree */
Fts3Expr *pPrev, /* Node most recently inserted into the tree */
Fts3Expr *pNew /* New binary node to insert into expression tree */
){
Fts3Expr *pSplit = pPrev;
while( pSplit->pParent && opPrecedence(pSplit->pParent)<=opPrecedence(pNew) ){
pSplit = pSplit->pParent;
}
if( pSplit->pParent ){
assert( pSplit->pParent->pRight==pSplit );
pSplit->pParent->pRight = pNew;
pNew->pParent = pSplit->pParent;
}else{
*ppHead = pNew;
}
pNew->pLeft = pSplit;
pSplit->pParent = pNew;
}
/*
** Parse the fts3 query expression found in buffer z, length n. This function
** returns either when the end of the buffer is reached or an unmatched
** closing bracket - ')' - is encountered.
**
** If successful, SQLITE_OK is returned, *ppExpr is set to point to the
** parsed form of the expression and *pnConsumed is set to the number of
** bytes read from buffer z. Otherwise, *ppExpr is set to 0 and SQLITE_NOMEM
** (out of memory error) or SQLITE_ERROR (parse error) is returned.
*/
static int fts3ExprParse(
ParseContext *pParse, /* fts3 query parse context */
const char *z, int n, /* Text of MATCH query */
Fts3Expr **ppExpr, /* OUT: Parsed query structure */
int *pnConsumed /* OUT: Number of bytes consumed */
){
Fts3Expr *pRet = 0;
Fts3Expr *pPrev = 0;
Fts3Expr *pNotBranch = 0; /* Only used in legacy parse mode */
int nIn = n;
const char *zIn = z;
int rc = SQLITE_OK;
int isRequirePhrase = 1;
while( rc==SQLITE_OK ){
Fts3Expr *p = 0;
int nByte = 0;
rc = getNextNode(pParse, zIn, nIn, &p, &nByte);
assert( nByte>0 || (rc!=SQLITE_OK && p==0) );
if( rc==SQLITE_OK ){
if( p ){
int isPhrase;
if( !sqlite3_fts3_enable_parentheses
&& p->eType==FTSQUERY_PHRASE && pParse->isNot
){
/* Create an implicit NOT operator. */
Fts3Expr *pNot = sqlite3Fts3MallocZero(sizeof(Fts3Expr));
if( !pNot ){
sqlite3Fts3ExprFree(p);
rc = SQLITE_NOMEM;
goto exprparse_out;
}
pNot->eType = FTSQUERY_NOT;
pNot->pRight = p;
p->pParent = pNot;
if( pNotBranch ){
pNot->pLeft = pNotBranch;
pNotBranch->pParent = pNot;
}
pNotBranch = pNot;
p = pPrev;
}else{
int eType = p->eType;
isPhrase = (eType==FTSQUERY_PHRASE || p->pLeft);
/* The isRequirePhrase variable is set to true if a phrase or
** an expression contained in parenthesis is required. If a
** binary operator (AND, OR, NOT or NEAR) is encounted when
** isRequirePhrase is set, this is a syntax error.
*/
if( !isPhrase && isRequirePhrase ){
sqlite3Fts3ExprFree(p);
rc = SQLITE_ERROR;
goto exprparse_out;
}
if( isPhrase && !isRequirePhrase ){
/* Insert an implicit AND operator. */
Fts3Expr *pAnd;
assert( pRet && pPrev );
pAnd = sqlite3Fts3MallocZero(sizeof(Fts3Expr));
if( !pAnd ){
sqlite3Fts3ExprFree(p);
rc = SQLITE_NOMEM;
goto exprparse_out;
}
pAnd->eType = FTSQUERY_AND;
insertBinaryOperator(&pRet, pPrev, pAnd);
pPrev = pAnd;
