Re: tsearch parser inefficiency if text includes urls or emails - new version

On Tuesday 08 December 2009 16:23:11 Kevin Grittner wrote:
> I wrote:
> > Frankly, I'd be amazed if there was a performance regression,
>
> OK, I'm amazed. While it apparently helps some cases dramatically
> (Andres had a case where run time was reduced by 93.2%), I found a
> pretty routine case where run time was increased by 3.1%. I tweaked
> the code and got that down to a 2.5% run time increase. I'm having
> troubles getting it any lower than that. And yes, this is real, not
> noise -- the slowest unpatched time for this test is faster than the
> fastest time with any version of the patch. :-(
>
> Andres, could you provide more information on the test which showed
> the dramatic improvement? In particular, info on OS, CPU, character
> set, encoding scheme, and what kind of data was used for the test.
>
> I'll do some more testing and try to figure out how the patch is
> slowing things down and post with details.
Could you show your testcase? I dont see why it could get slower?

I tested with various data, the one benefiting most was some changelog where
each entry was signed by an email.

OS: Debian Sid, Core2 Duo, UTF-8, and I tried both C and de_DE.UTF8.

Andres