Re: [HACKERS] Help: fmgr_info: function 0: cache lookup failed

Bruce Momjian <maillist(at)candle(dot)pha(dot)pa(dot)us> writes:
> Tom, was this dealth with?

What I originally did was the second choice (generate an error message)
but I had to back off to using a default when we discovered that the
rtree index operators don't have oprrest links in 6.5 :-(. I would
like to change it back after the rtree index entries are fixed, but
for the meanwhile you can mark this item done.

regards, tom lane

>> "D'Arcy" "J.M." Cain <darcy(at)druid(dot)net> writes:
>>>>>>> I tried just setting oprcanhash to true but that didn't do it. Can
>>>>>>> you suggest what fields I need to look at in pg_operator?
>>>>>
>>>>> oprrest and oprjoin
>>
>>>> OK, I did this and it worked. I'll go work on the documentation now.
>>
>> OK, I see the problem: btreesel() and friends blithely assume that the
>> operator used in an index will have a selectivity function (oprrest).
>>
>> I can see two reasonable fixes:
>> * Default to an 0.5 estimate if no oprrest link (this is what the
>> optimizer does for operators that have no oprrest).
>> * Generate an error message along the lines of "index operators must
>> have a restriction selectivity estimator", if we think that they
>> really really oughta.
>>
>> I'm not sure which way to jump. The former would be more friendly for
>> people just starting to develop index support for a new data type ...
>> but then they might never realize that lack of an estimator is hurting
>> performance for them. Comments?
>>
>> regards, tom lane