}
/* This test catches attempts to make either operand of a NEAR
** operator something other than a phrase. For example, either of
** the following:
**
** (bracketed expression) NEAR phrase
** phrase NEAR (bracketed expression)
**
** Return an error in either case.
*/
if( pPrev && (
(eType==FTSQUERY_NEAR && !isPhrase && pPrev->eType!=FTSQUERY_PHRASE)
|| (eType!=FTSQUERY_PHRASE && isPhrase && pPrev->eType==FTSQUERY_NEAR)
)){
sqlite3Fts3ExprFree(p);
rc = SQLITE_ERROR;
goto exprparse_out;
}
if( isPhrase ){
if( pRet ){
assert( pPrev && pPrev->pLeft && pPrev->pRight==0 );
pPrev->pRight = p;
p->pParent = pPrev;
}else{
pRet = p;
}
}else{
insertBinaryOperator(&pRet, pPrev, p);
}
isRequirePhrase = !isPhrase;
}
pPrev = p;
}
assert( nByte>0 );
}
assert( rc!=SQLITE_OK || (nByte>0 && nByte<=nIn) );
nIn -= nByte;
zIn += nByte;
}
if( rc==SQLITE_DONE && pRet && isRequirePhrase ){
rc = SQLITE_ERROR;
}
if( rc==SQLITE_DONE ){
rc = SQLITE_OK;
if( !sqlite3_fts3_enable_parentheses && pNotBranch ){
if( !pRet ){
rc = SQLITE_ERROR;
}else{
Fts3Expr *pIter = pNotBranch;
while( pIter->pLeft ){
pIter = pIter->pLeft;
}
pIter->pLeft = pRet;
pRet->pParent = pIter;
pRet = pNotBranch;
}
}
}
*pnConsumed = n - nIn;
exprparse_out:
if( rc!=SQLITE_OK ){
sqlite3Fts3ExprFree(pRet);
sqlite3Fts3ExprFree(pNotBranch);
pRet = 0;
}
*ppExpr = pRet;
return rc;
}
/*
** Return SQLITE_ERROR if the maximum depth of the expression tree passed
** as the only argument is more than nMaxDepth.
*/
static int fts3ExprCheckDepth(Fts3Expr *p, int nMaxDepth){
int rc = SQLITE_OK;
if( p ){
if( nMaxDepth<0 ){
rc = SQLITE_TOOBIG;
}else{
rc = fts3ExprCheckDepth(p->pLeft, nMaxDepth-1);
if( rc==SQLITE_OK ){
rc = fts3ExprCheckDepth(p->pRight, nMaxDepth-1);
}
}
}
return rc;
}
/*
** This function attempts to transform the expression tree at (*pp) to
** an equivalent but more balanced form. The tree is modified in place.
** If successful, SQLITE_OK is returned and (*pp) set to point to the
** new root expression node.
**
** nMaxDepth is the maximum allowable depth of the balanced sub-tree.
**
** Otherwise, if an error occurs, an SQLite error code is returned and
** expression (*pp) freed.
*/
static int fts3ExprBalance(Fts3Expr **pp, int nMaxDepth){
int rc = SQLITE_OK; /* Return code */
Fts3Expr *pRoot = *pp; /* Initial root node */
Fts3Expr *pFree = 0; /* List of free nodes. Linked by pParent. */
int eType = pRoot->eType; /* Type of node in this tree */
if( nMaxDepth==0 ){
rc = SQLITE_ERROR;
}
if( rc==SQLITE_OK ){
if( (eType==FTSQUERY_AND || eType==FTSQUERY_OR) ){
Fts3Expr **apLeaf;
apLeaf = (Fts3Expr **)sqlite3_malloc64(sizeof(Fts3Expr *) * nMaxDepth);
if( 0==apLeaf ){
rc = SQLITE_NOMEM;
}else{
memset(apLeaf, 0, sizeof(Fts3Expr *) * nMaxDepth);
}
if( rc==SQLITE_OK ){
int i;
Fts3Expr *p;
