Documentation Source Text

  <li> The estimated number of rows that SQLite expects the scan to visit.
</ul>

<p>
  For example, the following EXPLAIN QUERY PLAN command operates on a SELECT
  statement that is implemented by performing a full-table scan on table t1:
<pre>
    sqlite&gt; EXPLAIN QUERY PLAN SELECT a, b FROM t1 WHERE a=1; 
    0|0|0|SCAN TABLE t1 (~1000000 rows)
</pre>

<p>
  SQLite estimates that the full-table scan will visit approximately 
  1,000,000 records. If the query were able to use an index, then the SCAN
  record would include the name of the index and a description of how
  it is used by the query. For example:
<pre>
    sqlite&gt; CREATE INDEX i1 ON t1(a);
    sqlite&gt; EXPLAIN QUERY PLAN SELECT a, b FROM t1 WHERE a=1;
    0|0|0|SCAN TABLE t1 BY INDEX i1 (a=?) (~10 rows)
</pre>

<p>
  The output above shows that in this case, SQLite uses index "i1" to optimize
  a WHERE clause filter of the form (a=?) - in this case "a=1". SQLite 
  estimates that scanning the subset of index entries that match the "a=1"
  filter means scanning through approximately 10 records. In this case it is
  not possible to use index i1 as a [covering index], but if it were, the
  SCAN record would report that as well. For example:
<pre>
    sqlite&gt; CREATE INDEX i2 ON t1(a, b);
    sqlite&gt; EXPLAIN QUERY PLAN SELECT a, b FROM t1 WHERE a=1; 
    0|0|0|SCAN TABLE t1 BY COVERING INDEX i2 (a=?) (~10 rows)
</pre>

<p>
  All joins in SQLite are [join order|implemented using nested scans]. When a
  SELECT query that features a join is analyzed using EXPLAIN QUERY PLAN, one
  SCAN record is output for each nested scan. For example:
<pre>
    sqlite&gt; EXPLAIN QUERY PLAN SELECT t1.*, t2.* FROM t1, t2 WHERE t1.a=1 AND t1.b>2;
    0|0|0|SCAN TABLE t1 BY COVERING INDEX i2 (a=? AND b>?) (~3 rows)
    0|1|1|SCAN TABLE t2 (~1000000 rows)
</pre>

<p>
  The second column of output (column "order") indicates the nesting order. In
  this case, the scan of table t1 using index i2 is the outer loop (order=0)
  and the full-table scan of table t2 (order=1) is the inner loop. The third
  column (column "from"), indicates the position in the FROM clause of the
  SELECT statement that the table associated with each scan occurs in. In the
  case above, table t1 occupies the first position in the FROM clause, so the
  value of column "from" is 0 in the first SCAN record. Table t2 is in the
  second position, so the "from" column for the corresponding SCAN record is
  set to 1. In the following example, the positions of t1 and t2 in the FROM 
  clause of the SELECT are reversed. The query strategy remains the same, but
  the values in the "from" column of the output are adjusted accordingly.
<pre>
    sqlite&gt; EXPLAIN QUERY PLAN SELECT t1.*, t2.* FROM t2, t1 WHERE t1.a=1 AND t1.b>2;
    0|0|1|SCAN TABLE t1 BY COVERING INDEX i2 (a=? AND b>?) (~3 rows)
    0|1|0|SCAN TABLE t2 (~1000000 rows)
</pre>

<p>
  In the example above, SQLite estimates that the outer loop scan will visit
  approximately 3 rows, and the inner loop scan approximately 1,000,000. If
  you observe that SQLite's estimates are wildly inaccurate (and appear to be
  causing it to generate sub-optimal query plans), your queries may benefit
  from running the [ANALYZE] command on the database.

<p>
  If the WHERE clause of a query contains an OR expression, then SQLite might
  use the [or-connected-terms|"OR by union"] strategy (also described 
  [or optimization|here]). In this case there will be two SCAN records, one
  for each index scan, with the same values in both the "order" and "from" 
  columns. For example:
<pre>
    sqlite&gt; CREATE INDEX i3 ON t1(b);
    sqlite&gt; EXPLAIN QUERY PLAN SELECT * FROM t1 WHERE a=1 OR b=2;
    0|0|0|SCAN TABLE t1 BY INDEX i1 (a=?) (~10 rows)
    0|0|0|SCAN TABLE t1 BY INDEX i3 (b=?) (~10 rows)
</pre>

<h4>Temporary Sorting B-Trees</h4>

<p>
  If a SELECT query contains an ORDER BY, GROUP BY or DISTINCT clause, 
  SQLite may need to use a temporary b-tree structure to perform an 
  <a href="http://en.wikipedia.org/wiki/Insertion_sort">insertion sort</a> 
  of the output rows. Or, it may [sorting|use an index]. Using an index is 
  almost always much more efficient than performing an online insertion sort.
  If a temporary b-tree is required, a record is added to the EXPLAIN
  QUERY PLAN output with the "detail" field set to a string value of
  the form "USE TEMP B-TREE FOR xxx", where xxx is one of "ORDER BY",
  "GROUP BY" or "DISTINCT". For example:

<pre>
    sqlite&gt; EXPLAIN QUERY PLAN SELECT c, d FROM t2 ORDER BY c; 
    0|0|0|SCAN TABLE t2 (~1000000 rows)
    0|0|0|USE TEMP B-TREE FOR ORDER BY
</pre>

<p>
  In this case using the temporary b-tree can be avoided by creating an index
  on t2(c), as follows:

<pre>
    sqlite&gt; CREATE INDEX i4 ON t2(c);
    sqlite&gt; EXPLAIN QUERY PLAN SELECT c, d FROM t2 ORDER BY c; 
    0|0|0|SCAN TABLE t2 BY INDEX i4 (~1000000 rows)
</pre>

<h4>Subqueries</h4>

<p>
  In all the examples above, the first column (column "selectid") is always
  set to 0. If a query contains sub-selects, either as part of the FROM
  clause or as part of SQL expressions, then the output of EXPLAIN QUERY
  PLAN also includes a report for each sub-select. Each sub-select is assigned
  a distinct, non-zero "selectid" value. The top-level SELECT statement is
  always assigned the selectid value 0. For example:

<pre>
    sqlite&gt; EXPLAIN QUERY PLAN SELECT (SELECT b FROM t1 WHERE a=0), (SELECT a FROM t1 WHERE b=t2.c) FROM t2;
    0|0|0|SCAN TABLE t2 (~1000000 rows)
    0|0|0|EXECUTE SCALAR SUBQUERY 1
    1|0|0|SCAN TABLE t1 BY COVERING INDEX i2 (a=?) (~10 rows)
    0|0|0|EXECUTE CORRELATED SCALAR SUBQUERY 2
    2|0|0|SCAN TABLE t1 BY INDEX i3 (b=?) (~10 rows)
</pre>

  stores the results in a temporary table. It then uses the contents of the 
  temporary table in place of the subquery to execute the parent query. This
  is shown in the output of EXPLAIN QUERY PLAN by substituting a 
  "SCAN SUBQUERY" record for the "SCAN TABLE" record that normally appears
  for each element in the FROM clause. For example:

<pre>
    sqlite&gt; EXPLAIN QUERY PLAN SELECT count(*) FROM (SELECT max(b) AS x FROM t1 GROUP BY a) GROUP BY x;
    1|0|0|SCAN TABLE t1 BY COVERING INDEX i2 (~1000000 rows)
    0|0|0|SCAN SUBQUERY 1 (~1000000 rows)
    0|0|0|USE TEMP B-TREE FOR GROUP BY
</pre>

<p>
  If the [flattening optimization] is used on a subquery in the FROM clause
  of a SELECT statement, then the output of EXPLAIN QUERY PLAN reflects this.
  For example, in the following there is no "SCAN SUBQUERY" record even though
  there is a subquery in the FROM clause of the top level SELECT. Instead, since
  the flattening optimization does apply in this case, the EXPLAIN QUERY PLAN
  report shows that the top level query is implemented using a nested loop join
  of tables t1 and t2.

<pre>
    sqlite&gt; EXPLAIN QUERY PLAN SELECT * FROM (SELECT * FROM t2 WHERE c=1), t1;
    0|0|0|SCAN TABLE t2 BY INDEX i4 (c=?) (~10 rows)
    0|1|1|SCAN TABLE t1 (~1000000 rows)
</pre>

<h4>Compound Queries</h4>

<p>
  Each component query of a [compound query] (UNION, UNION ALL, EXCEPT or 
  INTERSECT) is assigned its own selectid and reported on separately. A
  single record is output for the parent (compound query) identifying the
  operation, and whether or not a temporary b-tree is used to implement
  it. For example:

<pre>
    sqlite&gt; EXPLAIN QUERY PLAN SELECT a FROM t1 UNION SELECT c FROM t2;
    1|0|0|SCAN TABLE t1 (~1000000 rows)
    2|0|0|SCAN TABLE t2 (~1000000 rows)
    0|0|0|COMPOUND SUBQUERIES 1 AND 2 USING TEMP B-TREE (UNION)
</pre>

<p>
  The "USING TEMP B-TREE" clause in the above output indicates that a 
  temporary b-tree structure is used to implement the UNION of the results
  of the two sub-selects. If the temporary b-tree were not required, as
  in the following example, the clause is not present.

<pre>
    sqlite&gt; EXPLAIN QUERY PLAN SELECT a FROM t1 EXCEPT SELECT d FROM t2 ORDER BY 1;
    1|0|0|SCAN TABLE t1 BY COVERING INDEX i2 (~1000000 rows)
    2|0|0|SCAN TABLE t2 (~1000000 rows)
    2|0|0|USE TEMP B-TREE FOR ORDER BY
    0|0|0|COMPOUND SUBQUERIES 1 AND 2 (EXCEPT)
</pre>


<tcl>
##############################################################################
Section expression expr {*expression {expression syntax}}

BubbleDiagram expr 1

DISTINCT are present, then the behaviour is as if ALL were specified. 
^If the simple SELECT is a SELECT DISTINCT, then duplicate rows are removed
from the set of result rows before it is returned. ^For the purposes of
detecting duplicate rows, two NULL values are considered to be equal. ^The
normal rules for selecting a collation sequence to compare text values with
apply.

<h3>Compound Select Statements
<tcl>hd_fragment compound</tcl>
<tcl>hd_keywords {compound select} {compound query}</tcl>
</h3>

<p>Two or more simple SELECT statements may be connected together to form
a compound SELECT using the UNION, UNION ALL, INTERSECT or EXCEPT operator.
^In a compound SELECT, all the constituent SELECTs must return the same 
number of result columns. ^As the components of a compound SELECT must
be simple SELECT statements, they may not contain ORDER BY or LIMIT clauses.
^ORDER BY and LIMIT clauses may only occur at the end of the entire compound