On Thu, Aug 3, 2023 at 6:04 PM Matthias van de Meent
<boekewurm+postg...@gmail.com> wrote:
>
>
> Is it? Detoasting only happens if the argument was toasted, and I have
> serious doubts that the result of (a->'a') will be toasted in our
> current system. Sure, we do need to allocate an intermediate result,
> but that's in a temporary memory context that should be trivially
> cheap to free.
>
> > /*
> > * v.val.numeric points into jsonb body, so we need to make a copy to
> > * return
> > */
> > retValue = DatumGetNumericCopy(NumericGetDatum(v.val.numeric));
> >
> > At last this method needs 1 extra FuncExpr than my method, this would
> > cost some expression execution effort. I'm not saying we need to avoid
> > expression execution generally, but extracting numeric fields from jsonb
> > looks a reasonable case.
>
> What's tb here?
>
>
> Kind regards,
>
> Matthias van de Meent
> Neon (https://neon.tech)
>
>
can confirm the patch's jsonb_object_field_numeric is faster than
pg_catalog."numeric"(jsonb).
also it works accurately either jsonb is in the page or in toast schema chunks.
I don't understand why we need to allocate an intermediate result
part. since you cannot directly update a jsonb value field.
This function is not easy to find out...
select numeric('{"a":11}'->'a'); --fail.
select jsonb_numeric(jsonb'{"a":11}'->'a'); --fail
select "numeric"('{"a":11}'::jsonb->'a'); --ok
