I noticed a small bug in the "Sort Method" code:

postgres=# explain analyze select * from test order by random() limit 1;
                                                   QUERY PLAN                   
                                 
-----------------------------------------------------------------------------------------------------------------
 Limit  (cost=21.50..21.50 rows=1 width=4) (actual time=3.649..3.651 rows=1 
loops=1)
   ->  Sort  (cost=21.50..24.00 rows=1000 width=4) (actual time=3.646..3.646 
rows=1 loops=1)
         Sort Key: (random())
         Sort Method:  quicksort  Memory: 17kB
         ->  Seq Scan on test  (cost=0.00..16.50 rows=1000 width=4) (actual 
time=0.021..1.707 rows=1000 loops=1)
 Total runtime: 3.704 ms
(6 rows)

It's printing "quicksort" even though it used a heap. This happens because we
don't bother deheapifying a singleton heap so the boundUsed flag never gets
set. The patch below just moves setting that flag to when the heap is made
instead of when it's deheapified.

One could make the argument that we should distinguish the noop sort from
quicksort or the half-hearted singleton heapsort from the full heapsort but
that seems like gilding. But distinguishing between heapsort and quicksort is
important since it really could be either depending on how many inputs there
were.


Index: src/backend/utils/sort/tuplesort.c
===================================================================
RCS file: /home/stark/src/REPOSITORY/pgsql/src/backend/utils/sort/tuplesort.c,v
retrieving revision 1.77
diff -u -r1.77 tuplesort.c
--- src/backend/utils/sort/tuplesort.c  7 Jun 2007 19:19:57 -0000       1.77
+++ src/backend/utils/sort/tuplesort.c  1 Sep 2007 17:17:25 -0000
@@ -2247,6 +2247,7 @@
        }
 
        Assert(state->memtupcount == state->bound);
+       state->boundUsed = true;
        state->status = TSS_BOUNDED;
 }
 
@@ -2284,7 +2285,6 @@
        REVERSEDIRECTION(state);
 
        state->status = TSS_SORTEDINMEM;
-       state->boundUsed = true;
 }
 
 /*


-- 
  Gregory Stark
  EnterpriseDB          http://www.enterprisedb.com

---------------------------(end of broadcast)---------------------------
TIP 1: if posting/reading through Usenet, please send an appropriate
       subscribe-nomail command to [EMAIL PROTECTED] so that your
       message can get through to the mailing list cleanly

Reply via email to