"David G. Johnston" <david.g.johnston@gmail.com> writes:
> On Sun, May 1, 2022 at 10:08 AM Tom Lane <tgl@sss.pgh.pa.us> wrote:
>> Maybe we could improve this situation by treating a "record" parameter
>> as polymorphic, though that might cause some odd inconsistencies with
>> plpgsql's historical treatment of "record" local variables.
> The extent of needing to treat "record" as polymorphic-like seems like it
> would be limited to resolve_polymorphic_argtype in funcapi.c. Namely, in
> computing the hash key for the compiled hash entry for the function.
> Similar to how we append the trigger oid in compute_function_hashkey in
> pl.compile (which ultimately calls the former) so trigger invocations
> become per-table.
I'm hesitant to touch funcapi.c for this; the scope of potential
side-effects becomes enormous as soon as you do.
regards, tom lane