Andrew Dunstan <andrew@dunslane.net> writes:
> On 2023-12-06 We 10:44, Tom Lane wrote:
>> In particular, has anyone done any performance testing?
>> I'm concerned about that because composite_to_json() has
>> zero capability to cache any metadata across calls, meaning
>> there is going to be a large amount of duplicated work
>> per row.
> Yeah, that's hard to deal with, too, as it can be called recursively.
Right. On the plus side, if we did improve this it would presumably
also benefit other callers of composite_to_json[b].
> OTOH I'd rather have a version of this that worked slowly than none at all.
It might be acceptable to plan on improving the performance later,
depending on just how bad it is now.
regards, tom lane