From: | Tom Lane <tgl(at)sss(dot)pgh(dot)pa(dot)us> |
---|---|
To: | Andrew Dunstan <andrew(at)dunslane(dot)net> |
Cc: | PostgreSQL-development <pgsql-hackers(at)postgresql(dot)org> |
Subject: | Re: join regression failure on cygwin |
Date: | 2009-07-24 14:22:23 |
Message-ID: | 26720.1248445343@sss.pgh.pa.us |
Views: | Raw Message | Whole Thread | Download mbox | Resend email |
Thread: | |
Lists: | pgsql-hackers |
I wrote:
> Andrew Dunstan <andrew(at)dunslane(dot)net> writes:
>> It appears on Googling a bit that the erand48() is buggy in that it
>> requires the seed to have been initialized with srand48() or it will
>> constantly return 0.0.
> Huh, and that sends us into an infinite loop? I'll take a look at that.
> Even though it's surely nonrandom, it doesn't seem like pathological
> behavior of the RNG should wedge us completely.
The answer is that a constant RNG result sends this bit of
geqo_selection() into a tight loop:
int first,
second;
first = linear(root, pool->size, bias);
second = linear(root, pool->size, bias);
if (pool->size > 1)
{
while (first == second)
second = linear(root, pool->size, bias);
}
Not sure if it's worth trying to do something about that, or exactly
what we'd do anyway. Even if we hacked this up somehow, a constant RNG
result would pretty much break GEQO for any useful purpose. So it could
be argued that having the regression tests fail here is a good thing...
regards, tom lane
From | Date | Subject | |
---|---|---|---|
Next Message | Andrew Dunstan | 2009-07-24 14:27:54 | Re: join regression failure on cygwin |
Previous Message | Kevin Grittner | 2009-07-24 14:02:38 | Re: query decorrelation in postgres |