Re: [HACKERS] Re: [BUGS] General Bug Report: Bug in optimizer

> Bruce Momjian wrote:
> >
> > Let me tell you why I don't think this is a bug. The optimizer will
> > choose ordered results over unordered results if the costs are the same.
> > In this case, the cost of the query is zero, so it chose to use the
> > index because the index produces an ordered result.
> >
> > This works well for un-vacuumed tables, because it thinks everything is
> > zero cost, and chooses the index.
>
> Agreed, this is ok as long as
>
> vac=> create table table1 (field1 int);
> CREATE
> vac=> insert into table1 values (1);
> INSERT 1583349 1
> vac=> create index i_table1__field1 on table1 (field1);
> CREATE
> vac=> explain select * from table1 where field1 = 1;
> NOTICE: QUERY PLAN:
>
> Seq Scan on table1 (cost=1.03 size=1 width=4)
>
> - SeqScan is used for small tables.
>
> So, only bug reported is left.
>
> Vadim
>

Fixed:

test=> explain select * from table1 where field1 = 1;
NOTICE: QUERY PLAN:

Index Scan using i_table1__field1 on table1 (cost=0.00 size=0 width=4)

EXPLAIN
test=> explain select * from table1 where field1 = -1;
NOTICE: QUERY PLAN:

Index Scan using i_table1__field1 on table1 (cost=0.00 size=0 width=4)

The function fixing it is in gram.y:

static Node *doNegate(Node *n)
{
if (IsA(n, A_Const))
{
A_Const *con = (A_Const *)n;

if (con->val.type == T_Integer)
{
con->val.val.ival = -con->val.val.ival;
return n;
}
if (con->val.type == T_Float)
{
con->val.val.dval = -con->val.val.dval;
return n;
}
}

return makeA_Expr(OP, "-", NULL, n);
}

It tries to merge the negative into the constant. We already had
special '-' handling in the grammer, so I just call this function,
rather than doing makeA_Expr in all cases.

Committed.

--
Bruce Momjian | http://www.op.net/~candle
maillist(at)candle(dot)pha(dot)pa(dot)us | (610) 853-3000
+ If your life is a hard drive, | 830 Blythe Avenue
+ Christ can be your backup. | Drexel Hill, Pennsylvania 19026