Fabien COELHO <coelho@cri.ensmp.fr> writes:
> This patch adds a "notice" at constraint creation time if the referential
> integrity check is to be "costly", that is the comparison operator
> involves some coercion.
This is not the correct test at all. The correct test would involve
looking in pg_amop to see if the selected operator is a member of the
opclass for the referenced column's index. (IIRC there are helper
functions in lsyscache.c that will make this easier.)
regards, tom lane