I wrote:
> WindowFuncs should be disallowed in qual clauses, so I think it's okay
> to leave those flags out. An Aggref could occur in a HAVING qual though.
> I'm not sure if this code could get applied to HAVING ... but it's
> not immediately clear that it can't. I'd be inclined to add
> PVC_RECURSE_AGGREGATES, as that seems more likely to be okay than not.
Actually, on closer look: why don't we just nuke that pull_var_clause
call entirely, along with the following loop inspecting its result?
The subsequent loop that looks for a matching StatisticExtInfo
expression will do just fine at rejecting any expression that
contains Vars of the wrong relation. Maybe there is some performance
argument why the pull_var_clause precheck is worth the trouble,
but I'm inclined to bet that it's actually a net loss.
regards, tom lane