1996-07-09 06:22:35 +00:00
|
|
|
/*-------------------------------------------------------------------------
|
|
|
|
*
|
1999-02-13 23:22:53 +00:00
|
|
|
* planner.c
|
1997-09-07 05:04:48 +00:00
|
|
|
* The query optimizer external interface.
|
1996-07-09 06:22:35 +00:00
|
|
|
*
|
|
|
|
* Copyright (c) 1994, Regents of the University of California
|
|
|
|
*
|
|
|
|
*
|
|
|
|
* IDENTIFICATION
|
1999-07-15 22:40:16 +00:00
|
|
|
* $Header: /cvsroot/pgsql/src/backend/optimizer/plan/planner.c,v 1.59 1999/07/15 22:39:27 momjian Exp $
|
1996-07-09 06:22:35 +00:00
|
|
|
*
|
|
|
|
*-------------------------------------------------------------------------
|
|
|
|
*/
|
1996-10-31 10:59:42 +00:00
|
|
|
#include <sys/types.h>
|
1998-02-13 03:37:04 +00:00
|
|
|
#include <string.h>
|
1996-10-31 10:59:42 +00:00
|
|
|
|
1996-07-09 06:22:35 +00:00
|
|
|
#include "postgres.h"
|
|
|
|
|
1999-01-25 12:01:19 +00:00
|
|
|
#include "nodes/makefuncs.h"
|
|
|
|
#include "catalog/pg_type.h"
|
1997-11-25 22:07:18 +00:00
|
|
|
#include "parser/parse_expr.h"
|
1996-07-09 06:22:35 +00:00
|
|
|
|
|
|
|
#include "utils/lsyscache.h"
|
|
|
|
#include "access/heapam.h"
|
|
|
|
|
|
|
|
#include "optimizer/internal.h"
|
|
|
|
#include "optimizer/planner.h"
|
|
|
|
#include "optimizer/prep.h"
|
|
|
|
#include "optimizer/planmain.h"
|
1998-02-13 03:37:04 +00:00
|
|
|
#include "optimizer/subselect.h"
|
1996-07-09 06:22:35 +00:00
|
|
|
|
|
|
|
/* DATA STRUCTURE CREATION/MANIPULATION ROUTINES */
|
|
|
|
#include "optimizer/clauses.h"
|
|
|
|
#include "optimizer/tlist.h"
|
|
|
|
#include "optimizer/var.h"
|
|
|
|
|
|
|
|
#include "executor/executor.h"
|
|
|
|
|
1999-02-03 19:31:24 +00:00
|
|
|
#include "utils/builtins.h"
|
|
|
|
#include "utils/syscache.h"
|
|
|
|
#include "access/genam.h"
|
|
|
|
#include "parser/parse_oper.h"
|
|
|
|
|
1999-05-03 00:38:44 +00:00
|
|
|
static List *make_subplanTargetList(Query *parse, List *tlist,
|
1999-05-25 16:15:34 +00:00
|
|
|
AttrNumber **groupColIdx);
|
1999-05-03 00:38:44 +00:00
|
|
|
static Plan *make_groupplan(List *group_tlist, bool tuplePerGroup,
|
1999-05-25 16:15:34 +00:00
|
|
|
List *groupClause, AttrNumber *grpColIdx,
|
|
|
|
Plan *subplan);
|
1999-02-03 19:31:24 +00:00
|
|
|
static bool need_sortplan(List *sortcls, Plan *plan);
|
1997-09-08 21:56:23 +00:00
|
|
|
static Plan *make_sortplan(List *tlist, List *sortcls, Plan *plannode);
|
1996-07-09 06:22:35 +00:00
|
|
|
|
|
|
|
/*****************************************************************************
|
|
|
|
*
|
1997-09-07 05:04:48 +00:00
|
|
|
* Query optimizer entry point
|
|
|
|
*
|
1996-07-09 06:22:35 +00:00
|
|
|
*****************************************************************************/
|
1998-02-26 04:46:47 +00:00
|
|
|
Plan *
|
1998-02-13 03:37:04 +00:00
|
|
|
planner(Query *parse)
|
|
|
|
{
|
|
|
|
Plan *result_plan;
|
1998-02-26 04:46:47 +00:00
|
|
|
|
1999-06-21 01:20:57 +00:00
|
|
|
/* Initialize state for subselects */
|
1998-02-13 03:37:04 +00:00
|
|
|
PlannerQueryLevel = 1;
|
|
|
|
PlannerInitPlan = NULL;
|
1999-06-21 01:20:57 +00:00
|
|
|
PlannerParamVar = NULL;
|
1998-02-13 03:37:04 +00:00
|
|
|
PlannerPlanId = 0;
|
1998-02-26 04:46:47 +00:00
|
|
|
|
1998-09-03 02:34:35 +00:00
|
|
|
transformKeySetQuery(parse);
|
1998-02-26 04:46:47 +00:00
|
|
|
result_plan = union_planner(parse);
|
|
|
|
|
|
|
|
Assert(PlannerQueryLevel == 1);
|
|
|
|
if (PlannerPlanId > 0)
|
1998-02-13 03:37:04 +00:00
|
|
|
{
|
|
|
|
result_plan->initPlan = PlannerInitPlan;
|
1998-02-26 04:46:47 +00:00
|
|
|
(void) SS_finalize_plan(result_plan);
|
1998-02-13 03:37:04 +00:00
|
|
|
}
|
1998-02-26 04:46:47 +00:00
|
|
|
result_plan->nParamExec = length(PlannerParamVar);
|
|
|
|
|
1998-09-01 03:29:17 +00:00
|
|
|
return result_plan;
|
1998-02-13 03:37:04 +00:00
|
|
|
}
|
1996-07-09 06:22:35 +00:00
|
|
|
|
1997-09-07 05:04:48 +00:00
|
|
|
/*
|
1999-02-13 23:22:53 +00:00
|
|
|
* union_planner
|
1997-09-07 05:04:48 +00:00
|
|
|
*
|
|
|
|
* Invokes the planner on union queries if there are any left,
|
|
|
|
* recursing if necessary to get them all, then processes normal plans.
|
|
|
|
*
|
1996-07-09 06:22:35 +00:00
|
|
|
* Returns a query plan.
|
1997-09-07 05:04:48 +00:00
|
|
|
*
|
1996-07-09 06:22:35 +00:00
|
|
|
*/
|
1998-02-26 04:46:47 +00:00
|
|
|
Plan *
|
1998-02-13 03:37:04 +00:00
|
|
|
union_planner(Query *parse)
|
1996-07-09 06:22:35 +00:00
|
|
|
{
|
1997-09-08 02:41:22 +00:00
|
|
|
List *tlist = parse->targetList;
|
|
|
|
List *rangetable = parse->rtable;
|
|
|
|
Plan *result_plan = (Plan *) NULL;
|
1999-05-03 00:38:44 +00:00
|
|
|
AttrNumber *groupColIdx = NULL;
|
Hi!
INTERSECT and EXCEPT is available for postgresql-v6.4!
The patch against v6.4 is included at the end of the current text
(in uuencoded form!)
I also included the text of my Master's Thesis. (a postscript
version). I hope that you find something of it useful and would be
happy if parts of it find their way into the PostgreSQL documentation
project (If so, tell me, then I send the sources of the document!)
The contents of the document are:
-) The first chapter might be of less interest as it gives only an
overview on SQL.
-) The second chapter gives a description on much of PostgreSQL's
features (like user defined types etc. and how to use these features)
-) The third chapter starts with an overview of PostgreSQL's internal
structure with focus on the stages a query has to pass (i.e. parser,
planner/optimizer, executor). Then a detailed description of the
implementation of the Having clause and the Intersect/Except logic is
given.
Originally I worked on v6.3.2 but never found time enough to prepare
and post a patch. Now I applied the changes to v6.4 to get Intersect
and Except working with the new version. Chapter 3 of my documentation
deals with the changes against v6.3.2, so keep that in mind when
comparing the parts of the code printed there with the patched sources
of v6.4.
Here are some remarks on the patch. There are some things that have
still to be done but at the moment I don't have time to do them
myself. (I'm doing my military service at the moment) Sorry for that
:-(
-) I used a rewrite technique for the implementation of the Except/Intersect
logic which rewrites the query to a semantically equivalent query before
it is handed to the rewrite system (for views, rules etc.), planner,
executor etc.
-) In v6.3.2 the types of the attributes of two select statements
connected by the UNION keyword had to match 100%. In v6.4 the types
only need to be familiar (i.e. int and float can be mixed). Since this
feature did not exist when I worked on Intersect/Except it
does not work correctly for Except/Intersect queries WHEN USED IN
COMBINATION WITH UNIONS! (i.e. sometimes the wrong type is used for the
resulting table. This is because until now the types of the attributes of
the first select statement have been used for the resulting table.
When Intersects and/or Excepts are used in combination with Unions it
might happen, that the first select statement of the original query
appears at another position in the query which will be executed. The reason
for this is the technique used for the implementation of
Except/Intersect which does a query rewrite!)
NOTE: It is NOT broken for pure UNION queries and pure INTERSECT/EXCEPT
queries!!!
