fulltext.c

最新的sqlite3.6.2源代码

static int fulltextNext(sqlite3_vtab_cursor *pCursor){
  fulltext_cursor *c = (fulltext_cursor *) pCursor;
  sqlite_int64 iDocid;
  int rc;

  switch( c->iCursorType ){
    case QUERY_GENERIC:
      /* TODO(shess) Handle SQLITE_SCHEMA AND SQLITE_BUSY. */
      rc = sqlite3_step(c->pStmt);
      switch( rc ){
        case SQLITE_ROW:
          c->eof = 0;
          return SQLITE_OK;
        case SQLITE_DONE:
          c->eof = 1;
          return SQLITE_OK;
        default:
          c->eof = 1;
          return rc;
      }
    case QUERY_FULLTEXT:
      rc = sqlite3_reset(c->pStmt);
      if( rc!=SQLITE_OK ) return rc;

      if( readerAtEnd(&c->result)){
        c->eof = 1;
        return SQLITE_OK;
      }
      iDocid = readDocid(&c->result);
      rc = sqlite3_bind_int64(c->pStmt, 1, iDocid);
      if( rc!=SQLITE_OK ) return rc;
      /* TODO(shess) Handle SQLITE_SCHEMA AND SQLITE_BUSY. */
      rc = sqlite3_step(c->pStmt);
      if( rc==SQLITE_ROW ){   /* the case we expect */
        c->eof = 0;
        return SQLITE_OK;
      }
      /* an error occurred; abort */
      return rc==SQLITE_DONE ? SQLITE_ERROR : rc;
    default:
      assert( 0 );
      return SQLITE_ERROR;  /* not reached */
  }
}

static int term_select_doclist(fulltext_vtab *v, const char *pTerm, int nTerm,
                               sqlite3_stmt **ppStmt){
  int rc;
  if( *ppStmt ){
    rc = sqlite3_reset(*ppStmt);
  } else {
    rc = sql_prepare(v->db, v->zName, ppStmt,
      "select doclist from %_term where term = ? order by first");
  }
  if( rc!=SQLITE_OK ) return rc;

  rc = sqlite3_bind_text(*ppStmt, 1, pTerm, nTerm, SQLITE_TRANSIENT);
  if( rc!=SQLITE_OK ) return rc;

  return sqlite3_step(*ppStmt);   /* TODO(adamd): handle schema error */
}

/* Read the posting list for [zTerm]; AND it with the doclist [in] to
 * produce the doclist [out], using the given offset [iOffset] for phrase
 * matching.
 * (*pSelect) is used to hold an SQLite statement used inside this function;
 * the caller should initialize *pSelect to NULL before the first call.
 */
static int query_merge(fulltext_vtab *v, sqlite3_stmt **pSelect,
                       const char *zTerm,
                       DocList *pIn, int iOffset, DocList *out){
  int rc;
  DocListMerge merge;

  if( pIn!=NULL && !pIn->nData ){
    /* If [pIn] is already empty, there's no point in reading the
     * posting list to AND it in; return immediately. */
      return SQLITE_OK;
  }

  rc = term_select_doclist(v, zTerm, -1, pSelect);
  if( rc!=SQLITE_ROW && rc!=SQLITE_DONE ) return rc;

  mergeInit(&merge, pIn, iOffset, out);
  while( rc==SQLITE_ROW ){
    DocList block;
    docListInit(&block, DL_POSITIONS_OFFSETS,
                sqlite3_column_blob(*pSelect, 0),
                sqlite3_column_bytes(*pSelect, 0));
    mergeBlock(&merge, &block);
    docListDestroy(&block);

    rc = sqlite3_step(*pSelect);
    if( rc!=SQLITE_ROW && rc!=SQLITE_DONE ){
      return rc;
    }
  }
  
  return SQLITE_OK;
}

typedef struct QueryTerm {
  int is_phrase;    /* true if this term begins a new phrase */
  const char *zTerm;
} QueryTerm;

/* A parsed query.
 *
 * As an example, parsing the query ["four score" years "new nation"] will
 * yield a Query with 5 terms:
 *   "four",   is_phrase = 1
 *   "score",  is_phrase = 0
 *   "years",  is_phrase = 1
 *   "new",    is_phrase = 1
 *   "nation", is_phrase = 0
 */
typedef struct Query {
  int nTerms;
  QueryTerm *pTerm;
} Query;

static void query_add(Query *q, int is_phrase, const char *zTerm){
  QueryTerm *t;
  ++q->nTerms;
  q->pTerm = realloc(q->pTerm, q->nTerms * sizeof(q->pTerm[0]));
  t = &q->pTerm[q->nTerms - 1];
  t->is_phrase = is_phrase;
  t->zTerm = zTerm;
}
    
static void query_free(Query *q){
  int i;
  for(i = 0; i < q->nTerms; ++i){
    free((void *) q->pTerm[i].zTerm);
  }
  free(q->pTerm);
}

