fts3.c

最新的sqlite3.6.2源代码

        dataBufferDestroy(&two);
      }
      dlrStep(&left);
      dlrStep(&right);
    }
  }

  dlrDestroy(&left);
  dlrDestroy(&right);
  dlwDestroy(&writer);
}

/* We have two DL_DOCIDS doclists:  pLeft and pRight.
** Write the intersection of these two doclists into pOut as a
** DL_DOCIDS doclist.
*/
static void docListAndMerge(
  const char *pLeft, int nLeft,
  const char *pRight, int nRight,
  DataBuffer *pOut      /* Write the combined doclist here */
){
  DLReader left, right;
  DLWriter writer;

  if( nLeft==0 || nRight==0 ) return;

  dlrInit(&left, DL_DOCIDS, pLeft, nLeft);
  dlrInit(&right, DL_DOCIDS, pRight, nRight);
  dlwInit(&writer, DL_DOCIDS, pOut);

  while( !dlrAtEnd(&left) && !dlrAtEnd(&right) ){
    if( dlrDocid(&left)<dlrDocid(&right) ){
      dlrStep(&left);
    }else if( dlrDocid(&right)<dlrDocid(&left) ){
      dlrStep(&right);
    }else{
      dlwAdd(&writer, dlrDocid(&left));
      dlrStep(&left);
      dlrStep(&right);
    }
  }

  dlrDestroy(&left);
  dlrDestroy(&right);
  dlwDestroy(&writer);
}

/* We have two DL_DOCIDS doclists:  pLeft and pRight.
** Write the union of these two doclists into pOut as a
** DL_DOCIDS doclist.
*/
static void docListOrMerge(
  const char *pLeft, int nLeft,
  const char *pRight, int nRight,
  DataBuffer *pOut      /* Write the combined doclist here */
){
  DLReader left, right;
  DLWriter writer;

  if( nLeft==0 ){
    if( nRight!=0 ) dataBufferAppend(pOut, pRight, nRight);
    return;
  }
  if( nRight==0 ){
    dataBufferAppend(pOut, pLeft, nLeft);
    return;
  }

  dlrInit(&left, DL_DOCIDS, pLeft, nLeft);
  dlrInit(&right, DL_DOCIDS, pRight, nRight);
  dlwInit(&writer, DL_DOCIDS, pOut);

  while( !dlrAtEnd(&left) || !dlrAtEnd(&right) ){
    if( dlrAtEnd(&right) ){
      dlwAdd(&writer, dlrDocid(&left));
      dlrStep(&left);
    }else if( dlrAtEnd(&left) ){
      dlwAdd(&writer, dlrDocid(&right));
      dlrStep(&right);
    }else if( dlrDocid(&left)<dlrDocid(&right) ){
      dlwAdd(&writer, dlrDocid(&left));
      dlrStep(&left);
    }else if( dlrDocid(&right)<dlrDocid(&left) ){
      dlwAdd(&writer, dlrDocid(&right));
      dlrStep(&right);
    }else{
      dlwAdd(&writer, dlrDocid(&left));
      dlrStep(&left);
      dlrStep(&right);
    }
  }

  dlrDestroy(&left);
  dlrDestroy(&right);
  dlwDestroy(&writer);
}

/* We have two DL_DOCIDS doclists:  pLeft and pRight.
** Write into pOut as DL_DOCIDS doclist containing all documents that
** occur in pLeft but not in pRight.
*/
static void docListExceptMerge(
  const char *pLeft, int nLeft,
  const char *pRight, int nRight,
  DataBuffer *pOut      /* Write the combined doclist here */
){
  DLReader left, right;
  DLWriter writer;

  if( nLeft==0 ) return;
  if( nRight==0 ){
    dataBufferAppend(pOut, pLeft, nLeft);
    return;
  }

  dlrInit(&left, DL_DOCIDS, pLeft, nLeft);
  dlrInit(&right, DL_DOCIDS, pRight, nRight);
  dlwInit(&writer, DL_DOCIDS, pOut);

  while( !dlrAtEnd(&left) ){
    while( !dlrAtEnd(&right) && dlrDocid(&right)<dlrDocid(&left) ){
      dlrStep(&right);
    }
    if( dlrAtEnd(&right) || dlrDocid(&left)<dlrDocid(&right) ){
      dlwAdd(&writer, dlrDocid(&left));
    }
    dlrStep(&left);
  }

  dlrDestroy(&left);
  dlrDestroy(&right);
  dlwDestroy(&writer);
}

static char *string_dup_n(const char *s, int n){
  char *str = sqlite3_malloc(n + 1);
  memcpy(str, s, n);
  str[n] = '\0';
  return str;
}

/* Duplicate a string; the caller must free() the returned string.
 * (We don't use strdup() since it is not part of the standard C library and
 * may not be available everywhere.) */
static char *string_dup(const char *s){
  return string_dup_n(s, strlen(s));
}

