On Fri, Nov 11, 2022 at 11:24 PM Tom Lane <tgl@sss.pgh.pa.us> wrote:
Richard Guo <guofenglinux@gmail.com> writes: > I'm wondering whether we need to insist on being strict for the lower > OJ's min_righthand. What if we instead check strictness for its whole > syn_righthand?
Surely not. What if the only point of strictness is for a rel that isn't part of the min_righthand? Then we could end up re-ordering based on a condition that isn't actually strict for what we've chosen as the join's RHS.
I think I've got your point. You're right. And doing so would cause another problem about ordering restriction as I observed. For the following two forms
(A leftjoin (B leftjoin C on (Pbc)) on (Pab)) left join D on (Pbcd)
AND
((A leftjoin B on (Pab)) leftjoin C on (Pbc)) left join D on (Pbcd)
Assume Pbc is strict for B, Pbcd is strict for C but not strict for B.
After applying this change, in the first form the (BC)/D join will be legal, while in the second form it is not.
It might be possible to change the other part of the equation and consider the A/B join's min_righthand instead of syn_righthand while checking if Pcd uses A/B's RHS; but I'm not real sure about that either.
This seems a more plausible change. I tried this way and didn't find any abnormal behaviour. But I'm not sure either. Maybe I need to try harder.