-) I had to add the field intersect_clause to some data structures
but did not find time to implement printfuncs for the new field.
This does NOT break the debug modes but when an Except/Intersect
is used the query debug output will be the already rewritten query.
-) Massive changes to the grammar rules for SELECT and INSERT statements
have been necessary (see comments in gram.y and documentation for
deatails) in order to be able to use mixed queries like
(SELECT ... UNION (SELECT ... EXCEPT SELECT)) INTERSECT SELECT...;
-) When using UNION/EXCEPT/INTERSECT you will get:
NOTICE: equal: "Don't know if nodes of type xxx are equal".
I did not have time to add comparsion support for all the needed nodes,
but the default behaviour of the function equal met my requirements.
I did not dare to supress this message!
That's the reason why the regression test for union will fail: These
messages are also included in the union.out file!
-) Somebody of you changed the union_planner() function for v6.4
(I copied the targetlist to new_tlist and that was removed and
replaced by a cleanup of the original targetlist). These chnages
violated some having queries executed against views so I changed
it back again. I did not have time to examine the differences between the
two versions but now it works :-)
If you want to find out, try the file queries/view_having.sql on
both versions and compare the results . Two queries won't produce a
correct result with your version.
regards
Stefan
1999-01-18 00:10:17 +00:00
|
|
|
Index rt_index;
|
1997-09-07 05:04:48 +00:00
|
|
|
|
1997-12-24 06:06:58 +00:00
|
|
|
if (parse->unionClause)
|
|
|
|
{
|
1999-05-25 16:15:34 +00:00
|
|
|
result_plan = (Plan *) plan_union_queries(parse);
|
|
|
|
/* XXX do we need to do this? bjm 12/19/97 */
|
|
|
|
tlist = preprocess_targetlist(tlist,
|
|
|
|
parse->commandType,
|
|
|
|
parse->resultRelation,
|
|
|
|
parse->rtable);
|
1997-12-24 06:06:58 +00:00
|
|
|
}
|
1999-02-03 21:18:02 +00:00
|
|
|
else if ((rt_index = first_inherit_rt_entry(rangetable)) != -1)
|
1997-09-07 05:04:48 +00:00
|
|
|
{
|
1999-06-06 17:38:11 +00:00
|
|
|
List *sub_tlist;
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Generate appropriate target list for subplan; may be different
|
|
|
|
* from tlist if grouping or aggregation is needed.
|
|
|
|
*/
|
|
|
|
sub_tlist = make_subplanTargetList(parse, tlist, &groupColIdx);
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Recursively plan the subqueries needed for inheritance
|
|
|
|
*/
|
|
|
|
result_plan = (Plan *) plan_inherit_queries(parse, sub_tlist,
|
|
|
|
rt_index);
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Fix up outer target list. NOTE: unlike the case for non-inherited
|
|
|
|
* query, we pass the unfixed tlist to subplans, which do their own
|
|
|
|
* fixing. But we still want to fix the outer target list afterwards.
|
|
|
|
* I *think* this is correct --- doing the fix before recursing is
|
|
|
|
* definitely wrong, because preprocess_targetlist() will do the
|
|
|
|
* wrong thing if invoked twice on the same list. Maybe that is a bug?
|
|
|
|
* tgl 6/6/99
|
|
|
|
*/
|
1997-12-20 07:59:44 +00:00
|
|
|
tlist = preprocess_targetlist(tlist,
|
1999-05-25 16:15:34 +00:00
|
|
|
parse->commandType,
|
|
|
|
parse->resultRelation,
|
|
|
|
parse->rtable);
|
1999-06-06 17:38:11 +00:00
|
|
|
|
|
|
|
if (parse->rowMark != NULL)
|
|
|
|
elog(ERROR, "SELECT FOR UPDATE is not supported for inherit queries");
|
1997-09-07 05:04:48 +00:00
|
|
|
}
|
|
|
|
else
|
1997-12-18 19:41:44 +00:00
|
|
|
{
|
1999-05-25 16:15:34 +00:00
|
|
|
List *sub_tlist;
|
1999-04-19 01:43:12 +00:00
|
|
|
|
1999-05-25 16:15:34 +00:00
|
|
|
/* Preprocess targetlist in case we are inside an INSERT/UPDATE. */
|
|
|
|
tlist = preprocess_targetlist(tlist,
|
|
|
|
parse->commandType,
|
|
|
|
parse->resultRelation,
|
|
|
|
parse->rtable);
|
1999-01-25 12:01:19 +00:00
|
|
|
|
1999-05-25 16:15:34 +00:00
|
|
|
/*
|
|
|
|
* Add row-mark targets for UPDATE (should this be done in
|
|
|
|
* preprocess_targetlist?)
|
|
|
|
*/
|
|
|
|
if (parse->rowMark != NULL)
|
|
|
|
{
|
|
|
|
List *l;
|
|
|
|
|
|
|
|
foreach(l, parse->rowMark)
|
|
|
|
{
|
|
|
|
RowMark *rowmark = (RowMark *) lfirst(l);
|
|
|
|
TargetEntry *ctid;
|
|
|
|
Resdom *resdom;
|
|
|
|
Var *var;
|
|
|
|
char *resname;
|
|
|
|
|
|
|
|
if (!(rowmark->info & ROW_MARK_FOR_UPDATE))
|
|
|
|
continue;
|
|
|
|
|
|
|
|
resname = (char *) palloc(32);
|
|
|
|
sprintf(resname, "ctid%u", rowmark->rti);
|
|
|
|
resdom = makeResdom(length(tlist) + 1,
|
|
|
|
TIDOID,
|
|
|
|
-1,
|
|
|
|
resname,
|
|
|
|
0,
|
|
|
|
0,
|
|
|
|
true);
|
|
|
|
|
|
|
|
var = makeVar(rowmark->rti, -1, TIDOID,
|
|
|
|
-1, 0, rowmark->rti, -1);
|
|
|
|
|
|
|
|
ctid = makeTargetEntry(resdom, (Node *) var);
|
|
|
|
tlist = lappend(tlist, ctid);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Generate appropriate target list for subplan; may be different
|
|
|
|
* from tlist if grouping or aggregation is needed.
|
|
|
|
*/
|
|
|
|
sub_tlist = make_subplanTargetList(parse, tlist, &groupColIdx);
|
1999-01-25 12:01:19 +00:00
|
|
|
|
1999-05-25 16:15:34 +00:00
|
|
|
/* Generate the (sub) plan */
|
|
|
|
result_plan = query_planner(parse,
|
|
|
|
parse->commandType,
|
|
|
|
sub_tlist,
|
|
|
|
(List *) parse->qual);
|
1997-12-18 19:41:44 +00:00
|
|
|
}
|
1999-05-25 16:15:34 +00:00
|
|
|
|
1999-06-12 19:27:41 +00:00
|
|
|
/* query_planner returns NULL if it thinks plan is bogus */
|
|
|
|
if (! result_plan)
|
|
|
|
elog(ERROR, "union_planner: failed to create plan");
|
|
|
|
|
1997-12-20 07:59:44 +00:00
|
|
|
/*
|
|
|
|
* If we have a GROUP BY clause, insert a group node (with the
|
|
|
|
* appropriate sort node.)
|
|
|
|
*/
|
1998-01-15 19:00:16 +00:00
|
|
|
if (parse->groupClause)
|
1997-12-20 07:59:44 +00:00
|
|
|
{
|
|
|
|
bool tuplePerGroup;
|
1999-05-03 00:38:44 +00:00
|
|
|
List *group_tlist;
|
1997-12-20 07:59:44 +00:00
|
|
|
|
|
|
|
/*
|
1999-05-03 00:38:44 +00:00
|
|
|
* Decide whether how many tuples per group the Group node needs
|
1997-12-20 07:59:44 +00:00
|
|
|
* to return. (Needs only one tuple per group if no aggregate is
|
|
|
|
* present. Otherwise, need every tuple from the group to do the
|
1999-05-03 00:38:44 +00:00
|
|
|
* aggregation.) Note tuplePerGroup is named backwards :-(
|
1997-12-20 07:59:44 +00:00
|
|
|
*/
|
1998-01-15 19:00:16 +00:00
|
|
|
tuplePerGroup = parse->hasAggs;
|
1997-12-20 07:59:44 +00:00
|
|
|
|
1999-05-25 16:15:34 +00:00
|
|
|
/*
|
|
|
|
* If there are aggregates then the Group node should just return
|
1999-05-03 00:38:44 +00:00
|
|
|
* the same (simplified) tlist as the subplan, which we indicate
|
|
|
|
* to make_groupplan by passing NIL. If there are no aggregates
|
|
|
|
* then the Group node had better compute the final tlist.
|
|
|
|
*/
|
|
|
|
group_tlist = parse->hasAggs ? NIL : tlist;
|
|
|
|
|
|
|
|
result_plan = make_groupplan(group_tlist,
|
|
|
|
tuplePerGroup,
|
|
|
|
parse->groupClause,
|
|
|
|
groupColIdx,
|
|
|
|
result_plan);
|
1997-12-20 07:59:44 +00:00
|
|
|
}
|
|
|
|
|
1999-04-19 01:43:12 +00:00
|
|
|
/*
|
|
|
|
* If we have a HAVING clause, do the necessary things with it.
