Re: BUG #18953: Planner fails to build plan for complex query with LATERAL references

I wrote:
> My thought is that have_dangerous_phv() was never more than a
> quick-n-dirty kludge, and what we really ought to do is remove it.

Here's a WIP patch along that line. It's unfinished in that, for
testing purposes, I just lobotomized have_dangerous_phv() to return
constant false rather than taking it out entirely. But of course
we'd want to clean up all the dead code if we go this way.

I have mixed feelings about whether to back-patch or just make
this change in HEAD. While we're clearly fixing a bug here,
the bug's been there for 10 years, so the lack of field reports
suggests strongly that this is not something ordinary users write.
Two arguments against back-patching are that
(1) the odds of introducing a new bug aren't zero;
(2) removing the have_dangerous_phv() restriction will change
some plan choices, which we generally dislike doing in stable
branches.

I'm still comfortable with shoving this into v18, but maybe we
should leave the back branches alone.

regards, tom lane