/* Set $p to point to the left-most leaf in the tree of eType nodes. */
for(p=pRoot; p->eType==eType; p=p->pLeft){
assert( p->pParent==0 || p->pParent->pLeft==p );
assert( p->pLeft && p->pRight );
}
/* This loop runs once for each leaf in the tree of eType nodes. */
while( 1 ){
int iLvl;
Fts3Expr *pParent = p->pParent; /* Current parent of p */
assert( pParent==0 || pParent->pLeft==p );
p->pParent = 0;
if( pParent ){
pParent->pLeft = 0;
}else{
pRoot = 0;
}
rc = fts3ExprBalance(&p, nMaxDepth-1);
if( rc!=SQLITE_OK ) break;
for(iLvl=0; p && iLvl<nMaxDepth; iLvl++){
if( apLeaf[iLvl]==0 ){
apLeaf[iLvl] = p;
p = 0;
}else{
assert( pFree );
pFree->pLeft = apLeaf[iLvl];
pFree->pRight = p;
pFree->pLeft->pParent = pFree;
pFree->pRight->pParent = pFree;
p = pFree;
pFree = pFree->pParent;
p->pParent = 0;
apLeaf[iLvl] = 0;
}
}
if( p ){
sqlite3Fts3ExprFree(p);
rc = SQLITE_TOOBIG;
break;
}
/* If that was the last leaf node, break out of the loop */
if( pParent==0 ) break;
/* Set $p to point to the next leaf in the tree of eType nodes */
for(p=pParent->pRight; p->eType==eType; p=p->pLeft);
/* Remove pParent from the original tree. */
assert( pParent->pParent==0 || pParent->pParent->pLeft==pParent );
pParent->pRight->pParent = pParent->pParent;
if( pParent->pParent ){
pParent->pParent->pLeft = pParent->pRight;
}else{
assert( pParent==pRoot );
pRoot = pParent->pRight;
}
/* Link pParent into the free node list. It will be used as an
** internal node of the new tree. */
pParent->pParent = pFree;
pFree = pParent;
}
if( rc==SQLITE_OK ){
p = 0;
for(i=0; i<nMaxDepth; i++){
if( apLeaf[i] ){
if( p==0 ){
p = apLeaf[i];
p->pParent = 0;
}else{
assert( pFree!=0 );
pFree->pRight = p;
pFree->pLeft = apLeaf[i];
pFree->pLeft->pParent = pFree;
pFree->pRight->pParent = pFree;
p = pFree;
pFree = pFree->pParent;
p->pParent = 0;
}
}
}
pRoot = p;
}else{
/* An error occurred. Delete the contents of the apLeaf[] array
** and pFree list. Everything else is cleaned up by the call to
** sqlite3Fts3ExprFree(pRoot) below. */
Fts3Expr *pDel;
for(i=0; i<nMaxDepth; i++){
sqlite3Fts3ExprFree(apLeaf[i]);
}
while( (pDel=pFree)!=0 ){
pFree = pDel->pParent;
sqlite3_free(pDel);
}
}
assert( pFree==0 );
sqlite3_free( apLeaf );
}
}else if( eType==FTSQUERY_NOT ){
Fts3Expr *pLeft = pRoot->pLeft;
Fts3Expr *pRight = pRoot->pRight;
pRoot->pLeft = 0;
pRoot->pRight = 0;
pLeft->pParent = 0;
pRight->pParent = 0;
rc = fts3ExprBalance(&pLeft, nMaxDepth-1);
if( rc==SQLITE_OK ){
rc = fts3ExprBalance(&pRight, nMaxDepth-1);
}
if( rc!=SQLITE_OK ){
sqlite3Fts3ExprFree(pRight);
sqlite3Fts3ExprFree(pLeft);
}else{
assert( pLeft && pRight );
pRoot->pLeft = pLeft;
pLeft->pParent = pRoot;
pRoot->pRight = pRight;
pRight->pParent = pRoot;
}
}
}
if( rc!=SQLITE_OK ){
sqlite3Fts3ExprFree(pRoot);
pRoot = 0;
}
*pp = pRoot;
return rc;
