Re: [HACKERS] Optimization for index-only scans with filter conditions

Mateusz Stefek <mateusz.stefek@gmail.com> writes:
> Attached is a patch, which changes the order of the checks. I.e. the
> visibility of the row is confirmed only after the qual condition is
> evaluated to true.

This cannot be accepted, because it's unsafe to try to apply user
code to a dead tuple.  The tuple might reference toast data that's
not there anymore, or in the worst case it might not even match
the table rowtype anymore.

Even without that problem, I would think that whether it is a win or
not would be mighty workload-dependent.
        regards, tom lane