Tom Lane wrote:
> Joe Conway <mail@joeconway.com> writes:
>>What if we modify equality_oper() and friends to do the check for us
>>when argtype is a varlena array?
>
>>If an array type is checked for an ordering operator, its element type
>>now *must* have an ordering operator in order for the result to be
>>meaningful. Same goes for ordering_oper().
>
> That sounds more like a solution. Is there anyplace else we need to
> push knowledge of this into? (Offhand I can't think of any, but ...)
>
Poking around:
1) I found that addTargetToSortList() uses compatible_oper_opid() if it
is given a specific opname, but ordering_oper_opid (which in turn uses
ordering_oper()) if not. Do you know in what cases will
addTargetToSortList() get called with a specific opname? If it's only
for "non-standard" operators, then we should be fine here I'd think.
2) AlterTableAddForeignKeyConstraint() uses oper() to check directly for
an "=" operator. This should probably be changed anyway to
equality_oper(), shouldn't it?
3) process_implied_equality() uses compatible_oper() to directly obtain
"=" operator. This should also be changed anyway to equality_oper(),
shouldn't it?
4) SelectSortFunction() might need work -- I'll have to study that one
more.
That's all I could find that looked suspect.
Joe