|
|
|
|
*/
|
|
|
|
if (parse->havingQual)
|
|
|
|
{
|
|
|
|
/* convert the havingQual to conjunctive normal form (cnf) */
|
|
|
|
parse->havingQual = (Node *) cnfify((Expr *) parse->havingQual, true);
|
|
|
|
|
|
|
|
if (parse->hasSubLinks)
|
|
|
|
{
|
1999-05-25 16:15:34 +00:00
|
|
|
/*
|
1999-06-21 01:20:57 +00:00
|
|
|
* There may be a subselect in the havingQual, so we have to
|
1999-05-25 16:15:34 +00:00
|
|
|
* process it using the same function as for a subselect in
|
|
|
|
* 'where'
|
1999-04-19 01:43:12 +00:00
|
|
|
*/
|
1999-06-21 01:20:57 +00:00
|
|
|
parse->havingQual = SS_process_sublinks(parse->havingQual);
|
1999-05-25 16:15:34 +00:00
|
|
|
|
|
|
|
/*
|
|
|
|
* Check for ungrouped variables passed to subplans. (Probably
|
1999-06-21 01:20:57 +00:00
|
|
|
* this should be done for the targetlist as well???)
|
1999-04-19 01:43:12 +00:00
|
|
|
*/
|
|
|
|
check_having_for_ungrouped_vars(parse->havingQual,
|
1999-05-12 15:02:39 +00:00
|
|
|
parse->groupClause,
|
|
|
|
parse->targetList);
|
1999-04-19 01:43:12 +00:00
|
|
|
}
|
|
|
|
|
|
|
|
/* Calculate the opfids from the opnos */
|
|
|
|
parse->havingQual = (Node *) fix_opids((List *) parse->havingQual);
|
|
|
|
}
|
|
|
|
|
1997-12-20 07:59:44 +00:00
|
|
|
/*
|
|
|
|
* If aggregate is present, insert the agg node
|
|
|
|
*/
|
1998-01-15 19:00:16 +00:00
|
|
|
if (parse->hasAggs)
|
1997-12-20 07:59:44 +00:00
|
|
|
{
|
1998-02-26 04:46:47 +00:00
|
|
|
result_plan = (Plan *) make_agg(tlist, result_plan);
|
1997-12-20 07:59:44 +00:00
|
|
|
|
1999-04-19 01:43:12 +00:00
|
|
|
/* HAVING clause, if any, becomes qual of the Agg node */
|
|
|
|
result_plan->qual = (List *) parse->havingQual;
|
|
|
|
|
1997-12-20 07:59:44 +00:00
|
|
|
/*
|
1999-05-25 16:15:34 +00:00
|
|
|
* Update vars to refer to subplan result tuples, find Aggrefs,
|
|
|
|
* make sure there is an Aggref in every HAVING clause.
|
1997-12-20 07:59:44 +00:00
|
|
|
*/
|
1999-05-25 16:15:34 +00:00
|
|
|
if (!set_agg_tlist_references((Agg *) result_plan))
|
1999-04-19 01:43:12 +00:00
|
|
|
elog(ERROR, "SELECT/HAVING requires aggregates to be valid");
|
1999-05-03 00:38:44 +00:00
|
|
|
|
|
|
|
/*
|
|
|
|
* Check that we actually found some aggregates, else executor
|
1999-06-21 01:20:57 +00:00
|
|
|
* will die unpleasantly. (This defends against possible bugs in
|
|
|
|
* parser or rewrite that might cause hasAggs to be incorrectly
|
|
|
|
* set 'true'. It's not easy to recover here, since we've already
|
|
|
|
* made decisions assuming there will be an Agg node.)
|
1999-05-03 00:38:44 +00:00
|
|
|
*/
|
|
|
|
if (((Agg *) result_plan)->aggs == NIL)
|
|
|
|
elog(ERROR, "union_planner: query is marked hasAggs, but I don't see any");
|
1999-05-25 16:15:34 +00:00
|
|
|
}
|
1999-01-25 12:01:19 +00:00
|
|
|
|
1997-09-07 05:04:48 +00:00
|
|
|
/*
|
|
|
|
* For now, before we hand back the plan, check to see if there is a
|
|
|
|
* user-specified sort that needs to be done. Eventually, this will
|
|
|
|
* be moved into the guts of the planner s.t. user specified sorts
|
|
|
|
* will be considered as part of the planning process. Since we can
|
|
|
|
* only make use of user-specified sorts in special cases, we can do
|
|
|
|
* the optimization step later.
|
|
|
|
*/
|
|
|
|
|
1997-12-27 06:41:41 +00:00
|
|
|
if (parse->uniqueFlag)
|
1997-09-07 05:04:48 +00:00
|
|
|
{
|
1997-12-27 06:41:41 +00:00
|
|
|
Plan *sortplan = make_sortplan(tlist, parse->sortClause, result_plan);
|
1997-09-07 05:04:48 +00:00
|
|
|
|
Hi!
INTERSECT and EXCEPT is available for postgresql-v6.4!
The patch against v6.4 is included at the end of the current text
(in uuencoded form!)
I also included the text of my Master's Thesis. (a postscript
version). I hope that you find something of it useful and would be
happy if parts of it find their way into the PostgreSQL documentation
project (If so, tell me, then I send the sources of the document!)
The contents of the document are:
-) The first chapter might be of less interest as it gives only an
overview on SQL.
-) The second chapter gives a description on much of PostgreSQL's
features (like user defined types etc. and how to use these features)
-) The third chapter starts with an overview of PostgreSQL's internal
structure with focus on the stages a query has to pass (i.e. parser,
planner/optimizer, executor). Then a detailed description of the
implementation of the Having clause and the Intersect/Except logic is
given.
Originally I worked on v6.3.2 but never found time enough to prepare
and post a patch. Now I applied the changes to v6.4 to get Intersect
and Except working with the new version. Chapter 3 of my documentation
deals with the changes against v6.3.2, so keep that in mind when
comparing the parts of the code printed there with the patched sources
of v6.4.
Here are some remarks on the patch. There are some things that have
still to be done but at the moment I don't have time to do them
myself. (I'm doing my military service at the moment) Sorry for that
:-(
-) I used a rewrite technique for the implementation of the Except/Intersect
logic which rewrites the query to a semantically equivalent query before
it is handed to the rewrite system (for views, rules etc.), planner,
executor etc.
-) In v6.3.2 the types of the attributes of two select statements
connected by the UNION keyword had to match 100%. In v6.4 the types
only need to be familiar (i.e. int and float can be mixed). Since this
feature did not exist when I worked on Intersect/Except it
does not work correctly for Except/Intersect queries WHEN USED IN
COMBINATION WITH UNIONS! (i.e. sometimes the wrong type is used for the
resulting table. This is because until now the types of the attributes of
the first select statement have been used for the resulting table.
When Intersects and/or Excepts are used in combination with Unions it
might happen, that the first select statement of the original query
appears at another position in the query which will be executed. The reason
for this is the technique used for the implementation of
Except/Intersect which does a query rewrite!)
NOTE: It is NOT broken for pure UNION queries and pure INTERSECT/EXCEPT
queries!!!
-) I had to add the field intersect_clause to some data structures
but did not find time to implement printfuncs for the new field.
This does NOT break the debug modes but when an Except/Intersect
is used the query debug output will be the already rewritten query.
-) Massive changes to the grammar rules for SELECT and INSERT statements
have been necessary (see comments in gram.y and documentation for
deatails) in order to be able to use mixed queries like
(SELECT ... UNION (SELECT ... EXCEPT SELECT)) INTERSECT SELECT...;
-) When using UNION/EXCEPT/INTERSECT you will get:
NOTICE: equal: "Don't know if nodes of type xxx are equal".
I did not have time to add comparsion support for all the needed nodes,
but the default behaviour of the function equal met my requirements.
I did not dare to supress this message!
That's the reason why the regression test for union will fail: These
messages are also included in the union.out file!
-) Somebody of you changed the union_planner() function for v6.4
(I copied the targetlist to new_tlist and that was removed and
replaced by a cleanup of the original targetlist). These chnages
violated some having queries executed against views so I changed
it back again. I did not have time to examine the differences between the
two versions but now it works :-)
If you want to find out, try the file queries/view_having.sql on
both versions and compare the results . Two queries won't produce a
correct result with your version.
regards
Stefan
1999-01-18 00:10:17 +00:00
|
|
|
return ((Plan *) make_unique(tlist, sortplan, parse->uniqueFlag));
|
1997-09-07 05:04:48 +00:00
|
|
|
}
|
|
|
|
else
|
|
|
|
{
|
1999-02-03 19:31:24 +00:00
|
|
|
if (parse->sortClause && need_sortplan(parse->sortClause, result_plan))
|
Hi!
INTERSECT and EXCEPT is available for postgresql-v6.4!
The patch against v6.4 is included at the end of the current text
(in uuencoded form!)