static int tokenize_segment(sqlite3_tokenizer *pTokenizer,
                            const char *zQuery, int in_phrase,
                            Query *pQuery){
  sqlite3_tokenizer_module *pModule = pTokenizer->pModule;
  sqlite3_tokenizer_cursor *pCursor;
  int is_first = 1;
  
  int rc = pModule->xOpen(pTokenizer, zQuery, -1, &pCursor);
  if( rc!=SQLITE_OK ) return rc;
  pCursor->pTokenizer = pTokenizer;

  while( 1 ){
    const char *zToken;
    int nToken, iStartOffset, iEndOffset, dummy_pos;

    rc = pModule->xNext(pCursor,
                        &zToken, &nToken,
                        &iStartOffset, &iEndOffset,
                        &dummy_pos);
    if( rc!=SQLITE_OK ) break;
    query_add(pQuery, !in_phrase || is_first, string_dup_n(zToken, nToken));
    is_first = 0;
  }

  return pModule->xClose(pCursor);
}

/* Parse a query string, yielding a Query object. */
static int parse_query(fulltext_vtab *v, const char *zQuery, Query *pQuery){
  char *zQuery1 = string_dup(zQuery);
  int in_phrase = 0;
  char *s = zQuery1;
  pQuery->nTerms = 0;
  pQuery->pTerm = NULL;

  while( *s ){
    char *t = s;
    while( *t ){
      if( *t=='"' ){
        *t++ = '\0';
        break;
      }
      ++t;
    }
    if( *s ){
      tokenize_segment(v->pTokenizer, s, in_phrase, pQuery);
    }
    s = t;
    in_phrase = !in_phrase;
  }
  
  free(zQuery1);
  return SQLITE_OK;
}

/* 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;
}

static int fulltextFilter(sqlite3_vtab_cursor *pCursor,
                          int idxNum, const char *idxStr,
                          int argc, sqlite3_value **argv){
  fulltext_cursor *c = (fulltext_cursor *) pCursor;
  fulltext_vtab *v = cursor_vtab(c);
  int rc;
  const char *zStatement;

  c->iCursorType = idxNum;
  switch( idxNum ){
    case QUERY_GENERIC:
      zStatement = "select rowid, content from %_content";
      break;

    case QUERY_FULLTEXT:   /* full-text search */
    {
      const char *zQuery = (const char *)sqlite3_value_text(argv[0]);
      DocList *pResult;
      assert( argc==1 );
      rc = fulltext_query(v, zQuery, &pResult);
      if( rc!=SQLITE_OK ) return rc;
      readerInit(&c->result, pResult);
      zStatement = "select rowid, content from %_content where rowid = ?";
      break;
    }

    default:
      assert( 0 );
  }

  rc = sql_prepare(v->db, v->zName, &c->pStmt, zStatement);
  if( rc!=SQLITE_OK ) return rc;

  return fulltextNext(pCursor);
}

static int fulltextEof(sqlite3_vtab_cursor *pCursor){
  fulltext_cursor *c = (fulltext_cursor *) pCursor;
  return c->eof;
}

static int fulltextColumn(sqlite3_vtab_cursor *pCursor,
                          sqlite3_context *pContext, int idxCol){
  fulltext_cursor *c = (fulltext_cursor *) pCursor;
  const char *s;

  assert( idxCol==0 );
  s = (const char *) sqlite3_column_text(c->pStmt, 1);
  sqlite3_result_text(pContext, s, -1, SQLITE_TRANSIENT);

  return SQLITE_OK;
}

static int fulltextRowid(sqlite3_vtab_cursor *pCursor, sqlite_int64 *pRowid){
  fulltext_cursor *c = (fulltext_cursor *) pCursor;

  *pRowid = sqlite3_column_int64(c->pStmt, 0);
  return SQLITE_OK;
}

/* 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;
}
/* Update the %_terms table to map the term [zTerm] to the given rowid. */
static int index_insert_term(fulltext_vtab *v, const char *zTerm, int nTerm,
                             sqlite_int64 iDocid, DocList *p){
  sqlite_int64 iFirst;
  sqlite_int64 iIndexRow;
  DocList doclist;

  int rc = term_chunk_select(v, zTerm, nTerm, iDocid, &iFirst);
  if( rc==SQLITE_DONE ){
    docListInit(&doclist, DL_POSITIONS_OFFSETS, 0, 0);
    if( docListUpdate(&doclist, iDocid, p) ){
      rc = term_insert(v, zTerm, nTerm, iDocid, &doclist);
      docListDestroy(&doclist);
      return rc;
    }
    return SQLITE_OK;
  }
  if( rc!=SQLITE_ROW ) return SQLITE_ERROR;

