On Wed, Apr 3, 2013 at 12:36 AM, Erikjan Rijkers <er@xs4all.nl> wrote:
On Mon, April 1, 2013 23:15, Alexander Korotkov wrote:
[trgm-regexp-0.14.patch.gz]
Hi Alexander,
Hi Erik!
Something went wrong in this version of the patch: many (most) queries that were earlier spectacularly fast have become slow, often slower than a seqscan or only marginally faster. See the attached numbers; it compares head(seqscan) with trgm-regex patch versions 13 and 14.
I did not even complete the test-run because version 14 is so clearly inferior to 13 (and earlier, as far as I can remember).
(let me know if you want the whole result, I can run it overnight)
Yes, there was bug in new color map scan implementation. It was fixed in attached version. Could you re-run tests please?
Attached version of patch also addresses problem that we now using API it's not correct to use special values of color number. It introduces additional structure ExtendedColor which contain ExtendedColorClass field for handling additional "virtual" colors introduced by analysis.