I also included the text of my Master's Thesis. (a postscript
version). I hope that you find something of it useful and would be
happy if parts of it find their way into the PostgreSQL documentation
project (If so, tell me, then I send the sources of the document!)
The contents of the document are:
-) The first chapter might be of less interest as it gives only an
overview on SQL.
-) The second chapter gives a description on much of PostgreSQL's
features (like user defined types etc. and how to use these features)
-) The third chapter starts with an overview of PostgreSQL's internal
structure with focus on the stages a query has to pass (i.e. parser,
planner/optimizer, executor). Then a detailed description of the
implementation of the Having clause and the Intersect/Except logic is
given.
Originally I worked on v6.3.2 but never found time enough to prepare
and post a patch. Now I applied the changes to v6.4 to get Intersect
and Except working with the new version. Chapter 3 of my documentation
deals with the changes against v6.3.2, so keep that in mind when
comparing the parts of the code printed there with the patched sources
of v6.4.
Here are some remarks on the patch. There are some things that have
still to be done but at the moment I don't have time to do them
myself. (I'm doing my military service at the moment) Sorry for that
:-(
-) I used a rewrite technique for the implementation of the Except/Intersect
logic which rewrites the query to a semantically equivalent query before
it is handed to the rewrite system (for views, rules etc.), planner,
executor etc.
-) In v6.3.2 the types of the attributes of two select statements
connected by the UNION keyword had to match 100%. In v6.4 the types
only need to be familiar (i.e. int and float can be mixed). Since this
feature did not exist when I worked on Intersect/Except it
does not work correctly for Except/Intersect queries WHEN USED IN
COMBINATION WITH UNIONS! (i.e. sometimes the wrong type is used for the
resulting table. This is because until now the types of the attributes of
the first select statement have been used for the resulting table.
When Intersects and/or Excepts are used in combination with Unions it
might happen, that the first select statement of the original query
appears at another position in the query which will be executed. The reason
for this is the technique used for the implementation of
Except/Intersect which does a query rewrite!)
NOTE: It is NOT broken for pure UNION queries and pure INTERSECT/EXCEPT
queries!!!
-) I had to add the field intersect_clause to some data structures
but did not find time to implement printfuncs for the new field.
This does NOT break the debug modes but when an Except/Intersect
is used the query debug output will be the already rewritten query.
-) Massive changes to the grammar rules for SELECT and INSERT statements
have been necessary (see comments in gram.y and documentation for
deatails) in order to be able to use mixed queries like
(SELECT ... UNION (SELECT ... EXCEPT SELECT)) INTERSECT SELECT...;
-) When using UNION/EXCEPT/INTERSECT you will get:
NOTICE: equal: "Don't know if nodes of type xxx are equal".
I did not have time to add comparsion support for all the needed nodes,
but the default behaviour of the function equal met my requirements.
I did not dare to supress this message!
That's the reason why the regression test for union will fail: These
messages are also included in the union.out file!
-) Somebody of you changed the union_planner() function for v6.4
(I copied the targetlist to new_tlist and that was removed and
replaced by a cleanup of the original targetlist). These chnages
violated some having queries executed against views so I changed
it back again. I did not have time to examine the differences between the
two versions but now it works :-)
If you want to find out, try the file queries/view_having.sql on
both versions and compare the results . Two queries won't produce a
correct result with your version.
regards
Stefan
1999-01-18 00:10:17 +00:00
|
|
|
return (make_sortplan(tlist, parse->sortClause, result_plan));
|
1997-09-07 05:04:48 +00:00
|
|
|
else
|
Hi!
INTERSECT and EXCEPT is available for postgresql-v6.4!
The patch against v6.4 is included at the end of the current text
(in uuencoded form!)
I also included the text of my Master's Thesis. (a postscript
version). I hope that you find something of it useful and would be
happy if parts of it find their way into the PostgreSQL documentation
project (If so, tell me, then I send the sources of the document!)
The contents of the document are:
-) The first chapter might be of less interest as it gives only an
overview on SQL.
-) The second chapter gives a description on much of PostgreSQL's
features (like user defined types etc. and how to use these features)
-) The third chapter starts with an overview of PostgreSQL's internal
structure with focus on the stages a query has to pass (i.e. parser,
planner/optimizer, executor). Then a detailed description of the
implementation of the Having clause and the Intersect/Except logic is
given.
Originally I worked on v6.3.2 but never found time enough to prepare
and post a patch. Now I applied the changes to v6.4 to get Intersect
and Except working with the new version. Chapter 3 of my documentation
deals with the changes against v6.3.2, so keep that in mind when
comparing the parts of the code printed there with the patched sources
of v6.4.
Here are some remarks on the patch. There are some things that have
still to be done but at the moment I don't have time to do them
myself. (I'm doing my military service at the moment) Sorry for that
:-(
-) I used a rewrite technique for the implementation of the Except/Intersect
logic which rewrites the query to a semantically equivalent query before
it is handed to the rewrite system (for views, rules etc.), planner,
executor etc.
-) In v6.3.2 the types of the attributes of two select statements
connected by the UNION keyword had to match 100%. In v6.4 the types
only need to be familiar (i.e. int and float can be mixed). Since this
feature did not exist when I worked on Intersect/Except it
does not work correctly for Except/Intersect queries WHEN USED IN
COMBINATION WITH UNIONS! (i.e. sometimes the wrong type is used for the
resulting table. This is because until now the types of the attributes of
the first select statement have been used for the resulting table.
When Intersects and/or Excepts are used in combination with Unions it
might happen, that the first select statement of the original query
appears at another position in the query which will be executed. The reason
for this is the technique used for the implementation of
Except/Intersect which does a query rewrite!)
NOTE: It is NOT broken for pure UNION queries and pure INTERSECT/EXCEPT
queries!!!
-) I had to add the field intersect_clause to some data structures
but did not find time to implement printfuncs for the new field.
This does NOT break the debug modes but when an Except/Intersect
is used the query debug output will be the already rewritten query.
-) Massive changes to the grammar rules for SELECT and INSERT statements
have been necessary (see comments in gram.y and documentation for
deatails) in order to be able to use mixed queries like
(SELECT ... UNION (SELECT ... EXCEPT SELECT)) INTERSECT SELECT...;
-) When using UNION/EXCEPT/INTERSECT you will get:
NOTICE: equal: "Don't know if nodes of type xxx are equal".
I did not have time to add comparsion support for all the needed nodes,
but the default behaviour of the function equal met my requirements.
I did not dare to supress this message!
That's the reason why the regression test for union will fail: These
messages are also included in the union.out file!
-) Somebody of you changed the union_planner() function for v6.4
(I copied the targetlist to new_tlist and that was removed and
replaced by a cleanup of the original targetlist). These chnages
violated some having queries executed against views so I changed
it back again. I did not have time to examine the differences between the
two versions but now it works :-)
If you want to find out, try the file queries/view_having.sql on
both versions and compare the results . Two queries won't produce a
correct result with your version.
regards
Stefan
1999-01-18 00:10:17 +00:00
|
|
|
return ((Plan *) result_plan);
|
1997-09-07 05:04:48 +00:00
|
|
|
}
|
1996-07-09 06:22:35 +00:00
|
|
|
|
|
|
|
}
|
|
|
|
|
1999-05-03 00:38:44 +00:00
|
|
|
/*---------------
|
|
|
|
* make_subplanTargetList
|
|
|
|
* Generate appropriate target lists when grouping is required.
|
|
|
|
*
|
|
|
|
* When union_planner inserts Aggregate and/or Group/Sort plan nodes above
|
|
|
|
* the result of query_planner, we typically need to pass a different
|
|
|
|
* target list to query_planner than the outer plan nodes should have.
|
|
|
|
* This routine generates the correct target list for the subplan, and
|
|
|
|
* if necessary modifies the target list for the inserted nodes as well.
|
|
|
|
*
|
|
|
|
* The initial target list passed from the parser already contains entries
|
|
|
|
* for all ORDER BY and GROUP BY expressions, but it will not have entries
|
|
|
|
* for variables used only in HAVING clauses; so we need to add those
|
|
|
|
* variables to the subplan target list. Also, if we are doing either
|
|
|
|
* grouping or aggregation, we flatten all expressions except GROUP BY items
|
|
|
|
* into their component variables; the other expressions will be computed by
|
|
|
|
* the inserted nodes rather than by the subplan. For example,
|
|
|
|
* given a query like
|
|
|
|
* SELECT a+b,SUM(c+d) FROM table GROUP BY a+b;
|
|
|
|
* we want to pass this targetlist to the subplan:
|
|
|
|
* a+b,c,d
|
|
|
|
* where the a+b target will be used by the Sort/Group steps, and the
|
|
|
|
* c and d targets will be needed to compute the aggregate results.
|
|
|
|
*
|
|
|
|
* 'parse' is the query being processed.
|
|
|
|
* 'tlist' is the query's target list. CAUTION: list elements may be
|
|
|
|
* modified by this routine!
|
|
|
|
* 'groupColIdx' receives an array of column numbers for the GROUP BY
|
|
|
|
* expressions (if there are any) in the subplan's target list.
