This is a review of the patch at this CF location:
as posted here:
This patch applied cleanly and compiled without warning. It
performed correctly. Since the patch modifies one function which has
one input and one output (through a pointer parameter), with no other
side effects, it has low risk of surprising problems.
It is strictly a performance patch, and accomplishes its goals. I
ran the entire dictionary through the patched and unpatched versions
five times each (using the supplied debugging patch) and saw no
changes to the behavior of the function -- identical results every
time. I ran performance tests with building and reindexing the
sorted dictionary, a randomly ordered dictionary, a randomly ordered
dictionary with the entries randomly truncated to 0 to 3 characters,
and a large README file -- five times each with and without the
patch. The patch was a clear winner with all of those except the
truncated dictionary, where the difference we well within the noise
(0.05% difference summing five runs when individual runs could vary
by up to about 4%).
While there was no reason to believe it could affect search
performance, I tested that anyway, and found no difference.
Once this message shows up on the list (so I can retrieve the message
ID) I will mark this "Ready for Committer".
pgsql-hackers by date
|Next:||From: Robert Haas||Date: 2011-09-26 00:46:22|
|Subject: Re: index-only scans|
|Previous:||From: Bruce Momjian||Date: 2011-09-25 21:12:34|
|Subject: Re: fix for pg_upgrade|