Alvaro Herrera <alvherre@alvh.no-ip.org> writes:
> On 2025-Apr-05, Tom Lane wrote:
>> So what we should be doing is building a parse-analysis result
>> that deparses into something that looks like the input; probably,
>> a JsonArrayQueryConstructor node with an analyzed EXPR_SUBLINK
>> SubLink below it. Then we can make this tlist-length check against
>> the analyzed SubLink, removing the problem of premature errors that
>> are not spelled the way we want.
> Sounds reasonable, I can try to find better coding for this, but it
> won't be soon. I won't be sad if somebody else wants to do it, so if
> you feel like it, please be my guest.
Yeah, it's not high priority for me either. Maybe someone else
will be interested in the project.
>> Anyway, that idea is far too invasive to be back-patchable,
>> and IMO it's too late to consider even getting it into v18.
>> So what I'm thinking is we should just apply the copyObject
>> hack for now, and resolve to reconsider this code later.
> Sounds reasonable.
OK, I'll get that done.
regards, tom lane
