Alexander Korotkov <aekorotkov@gmail.com> writes:
> On Sun, Nov 16, 2025 at 7:27 PM Tom Lane <tgl@sss.pgh.pa.us> wrote:
>> I dug into the history a little and could not find anything except
>> [1], which says
>>
>> I have made some changes (attachment).
>> * if the operator expression left or right side type category is
>> {array | domain | composite}, then don't do the transformation.
>> (i am not 10% sure with composite)
>>
>> with no further justification than that. There were even messages
>> later in the thread questioning the need for it, but nobody did
>> anything about it. The transformation does work perfectly well
>> if enabled, as shown by the second part of the attached test script.
> I think another email to reference is [1]. It analyses the problems
> with row expressions, but finally it mistakenly generalizes that for
> composite types. So, yes, thread didn't show any problems with
> composites.
Thanks for confirming that.
>> So I end with v3, now with a full-dress commit message.
> It looks very good, thank you so much for dedicating your time on fixing this.
Pushed v3, thank you for looking at it.
regards, tom lane
