Re: [PATCH] Opclass parameters

On 11.09.2019 1:03, Tomas Vondra wrote:

> On Tue, Sep 10, 2019 at 04:30:41AM +0300, Nikita Glukhov wrote:
>
>> 2. New AM method amattoptions().
>>
>>   amattoptions() is used to specify per-column AM-specific options.
>>   The example is signature length for bloom indexes (patch #3).
>>
>
> I'm somewhat confused how am I supposed to use this, considering the 
> patch
> set only defines this for the contrib/bloom index AM. So let's say I want
> to create a custom BRIN opclass with per-attribute options (like the two
> BRIN opclasses I work on in the other thread). Clearly, I can't tweak the
> IndexAmRoutine from the extension. ISTM the patch series should modify 
> all
> existing index AMs to have a valid amattoptions() implementation, calling
> the new amproc if defined.
>
> Or what is the correct way to define custom opclass for existing index AM
> (e.g. BRIN) with attribute options?
>
Per-attribute opclass options are implemented independently from per-attribute
AM options.  amattoptions() is optional and needs to be defined only if AM has
per-attribute options.  amproc #0 is called regardless of whether amattoptions
is defined or not.  That was the main reason why uniform procnum 0 was picked.

You should simply define function like that and use it as amproc #0:
  

Datum
brin_bloom_options(PG_FUNCTION_ARGS)
{
     local_relopts *relopts = (local_relopts *) PG_GETARG_POINTER(0);
     BloomOptions *blopts = NULL;

     extend_local_reloptions(relopts, blopts, sizeof(*blopts));

     add_local_real_reloption(relopts, "n_distinct_per_range", "desc",
                              -0.1, -1.0, INT_MAX, &blopts->nDistinctPerRange);

     add_local_real_reloption(relopts, "false_positive_rate", "desc",
                              0.01, 0.001, 1.0, &blopts->falsePositiveRate);

     PG_RETURN_VOID();
}

-- 
Nikita Glukhov
Postgres Professional: http://www.postgrespro.com
The Russian Postgres Company