|
|
|
|
*
|
|
|
|
* The result is the targetlist to be passed to the subplan. Also,
|
|
|
|
* the parent tlist is modified so that any nontrivial targetlist items that
|
|
|
|
* exactly match GROUP BY items are replaced by simple Var nodes referencing
|
|
|
|
* those outputs of the subplan. This avoids redundant recalculations in
|
|
|
|
* cases like
|
|
|
|
* SELECT a+1, ... GROUP BY a+1
|
|
|
|
* Note, however, that other varnodes in the parent's targetlist (and
|
|
|
|
* havingQual, if any) will still need to be updated to refer to outputs
|
1999-05-25 16:15:34 +00:00
|
|
|
* of the subplan. This routine is quite large enough already, so we do
|
1999-05-03 00:38:44 +00:00
|
|
|
* that later.
|
|
|
|
*---------------
|
|
|
|
*/
|
|
|
|
static List *
|
|
|
|
make_subplanTargetList(Query *parse,
|
|
|
|
List *tlist,
|
|
|
|
AttrNumber **groupColIdx)
|
|
|
|
{
|
|
|
|
List *sub_tlist;
|
|
|
|
List *prnt_tlist;
|
|
|
|
List *sl,
|
|
|
|
*gl;
|
|
|
|
List *glc = NIL;
|
|
|
|
List *extravars = NIL;
|
|
|
|
int numCols;
|
|
|
|
AttrNumber *grpColIdx = NULL;
|
|
|
|
int next_resno = 1;
|
|
|
|
|
|
|
|
*groupColIdx = NULL;
|
|
|
|
|
1999-05-25 16:15:34 +00:00
|
|
|
/*
|
|
|
|
* If we're not grouping or aggregating, nothing to do here;
|
1999-05-03 00:38:44 +00:00
|
|
|
* query_planner should receive the unmodified target list.
|
|
|
|
*/
|
|
|
|
if (!parse->hasAggs && !parse->groupClause && !parse->havingQual)
|
|
|
|
return tlist;
|
|
|
|
|
1999-05-25 16:15:34 +00:00
|
|
|
/*
|
|
|
|
* If grouping, make a working copy of groupClause list (which we use
|
1999-05-03 00:38:44 +00:00
|
|
|
* just to verify that we found all the groupClause items in tlist).
|
|
|
|
* Also allocate space to remember where the group columns are in the
|
|
|
|
* subplan tlist.
|
|
|
|
*/
|
|
|
|
numCols = length(parse->groupClause);
|
|
|
|
if (numCols > 0)
|
|
|
|
{
|
|
|
|
glc = listCopy(parse->groupClause);
|
|
|
|
grpColIdx = (AttrNumber *) palloc(sizeof(AttrNumber) * numCols);
|
|
|
|
*groupColIdx = grpColIdx;
|
|
|
|
}
|
|
|
|
|
1999-05-25 16:15:34 +00:00
|
|
|
sub_tlist = new_unsorted_tlist(tlist); /* make a modifiable copy */
|
1999-05-03 00:38:44 +00:00
|
|
|
|
|
|
|
/*
|
|
|
|
* Step 1: build grpColIdx by finding targetlist items that match
|
|
|
|
* GroupBy entries. If there are aggregates, remove non-GroupBy items
|
|
|
|
* from sub_tlist, and reset its resnos accordingly. When we leave an
|
1999-05-25 16:15:34 +00:00
|
|
|
* expression in the subplan tlist, modify the parent tlist to copy
|
|
|
|
* the value from the subplan output rather than re-evaluating it.
|
1999-05-03 00:38:44 +00:00
|
|
|
*/
|
|
|
|
prnt_tlist = tlist; /* scans parent tlist in sync with sl */
|
|
|
|
foreach(sl, sub_tlist)
|
|
|
|
{
|
|
|
|
TargetEntry *te = (TargetEntry *) lfirst(sl);
|
|
|
|
TargetEntry *parentte = (TargetEntry *) lfirst(prnt_tlist);
|
|
|
|
Resdom *resdom = te->resdom;
|
|
|
|
bool keepInSubPlan = true;
|
|
|
|
bool foundGroupClause = false;
|
|
|
|
int keyno = 0;
|
|
|
|
|
|
|
|
foreach(gl, parse->groupClause)
|
|
|
|
{
|
|
|
|
GroupClause *grpcl = (GroupClause *) lfirst(gl);
|
|
|
|
|
|
|
|
keyno++; /* sort key # for this GroupClause */
|
1999-05-12 15:02:39 +00:00
|
|
|
if (grpcl->tleGroupref == resdom->resgroupref)
|
1999-05-03 00:38:44 +00:00
|
|
|
{
|
|
|
|
/* Found a matching groupclause; record info for sorting */
|
|
|
|
foundGroupClause = true;
|
|
|
|
resdom->reskey = keyno;
|
|
|
|
resdom->reskeyop = get_opcode(grpcl->grpOpoid);
|
|
|
|
grpColIdx[keyno - 1] = next_resno;
|
1999-05-25 16:15:34 +00:00
|
|
|
|
|
|
|
/*
|
|
|
|
* Remove groupclause from our list of unmatched
|
|
|
|
* groupclauses. NB: this depends on having used a shallow
|
|
|
|
* listCopy() above.
|
1999-05-03 00:38:44 +00:00
|
|
|
*/
|
1999-05-25 16:15:34 +00:00
|
|
|
glc = lremove((void *) grpcl, glc);
|
1999-05-03 00:38:44 +00:00
|
|
|
break;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
1999-05-25 16:15:34 +00:00
|
|
|
if (!foundGroupClause)
|
1999-05-03 00:38:44 +00:00
|
|
|
{
|
1999-05-25 16:15:34 +00:00
|
|
|
|
1999-05-03 00:38:44 +00:00
|
|
|
/*
|
|
|
|
* Non-GroupBy entry: remove it from subplan if there are
|
1999-05-25 16:15:34 +00:00
|
|
|
* aggregates in query - it will be evaluated by Aggregate
|
|
|
|
* plan. But do not remove simple-Var entries; we'd just have
|
|
|
|
* to add them back anyway, and we risk confusing
|
|
|
|
* INSERT/UPDATE.
|
1999-05-03 00:38:44 +00:00
|
|
|
*/
|
1999-05-25 16:15:34 +00:00
|
|
|
if (parse->hasAggs && !IsA(te->expr, Var))
|
1999-05-03 00:38:44 +00:00
|
|
|
keepInSubPlan = false;
|
|
|
|
}
|
|
|
|
|
|
|
|
if (keepInSubPlan)
|
|
|
|
{
|
|
|
|
/* Assign new sequential resnos to subplan tlist items */
|
|
|
|
resdom->resno = next_resno++;
|
1999-05-25 16:15:34 +00:00
|
|
|
if (!IsA(parentte->expr, Var))
|
1999-05-03 00:38:44 +00:00
|
|
|
{
|
1999-05-25 16:15:34 +00:00
|
|
|
|
|
|
|
/*
|
|
|
|
* Since the item is being computed in the subplan, we can
|
|
|
|
* just make a Var node to reference it in the outer plan,
|
|
|
|
* rather than recomputing it there. Note we use varnoold
|
|
|
|
* = -1 as a flag to let replace_vars_with_subplan_refs
|
|
|
|
* know it needn't change this Var node. If it's only a
|
|
|
|
* Var anyway, we leave it alone for now;
|
1999-05-03 00:38:44 +00:00
|
|
|
* replace_vars_with_subplan_refs will fix it later.
|
|
|
|
*/
|
|
|
|
parentte->expr = (Node *) makeVar(1, resdom->resno,
|
|
|
|
resdom->restype,
|
|
|
|
resdom->restypmod,
|
|
|
|
0, -1, resdom->resno);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
else
|
|
|
|
{
|
1999-05-25 16:15:34 +00:00
|
|
|
|
|
|
|
/*
|
|
|
|
* Remove this tlist item from the subplan, but remember the
|
|
|
|
* vars it needs. The outer tlist item probably needs
|
|
|
|
* changes, but that will happen later.
|
1999-05-03 00:38:44 +00:00
|
|
|
*/
|
|
|
|
sub_tlist = lremove(te, sub_tlist);
|
|
|
|
extravars = nconc(extravars, pull_var_clause(te->expr));
|
|
|
|
}
|
|
|
|
|
|
|
|
prnt_tlist = lnext(prnt_tlist);
|
|
|
|
}
|
|
|
|
|
|
|
|
/* We should have found all the GROUP BY clauses in the tlist. */
|
|
|
|
if (length(glc) != 0)
|
|
|
|
elog(ERROR, "make_subplanTargetList: GROUP BY attribute not found in target list");
|
|
|
|
|
|
|
|
/*
|
1999-05-25 16:15:34 +00:00
|
|
|
* Add subplan targets for any variables needed by removed tlist
|
|
|
|
* entries that aren't otherwise mentioned in the subplan target list.
|
1999-05-03 00:38:44 +00:00
|
|
|
* We'll also need targets for any variables seen only in HAVING.