/* Format a string, replacing each occurrence of the % character with
 * zDb.zName.  This may be more convenient than sqlite_mprintf()
 * when one string is used repeatedly in a format string.
 * The caller must free() the returned string. */
static char *string_format(const char *zFormat,
                           const char *zDb, const char *zName){
  const char *p;
  size_t len = 0;
  size_t nDb = strlen(zDb);
  size_t nName = strlen(zName);
  size_t nFullTableName = nDb+1+nName;
  char *result;
  char *r;

  /* first compute length needed */
  for(p = zFormat ; *p ; ++p){
    len += (*p=='%' ? nFullTableName : 1);
  }
  len += 1;  /* for null terminator */

  r = result = sqlite3_malloc(len);
  for(p = zFormat; *p; ++p){
    if( *p=='%' ){
      memcpy(r, zDb, nDb);
      r += nDb;
      *r++ = '.';
      memcpy(r, zName, nName);
      r += nName;
    } else {
      *r++ = *p;
    }
  }
  *r++ = '\0';
  assert( r == result + len );
  return result;
}

static int sql_exec(sqlite3 *db, const char *zDb, const char *zName,
                    const char *zFormat){
  char *zCommand = string_format(zFormat, zDb, zName);
  int rc;
  FTSTRACE(("FTS3 sql: %s\n", zCommand));
  rc = sqlite3_exec(db, zCommand, NULL, 0, NULL);
  sqlite3_free(zCommand);
  return rc;
}

static int sql_prepare(sqlite3 *db, const char *zDb, const char *zName,
                       sqlite3_stmt **ppStmt, const char *zFormat){
  char *zCommand = string_format(zFormat, zDb, zName);
  int rc;
  FTSTRACE(("FTS3 prepare: %s\n", zCommand));
  rc = sqlite3_prepare_v2(db, zCommand, -1, ppStmt, NULL);
  sqlite3_free(zCommand);
  return rc;
}

/* end utility functions */

/* Forward reference */
typedef struct fulltext_vtab fulltext_vtab;

/* A single term in a query is represented by an instances of
** the following structure. Each word which may match against
** document content is a term. Operators, like NEAR or OR, are
** not terms. Query terms are organized as a flat list stored
** in the Query.pTerms array.
**
** If the QueryTerm.nPhrase variable is non-zero, then the QueryTerm
** is the first in a contiguous string of terms that are either part
** of the same phrase, or connected by the NEAR operator.
**
** If the QueryTerm.nNear variable is non-zero, then the token is followed 
** by a NEAR operator with span set to (nNear-1). For example, the 
** following query:
**
** The QueryTerm.iPhrase variable stores the index of the token within
** its phrase, indexed starting at 1, or 1 if the token is not part 
** of any phrase.
**
** For example, the data structure used to represent the following query:
**
**     ... MATCH 'sqlite NEAR/5 google NEAR/2 "search engine"'
**
** is:
**
**     {nPhrase=4, iPhrase=1, nNear=6, pTerm="sqlite"},
**     {nPhrase=0, iPhrase=1, nNear=3, pTerm="google"},
**     {nPhrase=0, iPhrase=1, nNear=0, pTerm="search"},
**     {nPhrase=0, iPhrase=2, nNear=0, pTerm="engine"},
**
** compiling the FTS3 syntax to Query structures is done by the parseQuery()
** function.
*/
typedef struct QueryTerm {
  short int nPhrase; /* How many following terms are part of the same phrase */
  short int iPhrase; /* This is the i-th term of a phrase. */
  short int iColumn; /* Column of the index that must match this term */
  signed char nNear; /* term followed by a NEAR operator with span=(nNear-1) */
  signed char isOr;  /* this term is preceded by "OR" */
  signed char isNot; /* this term is preceded by "-" */
  signed char isPrefix; /* this term is followed by "*" */
  char *pTerm;       /* text of the term.  '\000' terminated.  malloced */
  int nTerm;         /* Number of bytes in pTerm[] */
} QueryTerm;


/* A query string is parsed into a Query structure.
 *
 * We could, in theory, allow query strings to be complicated
 * nested expressions with precedence determined by parentheses.
 * But none of the major search engines do this.  (Perhaps the
 * feeling is that an parenthesized expression is two complex of
 * an idea for the average user to grasp.)  Taking our lead from
 * the major search engines, we will allow queries to be a list
 * of terms (with an implied AND operator) or phrases in double-quotes,
 * with a single optional "-" before each non-phrase term to designate
 * negation and an optional OR connector.
 *
 * OR binds more tightly than the implied AND, which is what the
 * major search engines seem to do.  So, for example:
 * 
 *    [one two OR three]     ==>    one AND (two OR three)
 *    [one OR two three]     ==>    (one OR two) AND three
 *
 * A "-" before a term matches all entries that lack that term.
 * The "-" must occur immediately before the term with in intervening
 * space.  This is how the search engines do it.
 *
 * A NOT term cannot be the right-hand operand of an OR.  If this
 * occurs in the query string, the NOT is ignored:
 *
 *    [one OR -two]          ==>    one OR two
 *
 */
typedef struct Query {
  fulltext_vtab *pFts;  /* The full text index */
  int nTerms;           /* Number of terms in the query */
  QueryTerm *pTerms;    /* Array of terms.  Space obtained from malloc() */
  int nextIsOr;         /* Set the isOr flag on the next inserted term */
  int nextIsNear;       /* Set the isOr flag on the next inserted term */
  int nextColumn;       /* Next word parsed must be in this column */
  int dfltColumn;       /* The default column */
} Query;