}
/*
** This function is similar to sqlite3Fts3ExprParse(), with the following
** differences:
**
** 1. It does not do expression rebalancing.
** 2. It does not check that the expression does not exceed the
** maximum allowable depth.
** 3. Even if it fails, *ppExpr may still be set to point to an
** expression tree. It should be deleted using sqlite3Fts3ExprFree()
** in this case.
*/
static int fts3ExprParseUnbalanced(
sqlite3_tokenizer *pTokenizer, /* Tokenizer module */
int iLangid, /* Language id for tokenizer */
char **azCol, /* Array of column names for fts3 table */
int bFts4, /* True to allow FTS4-only syntax */
int nCol, /* Number of entries in azCol[] */
int iDefaultCol, /* Default column to query */
const char *z, int n, /* Text of MATCH query */
Fts3Expr **ppExpr /* OUT: Parsed query structure */
){
int nParsed;
int rc;
ParseContext sParse;
memset(&sParse, 0, sizeof(ParseContext));
sParse.pTokenizer = pTokenizer;
sParse.iLangid = iLangid;
sParse.azCol = (const char **)azCol;
sParse.nCol = nCol;
sParse.iDefaultCol = iDefaultCol;
sParse.bFts4 = bFts4;
if( z==0 ){
*ppExpr = 0;
return SQLITE_OK;
}
if( n<0 ){
n = (int)strlen(z);
}
rc = fts3ExprParse(&sParse, z, n, ppExpr, &nParsed);
assert( rc==SQLITE_OK || *ppExpr==0 );
/* Check for mismatched parenthesis */
if( rc==SQLITE_OK && sParse.nNest ){
rc = SQLITE_ERROR;
}
return rc;
}
/*
** Parameters z and n contain a pointer to and length of a buffer containing
** an fts3 query expression, respectively. This function attempts to parse the
** query expression and create a tree of Fts3Expr structures representing the
** parsed expression. If successful, *ppExpr is set to point to the head
** of the parsed expression tree and SQLITE_OK is returned. If an error
** occurs, either SQLITE_NOMEM (out-of-memory error) or SQLITE_ERROR (parse
** error) is returned and *ppExpr is set to 0.
**
** If parameter n is a negative number, then z is assumed to point to a
** nul-terminated string and the length is determined using strlen().
**
** The first parameter, pTokenizer, is passed the fts3 tokenizer module to
** use to normalize query tokens while parsing the expression. The azCol[]
** array, which is assumed to contain nCol entries, should contain the names
** of each column in the target fts3 table, in order from left to right.
** Column names must be nul-terminated strings.
**
** The iDefaultCol parameter should be passed the index of the table column
** that appears on the left-hand-side of the MATCH operator (the default
** column to match against for tokens for which a column name is not explicitly
** specified as part of the query string), or -1 if tokens may by default
** match any table column.
*/
int sqlite3Fts3ExprParse(
sqlite3_tokenizer *pTokenizer, /* Tokenizer module */
int iLangid, /* Language id for tokenizer */
char **azCol, /* Array of column names for fts3 table */
int bFts4, /* True to allow FTS4-only syntax */
int nCol, /* Number of entries in azCol[] */
int iDefaultCol, /* Default column to query */
const char *z, int n, /* Text of MATCH query */
Fts3Expr **ppExpr, /* OUT: Parsed query structure */
char **pzErr /* OUT: Error message (sqlite3_malloc) */
){
int rc = fts3ExprParseUnbalanced(
pTokenizer, iLangid, azCol, bFts4, nCol, iDefaultCol, z, n, ppExpr
);
/* Rebalance the expression. And check that its depth does not exceed
** SQLITE_FTS3_MAX_EXPR_DEPTH. */
if( rc==SQLITE_OK && *ppExpr ){
rc = fts3ExprBalance(ppExpr, SQLITE_FTS3_MAX_EXPR_DEPTH);
if( rc==SQLITE_OK ){
rc = fts3ExprCheckDepth(*ppExpr, SQLITE_FTS3_MAX_EXPR_DEPTH);
}
}
if( rc!=SQLITE_OK ){
sqlite3Fts3ExprFree(*ppExpr);
*ppExpr = 0;
if( rc==SQLITE_TOOBIG ){
sqlite3Fts3ErrMsg(pzErr,
"FTS expression tree is too large (maximum depth %d)",
SQLITE_FTS3_MAX_EXPR_DEPTH
);
rc = SQLITE_ERROR;
}else if( rc==SQLITE_ERROR ){
sqlite3Fts3ErrMsg(pzErr, "malformed MATCH expression: [%s]", z);
}
}
return rc;
