Re: Should we allow ALTER OPERATOR CLASS to ADD/DROP operators and procedures?

On 2/7/25 01:15, Tom Lane wrote:
> Tomas Vondra <tomas@vondra.me> writes:
>> Which is why PostGIS devs chose a different path - updating the catalogs
>> to modify the opclass definition [3], using this script [4].
> 
>> That's a bit ... terrifying.
> 
> Indeed.  Did it not occur to them to use ALTER OPERATOR FAMILY?
> 

No, it did not. But that may be my fault - I've reported the bug, and
I've been discussing possible fixes with them, and it didn't occur to me
either :-(

It seems quite non-intuitive to fix a bug in OPERATOR CLASS by doing
ALTER OPERATOR FAMILY.

> The end result of doing so is that the added opfamily members
> wouldn't have been bound tightly to the opclass, but that does
> not seem like a huge deal in context.  They'd still have worked
> fine AFAIK.
> 

I haven't tried but I assume you're suggesting adding the MERGE proc to
the opfamily would be enough for the opclass to pick it up?

However, I'm not sure this would be a complete fix, because what would
they do with the existing ADD_VALUE proc? Sure, they could just tweak
the C code to "redirect" the call to the built-in one. But that requires
the MERGE proc being defined, so it'd crash if someone installs the new
version .so but doesn't do ALTER EXTENSION ... UPDATE.


regards

-- 
Tomas Vondra