/*
** An instance of the following structure keeps track of generated
** matching-word offset information and snippets.
*/
typedef struct Snippet {
  int nMatch;     /* Total number of matches */
  int nAlloc;     /* Space allocated for aMatch[] */
  struct snippetMatch { /* One entry for each matching term */
    char snStatus;       /* Status flag for use while constructing snippets */
    short int iCol;      /* The column that contains the match */
    short int iTerm;     /* The index in Query.pTerms[] of the matching term */
    int iToken;          /* The index of the matching document token */
    short int nByte;     /* Number of bytes in the term */
    int iStart;          /* The offset to the first character of the term */
  } *aMatch;      /* Points to space obtained from malloc */
  char *zOffset;  /* Text rendering of aMatch[] */
  int nOffset;    /* strlen(zOffset) */
  char *zSnippet; /* Snippet text */
  int nSnippet;   /* strlen(zSnippet) */
} Snippet;


typedef enum QueryType {
  QUERY_GENERIC,   /* table scan */
  QUERY_DOCID,     /* lookup by docid */
  QUERY_FULLTEXT   /* QUERY_FULLTEXT + [i] is a full-text search for column i*/
} QueryType;

typedef enum fulltext_statement {
  CONTENT_INSERT_STMT,
  CONTENT_SELECT_STMT,
  CONTENT_UPDATE_STMT,
  CONTENT_DELETE_STMT,
  CONTENT_EXISTS_STMT,

  BLOCK_INSERT_STMT,
  BLOCK_SELECT_STMT,
  BLOCK_DELETE_STMT,
  BLOCK_DELETE_ALL_STMT,

  SEGDIR_MAX_INDEX_STMT,
  SEGDIR_SET_STMT,
  SEGDIR_SELECT_LEVEL_STMT,
  SEGDIR_SPAN_STMT,
  SEGDIR_DELETE_STMT,
  SEGDIR_SELECT_SEGMENT_STMT,
  SEGDIR_SELECT_ALL_STMT,
  SEGDIR_DELETE_ALL_STMT,
  SEGDIR_COUNT_STMT,

  MAX_STMT                     /* Always at end! */
} fulltext_statement;

/* These must exactly match the enum above. */
/* TODO(shess): Is there some risk that a statement will be used in two
** cursors at once, e.g.  if a query joins a virtual table to itself?
** If so perhaps we should move some of these to the cursor object.
*/
static const char *const fulltext_zStatement[MAX_STMT] = {
  /* CONTENT_INSERT */ NULL,  /* generated in contentInsertStatement() */
  /* CONTENT_SELECT */ NULL,  /* generated in contentSelectStatement() */
  /* CONTENT_UPDATE */ NULL,  /* generated in contentUpdateStatement() */
  /* CONTENT_DELETE */ "delete from %_content where docid = ?",
  /* CONTENT_EXISTS */ "select docid from %_content limit 1",

  /* BLOCK_INSERT */
  "insert into %_segments (blockid, block) values (null, ?)",
  /* BLOCK_SELECT */ "select block from %_segments where blockid = ?",
  /* BLOCK_DELETE */ "delete from %_segments where blockid between ? and ?",
  /* BLOCK_DELETE_ALL */ "delete from %_segments",

  /* SEGDIR_MAX_INDEX */ "select max(idx) from %_segdir where level = ?",
  /* SEGDIR_SET */ "insert into %_segdir values (?, ?, ?, ?, ?, ?)",
  /* SEGDIR_SELECT_LEVEL */
  "select start_block, leaves_end_block, root from %_segdir "
  " where level = ? order by idx",
  /* SEGDIR_SPAN */
  "select min(start_block), max(end_block) from %_segdir "
  " where level = ? and start_block <> 0",
  /* SEGDIR_DELETE */ "delete from %_segdir where level = ?",

  /* NOTE(shess): The first three results of the following two
  ** statements must match.
  */
  /* SEGDIR_SELECT_SEGMENT */
  "select start_block, leaves_end_block, root from %_segdir "
  " where level = ? and idx = ?",
  /* SEGDIR_SELECT_ALL */
  "select start_block, leaves_end_block, root from %_segdir "
  " order by level desc, idx asc",
  /* SEGDIR_DELETE_ALL */ "delete from %_segdir",
  /* SEGDIR_COUNT */ "select count(*), ifnull(max(level),0) from %_segdir",
};

/*
** A connection to a fulltext index is an instance of the following
** structure.  The