Nicolas Adenis-Lamarre <nicolas.adenis.lamarre@gmail.com> writes:
> would it be accepted to do a patch to replace coalesce by and/or,
> mainly to fix related estimations.
Almost certainly not. It'd be very hard to do that while preserving
the expected semantics of COALESCE: no argument is to be evaluated
more than once, and people sometimes expect strict left-to-right
evaluation. I've even seen it used as an intentional optimization
fence.
If you think you can improve the estimation around it, I'd suggest
tackling that directly.
regards, tom lane