|
|
|
|
*/
|
|
|
|
extravars = nconc(extravars, pull_var_clause(parse->havingQual));
|
|
|
|
|
|
|
|
foreach(gl, extravars)
|
|
|
|
{
|
|
|
|
Var *v = (Var *) lfirst(gl);
|
|
|
|
|
|
|
|
if (tlist_member(v, sub_tlist) == NULL)
|
|
|
|
{
|
1999-05-25 16:15:34 +00:00
|
|
|
|
|
|
|
/*
|
|
|
|
* Make sure sub_tlist element is a fresh object not shared
|
|
|
|
* with any other structure; not sure if anything will break
|
|
|
|
* if it is shared, but better to be safe...
|
1999-05-04 00:00:20 +00:00
|
|
|
*/
|
1999-05-03 00:38:44 +00:00
|
|
|
sub_tlist = lappend(sub_tlist,
|
1999-05-04 00:00:20 +00:00
|
|
|
create_tl_element((Var *) copyObject(v),
|
|
|
|
next_resno));
|
1999-05-03 00:38:44 +00:00
|
|
|
next_resno++;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
return sub_tlist;
|
|
|
|
}
|
|
|
|
|
|
|
|
static Plan *
|
|
|
|
make_groupplan(List *group_tlist,
|
|
|
|
bool tuplePerGroup,
|
|
|
|
List *groupClause,
|
|
|
|
AttrNumber *grpColIdx,
|
|
|
|
Plan *subplan)
|
|
|
|
{
|
|
|
|
List *sort_tlist;
|
|
|
|
List *sl;
|
|
|
|
Sort *sortplan;
|
|
|
|
Group *grpplan;
|
|
|
|
int numCols = length(groupClause);
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Make the targetlist for the Sort node; it always just references
|
|
|
|
* each of the corresponding target items of the subplan. We need to
|
1999-05-25 16:15:34 +00:00
|
|
|
* ensure that simple Vars in the subplan's target list are
|
|
|
|
* recognizable by replace_vars_with_subplan_refs when it's applied to
|
|
|
|
* the Sort/Group target list, so copy up their varnoold/varoattno.
|
1999-05-03 00:38:44 +00:00
|
|
|
*/
|
|
|
|
sort_tlist = NIL;
|
|
|
|
foreach(sl, subplan->targetlist)
|
|
|
|
{
|
|
|
|
TargetEntry *te = (TargetEntry *) lfirst(sl);
|
|
|
|
Resdom *resdom = te->resdom;
|
|
|
|
Var *newvar;
|
|
|
|
|
|
|
|
if (IsA(te->expr, Var))
|
|
|
|
{
|
1999-05-25 16:15:34 +00:00
|
|
|
Var *subvar = (Var *) te->expr;
|
|
|
|
|
1999-05-03 00:38:44 +00:00
|
|
|
newvar = makeVar(1, resdom->resno,
|
|
|
|
resdom->restype, resdom->restypmod,
|
|
|
|
0, subvar->varnoold, subvar->varoattno);
|
|
|
|
}
|
|
|
|
else
|
|
|
|
{
|
|
|
|
newvar = makeVar(1, resdom->resno,
|
|
|
|
resdom->restype, resdom->restypmod,
|
|
|
|
0, -1, resdom->resno);
|
|
|
|
}
|
|
|
|
|
|
|
|
sort_tlist = lappend(sort_tlist,
|
1999-05-25 16:15:34 +00:00
|
|
|
makeTargetEntry((Resdom *) copyObject(resdom),
|
|
|
|
(Node *) newvar));
|
1999-05-03 00:38:44 +00:00
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Make the Sort node
|
|
|
|
*/
|
|
|
|
sortplan = make_sort(sort_tlist,
|
|
|
|
_NONAME_RELATION_ID_,
|
|
|
|
subplan,
|
|
|
|
numCols);
|
|
|
|
sortplan->plan.cost = subplan->cost; /* XXX assume no cost */
|
|
|
|
|
|
|
|
/*
|
1999-05-25 16:15:34 +00:00
|
|
|
* If the caller gave us a target list, use it after fixing the
|
|
|
|
* variables. If not, we need the same sort of "repeater" tlist as for
|
|
|
|
* the Sort node.
|
1999-05-03 00:38:44 +00:00
|
|
|
*/
|
|
|
|
if (group_tlist)
|
|
|
|
{
|
1999-05-25 16:15:34 +00:00
|
|
|
group_tlist = copyObject(group_tlist); /* necessary?? */
|
1999-05-03 00:38:44 +00:00
|
|
|
replace_tlist_with_subplan_refs(group_tlist,
|
|
|
|
(Index) 0,
|
|
|
|
subplan->targetlist);
|
|
|
|
}
|
|
|
|
else
|
|
|
|
group_tlist = copyObject(sort_tlist);
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Make the Group node
|
|
|
|
*/
|
|
|
|
grpplan = make_group(group_tlist, tuplePerGroup, numCols,
|
|
|
|
grpColIdx, sortplan);
|
|
|
|
|
|
|
|
return (Plan *) grpplan;
|
|
|
|
}
|
|
|
|
|
1996-07-09 06:22:35 +00:00
|
|
|
/*
|
1999-02-13 23:22:53 +00:00
|
|
|
* make_sortplan
|
1997-09-07 05:04:48 +00:00
|
|
|
* Returns a sortplan which is basically a SORT node attached to the
|
|
|
|
* top of the plan returned from the planner. It also adds the
|
|
|
|
* cost of sorting into the plan.
|
|
|
|
*
|
1996-07-09 06:22:35 +00:00
|
|
|
* sortkeys: ( resdom1 resdom2 resdom3 ...)
|
|
|
|
* sortops: (sortop1 sortop2 sortop3 ...)
|
|
|
|
*/
|
1997-09-08 02:41:22 +00:00
|
|
|
static Plan *
|
1997-09-08 21:56:23 +00:00
|
|
|
make_sortplan(List *tlist, List *sortcls, Plan *plannode)
|
1996-07-09 06:22:35 +00:00
|
|
|
{
|
1997-09-08 02:41:22 +00:00
|
|
|
Plan *sortplan = (Plan *) NULL;
|
|
|
|
List *temp_tlist = NIL;
|
|
|
|
List *i = NIL;
|
|
|
|
Resdom *resnode = (Resdom *) NULL;
|
|
|
|
Resdom *resdom = (Resdom *) NULL;
|
|
|
|
int keyno = 1;
|
1997-09-07 05:04:48 +00:00
|
|
|
|
|
|
|
/*
|
|
|
|
* First make a copy of the tlist so that we don't corrupt the the
|
|
|
|
* original .
|
|
|
|
*/
|
|
|
|
|
|
|
|
temp_tlist = new_unsorted_tlist(tlist);
|
|
|
|
|
|
|
|
foreach(i, sortcls)
|
|
|
|
{
|
1997-09-08 02:41:22 +00:00
|
|
|
SortClause *sortcl = (SortClause *) lfirst(i);
|
1997-09-07 05:04:48 +00:00
|
|
|
|
|
|
|
resnode = sortcl->resdom;
|
|
|
|
resdom = tlist_resdom(temp_tlist, resnode);
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Order the resdom keys and replace the operator OID for each key
|
|
|
|
* with the regproc OID.
|
|
|
|
*/
|
|
|
|
resdom->reskey = keyno;
|
|
|
|
resdom->reskeyop = get_opcode(sortcl->opoid);
|
|
|
|
keyno += 1;
|
|
|
|
}
|
|
|
|
|
|
|
|
sortplan = (Plan *) make_sort(temp_tlist,
|
1999-02-09 17:03:14 +00:00
|
|
|
_NONAME_RELATION_ID_,
|
1997-09-07 05:04:48 +00:00
|
|
|
(Plan *) plannode,
|
|
|
|
length(sortcls));
|
|
|
|
|
|
|
|
/*
|
|
|
|
* XXX Assuming that an internal sort has no. cost. This is wrong, but
|
|
|
|
* given that at this point, we don't know the no. of tuples returned,
|
|
|
|
* etc, we can't do better than to add a constant cost. This will be
|
|
|
|
* fixed once we move the sort further into the planner, but for now
|
|
|
|
* ... functionality....
|
1996-07-09 06:22:35 +00:00
|
|
|
*/
|
1997-09-07 05:04:48 +00:00
|
|
|
|
|
|
|
sortplan->cost = plannode->cost;
|
|
|
|
|
1998-09-01 03:29:17 +00:00
|
|
|
return sortplan;
|
1996-07-09 06:22:35 +00:00
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* pg_checkretval() -- check return value of a list of sql parse
|
1997-09-07 05:04:48 +00:00
|
|
|
* trees.
|
1996-07-09 06:22:35 +00:00
|
|
|
*
|
|
|
|
* The return value of a sql function is the value returned by
|
|
|
|
* the final query in the function. We do some ad-hoc define-time
|
|
|
|
* type checking here to be sure that the user is returning the
|
|
|
|
* type he claims.
|
1999-05-03 00:38:44 +00:00
|
|
|
*
|
|
|
|
* XXX Why is this function in this module?
