Re: Performance problem in textanycat/anytextcat

On Mon, May 17, 2010 at 9:23 PM, Tom Lane <tgl(at)sss(dot)pgh(dot)pa(dot)us> wrote:
> Robert Haas <robertmhaas(at)gmail(dot)com> writes:
>> On Mon, May 17, 2010 at 4:01 PM, Tom Lane <tgl(at)sss(dot)pgh(dot)pa(dot)us> wrote:
>>> Perhaps this is a backpatchable bug fix.  Comments?
>
>> I can't say whether this is safe enough to back-patch, but the way
>> this is set up, don't we also need to fix some catalog entries and, if
>> yes, isn't that problematic?
>
> The only catalog entries at issue, AFAICT, are the textanycat/anytextcat
> ones.  I am not sure whether we should attempt to back-patch changes for
> them, but this patch wouldn't make the situation in the back branches
> worse.  In particular, if we apply this patch but don't change the
> catalog entries, then nothing would change at all about the problematic
> cases, because the planner would decide it couldn't safely inline the
> function.  The only cases where inlining will happen is where the
> expression's apparent volatility stays the same or decreases, so as far
> as that issue is concerned this patch will never make CREATE INDEX
> reject a case it would have accepted otherwise.  The patch *will* make
> CREATE INDEX reject cases with volatile default arguments hiding under
> non-volatile functions, but that's got nothing to do with any built-in
> functions; and that's the case I claim is clearly a bug fix.

Well, I guess it boils down to what you think the chances that this
change will unintentionally break something are, then. I don't have a
good feeling for that.

--
Robert Haas
EnterpriseDB: http://www.enterprisedb.com
The Enterprise Postgres Company