I happened to notice this comment in the logic in
ATAddForeignKeyConstraint that tries to decide if it can skip
revalidating a foreign-key constraint after a DDL change:
* Since we require that all collations share the same notion of
* equality (which they do, because texteq reduces to bitwise
* equality), we don't compare collation here.
Hasn't this been broken by the introduction of nondeterministic
collations?
regards, tom lane