|
1996-07-09 06:22:35 +00:00
|
|
|
*/
|
|
|
|
void
|
1999-05-13 07:29:22 +00:00
|
|
|
pg_checkretval(Oid rettype, List *queryTreeList)
|
1996-07-09 06:22:35 +00:00
|
|
|
{
|
1997-09-08 02:41:22 +00:00
|
|
|
Query *parse;
|
|
|
|
List *tlist;
|
|
|
|
List *rt;
|
|
|
|
int cmd;
|
|
|
|
Type typ;
|
|
|
|
Resdom *resnode;
|
|
|
|
Relation reln;
|
|
|
|
Oid relid;
|
|
|
|
Oid tletype;
|
|
|
|
int relnatts;
|
|
|
|
int i;
|
1997-09-07 05:04:48 +00:00
|
|
|
|
|
|
|
/* find the final query */
|
1999-05-25 16:15:34 +00:00
|
|
|
parse = (Query *) nth(length(queryTreeList) - 1, queryTreeList);
|
1997-09-07 05:04:48 +00:00
|
|
|
|
|
|
|
/*
|
|
|
|
* test 1: if the last query is a utility invocation, then there had
|
|
|
|
* better not be a return value declared.
|
|
|
|
*/
|
|
|
|
if (parse->commandType == CMD_UTILITY)
|
|
|
|
{
|
|
|
|
if (rettype == InvalidOid)
|
|
|
|
return;
|
|
|
|
else
|
1998-01-07 21:07:04 +00:00
|
|
|
elog(ERROR, "return type mismatch in function decl: final query is a catalog utility");
|
1997-09-07 05:04:48 +00:00
|
|
|
}
|
|
|
|
|
|
|
|
/* okay, it's an ordinary query */
|
|
|
|
tlist = parse->targetList;
|
|
|
|
rt = parse->rtable;
|
|
|
|
cmd = parse->commandType;
|
|
|
|
|
|
|
|
/*
|
|
|
|
* test 2: if the function is declared to return no value, then the
|
|
|
|
* final query had better not be a retrieve.
|
|
|
|
*/
|
1996-07-09 06:22:35 +00:00
|
|
|
if (rettype == InvalidOid)
|
1997-09-07 05:04:48 +00:00
|
|
|
{
|
|
|
|
if (cmd == CMD_SELECT)
|
1998-01-07 21:07:04 +00:00
|
|
|
elog(ERROR,
|
1997-09-07 05:04:48 +00:00
|
|
|
"function declared with no return type, but final query is a retrieve");
|
|
|
|
else
|
|
|
|
return;
|
|
|
|
}
|
|
|
|
|
|
|
|
/* by here, the function is declared to return some type */
|
1997-11-25 22:07:18 +00:00
|
|
|
if ((typ = typeidType(rettype)) == NULL)
|
1999-05-10 00:46:32 +00:00
|
|
|
elog(ERROR, "can't find return type %u for function\n", rettype);
|
1997-09-07 05:04:48 +00:00
|
|
|
|
|
|
|
/*
|
|
|
|
* test 3: if the function is declared to return a value, then the
|
|
|
|
* final query had better be a retrieve.
|
|
|
|
*/
|
|
|
|
if (cmd != CMD_SELECT)
|
1998-01-07 21:07:04 +00:00
|
|
|
elog(ERROR, "function declared to return type %s, but final query is not a retrieve", typeTypeName(typ));
|
1997-09-07 05:04:48 +00:00
|
|
|
|
|
|
|
/*
|
|
|
|
* test 4: for base type returns, the target list should have exactly
|
|
|
|
* one entry, and its type should agree with what the user declared.
|
|
|
|
*/
|
|
|
|
|
1997-11-25 22:07:18 +00:00
|
|
|
if (typeTypeRelid(typ) == InvalidOid)
|
1997-09-07 05:04:48 +00:00
|
|
|
{
|
1998-07-20 21:18:35 +00:00
|
|
|
if (ExecTargetListLength(tlist) > 1)
|
1998-01-07 21:07:04 +00:00
|
|
|
elog(ERROR, "function declared to return %s returns multiple values in final retrieve", typeTypeName(typ));
|
1997-09-07 05:04:48 +00:00
|
|
|
|
|
|
|
resnode = (Resdom *) ((TargetEntry *) lfirst(tlist))->resdom;
|
|
|
|
if (resnode->restype != rettype)
|
1998-01-07 21:07:04 +00:00
|
|
|
elog(ERROR, "return type mismatch in function: declared to return %s, returns %s", typeTypeName(typ), typeidTypeName(resnode->restype));
|
1997-09-07 05:04:48 +00:00
|
|
|
|
|
|
|
/* by here, base return types match */
|
|
|
|
return;
|
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* If the target list is of length 1, and the type of the varnode in
|
|
|
|
* the target list is the same as the declared return type, this is
|
|
|
|
* okay. This can happen, for example, where the body of the function
|
|
|
|
* is 'retrieve (x = func2())', where func2 has the same return type
|
|
|
|
* as the function that's calling it.
|
|
|
|
*/
|
1998-07-20 21:18:35 +00:00
|
|
|
if (ExecTargetListLength(tlist) == 1)
|
1997-09-07 05:04:48 +00:00
|
|
|
{
|
|
|
|
resnode = (Resdom *) ((TargetEntry *) lfirst(tlist))->resdom;
|
|
|
|
if (resnode->restype == rettype)
|
|
|
|
return;
|
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* By here, the procedure returns a (set of) tuples. This part of the
|
|
|
|
* typechecking is a hack. We look up the relation that is the
|
|
|
|
* declared return type, and be sure that attributes 1 .. n in the
|
|
|
|
* target list match the declared types.
|
|
|
|
*/
|
1997-11-25 22:07:18 +00:00
|
|
|
reln = heap_open(typeTypeRelid(typ));
|
1997-09-07 05:04:48 +00:00
|
|
|
|
|
|
|
if (!RelationIsValid(reln))
|
1999-05-10 00:46:32 +00:00
|
|
|
elog(ERROR, "cannot open relation relid %u", typeTypeRelid(typ));
|
1997-09-07 05:04:48 +00:00
|
|
|
|
|
|
|
relid = reln->rd_id;
|
|
|
|
relnatts = reln->rd_rel->relnatts;
|
|
|
|
|
1998-07-20 21:18:35 +00:00
|
|
|
if (ExecTargetListLength(tlist) != relnatts)
|
1998-01-07 21:07:04 +00:00
|
|
|
elog(ERROR, "function declared to return type %s does not retrieve (%s.*)", typeTypeName(typ), typeTypeName(typ));
|
1997-09-07 05:04:48 +00:00
|
|
|
|
|
|
|
/* expect attributes 1 .. n in order */
|
|
|
|
for (i = 1; i <= relnatts; i++)
|
|
|
|
{
|
1997-09-08 02:41:22 +00:00
|
|
|
TargetEntry *tle = lfirst(tlist);
|
|
|
|
Node *thenode = tle->expr;
|
1997-09-07 05:04:48 +00:00
|
|
|
|
|
|
|
tlist = lnext(tlist);
|
|
|
|
tletype = exprType(thenode);
|
|
|
|
|
1999-05-25 16:15:34 +00:00
|
|
|
#ifdef NOT_USED /* fix me */
|
1997-09-07 05:04:48 +00:00
|
|
|
/* this is tedious */
|
|
|
|
if (IsA(thenode, Var))
|
|
|
|
tletype = (Oid) ((Var *) thenode)->vartype;
|
|
|
|
else if (IsA(thenode, Const))
|
|
|
|
tletype = (Oid) ((Const *) thenode)->consttype;
|
|
|
|
else if (IsA(thenode, Param))
|
|
|
|
tletype = (Oid) ((Param *) thenode)->paramtype;
|
|
|
|
else if (IsA(thenode, Expr))
|
|
|
|
tletype = Expr;
|
1997-09-08 21:56:23 +00:00
|
|
|
|
1997-09-07 05:04:48 +00:00
|
|
|
else if (IsA(thenode, LispList))
|
|
|
|
{
|
|
|
|
thenode = lfirst(thenode);
|
|
|
|
if (IsA(thenode, Oper))
|
|
|
|
tletype = (Oid) get_opresulttype((Oper *) thenode);
|
|
|
|
else if (IsA(thenode, Func))
|
|
|
|
tletype = (Oid) get_functype((Func *) thenode);
|
|
|
|
else
|
1998-01-07 21:07:04 +00:00
|
|
|
elog(ERROR, "function declared to return type %s does not retrieve (%s.all)", typeTypeName(typ), typeTypeName(typ));
|
1997-09-07 05:04:48 +00:00
|
|
|
}
|
|
|
|
else
|
1998-01-07 21:07:04 +00:00
|
|
|
elog(ERROR, "function declared to return type %s does not retrieve (%s.all)", typeTypeName(typ), typeTypeName(typ));
|
1997-09-05 20:20:56 +00:00
|
|
|
#endif
|
1997-09-07 05:04:48 +00:00
|
|
|
/* reach right in there, why don't you? */
|
|
|
|
if (tletype != reln->rd_att->attrs[i - 1]->atttypid)
|
1998-01-07 21:07:04 +00:00
|
|
|
elog(ERROR, "function declared to return type %s does not retrieve (%s.all)", typeTypeName(typ), typeTypeName(typ));
|
1997-09-07 05:04:48 +00:00
|
|
|
}
|
1996-07-09 06:22:35 +00:00
|
|
|
|
1997-09-07 05:04:48 +00:00
|
|
|
heap_close(reln);
|
1996-07-09 06:22:35 +00:00
|
|
|
|
1997-09-07 05:04:48 +00:00
|
|
|
/* success */
|
|
|
|
return;
|
1996-07-09 06:22:35 +00:00
|
|
|
}
|
1999-02-03 19:31:24 +00:00
|
|
|
|
|
|
|
|
|
|
|
/* ----------
|
|
|
|
* Support function for need_sortplan
|
|
|
|
* ----------
|
|
|
|
*/
|
|
|
|
static TargetEntry *
|
|
|
|
get_matching_tle(Plan *plan, Resdom *resdom)
|
|
|
|
{
|
1999-05-25 16:15:34 +00:00
|
|
|
List *i;
|
|
|
|
TargetEntry *tle;
|
1999-02-03 19:31:24 +00:00
|
|
|
|
1999-05-25 16:15:34 +00:00
|
|
|
foreach(i, plan->targetlist)
|
|
|
|
{
|
|
|
|
tle = (TargetEntry *) lfirst(i);
|
1999-02-03 19:31:24 +00:00
|
|
|
if (tle->resdom->resno == resdom->resno)
|
|
|
|
return tle;
|
|
|
|
}
|
|
|
|
return NULL;
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
|
|
/* ----------
|
|
|
|
* Check if a user requested ORDER BY is already satisfied by
|
|
|
|
* the choosen index scan.