}
/*
** Free a single node of an expression tree.
*/
static void fts3FreeExprNode(Fts3Expr *p){
assert( p->eType==FTSQUERY_PHRASE || p->pPhrase==0 );
sqlite3Fts3EvalPhraseCleanup(p->pPhrase);
sqlite3_free(p->aMI);
sqlite3_free(p);
}
/*
** Free a parsed fts3 query expression allocated by sqlite3Fts3ExprParse().
**
** This function would be simpler if it recursively called itself. But
** that would mean passing a sufficiently large expression to ExprParse()
** could cause a stack overflow.
*/
void sqlite3Fts3ExprFree(Fts3Expr *pDel){
Fts3Expr *p;
assert( pDel==0 || pDel->pParent==0 );
for(p=pDel; p && (p->pLeft||p->pRight); p=(p->pLeft ? p->pLeft : p->pRight)){
assert( p->pParent==0 || p==p->pParent->pRight || p==p->pParent->pLeft );
}
while( p ){
Fts3Expr *pParent = p->pParent;
fts3FreeExprNode(p);
if( pParent && p==pParent->pLeft && pParent->pRight ){
p = pParent->pRight;
while( p && (p->pLeft || p->pRight) ){
assert( p==p->pParent->pRight || p==p->pParent->pLeft );
p = (p->pLeft ? p->pLeft : p->pRight);
}
}else{
p = pParent;
}
}
}
/****************************************************************************
*****************************************************************************
** Everything after this point is just test code.
*/
#ifdef SQLITE_TEST
#include <stdio.h>
/*
** Return a pointer to a buffer containing a text representation of the
** expression passed as the first argument. The buffer is obtained from
** sqlite3_malloc(). It is the responsibility of the caller to use
** sqlite3_free() to release the memory. If an OOM condition is encountered,
** NULL is returned.
**
** If the second argument is not NULL, then its contents are prepended to
** the returned expression text and then freed using sqlite3_free().
*/
static char *exprToString(Fts3Expr *pExpr, char *zBuf){
if( pExpr==0 ){
return sqlite3_mprintf("");
}
switch( pExpr->eType ){
case FTSQUERY_PHRASE: {
Fts3Phrase *pPhrase = pExpr->pPhrase;
int i;
zBuf = sqlite3_mprintf(
"%zPHRASE %d 0", zBuf, pPhrase->iColumn);
for(i=0; zBuf && i<pPhrase->nToken; i++){
zBuf = sqlite3_mprintf("%z %.*s%s", zBuf,
pPhrase->aToken[i].n, pPhrase->aToken[i].z,
(pPhrase->aToken[i].isPrefix?"+":"")
);
}
return zBuf;
}
case FTSQUERY_NEAR:
zBuf = sqlite3_mprintf("%zNEAR/%d ", zBuf, pExpr->nNear);
break;
case FTSQUERY_NOT:
zBuf = sqlite3_mprintf("%zNOT ", zBuf);
break;
case FTSQUERY_AND:
zBuf = sqlite3_mprintf("%zAND ", zBuf);
break;
case FTSQUERY_OR:
zBuf = sqlite3_mprintf("%zOR ", zBuf);
break;
}
if( zBuf ) zBuf = sqlite3_mprintf("%z{", zBuf);
if( zBuf ) zBuf = exprToString(pExpr->pLeft, zBuf);
if( zBuf ) zBuf = sqlite3_mprintf("%z} {", zBuf);
if( zBuf ) zBuf = exprToString(pExpr->pRight, zBuf);
if( zBuf ) zBuf = sqlite3_mprintf("%z}", zBuf);
return zBuf;