  /* This word is in the index; add this document ID to its blob. */

  rc = term_select(v, zTerm, nTerm, iFirst, &iIndexRow, &doclist);
  if( rc!=SQLITE_OK ) return rc;

  if( docListUpdate(&doclist, iDocid, p) ){
    /* If the blob is too big, split it in half. */
    if( doclist.nData>CHUNK_MAX ){
      DocList half;
      if( docListSplit(&doclist, &half) ){
        rc = term_insert(v, zTerm, nTerm, firstDocid(&half), &half);
        docListDestroy(&half);
        if( rc!=SQLITE_OK ) goto err;
      }
    }
    rc = term_update(v, iIndexRow, &doclist);
  }

err:
  docListDestroy(&doclist);
  return rc;
}

/* Insert a row into the full-text index; set *piRowid to be the ID of the
 * new row. */
static int index_insert(fulltext_vtab *v,
                        sqlite3_value *pRequestRowid, const char *zText,
                        sqlite_int64 *piRowid){
  Hash terms;  /* maps term string -> PosList */
  HashElem *e;

  int rc = content_insert(v, pRequestRowid, zText, -1);
  if( rc!=SQLITE_OK ) return rc;
  *piRowid = sqlite3_last_insert_rowid(v->db);

  if( !zText ) return SQLITE_OK;   /* nothing to index */

  rc = build_terms(&terms, v->pTokenizer, zText, *piRowid);
  if( rc!=SQLITE_OK ) return rc;

  for(e=HashFirst(&terms); e; e=HashNext(e)){
    DocList *p = HashData(e);
    rc = index_insert_term(v, HashKey(e), HashKeysize(e), *piRowid, p);
    if( rc!=SQLITE_OK ) break;
  }

  for(e=HashFirst(&terms); e; e=HashNext(e)){
    DocList *p = HashData(e);
    docListDelete(p);
  }
  HashClear(&terms);
  return rc;
}

static int index_delete_term(fulltext_vtab *v, const char *zTerm, int nTerm,
                             sqlite_int64 iDocid){
  sqlite_int64 iFirst;
  sqlite_int64 iIndexRow;
  DocList doclist;

  int rc = term_chunk_select(v, zTerm, nTerm, iDocid, &iFirst);
  if( rc!=SQLITE_ROW ) return SQLITE_ERROR;

  rc = term_select(v, zTerm, nTerm, iFirst, &iIndexRow, &doclist);
  if( rc!=SQLITE_OK ) return rc;

  if( docListUpdate(&doclist, iDocid, NULL) ){
    if( doclist.nData>0 ){
      rc = term_update(v, iIndexRow, &doclist);
    } else {  /* empty posting list */
      rc = term_delete(v, iIndexRow);
    }
  }
  docListDestroy(&doclist);
  return rc;
}

/* Delete a row from the full-text index. */
static int index_delete(fulltext_vtab *v, sqlite_int64 iRow){
  char *zText;
  Hash terms;
  HashElem *e;

  int rc = content_select(v, iRow, &zText);
  if( rc!=SQLITE_OK ) return rc;

  rc = build_terms(&terms, v->pTokenizer, zText, iRow);
  free(zText);
  if( rc!=SQLITE_OK ) return rc;

  for(e=HashFirst(&terms); e; e=HashNext(e)){
    rc = index_delete_term(v, HashKey(e), HashKeysize(e), iRow);
    if( rc!=SQLITE_OK ) break;
  }
  for(e=HashFirst(&terms); e; e=HashNext(e)){
    DocList *p = HashData(e);
    docListDelete(p);
  }
  HashClear(&terms);

  return content_delete(v, iRow);
}

static int fulltextUpdate(sqlite3_vtab *pVtab, int nArg, sqlite3_value **ppArg,
                   sqlite_int64 *pRowid){
  fulltext_vtab *v = (fulltext_vtab *) pVtab;

  if( nArg<2 ){
    return index_delete(v, sqlite3_value_int64(ppArg[0]));
  }

  if( sqlite3_value_type(ppArg[0]) != SQLITE_NULL ){
    return SQLITE_ERROR;   /* an update; not yet supported */
  }

  assert( nArg==3 );    /* ppArg[1] = rowid, ppArg[2] = content */
  return index_insert(v, ppArg[1],
                      (const char *)sqlite3_value_text(ppArg[2]), pRowid);
}

static sqlite3_module fulltextModule = {
  0,
  fulltextCreate,
  fulltextConnect,
  fulltextBestIndex,
  fulltextDisconnect,
  fulltextDestroy,
  fulltextOpen,
  fulltextClose,
  fulltextFilter,
  fulltextNext,
  fulltextEof,
  fulltextColumn,
  fulltextRowid,
  fulltextUpdate
};

int fulltext_init(sqlite3 *db){
 return sqlite3_create_module(db, "fulltext", &fulltextModule, 0);
}

#if !SQLITE_CORE
int sqlite3_extension_init(sqlite3 *db, char **pzErrMsg,
                           const sqlite3_api_routines *pApi){
 SQLITE_EXTENSION_INIT2(pApi)
 return fulltext_init(db);
}
#endif