|
|
|
|
*
|
|
|
|
* Returns TRUE if sort is required, FALSE if can be omitted.
|
|
|
|
* ----------
|
|
|
|
*/
|
|
|
|
static bool
|
|
|
|
need_sortplan(List *sortcls, Plan *plan)
|
|
|
|
{
|
|
|
|
Relation indexRel;
|
1999-05-25 16:15:34 +00:00
|
|
|
IndexScan *indexScan;
|
|
|
|
Oid indexId;
|
|
|
|
List *i;
|
1999-02-03 19:31:24 +00:00
|
|
|
HeapTuple htup;
|
1999-05-25 16:15:34 +00:00
|
|
|
Form_pg_index index_tup;
|
|
|
|
int key_no = 0;
|
1999-02-03 19:31:24 +00:00
|
|
|
|
|
|
|
/* ----------
|
|
|
|
* Must be an IndexScan
|
|
|
|
* ----------
|
|
|
|
*/
|
1999-05-25 16:15:34 +00:00
|
|
|
if (nodeTag(plan) != T_IndexScan)
|
1999-02-03 19:31:24 +00:00
|
|
|
return TRUE;
|
|
|
|
|
1999-05-25 16:15:34 +00:00
|
|
|
indexScan = (IndexScan *) plan;
|
1999-02-03 19:31:24 +00:00
|
|
|
|
|
|
|
/* ----------
|
|
|
|
* Should not have left- or righttree
|
|
|
|
* ----------
|
|
|
|
*/
|
1999-05-25 16:15:34 +00:00
|
|
|
if (plan->lefttree != NULL)
|
1999-02-03 19:31:24 +00:00
|
|
|
return TRUE;
|
1999-05-25 16:15:34 +00:00
|
|
|
if (plan->righttree != NULL)
|
1999-02-03 19:31:24 +00:00
|
|
|
return TRUE;
|
|
|
|
|
|
|
|
/* ----------
|
|
|
|
* Must be a single index scan
|
|
|
|
* ----------
|
|
|
|
*/
|
1999-05-25 16:15:34 +00:00
|
|
|
if (length(indexScan->indxid) != 1)
|
1999-02-03 19:31:24 +00:00
|
|
|
return TRUE;
|
|
|
|
|
|
|
|
/* ----------
|
|
|
|
* Indices can only have up to 8 attributes. So an ORDER BY using
|
|
|
|
* more that 8 attributes could never be satisfied by an index.
|
|
|
|
* ----------
|
|
|
|
*/
|
1999-05-25 16:15:34 +00:00
|
|
|
if (length(sortcls) > 8)
|
1999-02-03 19:31:24 +00:00
|
|
|
return TRUE;
|
|
|
|
|
|
|
|
/* ----------
|
|
|
|
* The choosen Index must be a btree
|
|
|
|
* ----------
|
|
|
|
*/
|
|
|
|
indexId = lfirsti(indexScan->indxid);
|
|
|
|
|
|
|
|
indexRel = index_open(indexId);
|
1999-05-25 16:15:34 +00:00
|
|
|
if (strcmp(nameout(&(indexRel->rd_am->amname)), "btree") != 0)
|
|
|
|
{
|
1999-02-03 19:31:24 +00:00
|
|
|
heap_close(indexRel);
|
|
|
|
return TRUE;
|
|
|
|
}
|
|
|
|
heap_close(indexRel);
|
|
|
|
|
|
|
|
/* ----------
|
|
|
|
* Fetch the index tuple
|
|
|
|
* ----------
|
|
|
|
*/
|
|
|
|
htup = SearchSysCacheTuple(INDEXRELID,
|
1999-05-25 16:15:34 +00:00
|
|
|
ObjectIdGetDatum(indexId), 0, 0, 0);
|
|
|
|
if (!HeapTupleIsValid(htup))
|
1999-05-10 00:46:32 +00:00
|
|
|
elog(ERROR, "cache lookup for index %u failed", indexId);
|
1999-02-03 19:31:24 +00:00
|
|
|
index_tup = (Form_pg_index) GETSTRUCT(htup);
|
|
|
|
|
|
|
|
/* ----------
|
|
|
|
* Check if all the sort clauses match the attributes in the index
|
|
|
|
* ----------
|
|
|
|
*/
|
1999-05-25 16:15:34 +00:00
|
|
|
foreach(i, sortcls)
|
|
|
|
{
|
|
|
|
SortClause *sortcl;
|
|
|
|
Resdom *resdom;
|
|
|
|
TargetEntry *tle;
|
|
|
|
Var *var;
|
1999-02-03 19:31:24 +00:00
|
|
|
|
|
|
|
sortcl = (SortClause *) lfirst(i);
|
|
|
|
|
|
|
|
resdom = sortcl->resdom;
|
|
|
|
tle = get_matching_tle(plan, resdom);
|
1999-05-25 16:15:34 +00:00
|
|
|
if (tle == NULL)
|
|
|
|
{
|
1999-02-03 19:31:24 +00:00
|
|
|
/* ----------
|
|
|
|
* Could this happen?
|
|
|
|
* ----------
|
|
|
|
*/
|
|
|
|
return TRUE;
|
|
|
|
}
|
1999-05-25 16:15:34 +00:00
|
|
|
if (nodeTag(tle->expr) != T_Var)
|
|
|
|
{
|
1999-02-03 19:31:24 +00:00
|
|
|
/* ----------
|
|
|
|
* The target list expression isn't a var, so it
|
|
|
|
* cannot be the indexed attribute
|
|
|
|
* ----------
|
|
|
|
*/
|
|
|
|
return TRUE;
|
|
|
|
}
|
1999-05-25 16:15:34 +00:00
|
|
|
var = (Var *) (tle->expr);
|
1999-02-03 19:31:24 +00:00
|
|
|
|
1999-05-25 16:15:34 +00:00
|
|
|
if (var->varno != indexScan->scan.scanrelid)
|
|
|
|
{
|
1999-02-03 19:31:24 +00:00
|
|
|
/* ----------
|
|
|
|
* This Var isn't from the scan relation. So it isn't
|
|
|
|
* that of the index
|
|
|
|
* ----------
|
|
|
|
*/
|
|
|
|
return TRUE;
|
|
|
|
}
|
|
|
|
|
1999-05-25 16:15:34 +00:00
|
|
|
if (var->varattno != index_tup->indkey[key_no])
|
|
|
|
{
|
1999-02-03 19:31:24 +00:00
|
|
|
/* ----------
|
|
|
|
* It isn't the indexed attribute.
|
|
|
|
* ----------
|
|
|
|
*/
|
|
|
|
return TRUE;
|
|
|
|
}
|
|
|
|
|
1999-05-25 16:15:34 +00:00
|
|
|
if (oprid(oper("<", resdom->restype, resdom->restype, FALSE)) != sortcl->opoid)
|
|
|
|
{
|
1999-02-03 19:31:24 +00:00
|
|
|
/* ----------
|
|
|
|
* Sort order isn't in ascending order.
|
|
|
|
* ----------
|
|
|
|
*/
|
|
|
|
return TRUE;
|
|
|
|
}
|
|
|
|
|
|
|
|
key_no++;
|
|
|
|
}
|
|
|
|
|
|
|
|
/* ----------
|
|
|
|
* Index matches ORDER BY - sort not required
|
|
|
|
* ----------
|
|
|
|
*/
|
|
|
|
return FALSE;
|
|
|
|
}
|