Martijn van Oosterhout <kleptog@svana.org> writes:
> Something that has been discussed in the past is allowing non-strict
> type input/output functions to be evaluated at query execution time,
> rather than during parse time. This would give the type input/output
> functions the Expr node they need to extract this info.
We could make that happen for literals used in queries (see comment in
coerce_type()), but it's not appealing to expect all of the ad-hoc I/O
function calls in the whole system to supply dummy expression trees.
That would be adding overhead to all cases that's only useful in a few.
I think requiring the functions that need this info to do extra work
is probably the right answer. (It's already possible to cache whatever
lookups you have to do, cf array_in or record_in, so the overhead isn't
*that* daunting.)
regards, tom lane
