mirror of
https://github.com/sqlite/sqlite.git
synced 2025-11-11 01:42:22 +03:00
Never flatten a CTE that is labeled MATERIALIZED. This is really something
of a bug fix because the documentation says that a MATERIALIZED CTE is an optimization barrier. FossilOrigin-Name: b7ef9796f548ce43e06673f86bbec68157aae5e4ee8451e1d87ee5f70af1bb27
This commit is contained in:
14
src/select.c
14
src/select.c
@@ -7314,7 +7314,14 @@ int sqlite3Select(
|
||||
** (a) The outer query has a different ORDER BY clause
|
||||
** (b) The subquery is part of a join
|
||||
** See forum post 062d576715d277c8
|
||||
**
|
||||
** The above cases might get an incorrect result if ORDER BY is omitted.
|
||||
** The following constraints are not required for correct answers, but are
|
||||
** included in order to give developers more control over when a sort
|
||||
** occurs:
|
||||
**
|
||||
** (6) The subquery is really a MATERIALIZED CTE
|
||||
** (7) The OmitOrderBy optimization is disabled
|
||||
*/
|
||||
if( pSub->pOrderBy!=0
|
||||
&& (p->pOrderBy!=0 || pTabList->nSrc>1) /* Condition (5) */
|
||||
@@ -7322,7 +7329,7 @@ int sqlite3Select(
|
||||
&& (pSub->selFlags & SF_OrderByReqd)==0 /* Condition (2) */
|
||||
&& (p->selFlags & SF_OrderByReqd)==0 /* Condition (3) and (4) */
|
||||
&& (pItem->fg.isCte==0 || pItem->u2.pCteUse->eM10d!=M10d_Yes) /* (6) */
|
||||
&& OptimizationEnabled(db, SQLITE_OmitOrderBy)
|
||||
&& OptimizationEnabled(db, SQLITE_OmitOrderBy) /* (7) */
|
||||
){
|
||||
TREETRACE(0x800,pParse,p,
|
||||
("omit superfluous ORDER BY on %r FROM-clause subquery\n",i+1));
|
||||
@@ -7359,6 +7366,11 @@ int sqlite3Select(
|
||||
continue;
|
||||
}
|
||||
|
||||
/* Do not flatten a MATERIALIZED CTE */
|
||||
if( pItem->fg.isCte && pItem->u2.pCteUse->eM10d==M10d_Yes ){
|
||||
continue;
|
||||
}
|
||||
|
||||
if( flattenSubquery(pParse, p, i, isAgg) ){
|
||||
if( pParse->nErr ) goto select_end;
|
||||
/* This subquery can be absorbed into its parent. */
|
||||
|
||||
Reference in New Issue
Block a user