diff options
| author | Tom Lane | 2014-01-05 17:28:39 +0000 |
|---|---|---|
| committer | Tom Lane | 2014-01-05 17:28:39 +0000 |
| commit | 8b49a6044d06b557047210dba2735081bb037e96 (patch) | |
| tree | 0728628b84691f01663256bceb4e648f3af6a4d6 /src/include | |
| parent | 92459e7a7f87f91fc3012bea9eef870cf464d91f (diff) | |
Cache catalog lookup data across groups in ordered-set aggregates.
The initial commit of ordered-set aggregates just did all the setup work
afresh each time the aggregate function is started up. But in a GROUP BY
query, the catalog lookups need not be repeated for each group, since the
column datatypes and sort information won't change. When there are many
small groups, this makes for a useful, though not huge, performance
improvement. Per suggestion from Andrew Gierth.
Profiling of these cases suggests that it might be profitable to avoid
duplicate lookups within tuplesort startup as well; but changing the
tuplesort APIs would have much broader impact, so I left that for
another day.
Diffstat (limited to 'src/include')
| -rw-r--r-- | src/include/catalog/pg_operator.h | 1 |
1 files changed, 1 insertions, 0 deletions
diff --git a/src/include/catalog/pg_operator.h b/src/include/catalog/pg_operator.h index a49cfdbdded..2fe9f935d46 100644 --- a/src/include/catalog/pg_operator.h +++ b/src/include/catalog/pg_operator.h @@ -128,6 +128,7 @@ DATA(insert OID = 95 ( "<" PGNSP PGUID b f f 21 21 16 520 524 int2lt scalar DESCR("less than"); DATA(insert OID = 96 ( "=" PGNSP PGUID b t t 23 23 16 96 518 int4eq eqsel eqjoinsel )); DESCR("equal"); +#define Int4EqualOperator 96 DATA(insert OID = 97 ( "<" PGNSP PGUID b f f 23 23 16 521 525 int4lt scalarltsel scalarltjoinsel )); DESCR("less than"); #define Int4LessOperator 97 |