David Rowley <dgrowleyml@gmail.com> writes:
> On Sun, 9 Mar 2025 at 04:50, Tom Lane <tgl@sss.pgh.pa.us> wrote:
>> Yeah. I don't think there is any way for array_agg_deserialize to
>> know the correct typmod, so what we have to do is disable using
>> partial aggregation in this case. Fortunately there's a
>> policy-setting function that can be taught that, as attached.
> The only way I can think of to get that would be to special-case
> array_agg_serialize() to have it serialize the typmod when the send
> function is record_send(), then add a similar special-case to
> array_agg_deserialize() to check for a record_recv() and deserialize
> the typmod there. That doesn't seem very pretty, so I'm happy to go
> with your fix to disable parallel aggregates for this case.
Yeah ... we could probably make that work if we had to, but the
ugliness would likely metastasize outside array_agg_[de]serialize.
Since it took more than a year to get a field report, I'm content
to just disable the optimization instead. Pushed that way.
regards, tom lane
