static int fulltext_query(fulltext_vtab *v, const char *zQuery,
DocList **pResult){
Query q;
int phrase_start = -1;
int i;
sqlite3_stmt *pSelect = NULL;
DocList *d = NULL;
int rc = parse_query(v, zQuery, &q);
if( rc!=SQLITE_OK ) return rc;
for(i = 0 ; i < q.nTerms ; ++i){
int need_positions = i<q.nTerms-1 && !q.pTerm[i+1].is_phrase;
DocList *next = docListNew(need_positions ? DL_POSITIONS : DL_DOCIDS);
if( q.pTerm[i].is_phrase ){
phrase_start = i;
}
rc = query_merge(v, &pSelect, q.pTerm[i].zTerm, d, i - phrase_start, next);
if( rc!=SQLITE_OK ) break;
if( d!=NULL ){
docListDelete(d);
}
d = next;
}
sqlite3_finalize(pSelect);
query_free(&q);
*pResult = d;
return rc;
}
/* Perform a full-text query; return a list of documents in [pResult]. */
static int fulltext_query(fulltext_vtab *v, const char *zQuery,
DocList **pResult){
Query q;
int phrase_start = -1;
int i;
sqlite3_stmt *pSelect = NULL;
DocList *d = NULL;
int rc = parse_query(v, zQuery, &q);
if( rc!=SQLITE_OK ) return rc;
/* Merge terms. */
for(i = 0 ; i < q.nTerms ; ++i){
/* In each merge step, we need to generate positions whenever we're
* processing a phrase which hasn't ended yet. */
int need_positions = i<q.nTerms-1 && !q.pTerm[i+1].is_phrase;
DocList *next = docListNew(need_positions ? DL_POSITIONS : DL_DOCIDS);
if( q.pTerm[i].is_phrase ){
phrase_start = i;
}
rc = query_merge(v, &pSelect, q.pTerm[i].zTerm, d, i - phrase_start, next);
if( rc!=SQLITE_OK ) break;
if( d!=NULL ){
docListDelete(d);
}
d = next;
}
sqlite3_finalize(pSelect);
query_free(&q);
*pResult = d;
return rc;
}
/* Build a hash table containing all terms in zText. */
static int build_terms(Hash *terms, sqlite3_tokenizer *pTokenizer,
const char *zText, sqlite_int64 iDocid){
sqlite3_tokenizer_cursor *pCursor;
const char *pToken;
int nTokenBytes;
int iStartOffset, iEndOffset, iPosition;
int rc = pTokenizer->pModule->xOpen(pTokenizer, zText, -1, &pCursor);
if( rc!=SQLITE_OK ) return rc;
pCursor->pTokenizer = pTokenizer;
HashInit(terms, HASH_STRING, 1);
while( SQLITE_OK==pTokenizer->pModule->xNext(pCursor,
&pToken, &nTokenBytes,
&iStartOffset, &iEndOffset,
&iPosition) ){
DocList *p;
/* Positions can't be negative; we use -1 as a terminator internally. */
if( iPosition<0 ) {
rc = SQLITE_ERROR;
goto err;
}
p = HashFind(terms, pToken, nTokenBytes);
if( p==NULL ){
p = docListNew(DL_POSITIONS_OFFSETS);
docListAddDocid(p, iDocid);
HashInsert(terms, pToken, nTokenBytes, p);
}
docListAddPosOffset(p, iPosition, iStartOffset, iEndOffset);
}
err:
/* TODO(shess) Check return? Should this be able to cause errors at
** this point? Actually, same question about sqlite3_finalize(),
** though one could argue that failure there means that the data is
** not durable. *ponder*
*/
pTokenizer->pModule->xClose(pCursor);
return rc;
}
static int build_terms(Hash *terms, sqlite3_tokenizer *pTokenizer,
const char *zText, sqlite_int64 iDocid){
sqlite3_tokenizer_cursor *pCursor;
const char *pToken;
int nTokenBytes;
int iStartOffset, iEndOffset, iPosition;
int rc = pTokenizer->pModule->xOpen(pTokenizer, zText, -1, &pCursor);
if( rc!=SQLITE_OK ) return rc;
pCursor->pTokenizer = pTokenizer;
HashInit(terms, HASH_STRING, 1);
while( SQLITE_OK==pTokenizer->pModule->xNext(pCursor,
&pToken, &nTokenBytes,
&iStartOffset, &iEndOffset,
&iPosition) ){
DocList *p;
if( iPosition<0 ) {
rc = SQLITE_ERROR;
goto err;
}
p = HashFind(terms, pToken, nTokenBytes);
if( p==NULL ){
p = docListNew(DL_POSITIONS_OFFSETS);
docListAddDocid(p, iDocid);
HashInsert(terms, pToken, nTokenBytes, p);
}
docListAddPosOffset(p, iPosition, iStartOffset, iEndOffset);
}
err:
pTokenizer->pModule->xClose(pCursor);
return rc;
}
