SQLite

  int iLower, iUpper;       /* Range of histogram regions containing pRhs */
  u8 aff;                   /* Column affinity */
  int rc;                   /* Subfunction return code */
  double nRowEst;           /* New estimate of the number of rows */

  assert( p->aSample!=0 );
  aff = p->pTable->aCol[p->aiColumn[0]].affinity;
  if( pExpr ){
    rc = valueFromExpr(pParse, pExpr, aff, &pRhs);
    if( rc ) goto whereEqualScanEst_cancel;
  }else{
    pRhs = sqlite3ValueNew(pParse->db);
  }
  if( pRhs==0 ) return SQLITE_NOTFOUND;
  rc = whereRangeRegion(pParse, p, pRhs, 0, &iLower);
  if( rc ) goto whereEqualScanEst_cancel;
  rc = whereRangeRegion(pParse, p, pRhs, 1, &iUpper);
  if( rc ) goto whereEqualScanEst_cancel;
  WHERETRACE(("equality scan regions: %d..%d\n", iLower, iUpper));
  if( iLower>=iUpper ){

#ifdef SQLITE_ENABLE_STAT2
    /* If the constraint is of the form x=VALUE and histogram
    ** data is available for column x, then it might be possible
    ** to get a better estimate on the number of rows based on
    ** VALUE and how common that value is according to the histogram.
    */
    if( nRow>(double)1 && nEq==1 && pFirstTerm!=0 ){
      if( pFirstTerm->eOperator & (WO_EQ|WO_ISNULL) ){
        testcase( pFirstTerm->eOperator==WO_EQ );
        testcase( pFirstTerm->pOperator==WO_ISNULL );
        whereEqualScanEst(pParse, pProbe, pFirstTerm->pExpr->pRight, &nRow);
      }else if( pFirstTerm->eOperator==WO_IN && bInEst==0 ){
        whereInScanEst(pParse, pProbe, pFirstTerm->pExpr->x.pList, &nRow);
      }
    }
#endif /* SQLITE_ENABLE_STAT2 */

  215  {z IN (-1,3)}         t1z  150
  216  {z=-1 OR z=3}         t1z  150

  300  {y=0}                 {}   100
  301  {y=1}                 t1y   50
  302  {y=0.1}               t1y   50

  400  {x IS NULL}           t1x  400

} {
  # Verify that the expected index is used with the expected row count
  do_test analyze5-1.${testid}a {
    set x [lindex [eqp "SELECT * FROM t1 WHERE $where"] 3]
    set idx {}
    regexp {INDEX (t1.) } $x all idx
    regexp {~([0-9]+) rows} $x all nrow
    list $idx $nrow
  } [list $index $rows]

  # Verify that the same result is achieved regardless of whether or not
  # the index is used
  do_test analyze5-1.${testid}b {
    set w2 [string map {y +y z +z} $where]
    set a1 [db eval "SELECT rowid FROM t1 NOT INDEXED WHERE $w2\
                     ORDER BY +rowid"]
    set a2 [db eval "SELECT rowid FROM t1 WHERE $where ORDER BY +rowid"]
    if {$a1==$a2} {
      set res ok
    } else {
      set res "a1=\[$a1\] a2=\[$a2\]"
    }
    set res
  } {ok}
}

# Increase the number of NULLs in column x
#
db eval {
   UPDATE t1 SET x=NULL;
   UPDATE t1 SET x=rowid
    WHERE rowid IN (SELECT rowid FROM t1 ORDER BY random() LIMIT 5);
   ANALYZE;
}

# Verify that range queries generate the correct row count estimates
#
foreach {testid where index rows} {
  500  {x IS NULL AND u='charlie'}         t1u  20
  501  {x=1 AND u='charlie'}               t1x   5
  502  {x IS NULL}                          {} 100
  503  {x=1}                               t1x  50
  504  {x IS NOT NULL}                     t1x  25

} {
  # Verify that the expected index is used with the expected row count
  do_test analyze5-1.${testid}a {
    set x [lindex [eqp "SELECT * FROM t1 WHERE $where"] 3]
    set idx {}
    regexp {INDEX (t1.) } $x all idx
    regexp {~([0-9]+) rows} $x all nrow

      set res ok
    } else {
      set res "a1=\[$a1\] a2=\[$a2\]"
    }
    set res
  } {ok}
}


finish_test