Alexander Lakhin <exclusion@gmail.com> writes:
> I've managed to discover one more anomaly introduced by a16ef313f, that is
> not fixed by fix-bug-18953-some-more:
Just to note that I am studying this. It looks to me like the
issue is that identify_current_nestloop_params() is handing back
a PHV with too few nullingrel bits set for the place that we want
to put it, as is acknowledged to be possible in its comments.
We thought we could get away with that, but in this context setrefs.c
will complain. I'm inclined to try to make it set the bits more
accurately, rather than further weaken setrefs.c's cross-checks.
I'm wondering again whether this isn't a case that is reachable
before a16ef313f. Perhaps the have_dangerous_phv check prevented
forming plan trees that could have this issue, but it's not real
clear why.
regards, tom lane