}
/*
** This is the implementation of a scalar SQL function used to test the
** expression parser. It should be called as follows:
**
** fts3_exprtest(<tokenizer>, <expr>, <column 1>, ...);
**
** The first argument, <tokenizer>, is the name of the fts3 tokenizer used
** to parse the query expression (see README.tokenizers). The second argument
** is the query expression to parse. Each subsequent argument is the name
** of a column of the fts3 table that the query expression may refer to.
** For example:
**
** SELECT fts3_exprtest('simple', 'Bill col2:Bloggs', 'col1', 'col2');
*/
static void fts3ExprTestCommon(
int bRebalance,
sqlite3_context *context,
int argc,
sqlite3_value **argv
){
sqlite3_tokenizer *pTokenizer = 0;
int rc;
char **azCol = 0;
const char *zExpr;
int nExpr;
int nCol;
int ii;
Fts3Expr *pExpr;
char *zBuf = 0;
Fts3Hash *pHash = (Fts3Hash*)sqlite3_user_data(context);
const char *zTokenizer = 0;
char *zErr = 0;
if( argc<3 ){
sqlite3_result_error(context,
"Usage: fts3_exprtest(tokenizer, expr, col1, ...", -1
);
return;
}
zTokenizer = (const char*)sqlite3_value_text(argv[0]);
rc = sqlite3Fts3InitTokenizer(pHash, zTokenizer, &pTokenizer, &zErr);
if( rc!=SQLITE_OK ){
if( rc==SQLITE_NOMEM ){
sqlite3_result_error_nomem(context);
}else{
sqlite3_result_error(context, zErr, -1);
}
sqlite3_free(zErr);
return;
}
zExpr = (const char *)sqlite3_value_text(argv[1]);
nExpr = sqlite3_value_bytes(argv[1]);
nCol = argc-2;
azCol = (char **)sqlite3_malloc64(nCol*sizeof(char *));
if( !azCol ){
sqlite3_result_error_nomem(context);
goto exprtest_out;
}
for(ii=0; ii<nCol; ii++){
azCol[ii] = (char *)sqlite3_value_text(argv[ii+2]);
}
if( bRebalance ){
char *zDummy = 0;
rc = sqlite3Fts3ExprParse(
pTokenizer, 0, azCol, 0, nCol, nCol, zExpr, nExpr, &pExpr, &zDummy
);
assert( rc==SQLITE_OK || pExpr==0 );
sqlite3_free(zDummy);
}else{
rc = fts3ExprParseUnbalanced(
pTokenizer, 0, azCol, 0, nCol, nCol, zExpr, nExpr, &pExpr
);
}
if( rc!=SQLITE_OK && rc!=SQLITE_NOMEM ){
sqlite3Fts3ExprFree(pExpr);
sqlite3_result_error(context, "Error parsing expression", -1);
}else if( rc==SQLITE_NOMEM || !(zBuf = exprToString(pExpr, 0)) ){
sqlite3_result_error_nomem(context);
}else{
sqlite3_result_text(context, zBuf, -1, SQLITE_TRANSIENT);
sqlite3_free(zBuf);
}
sqlite3Fts3ExprFree(pExpr);
exprtest_out:
if( pTokenizer ){
rc = pTokenizer->pModule->xDestroy(pTokenizer);
}
sqlite3_free(azCol);
}
static void fts3ExprTest(
sqlite3_context *context,
int argc,
sqlite3_value **argv
){
fts3ExprTestCommon(0, context, argc, argv);
}
static void fts3ExprTestRebalance(
sqlite3_context *context,
int argc,
sqlite3_value **argv
){
fts3ExprTestCommon(1, context, argc, argv);
}
/*
** Register the query expression parser test function fts3_exprtest()
** with database connection db.
*/
int sqlite3Fts3ExprInitTestInterface(sqlite3 *db, Fts3Hash *pHash){
int rc = sqlite3_create_function(
db, "fts3_exprtest", -1, SQLITE_UTF8, (void*)pHash, fts3ExprTest, 0, 0
);
if( rc==SQLITE_OK ){
rc = sqlite3_create_function(db, "fts3_exprtest_rebalance",
-1, SQLITE_UTF8, (void*)pHash, fts3ExprTestRebalance, 0, 0
);
}
return rc;
}
#endif
#endif /* !defined(SQLITE_CORE) || defined(SQLITE_ENABLE_FTS3) */