On Wed, May 13, 2015 at 09:34:53AM -0400, Robert Haas wrote:
> On Tue, May 12, 2015 at 7:22 PM, Tom Lane <tgl@sss.pgh.pa.us> wrote:
> > MinMaxExpr is an implicit invocation of a btree comparison function.
> > Are we supposing that all of those are necessarily leakproof?
>
> I suspect it's an oversight, because the comment gives no hint that
> any such intention was present. It's been more than three years since
> I committed that code (under a different function name) so my memory
> is a little fuzzy, but I believe it just didn't occur to me that
> MinMaxExpr could include a function call.
>
> I suspect it's safe in practice, but in theory it's probably a bug.
Agreed; it is formally a bug. We considered[1] special trust of operator
class members and decided against it. Since almost every btree opfamily
member is leakproof in practice, I doubt the bug has harmed anyone.
[1] http://www.postgresql.org/message-id/flat/20110707223526.GJ1840@tornado.leadboat.com