Re: Multivariate MCV stats can leak data to unprivileged users

Tomas Vondra <tomas(dot)vondra(at)2ndquadrant(dot)com> writes:
> On Sat, May 18, 2019 at 03:45:11PM -0400, Tom Lane wrote:
>> Tomas Vondra <tomas(dot)vondra(at)2ndquadrant(dot)com> writes:
>>> But that's not an issue intruduced by PG12, it works like that even for
>>> the extended statistics introduced in PG10.

>> Yeah, but no time like the present to fix it if it's wrong ...

> Sorry, not sure I understand. Are you saying we should try to rework
> this before the beta1 release, or that we don't have time to do that?

> I think we have four options - rework it before beta1, rework it after
> beta1, rework it in PG13 and leave it as it is now.

Yup, that's about what the options are. I'm just voting against
"change it in v13". If we're going to change it, then the fewer
major versions that have the bogus definition the better --- and
since we're changing that catalog in v12 anyway, users will see
fewer distinct behaviors if we do this change too.

It's very possibly too late to get it done before beta1,
unfortunately. But as Andres noted, post-beta1 catversion
bumps are hardly unusual, so I do not think "rework after
beta1" is unacceptable.

regards, tom lane